1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext2/inode.c
   4 *
   5 * Copyright (C) 1992, 1993, 1994, 1995
   6 * Remy Card (card@masi.ibp.fr)
   7 * Laboratoire MASI - Institut Blaise Pascal
   8 * Universite Pierre et Marie Curie (Paris VI)
   9 *
  10 *  from
  11 *
  12 *  linux/fs/minix/inode.c
  13 *
  14 *  Copyright (C) 1991, 1992  Linus Torvalds
  15 *
  16 *  Goal-directed block allocation by Stephen Tweedie
  17 * 	(sct@dcs.ed.ac.uk), 1993, 1998
  18 *  Big-endian to little-endian byte-swapping/bitmaps by
  19 *        David S. Miller (davem@caip.rutgers.edu), 1995
  20 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  21 * 	(jj@sunsite.ms.mff.cuni.cz)
  22 *
  23 *  Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
  24 */
  25
  26#include <linux/time.h>
  27#include <linux/highuid.h>
  28#include <linux/pagemap.h>
  29#include <linux/dax.h>
  30#include <linux/blkdev.h>
  31#include <linux/quotaops.h>
  32#include <linux/writeback.h>
  33#include <linux/buffer_head.h>
  34#include <linux/mpage.h>
  35#include <linux/fiemap.h>
  36#include <linux/iomap.h>
  37#include <linux/namei.h>
  38#include <linux/uio.h>
  39#include "ext2.h"
  40#include "acl.h"
  41#include "xattr.h"
  42
  43static int __ext2_write_inode(struct inode *inode, int do_sync);
  44
  45/*
  46 * Test whether an inode is a fast symlink.
  47 */
  48static inline int ext2_inode_is_fast_symlink(struct inode *inode)
  49{
  50	int ea_blocks = EXT2_I(inode)->i_file_acl ?
  51		(inode->i_sb->s_blocksize >> 9) : 0;
  52
  53	return (S_ISLNK(inode->i_mode) &&
  54		inode->i_blocks - ea_blocks == 0);
  55}
  56
  57static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
  58
  59static void ext2_write_failed(struct address_space *mapping, loff_t to)
  60{
  61	struct inode *inode = mapping->host;
  62
  63	if (to > inode->i_size) {
  64		truncate_pagecache(inode, inode->i_size);
  65		ext2_truncate_blocks(inode, inode->i_size);
  66	}
  67}
  68
  69/*
  70 * Called at the last iput() if i_nlink is zero.
  71 */
  72void ext2_evict_inode(struct inode * inode)
  73{
  74	struct ext2_block_alloc_info *rsv;
  75	int want_delete = 0;
  76
  77	if (!inode->i_nlink && !is_bad_inode(inode)) {
  78		want_delete = 1;
  79		dquot_initialize(inode);
  80	} else {
  81		dquot_drop(inode);
  82	}
  83
  84	truncate_inode_pages_final(&inode->i_data);
  85
  86	if (want_delete) {
  87		sb_start_intwrite(inode->i_sb);
  88		/* set dtime */
  89		EXT2_I(inode)->i_dtime	= ktime_get_real_seconds();
  90		mark_inode_dirty(inode);
  91		__ext2_write_inode(inode, inode_needs_sync(inode));
  92		/* truncate to 0 */
  93		inode->i_size = 0;
  94		if (inode->i_blocks)
  95			ext2_truncate_blocks(inode, 0);
  96		ext2_xattr_delete_inode(inode);
  97	}
  98
  99	invalidate_inode_buffers(inode);
 100	clear_inode(inode);
 101
 102	ext2_discard_reservation(inode);
 103	rsv = EXT2_I(inode)->i_block_alloc_info;
 104	EXT2_I(inode)->i_block_alloc_info = NULL;
 105	if (unlikely(rsv))
 106		kfree(rsv);
 107
 108	if (want_delete) {
 109		ext2_free_inode(inode);
 110		sb_end_intwrite(inode->i_sb);
 111	}
 112}
 113
 114typedef struct {
 115	__le32	*p;
 116	__le32	key;
 117	struct buffer_head *bh;
 118} Indirect;
 119
 120static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
 121{
 122	p->key = *(p->p = v);
 123	p->bh = bh;
 124}
 125
 126static inline int verify_chain(Indirect *from, Indirect *to)
 127{
 128	while (from <= to && from->key == *from->p)
 129		from++;
 130	return (from > to);
 131}
 132
 133/**
 134 *	ext2_block_to_path - parse the block number into array of offsets
 135 *	@inode: inode in question (we are only interested in its superblock)
 136 *	@i_block: block number to be parsed
 137 *	@offsets: array to store the offsets in
 138 *      @boundary: set this non-zero if the referred-to block is likely to be
 139 *             followed (on disk) by an indirect block.
 140 *	To store the locations of file's data ext2 uses a data structure common
 141 *	for UNIX filesystems - tree of pointers anchored in the inode, with
 142 *	data blocks at leaves and indirect blocks in intermediate nodes.
 143 *	This function translates the block number into path in that tree -
 144 *	return value is the path length and @offsets[n] is the offset of
 145 *	pointer to (n+1)th node in the nth one. If @block is out of range
 146 *	(negative or too large) warning is printed and zero returned.
 147 *
 148 *	Note: function doesn't find node addresses, so no IO is needed. All
 149 *	we need to know is the capacity of indirect blocks (taken from the
 150 *	inode->i_sb).
 151 */
 152
 153/*
 154 * Portability note: the last comparison (check that we fit into triple
 155 * indirect block) is spelled differently, because otherwise on an
 156 * architecture with 32-bit longs and 8Kb pages we might get into trouble
 157 * if our filesystem had 8Kb blocks. We might use long long, but that would
 158 * kill us on x86. Oh, well, at least the sign propagation does not matter -
 159 * i_block would have to be negative in the very beginning, so we would not
 160 * get there at all.
 161 */
 162
 163static int ext2_block_to_path(struct inode *inode,
 164			long i_block, int offsets[4], int *boundary)
 165{
 166	int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
 167	int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
 168	const long direct_blocks = EXT2_NDIR_BLOCKS,
 169		indirect_blocks = ptrs,
 170		double_blocks = (1 << (ptrs_bits * 2));
 171	int n = 0;
 172	int final = 0;
 173
 174	if (i_block < 0) {
 175		ext2_msg(inode->i_sb, KERN_WARNING,
 176			"warning: %s: block < 0", __func__);
 177	} else if (i_block < direct_blocks) {
 178		offsets[n++] = i_block;
 179		final = direct_blocks;
 180	} else if ( (i_block -= direct_blocks) < indirect_blocks) {
 181		offsets[n++] = EXT2_IND_BLOCK;
 182		offsets[n++] = i_block;
 183		final = ptrs;
 184	} else if ((i_block -= indirect_blocks) < double_blocks) {
 185		offsets[n++] = EXT2_DIND_BLOCK;
 186		offsets[n++] = i_block >> ptrs_bits;
 187		offsets[n++] = i_block & (ptrs - 1);
 188		final = ptrs;
 189	} else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
 190		offsets[n++] = EXT2_TIND_BLOCK;
 191		offsets[n++] = i_block >> (ptrs_bits * 2);
 192		offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
 193		offsets[n++] = i_block & (ptrs - 1);
 194		final = ptrs;
 195	} else {
 196		ext2_msg(inode->i_sb, KERN_WARNING,
 197			"warning: %s: block is too big", __func__);
 198	}
 199	if (boundary)
 200		*boundary = final - 1 - (i_block & (ptrs - 1));
 201
 202	return n;
 203}
 204
 205/**
 206 *	ext2_get_branch - read the chain of indirect blocks leading to data
 207 *	@inode: inode in question
 208 *	@depth: depth of the chain (1 - direct pointer, etc.)
 209 *	@offsets: offsets of pointers in inode/indirect blocks
 210 *	@chain: place to store the result
 211 *	@err: here we store the error value
 212 *
 213 *	Function fills the array of triples <key, p, bh> and returns %NULL
 214 *	if everything went OK or the pointer to the last filled triple
 215 *	(incomplete one) otherwise. Upon the return chain[i].key contains
 216 *	the number of (i+1)-th block in the chain (as it is stored in memory,
 217 *	i.e. little-endian 32-bit), chain[i].p contains the address of that
 218 *	number (it points into struct inode for i==0 and into the bh->b_data
 219 *	for i>0) and chain[i].bh points to the buffer_head of i-th indirect
 220 *	block for i>0 and NULL for i==0. In other words, it holds the block
 221 *	numbers of the chain, addresses they were taken from (and where we can
 222 *	verify that chain did not change) and buffer_heads hosting these
 223 *	numbers.
 224 *
 225 *	Function stops when it stumbles upon zero pointer (absent block)
 226 *		(pointer to last triple returned, *@err == 0)
 227 *	or when it gets an IO error reading an indirect block
 228 *		(ditto, *@err == -EIO)
 229 *	or when it notices that chain had been changed while it was reading
 230 *		(ditto, *@err == -EAGAIN)
 231 *	or when it reads all @depth-1 indirect blocks successfully and finds
 232 *	the whole chain, all way to the data (returns %NULL, *err == 0).
 233 */
 234static Indirect *ext2_get_branch(struct inode *inode,
 235				 int depth,
 236				 int *offsets,
 237				 Indirect chain[4],
 238				 int *err)
 239{
 240	struct super_block *sb = inode->i_sb;
 241	Indirect *p = chain;
 242	struct buffer_head *bh;
 243
 244	*err = 0;
 245	/* i_data is not going away, no lock needed */
 246	add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
 247	if (!p->key)
 248		goto no_block;
 249	while (--depth) {
 250		bh = sb_bread(sb, le32_to_cpu(p->key));
 251		if (!bh)
 252			goto failure;
 253		read_lock(&EXT2_I(inode)->i_meta_lock);
 254		if (!verify_chain(chain, p))
 255			goto changed;
 256		add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
 257		read_unlock(&EXT2_I(inode)->i_meta_lock);
 258		if (!p->key)
 259			goto no_block;
 260	}
 261	return NULL;
 262
 263changed:
 264	read_unlock(&EXT2_I(inode)->i_meta_lock);
 265	brelse(bh);
 266	*err = -EAGAIN;
 267	goto no_block;
 268failure:
 269	*err = -EIO;
 270no_block:
 271	return p;
 272}
 273
 274/**
 275 *	ext2_find_near - find a place for allocation with sufficient locality
 276 *	@inode: owner
 277 *	@ind: descriptor of indirect block.
 278 *
 279 *	This function returns the preferred place for block allocation.
 280 *	It is used when heuristic for sequential allocation fails.
 281 *	Rules are:
 282 *	  + if there is a block to the left of our position - allocate near it.
 283 *	  + if pointer will live in indirect block - allocate near that block.
 284 *	  + if pointer will live in inode - allocate in the same cylinder group.
 285 *
 286 * In the latter case we colour the starting block by the callers PID to
 287 * prevent it from clashing with concurrent allocations for a different inode
 288 * in the same block group.   The PID is used here so that functionally related
 289 * files will be close-by on-disk.
 290 *
 291 *	Caller must make sure that @ind is valid and will stay that way.
 292 */
 293
 294static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
 295{
 296	struct ext2_inode_info *ei = EXT2_I(inode);
 297	__le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
 298	__le32 *p;
 299	ext2_fsblk_t bg_start;
 300	ext2_fsblk_t colour;
 301
 302	/* Try to find previous block */
 303	for (p = ind->p - 1; p >= start; p--)
 304		if (*p)
 305			return le32_to_cpu(*p);
 306
 307	/* No such thing, so let's try location of indirect block */
 308	if (ind->bh)
 309		return ind->bh->b_blocknr;
 310
 311	/*
 312	 * It is going to be referred from inode itself? OK, just put it into
 313	 * the same cylinder group then.
 314	 */
 315	bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
 316	colour = (current->pid % 16) *
 317			(EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
 318	return bg_start + colour;
 319}
 320
 321/**
 322 *	ext2_find_goal - find a preferred place for allocation.
 323 *	@inode: owner
 324 *	@block:  block we want
 325 *	@partial: pointer to the last triple within a chain
 326 *
 327 *	Returns preferred place for a block (the goal).
 328 */
 329
 330static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
 331					  Indirect *partial)
 332{
 333	struct ext2_block_alloc_info *block_i;
 334
 335	block_i = EXT2_I(inode)->i_block_alloc_info;
 336
 337	/*
 338	 * try the heuristic for sequential allocation,
 339	 * failing that at least try to get decent locality.
 340	 */
 341	if (block_i && (block == block_i->last_alloc_logical_block + 1)
 342		&& (block_i->last_alloc_physical_block != 0)) {
 343		return block_i->last_alloc_physical_block + 1;
 344	}
 345
 346	return ext2_find_near(inode, partial);
 347}
 348
 349/**
 350 *	ext2_blks_to_allocate: Look up the block map and count the number
 351 *	of direct blocks need to be allocated for the given branch.
 352 *
 353 * 	@branch: chain of indirect blocks
 354 *	@k: number of blocks need for indirect blocks
 355 *	@blks: number of data blocks to be mapped.
 356 *	@blocks_to_boundary:  the offset in the indirect block
 357 *
 358 *	return the total number of blocks to be allocate, including the
 359 *	direct and indirect blocks.
 360 */
 361static int
 362ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
 363		int blocks_to_boundary)
 364{
 365	unsigned long count = 0;
 366
 367	/*
 368	 * Simple case, [t,d]Indirect block(s) has not allocated yet
 369	 * then it's clear blocks on that path have not allocated
 370	 */
 371	if (k > 0) {
 372		/* right now don't hanel cross boundary allocation */
 373		if (blks < blocks_to_boundary + 1)
 374			count += blks;
 375		else
 376			count += blocks_to_boundary + 1;
 377		return count;
 378	}
 379
 380	count++;
 381	while (count < blks && count <= blocks_to_boundary
 382		&& le32_to_cpu(*(branch[0].p + count)) == 0) {
 383		count++;
 384	}
 385	return count;
 386}
 387
 388/**
 389 *	ext2_alloc_blocks: multiple allocate blocks needed for a branch
 390 *	@indirect_blks: the number of blocks need to allocate for indirect
 391 *			blocks
 392 *
 393 *	@new_blocks: on return it will store the new block numbers for
 394 *	the indirect blocks(if needed) and the first direct block,
 395 *	@blks:	on return it will store the total number of allocated
 396 *		direct blocks
 397 */
 398static int ext2_alloc_blocks(struct inode *inode,
 399			ext2_fsblk_t goal, int indirect_blks, int blks,
 400			ext2_fsblk_t new_blocks[4], int *err)
 401{
 402	int target, i;
 403	unsigned long count = 0;
 404	int index = 0;
 405	ext2_fsblk_t current_block = 0;
 406	int ret = 0;
 407
 408	/*
 409	 * Here we try to allocate the requested multiple blocks at once,
 410	 * on a best-effort basis.
 411	 * To build a branch, we should allocate blocks for
 412	 * the indirect blocks(if not allocated yet), and at least
 413	 * the first direct block of this branch.  That's the
 414	 * minimum number of blocks need to allocate(required)
 415	 */
 416	target = blks + indirect_blks;
 417
 418	while (1) {
 419		count = target;
 420		/* allocating blocks for indirect blocks and direct blocks */
 421		current_block = ext2_new_blocks(inode,goal,&count,err);
 422		if (*err)
 423			goto failed_out;
 424
 425		target -= count;
 426		/* allocate blocks for indirect blocks */
 427		while (index < indirect_blks && count) {
 428			new_blocks[index++] = current_block++;
 429			count--;
 430		}
 431
 432		if (count > 0)
 433			break;
 434	}
 435
 436	/* save the new block number for the first direct block */
 437	new_blocks[index] = current_block;
 438
 439	/* total number of blocks allocated for direct blocks */
 440	ret = count;
 441	*err = 0;
 442	return ret;
 443failed_out:
 444	for (i = 0; i <index; i++)
 445		ext2_free_blocks(inode, new_blocks[i], 1);
 446	if (index)
 447		mark_inode_dirty(inode);
 448	return ret;
 449}
 450
 451/**
 452 *	ext2_alloc_branch - allocate and set up a chain of blocks.
 453 *	@inode: owner
 454 *	@indirect_blks: depth of the chain (number of blocks to allocate)
 455 *	@blks: number of allocated direct blocks
 456 *	@goal: preferred place for allocation
 457 *	@offsets: offsets (in the blocks) to store the pointers to next.
 458 *	@branch: place to store the chain in.
 459 *
 460 *	This function allocates @num blocks, zeroes out all but the last one,
 461 *	links them into chain and (if we are synchronous) writes them to disk.
 462 *	In other words, it prepares a branch that can be spliced onto the
 463 *	inode. It stores the information about that chain in the branch[], in
 464 *	the same format as ext2_get_branch() would do. We are calling it after
 465 *	we had read the existing part of chain and partial points to the last
 466 *	triple of that (one with zero ->key). Upon the exit we have the same
 467 *	picture as after the successful ext2_get_block(), except that in one
 468 *	place chain is disconnected - *branch->p is still zero (we did not
 469 *	set the last link), but branch->key contains the number that should
 470 *	be placed into *branch->p to fill that gap.
 471 *
 472 *	If allocation fails we free all blocks we've allocated (and forget
 473 *	their buffer_heads) and return the error value the from failed
 474 *	ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
 475 *	as described above and return 0.
 476 */
 477
 478static int ext2_alloc_branch(struct inode *inode,
 479			int indirect_blks, int *blks, ext2_fsblk_t goal,
 480			int *offsets, Indirect *branch)
 481{
 482	int blocksize = inode->i_sb->s_blocksize;
 483	int i, n = 0;
 484	int err = 0;
 485	struct buffer_head *bh;
 486	int num;
 487	ext2_fsblk_t new_blocks[4];
 488	ext2_fsblk_t current_block;
 489
 490	num = ext2_alloc_blocks(inode, goal, indirect_blks,
 491				*blks, new_blocks, &err);
 492	if (err)
 493		return err;
 494
 495	branch[0].key = cpu_to_le32(new_blocks[0]);
 496	/*
 497	 * metadata blocks and data blocks are allocated.
 498	 */
 499	for (n = 1; n <= indirect_blks;  n++) {
 500		/*
 501		 * Get buffer_head for parent block, zero it out
 502		 * and set the pointer to new one, then send
 503		 * parent to disk.
 504		 */
 505		bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
 506		if (unlikely(!bh)) {
 507			err = -ENOMEM;
 508			goto failed;
 509		}
 510		branch[n].bh = bh;
 511		lock_buffer(bh);
 512		memset(bh->b_data, 0, blocksize);
 513		branch[n].p = (__le32 *) bh->b_data + offsets[n];
 514		branch[n].key = cpu_to_le32(new_blocks[n]);
 515		*branch[n].p = branch[n].key;
 516		if ( n == indirect_blks) {
 517			current_block = new_blocks[n];
 518			/*
 519			 * End of chain, update the last new metablock of
 520			 * the chain to point to the new allocated
 521			 * data blocks numbers
 522			 */
 523			for (i=1; i < num; i++)
 524				*(branch[n].p + i) = cpu_to_le32(++current_block);
 525		}
 526		set_buffer_uptodate(bh);
 527		unlock_buffer(bh);
 528		mark_buffer_dirty_inode(bh, inode);
 529		/* We used to sync bh here if IS_SYNC(inode).
 530		 * But we now rely upon generic_write_sync()
 531		 * and b_inode_buffers.  But not for directories.
 532		 */
 533		if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
 534			sync_dirty_buffer(bh);
 535	}
 536	*blks = num;
 537	return err;
 538
 539failed:
 540	for (i = 1; i < n; i++)
 541		bforget(branch[i].bh);
 542	for (i = 0; i < indirect_blks; i++)
 543		ext2_free_blocks(inode, new_blocks[i], 1);
 544	ext2_free_blocks(inode, new_blocks[i], num);
 545	return err;
 546}
 547
 548/**
 549 * ext2_splice_branch - splice the allocated branch onto inode.
 550 * @inode: owner
 551 * @block: (logical) number of block we are adding
 552 * @where: location of missing link
 553 * @num:   number of indirect blocks we are adding
 554 * @blks:  number of direct blocks we are adding
 555 *
 556 * This function fills the missing link and does all housekeeping needed in
 557 * inode (->i_blocks, etc.). In case of success we end up with the full
 558 * chain to new block and return 0.
 559 */
 560static void ext2_splice_branch(struct inode *inode,
 561			long block, Indirect *where, int num, int blks)
 562{
 563	int i;
 564	struct ext2_block_alloc_info *block_i;
 565	ext2_fsblk_t current_block;
 566
 567	block_i = EXT2_I(inode)->i_block_alloc_info;
 568
 569	/* XXX LOCKING probably should have i_meta_lock ?*/
 570	/* That's it */
 571
 572	*where->p = where->key;
 573
 574	/*
 575	 * Update the host buffer_head or inode to point to more just allocated
 576	 * direct blocks blocks
 577	 */
 578	if (num == 0 && blks > 1) {
 579		current_block = le32_to_cpu(where->key) + 1;
 580		for (i = 1; i < blks; i++)
 581			*(where->p + i ) = cpu_to_le32(current_block++);
 582	}
 583
 584	/*
 585	 * update the most recently allocated logical & physical block
 586	 * in i_block_alloc_info, to assist find the proper goal block for next
 587	 * allocation
 588	 */
 589	if (block_i) {
 590		block_i->last_alloc_logical_block = block + blks - 1;
 591		block_i->last_alloc_physical_block =
 592				le32_to_cpu(where[num].key) + blks - 1;
 593	}
 594
 595	/* We are done with atomic stuff, now do the rest of housekeeping */
 596
 597	/* had we spliced it onto indirect block? */
 598	if (where->bh)
 599		mark_buffer_dirty_inode(where->bh, inode);
 600
 601	inode->i_ctime = current_time(inode);
 602	mark_inode_dirty(inode);
 603}
 604
 605/*
 606 * Allocation strategy is simple: if we have to allocate something, we will
 607 * have to go the whole way to leaf. So let's do it before attaching anything
 608 * to tree, set linkage between the newborn blocks, write them if sync is
 609 * required, recheck the path, free and repeat if check fails, otherwise
 610 * set the last missing link (that will protect us from any truncate-generated
 611 * removals - all blocks on the path are immune now) and possibly force the
 612 * write on the parent block.
 613 * That has a nice additional property: no special recovery from the failed
 614 * allocations is needed - we simply release blocks and do not touch anything
 615 * reachable from inode.
 616 *
 617 * `handle' can be NULL if create == 0.
 618 *
 619 * return > 0, # of blocks mapped or allocated.
 620 * return = 0, if plain lookup failed.
 621 * return < 0, error case.
 622 */
 623static int ext2_get_blocks(struct inode *inode,
 624			   sector_t iblock, unsigned long maxblocks,
 625			   u32 *bno, bool *new, bool *boundary,
 626			   int create)
 627{
 628	int err;
 629	int offsets[4];
 630	Indirect chain[4];
 631	Indirect *partial;
 632	ext2_fsblk_t goal;
 633	int indirect_blks;
 634	int blocks_to_boundary = 0;
 635	int depth;
 636	struct ext2_inode_info *ei = EXT2_I(inode);
 637	int count = 0;
 638	ext2_fsblk_t first_block = 0;
 639
 640	BUG_ON(maxblocks == 0);
 641
 642	depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
 643
 644	if (depth == 0)
 645		return -EIO;
 646
 647	partial = ext2_get_branch(inode, depth, offsets, chain, &err);
 648	/* Simplest case - block found, no allocation needed */
 649	if (!partial) {
 650		first_block = le32_to_cpu(chain[depth - 1].key);
 651		count++;
 652		/*map more blocks*/
 653		while (count < maxblocks && count <= blocks_to_boundary) {
 654			ext2_fsblk_t blk;
 655
 656			if (!verify_chain(chain, chain + depth - 1)) {
 657				/*
 658				 * Indirect block might be removed by
 659				 * truncate while we were reading it.
 660				 * Handling of that case: forget what we've
 661				 * got now, go to reread.
 662				 */
 663				err = -EAGAIN;
 664				count = 0;
 665				partial = chain + depth - 1;
 666				break;
 667			}
 668			blk = le32_to_cpu(*(chain[depth-1].p + count));
 669			if (blk == first_block + count)
 670				count++;
 671			else
 672				break;
 673		}
 674		if (err != -EAGAIN)
 675			goto got_it;
 676	}
 677
 678	/* Next simple case - plain lookup or failed read of indirect block */
 679	if (!create || err == -EIO)
 680		goto cleanup;
 681
 682	mutex_lock(&ei->truncate_mutex);
 683	/*
 684	 * If the indirect block is missing while we are reading
 685	 * the chain(ext2_get_branch() returns -EAGAIN err), or
 686	 * if the chain has been changed after we grab the semaphore,
 687	 * (either because another process truncated this branch, or
 688	 * another get_block allocated this branch) re-grab the chain to see if
 689	 * the request block has been allocated or not.
 690	 *
 691	 * Since we already block the truncate/other get_block
 692	 * at this point, we will have the current copy of the chain when we
 693	 * splice the branch into the tree.
 694	 */
 695	if (err == -EAGAIN || !verify_chain(chain, partial)) {
 696		while (partial > chain) {
 697			brelse(partial->bh);
 698			partial--;
 699		}
 700		partial = ext2_get_branch(inode, depth, offsets, chain, &err);
 701		if (!partial) {
 702			count++;
 703			mutex_unlock(&ei->truncate_mutex);
 704			if (err)
 705				goto cleanup;
 706			goto got_it;
 707		}
 708	}
 709
 710	/*
 711	 * Okay, we need to do block allocation.  Lazily initialize the block
 712	 * allocation info here if necessary
 713	*/
 714	if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
 715		ext2_init_block_alloc_info(inode);
 716
 717	goal = ext2_find_goal(inode, iblock, partial);
 718
 719	/* the number of blocks need to allocate for [d,t]indirect blocks */
 720	indirect_blks = (chain + depth) - partial - 1;
 721	/*
 722	 * Next look up the indirect map to count the total number of
 723	 * direct blocks to allocate for this branch.
 724	 */
 725	count = ext2_blks_to_allocate(partial, indirect_blks,
 726					maxblocks, blocks_to_boundary);
 727	/*
 728	 * XXX ???? Block out ext2_truncate while we alter the tree
 729	 */
 730	err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
 731				offsets + (partial - chain), partial);
 732
 733	if (err) {
 734		mutex_unlock(&ei->truncate_mutex);
 735		goto cleanup;
 736	}
 737
 738	if (IS_DAX(inode)) {
 739		/*
 740		 * We must unmap blocks before zeroing so that writeback cannot
 741		 * overwrite zeros with stale data from block device page cache.
 742		 */
 743		clean_bdev_aliases(inode->i_sb->s_bdev,
 744				   le32_to_cpu(chain[depth-1].key),
 745				   count);
 746		/*
 747		 * block must be initialised before we put it in the tree
 748		 * so that it's not found by another thread before it's
 749		 * initialised
 750		 */
 751		err = sb_issue_zeroout(inode->i_sb,
 752				le32_to_cpu(chain[depth-1].key), count,
 753				GFP_NOFS);
 754		if (err) {
 755			mutex_unlock(&ei->truncate_mutex);
 756			goto cleanup;
 757		}
 758	}
 759	*new = true;
 760
 761	ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
 762	mutex_unlock(&ei->truncate_mutex);
 763got_it:
 764	if (count > blocks_to_boundary)
 765		*boundary = true;
 766	err = count;
 767	/* Clean up and exit */
 768	partial = chain + depth - 1;	/* the whole chain */
 769cleanup:
 770	while (partial > chain) {
 771		brelse(partial->bh);
 772		partial--;
 773	}
 774	if (err > 0)
 775		*bno = le32_to_cpu(chain[depth-1].key);
 776	return err;
 777}
 778
 779int ext2_get_block(struct inode *inode, sector_t iblock,
 780		struct buffer_head *bh_result, int create)
 781{
 782	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
 783	bool new = false, boundary = false;
 784	u32 bno;
 785	int ret;
 786
 787	ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
 788			create);
 789	if (ret <= 0)
 790		return ret;
 791
 792	map_bh(bh_result, inode->i_sb, bno);
 793	bh_result->b_size = (ret << inode->i_blkbits);
 794	if (new)
 795		set_buffer_new(bh_result);
 796	if (boundary)
 797		set_buffer_boundary(bh_result);
 798	return 0;
 799
 800}
 801
 802#ifdef CONFIG_FS_DAX
 803static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 804		unsigned flags, struct iomap *iomap)
 805{
 806	unsigned int blkbits = inode->i_blkbits;
 807	unsigned long first_block = offset >> blkbits;
 808	unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
 809	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 810	bool new = false, boundary = false;
 811	u32 bno;
 812	int ret;
 813
 814	ret = ext2_get_blocks(inode, first_block, max_blocks,
 815			&bno, &new, &boundary, flags & IOMAP_WRITE);
 816	if (ret < 0)
 817		return ret;
 818
 819	iomap->flags = 0;
 820	iomap->bdev = inode->i_sb->s_bdev;
 821	iomap->offset = (u64)first_block << blkbits;
 822	iomap->dax_dev = sbi->s_daxdev;
 823
 824	if (ret == 0) {
 825		iomap->type = IOMAP_HOLE;
 826		iomap->addr = IOMAP_NULL_ADDR;
 827		iomap->length = 1 << blkbits;
 828	} else {
 829		iomap->type = IOMAP_MAPPED;
 830		iomap->addr = (u64)bno << blkbits;
 831		iomap->length = (u64)ret << blkbits;
 832		iomap->flags |= IOMAP_F_MERGED;
 833	}
 834
 835	if (new)
 836		iomap->flags |= IOMAP_F_NEW;
 837	return 0;
 838}
 839
 840static int
 841ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
 842		ssize_t written, unsigned flags, struct iomap *iomap)
 843{
 844	if (iomap->type == IOMAP_MAPPED &&
 845	    written < length &&
 846	    (flags & IOMAP_WRITE))
 847		ext2_write_failed(inode->i_mapping, offset + length);
 848	return 0;
 849}
 850
 851const struct iomap_ops ext2_iomap_ops = {
 852	.iomap_begin		= ext2_iomap_begin,
 853	.iomap_end		= ext2_iomap_end,
 854};
 855#else
 856/* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
 857const struct iomap_ops ext2_iomap_ops;
 858#endif /* CONFIG_FS_DAX */
 859
 860int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 861		u64 start, u64 len)
 862{
 863	return generic_block_fiemap(inode, fieinfo, start, len,
 864				    ext2_get_block);
 865}
 866
 867static int ext2_writepage(struct page *page, struct writeback_control *wbc)
 868{
 869	return block_write_full_page(page, ext2_get_block, wbc);
 870}
 871
 872static int ext2_readpage(struct file *file, struct page *page)
 873{
 874	return mpage_readpage(page, ext2_get_block);
 875}
 876
 877static int
 878ext2_readpages(struct file *file, struct address_space *mapping,
 879		struct list_head *pages, unsigned nr_pages)
 880{
 881	return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
 882}
 883
 884static int
 885ext2_write_begin(struct file *file, struct address_space *mapping,
 886		loff_t pos, unsigned len, unsigned flags,
 887		struct page **pagep, void **fsdata)
 888{
 889	int ret;
 890
 891	ret = block_write_begin(mapping, pos, len, flags, pagep,
 892				ext2_get_block);
 893	if (ret < 0)
 894		ext2_write_failed(mapping, pos + len);
 895	return ret;
 896}
 897
 898static int ext2_write_end(struct file *file, struct address_space *mapping,
 899			loff_t pos, unsigned len, unsigned copied,
 900			struct page *page, void *fsdata)
 901{
 902	int ret;
 903
 904	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
 905	if (ret < len)
 906		ext2_write_failed(mapping, pos + len);
 907	return ret;
 908}
 909
 910static int
 911ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
 912		loff_t pos, unsigned len, unsigned flags,
 913		struct page **pagep, void **fsdata)
 914{
 915	int ret;
 916
 917	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
 918			       ext2_get_block);
 919	if (ret < 0)
 920		ext2_write_failed(mapping, pos + len);
 921	return ret;
 922}
 923
 924static int ext2_nobh_writepage(struct page *page,
 925			struct writeback_control *wbc)
 926{
 927	return nobh_writepage(page, ext2_get_block, wbc);
 928}
 929
 930static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
 931{
 932	return generic_block_bmap(mapping,block,ext2_get_block);
 933}
 934
 935static ssize_t
 936ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 937{
 938	struct file *file = iocb->ki_filp;
 939	struct address_space *mapping = file->f_mapping;
 940	struct inode *inode = mapping->host;
 941	size_t count = iov_iter_count(iter);
 942	loff_t offset = iocb->ki_pos;
 943	ssize_t ret;
 944
 945	ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
 946	if (ret < 0 && iov_iter_rw(iter) == WRITE)
 947		ext2_write_failed(mapping, offset + count);
 948	return ret;
 949}
 950
 951static int
 952ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
 953{
 954	return mpage_writepages(mapping, wbc, ext2_get_block);
 955}
 956
 957static int
 958ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc)
 959{
 960	return dax_writeback_mapping_range(mapping,
 961			mapping->host->i_sb->s_bdev, wbc);
 962}
 963
 964const struct address_space_operations ext2_aops = {
 965	.readpage		= ext2_readpage,
 966	.readpages		= ext2_readpages,
 967	.writepage		= ext2_writepage,
 968	.write_begin		= ext2_write_begin,
 969	.write_end		= ext2_write_end,
 970	.bmap			= ext2_bmap,
 971	.direct_IO		= ext2_direct_IO,
 972	.writepages		= ext2_writepages,
 973	.migratepage		= buffer_migrate_page,
 974	.is_partially_uptodate	= block_is_partially_uptodate,
 975	.error_remove_page	= generic_error_remove_page,
 976};
 977
 978const struct address_space_operations ext2_nobh_aops = {
 979	.readpage		= ext2_readpage,
 980	.readpages		= ext2_readpages,
 981	.writepage		= ext2_nobh_writepage,
 982	.write_begin		= ext2_nobh_write_begin,
 983	.write_end		= nobh_write_end,
 984	.bmap			= ext2_bmap,
 985	.direct_IO		= ext2_direct_IO,
 986	.writepages		= ext2_writepages,
 987	.migratepage		= buffer_migrate_page,
 988	.error_remove_page	= generic_error_remove_page,
 989};
 990
 991static const struct address_space_operations ext2_dax_aops = {
 992	.writepages		= ext2_dax_writepages,
 993	.direct_IO		= noop_direct_IO,
 994	.set_page_dirty		= noop_set_page_dirty,
 995	.invalidatepage		= noop_invalidatepage,
 996};
 997
 998/*
 999 * Probably it should be a library function... search for first non-zero word
1000 * or memcmp with zero_page, whatever is better for particular architecture.
1001 * Linus?
1002 */
1003static inline int all_zeroes(__le32 *p, __le32 *q)
1004{
1005	while (p < q)
1006		if (*p++)
1007			return 0;
1008	return 1;
1009}
1010
1011/**
1012 *	ext2_find_shared - find the indirect blocks for partial truncation.
1013 *	@inode:	  inode in question
1014 *	@depth:	  depth of the affected branch
1015 *	@offsets: offsets of pointers in that branch (see ext2_block_to_path)
1016 *	@chain:	  place to store the pointers to partial indirect blocks
1017 *	@top:	  place to the (detached) top of branch
1018 *
1019 *	This is a helper function used by ext2_truncate().
1020 *
1021 *	When we do truncate() we may have to clean the ends of several indirect
1022 *	blocks but leave the blocks themselves alive. Block is partially
1023 *	truncated if some data below the new i_size is referred from it (and
1024 *	it is on the path to the first completely truncated data block, indeed).
1025 *	We have to free the top of that path along with everything to the right
1026 *	of the path. Since no allocation past the truncation point is possible
1027 *	until ext2_truncate() finishes, we may safely do the latter, but top
1028 *	of branch may require special attention - pageout below the truncation
1029 *	point might try to populate it.
1030 *
1031 *	We atomically detach the top of branch from the tree, store the block
1032 *	number of its root in *@top, pointers to buffer_heads of partially
1033 *	truncated blocks - in @chain[].bh and pointers to their last elements
1034 *	that should not be removed - in @chain[].p. Return value is the pointer
1035 *	to last filled element of @chain.
1036 *
1037 *	The work left to caller to do the actual freeing of subtrees:
1038 *		a) free the subtree starting from *@top
1039 *		b) free the subtrees whose roots are stored in
1040 *			(@chain[i].p+1 .. end of @chain[i].bh->b_data)
1041 *		c) free the subtrees growing from the inode past the @chain[0].p
1042 *			(no partially truncated stuff there).
1043 */
1044
1045static Indirect *ext2_find_shared(struct inode *inode,
1046				int depth,
1047				int offsets[4],
1048				Indirect chain[4],
1049				__le32 *top)
1050{
1051	Indirect *partial, *p;
1052	int k, err;
1053
1054	*top = 0;
1055	for (k = depth; k > 1 && !offsets[k-1]; k--)
1056		;
1057	partial = ext2_get_branch(inode, k, offsets, chain, &err);
1058	if (!partial)
1059		partial = chain + k-1;
1060	/*
1061	 * If the branch acquired continuation since we've looked at it -
1062	 * fine, it should all survive and (new) top doesn't belong to us.
1063	 */
1064	write_lock(&EXT2_I(inode)->i_meta_lock);
1065	if (!partial->key && *partial->p) {
1066		write_unlock(&EXT2_I(inode)->i_meta_lock);
1067		goto no_top;
1068	}
1069	for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1070		;
1071	/*
1072	 * OK, we've found the last block that must survive. The rest of our
1073	 * branch should be detached before unlocking. However, if that rest
1074	 * of branch is all ours and does not grow immediately from the inode
1075	 * it's easier to cheat and just decrement partial->p.
1076	 */
1077	if (p == chain + k - 1 && p > chain) {
1078		p->p--;
1079	} else {
1080		*top = *p->p;
1081		*p->p = 0;
1082	}
1083	write_unlock(&EXT2_I(inode)->i_meta_lock);
1084
1085	while(partial > p)
1086	{
1087		brelse(partial->bh);
1088		partial--;
1089	}
1090no_top:
1091	return partial;
1092}
1093
1094/**
1095 *	ext2_free_data - free a list of data blocks
1096 *	@inode:	inode we are dealing with
1097 *	@p:	array of block numbers
1098 *	@q:	points immediately past the end of array
1099 *
1100 *	We are freeing all blocks referred from that array (numbers are
1101 *	stored as little-endian 32-bit) and updating @inode->i_blocks
1102 *	appropriately.
1103 */
1104static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1105{
1106	unsigned long block_to_free = 0, count = 0;
1107	unsigned long nr;
1108
1109	for ( ; p < q ; p++) {
1110		nr = le32_to_cpu(*p);
1111		if (nr) {
1112			*p = 0;
1113			/* accumulate blocks to free if they're contiguous */
1114			if (count == 0)
1115				goto free_this;
1116			else if (block_to_free == nr - count)
1117				count++;
1118			else {
1119				ext2_free_blocks (inode, block_to_free, count);
1120				mark_inode_dirty(inode);
1121			free_this:
1122				block_to_free = nr;
1123				count = 1;
1124			}
1125		}
1126	}
1127	if (count > 0) {
1128		ext2_free_blocks (inode, block_to_free, count);
1129		mark_inode_dirty(inode);
1130	}
1131}
1132
1133/**
1134 *	ext2_free_branches - free an array of branches
1135 *	@inode:	inode we are dealing with
1136 *	@p:	array of block numbers
1137 *	@q:	pointer immediately past the end of array
1138 *	@depth:	depth of the branches to free
1139 *
1140 *	We are freeing all blocks referred from these branches (numbers are
1141 *	stored as little-endian 32-bit) and updating @inode->i_blocks
1142 *	appropriately.
1143 */
1144static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1145{
1146	struct buffer_head * bh;
1147	unsigned long nr;
1148
1149	if (depth--) {
1150		int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1151		for ( ; p < q ; p++) {
1152			nr = le32_to_cpu(*p);
1153			if (!nr)
1154				continue;
1155			*p = 0;
1156			bh = sb_bread(inode->i_sb, nr);
1157			/*
1158			 * A read failure? Report error and clear slot
1159			 * (should be rare).
1160			 */ 
1161			if (!bh) {
1162				ext2_error(inode->i_sb, "ext2_free_branches",
1163					"Read failure, inode=%ld, block=%ld",
1164					inode->i_ino, nr);
1165				continue;
1166			}
1167			ext2_free_branches(inode,
1168					   (__le32*)bh->b_data,
1169					   (__le32*)bh->b_data + addr_per_block,
1170					   depth);
1171			bforget(bh);
1172			ext2_free_blocks(inode, nr, 1);
1173			mark_inode_dirty(inode);
1174		}
1175	} else
1176		ext2_free_data(inode, p, q);
1177}
1178
1179/* dax_sem must be held when calling this function */
1180static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1181{
1182	__le32 *i_data = EXT2_I(inode)->i_data;
1183	struct ext2_inode_info *ei = EXT2_I(inode);
1184	int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1185	int offsets[4];
1186	Indirect chain[4];
1187	Indirect *partial;
1188	__le32 nr = 0;
1189	int n;
1190	long iblock;
1191	unsigned blocksize;
1192	blocksize = inode->i_sb->s_blocksize;
1193	iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1194
1195#ifdef CONFIG_FS_DAX
1196	WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1197#endif
1198
1199	n = ext2_block_to_path(inode, iblock, offsets, NULL);
1200	if (n == 0)
1201		return;
1202
1203	/*
1204	 * From here we block out all ext2_get_block() callers who want to
1205	 * modify the block allocation tree.
1206	 */
1207	mutex_lock(&ei->truncate_mutex);
1208
1209	if (n == 1) {
1210		ext2_free_data(inode, i_data+offsets[0],
1211					i_data + EXT2_NDIR_BLOCKS);
1212		goto do_indirects;
1213	}
1214
1215	partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1216	/* Kill the top of shared branch (already detached) */
1217	if (nr) {
1218		if (partial == chain)
1219			mark_inode_dirty(inode);
1220		else
1221			mark_buffer_dirty_inode(partial->bh, inode);
1222		ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1223	}
1224	/* Clear the ends of indirect blocks on the shared branch */
1225	while (partial > chain) {
1226		ext2_free_branches(inode,
1227				   partial->p + 1,
1228				   (__le32*)partial->bh->b_data+addr_per_block,
1229				   (chain+n-1) - partial);
1230		mark_buffer_dirty_inode(partial->bh, inode);
1231		brelse (partial->bh);
1232		partial--;
1233	}
1234do_indirects:
1235	/* Kill the remaining (whole) subtrees */
1236	switch (offsets[0]) {
1237		default:
1238			nr = i_data[EXT2_IND_BLOCK];
1239			if (nr) {
1240				i_data[EXT2_IND_BLOCK] = 0;
1241				mark_inode_dirty(inode);
1242				ext2_free_branches(inode, &nr, &nr+1, 1);
1243			}
1244			/* fall through */
1245		case EXT2_IND_BLOCK:
1246			nr = i_data[EXT2_DIND_BLOCK];
1247			if (nr) {
1248				i_data[EXT2_DIND_BLOCK] = 0;
1249				mark_inode_dirty(inode);
1250				ext2_free_branches(inode, &nr, &nr+1, 2);
1251			}
1252			/* fall through */
1253		case EXT2_DIND_BLOCK:
1254			nr = i_data[EXT2_TIND_BLOCK];
1255			if (nr) {
1256				i_data[EXT2_TIND_BLOCK] = 0;
1257				mark_inode_dirty(inode);
1258				ext2_free_branches(inode, &nr, &nr+1, 3);
1259			}
1260		case EXT2_TIND_BLOCK:
1261			;
1262	}
1263
1264	ext2_discard_reservation(inode);
1265
1266	mutex_unlock(&ei->truncate_mutex);
1267}
1268
1269static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1270{
1271	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1272	    S_ISLNK(inode->i_mode)))
1273		return;
1274	if (ext2_inode_is_fast_symlink(inode))
1275		return;
1276
1277	dax_sem_down_write(EXT2_I(inode));
1278	__ext2_truncate_blocks(inode, offset);
1279	dax_sem_up_write(EXT2_I(inode));
1280}
1281
1282static int ext2_setsize(struct inode *inode, loff_t newsize)
1283{
1284	int error;
1285
1286	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1287	    S_ISLNK(inode->i_mode)))
1288		return -EINVAL;
1289	if (ext2_inode_is_fast_symlink(inode))
1290		return -EINVAL;
1291	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1292		return -EPERM;
1293
1294	inode_dio_wait(inode);
1295
1296	if (IS_DAX(inode)) {
1297		error = iomap_zero_range(inode, newsize,
1298					 PAGE_ALIGN(newsize) - newsize, NULL,
1299					 &ext2_iomap_ops);
1300	} else if (test_opt(inode->i_sb, NOBH))
1301		error = nobh_truncate_page(inode->i_mapping,
1302				newsize, ext2_get_block);
1303	else
1304		error = block_truncate_page(inode->i_mapping,
1305				newsize, ext2_get_block);
1306	if (error)
1307		return error;
1308
1309	dax_sem_down_write(EXT2_I(inode));
1310	truncate_setsize(inode, newsize);
1311	__ext2_truncate_blocks(inode, newsize);
1312	dax_sem_up_write(EXT2_I(inode));
1313
1314	inode->i_mtime = inode->i_ctime = current_time(inode);
1315	if (inode_needs_sync(inode)) {
1316		sync_mapping_buffers(inode->i_mapping);
1317		sync_inode_metadata(inode, 1);
1318	} else {
1319		mark_inode_dirty(inode);
1320	}
1321
1322	return 0;
1323}
1324
1325static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1326					struct buffer_head **p)
1327{
1328	struct buffer_head * bh;
1329	unsigned long block_group;
1330	unsigned long block;
1331	unsigned long offset;
1332	struct ext2_group_desc * gdp;
1333
1334	*p = NULL;
1335	if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1336	    ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1337		goto Einval;
1338
1339	block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1340	gdp = ext2_get_group_desc(sb, block_group, NULL);
1341	if (!gdp)
1342		goto Egdp;
1343	/*
1344	 * Figure out the offset within the block group inode table
1345	 */
1346	offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1347	block = le32_to_cpu(gdp->bg_inode_table) +
1348		(offset >> EXT2_BLOCK_SIZE_BITS(sb));
1349	if (!(bh = sb_bread(sb, block)))
1350		goto Eio;
1351
1352	*p = bh;
1353	offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1354	return (struct ext2_inode *) (bh->b_data + offset);
1355
1356Einval:
1357	ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1358		   (unsigned long) ino);
1359	return ERR_PTR(-EINVAL);
1360Eio:
1361	ext2_error(sb, "ext2_get_inode",
1362		   "unable to read inode block - inode=%lu, block=%lu",
1363		   (unsigned long) ino, block);
1364Egdp:
1365	return ERR_PTR(-EIO);
1366}
1367
1368void ext2_set_inode_flags(struct inode *inode)
1369{
1370	unsigned int flags = EXT2_I(inode)->i_flags;
1371
1372	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1373				S_DIRSYNC | S_DAX);
1374	if (flags & EXT2_SYNC_FL)
1375		inode->i_flags |= S_SYNC;
1376	if (flags & EXT2_APPEND_FL)
1377		inode->i_flags |= S_APPEND;
1378	if (flags & EXT2_IMMUTABLE_FL)
1379		inode->i_flags |= S_IMMUTABLE;
1380	if (flags & EXT2_NOATIME_FL)
1381		inode->i_flags |= S_NOATIME;
1382	if (flags & EXT2_DIRSYNC_FL)
1383		inode->i_flags |= S_DIRSYNC;
1384	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1385		inode->i_flags |= S_DAX;
1386}
1387
1388void ext2_set_file_ops(struct inode *inode)
1389{
1390	inode->i_op = &ext2_file_inode_operations;
1391	inode->i_fop = &ext2_file_operations;
1392	if (IS_DAX(inode))
1393		inode->i_mapping->a_ops = &ext2_dax_aops;
1394	else if (test_opt(inode->i_sb, NOBH))
1395		inode->i_mapping->a_ops = &ext2_nobh_aops;
1396	else
1397		inode->i_mapping->a_ops = &ext2_aops;
1398}
1399
1400struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1401{
1402	struct ext2_inode_info *ei;
1403	struct buffer_head * bh = NULL;
1404	struct ext2_inode *raw_inode;
1405	struct inode *inode;
1406	long ret = -EIO;
1407	int n;
1408	uid_t i_uid;
1409	gid_t i_gid;
1410
1411	inode = iget_locked(sb, ino);
1412	if (!inode)
1413		return ERR_PTR(-ENOMEM);
1414	if (!(inode->i_state & I_NEW))
1415		return inode;
1416
1417	ei = EXT2_I(inode);
1418	ei->i_block_alloc_info = NULL;
1419
1420	raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1421	if (IS_ERR(raw_inode)) {
1422		ret = PTR_ERR(raw_inode);
1423 		goto bad_inode;
1424	}
1425
1426	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1427	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1428	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1429	if (!(test_opt (inode->i_sb, NO_UID32))) {
1430		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1431		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1432	}
1433	i_uid_write(inode, i_uid);
1434	i_gid_write(inode, i_gid);
1435	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1436	inode->i_size = le32_to_cpu(raw_inode->i_size);
1437	inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1438	inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1439	inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1440	inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1441	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1442	/* We now have enough fields to check if the inode was active or not.
1443	 * This is needed because nfsd might try to access dead inodes
1444	 * the test is that same one that e2fsck uses
1445	 * NeilBrown 1999oct15
1446	 */
1447	if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1448		/* this inode is deleted */
 
1449		ret = -ESTALE;
1450		goto bad_inode;
1451	}
1452	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1453	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1454	ext2_set_inode_flags(inode);
1455	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1456	ei->i_frag_no = raw_inode->i_frag;
1457	ei->i_frag_size = raw_inode->i_fsize;
1458	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1459	ei->i_dir_acl = 0;
1460
1461	if (ei->i_file_acl &&
1462	    !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1463		ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1464			   ei->i_file_acl);
 
1465		ret = -EFSCORRUPTED;
1466		goto bad_inode;
1467	}
1468
1469	if (S_ISREG(inode->i_mode))
1470		inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1471	else
1472		ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1473	if (i_size_read(inode) < 0) {
1474		ret = -EFSCORRUPTED;
1475		goto bad_inode;
1476	}
1477	ei->i_dtime = 0;
1478	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1479	ei->i_state = 0;
1480	ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1481	ei->i_dir_start_lookup = 0;
1482
1483	/*
1484	 * NOTE! The in-memory inode i_data array is in little-endian order
1485	 * even on big-endian machines: we do NOT byteswap the block numbers!
1486	 */
1487	for (n = 0; n < EXT2_N_BLOCKS; n++)
1488		ei->i_data[n] = raw_inode->i_block[n];
1489
1490	if (S_ISREG(inode->i_mode)) {
1491		ext2_set_file_ops(inode);
1492	} else if (S_ISDIR(inode->i_mode)) {
1493		inode->i_op = &ext2_dir_inode_operations;
1494		inode->i_fop = &ext2_dir_operations;
1495		if (test_opt(inode->i_sb, NOBH))
1496			inode->i_mapping->a_ops = &ext2_nobh_aops;
1497		else
1498			inode->i_mapping->a_ops = &ext2_aops;
1499	} else if (S_ISLNK(inode->i_mode)) {
1500		if (ext2_inode_is_fast_symlink(inode)) {
1501			inode->i_link = (char *)ei->i_data;
1502			inode->i_op = &ext2_fast_symlink_inode_operations;
1503			nd_terminate_link(ei->i_data, inode->i_size,
1504				sizeof(ei->i_data) - 1);
1505		} else {
1506			inode->i_op = &ext2_symlink_inode_operations;
1507			inode_nohighmem(inode);
1508			if (test_opt(inode->i_sb, NOBH))
1509				inode->i_mapping->a_ops = &ext2_nobh_aops;
1510			else
1511				inode->i_mapping->a_ops = &ext2_aops;
1512		}
1513	} else {
1514		inode->i_op = &ext2_special_inode_operations;
1515		if (raw_inode->i_block[0])
1516			init_special_inode(inode, inode->i_mode,
1517			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1518		else 
1519			init_special_inode(inode, inode->i_mode,
1520			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1521	}
1522	brelse (bh);
 
1523	unlock_new_inode(inode);
1524	return inode;
1525	
1526bad_inode:
1527	brelse(bh);
1528	iget_failed(inode);
1529	return ERR_PTR(ret);
1530}
1531
1532static int __ext2_write_inode(struct inode *inode, int do_sync)
1533{
1534	struct ext2_inode_info *ei = EXT2_I(inode);
1535	struct super_block *sb = inode->i_sb;
1536	ino_t ino = inode->i_ino;
1537	uid_t uid = i_uid_read(inode);
1538	gid_t gid = i_gid_read(inode);
1539	struct buffer_head * bh;
1540	struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1541	int n;
1542	int err = 0;
1543
1544	if (IS_ERR(raw_inode))
1545 		return -EIO;
1546
1547	/* For fields not not tracking in the in-memory inode,
1548	 * initialise them to zero for new inodes. */
1549	if (ei->i_state & EXT2_STATE_NEW)
1550		memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1551
1552	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1553	if (!(test_opt(sb, NO_UID32))) {
1554		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1555		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1556/*
1557 * Fix up interoperability with old kernels. Otherwise, old inodes get
1558 * re-used with the upper 16 bits of the uid/gid intact
1559 */
1560		if (!ei->i_dtime) {
1561			raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1562			raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1563		} else {
1564			raw_inode->i_uid_high = 0;
1565			raw_inode->i_gid_high = 0;
1566		}
1567	} else {
1568		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1569		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1570		raw_inode->i_uid_high = 0;
1571		raw_inode->i_gid_high = 0;
1572	}
1573	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1574	raw_inode->i_size = cpu_to_le32(inode->i_size);
1575	raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1576	raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1577	raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1578
1579	raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1580	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1581	raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1582	raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1583	raw_inode->i_frag = ei->i_frag_no;
1584	raw_inode->i_fsize = ei->i_frag_size;
1585	raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1586	if (!S_ISREG(inode->i_mode))
1587		raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1588	else {
1589		raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1590		if (inode->i_size > 0x7fffffffULL) {
1591			if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1592					EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1593			    EXT2_SB(sb)->s_es->s_rev_level ==
1594					cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1595			       /* If this is the first large file
1596				* created, add a flag to the superblock.
1597				*/
1598				spin_lock(&EXT2_SB(sb)->s_lock);
1599				ext2_update_dynamic_rev(sb);
1600				EXT2_SET_RO_COMPAT_FEATURE(sb,
1601					EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1602				spin_unlock(&EXT2_SB(sb)->s_lock);
1603				ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1604			}
1605		}
1606	}
1607	
1608	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1609	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1610		if (old_valid_dev(inode->i_rdev)) {
1611			raw_inode->i_block[0] =
1612				cpu_to_le32(old_encode_dev(inode->i_rdev));
1613			raw_inode->i_block[1] = 0;
1614		} else {
1615			raw_inode->i_block[0] = 0;
1616			raw_inode->i_block[1] =
1617				cpu_to_le32(new_encode_dev(inode->i_rdev));
1618			raw_inode->i_block[2] = 0;
1619		}
1620	} else for (n = 0; n < EXT2_N_BLOCKS; n++)
1621		raw_inode->i_block[n] = ei->i_data[n];
1622	mark_buffer_dirty(bh);
1623	if (do_sync) {
1624		sync_dirty_buffer(bh);
1625		if (buffer_req(bh) && !buffer_uptodate(bh)) {
1626			printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1627				sb->s_id, (unsigned long) ino);
1628			err = -EIO;
1629		}
1630	}
1631	ei->i_state &= ~EXT2_STATE_NEW;
1632	brelse (bh);
1633	return err;
1634}
1635
1636int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1637{
1638	return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1639}
1640
1641int ext2_getattr(const struct path *path, struct kstat *stat,
1642		u32 request_mask, unsigned int query_flags)
1643{
1644	struct inode *inode = d_inode(path->dentry);
1645	struct ext2_inode_info *ei = EXT2_I(inode);
1646	unsigned int flags;
1647
1648	flags = ei->i_flags & EXT2_FL_USER_VISIBLE;
1649	if (flags & EXT2_APPEND_FL)
1650		stat->attributes |= STATX_ATTR_APPEND;
1651	if (flags & EXT2_COMPR_FL)
1652		stat->attributes |= STATX_ATTR_COMPRESSED;
1653	if (flags & EXT2_IMMUTABLE_FL)
1654		stat->attributes |= STATX_ATTR_IMMUTABLE;
1655	if (flags & EXT2_NODUMP_FL)
1656		stat->attributes |= STATX_ATTR_NODUMP;
1657	stat->attributes_mask |= (STATX_ATTR_APPEND |
1658			STATX_ATTR_COMPRESSED |
1659			STATX_ATTR_ENCRYPTED |
1660			STATX_ATTR_IMMUTABLE |
1661			STATX_ATTR_NODUMP);
1662
1663	generic_fillattr(inode, stat);
1664	return 0;
1665}
1666
1667int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1668{
1669	struct inode *inode = d_inode(dentry);
1670	int error;
1671
1672	error = setattr_prepare(dentry, iattr);
1673	if (error)
1674		return error;
1675
1676	if (is_quota_modification(inode, iattr)) {
1677		error = dquot_initialize(inode);
1678		if (error)
1679			return error;
1680	}
1681	if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1682	    (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1683		error = dquot_transfer(inode, iattr);
1684		if (error)
1685			return error;
1686	}
1687	if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1688		error = ext2_setsize(inode, iattr->ia_size);
1689		if (error)
1690			return error;
1691	}
1692	setattr_copy(inode, iattr);
1693	if (iattr->ia_valid & ATTR_MODE)
1694		error = posix_acl_chmod(inode, inode->i_mode);
1695	mark_inode_dirty(inode);
1696
1697	return error;
1698}