Linux Audio

Check our new training course

Loading...
v3.1
   1/*
   2 *  linux/fs/ext4/inode.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  from
  10 *
  11 *  linux/fs/minix/inode.c
  12 *
  13 *  Copyright (C) 1991, 1992  Linus Torvalds
  14 *
  15 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  16 *	(jj@sunsite.ms.mff.cuni.cz)
  17 *
  18 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  19 */
  20
  21#include <linux/module.h>
  22#include <linux/fs.h>
  23#include <linux/time.h>
  24#include <linux/jbd2.h>
  25#include <linux/highuid.h>
  26#include <linux/pagemap.h>
 
  27#include <linux/quotaops.h>
  28#include <linux/string.h>
  29#include <linux/buffer_head.h>
  30#include <linux/writeback.h>
  31#include <linux/pagevec.h>
  32#include <linux/mpage.h>
  33#include <linux/namei.h>
  34#include <linux/uio.h>
  35#include <linux/bio.h>
  36#include <linux/workqueue.h>
  37#include <linux/kernel.h>
  38#include <linux/printk.h>
  39#include <linux/slab.h>
  40#include <linux/ratelimit.h>
 
  41
  42#include "ext4_jbd2.h"
  43#include "xattr.h"
  44#include "acl.h"
  45#include "ext4_extents.h"
  46#include "truncate.h"
  47
  48#include <trace/events/ext4.h>
  49
  50#define MPAGE_DA_EXTENT_TAIL 0x01
  51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  52static inline int ext4_begin_ordered_truncate(struct inode *inode,
  53					      loff_t new_size)
  54{
  55	trace_ext4_begin_ordered_truncate(inode, new_size);
  56	/*
  57	 * If jinode is zero, then we never opened the file for
  58	 * writing, so there's no need to call
  59	 * jbd2_journal_begin_ordered_truncate() since there's no
  60	 * outstanding writes we need to flush.
  61	 */
  62	if (!EXT4_I(inode)->jinode)
  63		return 0;
  64	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
  65						   EXT4_I(inode)->jinode,
  66						   new_size);
  67}
  68
  69static void ext4_invalidatepage(struct page *page, unsigned long offset);
  70static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
  71				   struct buffer_head *bh_result, int create);
  72static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
  73static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
  74static int __ext4_journalled_writepage(struct page *page, unsigned int len);
  75static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
 
 
  76
  77/*
  78 * Test whether an inode is a fast symlink.
  79 */
  80static int ext4_inode_is_fast_symlink(struct inode *inode)
  81{
  82	int ea_blocks = EXT4_I(inode)->i_file_acl ?
  83		(inode->i_sb->s_blocksize >> 9) : 0;
 
 
 
  84
  85	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
  86}
  87
  88/*
  89 * Restart the transaction associated with *handle.  This does a commit,
  90 * so before we call here everything must be consistently dirtied against
  91 * this transaction.
  92 */
  93int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
  94				 int nblocks)
  95{
  96	int ret;
  97
  98	/*
  99	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
 100	 * moment, get_block can be called only for blocks inside i_size since
 101	 * page cache has been already dropped and writes are blocked by
 102	 * i_mutex. So we can safely drop the i_data_sem here.
 103	 */
 104	BUG_ON(EXT4_JOURNAL(inode) == NULL);
 105	jbd_debug(2, "restarting handle %p\n", handle);
 106	up_write(&EXT4_I(inode)->i_data_sem);
 107	ret = ext4_journal_restart(handle, nblocks);
 108	down_write(&EXT4_I(inode)->i_data_sem);
 109	ext4_discard_preallocations(inode);
 110
 111	return ret;
 112}
 113
 114/*
 115 * Called at the last iput() if i_nlink is zero.
 116 */
 117void ext4_evict_inode(struct inode *inode)
 118{
 119	handle_t *handle;
 120	int err;
 121
 122	trace_ext4_evict_inode(inode);
 123
 124	ext4_ioend_wait(inode);
 125
 126	if (inode->i_nlink) {
 127		/*
 128		 * When journalling data dirty buffers are tracked only in the
 129		 * journal. So although mm thinks everything is clean and
 130		 * ready for reaping the inode might still have some pages to
 131		 * write in the running transaction or waiting to be
 132		 * checkpointed. Thus calling jbd2_journal_invalidatepage()
 133		 * (via truncate_inode_pages()) to discard these buffers can
 134		 * cause data loss. Also even if we did not discard these
 135		 * buffers, we would have no way to find them after the inode
 136		 * is reaped and thus user could see stale data if he tries to
 137		 * read them before the transaction is checkpointed. So be
 138		 * careful and force everything to disk here... We use
 139		 * ei->i_datasync_tid to store the newest transaction
 140		 * containing inode's data.
 141		 *
 142		 * Note that directories do not have this problem because they
 143		 * don't use page cache.
 144		 */
 145		if (ext4_should_journal_data(inode) &&
 
 146		    (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
 147			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 148			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 149
 150			jbd2_log_start_commit(journal, commit_tid);
 151			jbd2_log_wait_commit(journal, commit_tid);
 152			filemap_write_and_wait(&inode->i_data);
 153		}
 154		truncate_inode_pages(&inode->i_data, 0);
 
 155		goto no_delete;
 156	}
 157
 158	if (!is_bad_inode(inode))
 159		dquot_initialize(inode);
 
 160
 161	if (ext4_should_order_data(inode))
 162		ext4_begin_ordered_truncate(inode, 0);
 163	truncate_inode_pages(&inode->i_data, 0);
 164
 165	if (is_bad_inode(inode))
 166		goto no_delete;
 167
 168	handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
 
 
 
 
 
 
 169	if (IS_ERR(handle)) {
 170		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 171		/*
 172		 * If we're going to skip the normal cleanup, we still need to
 173		 * make sure that the in-core orphan linked list is properly
 174		 * cleaned up.
 175		 */
 176		ext4_orphan_del(NULL, inode);
 
 177		goto no_delete;
 178	}
 179
 180	if (IS_SYNC(inode))
 181		ext4_handle_sync(handle);
 182	inode->i_size = 0;
 183	err = ext4_mark_inode_dirty(handle, inode);
 184	if (err) {
 185		ext4_warning(inode->i_sb,
 186			     "couldn't mark inode dirty (err %d)", err);
 187		goto stop_handle;
 188	}
 189	if (inode->i_blocks)
 190		ext4_truncate(inode);
 
 
 
 
 
 
 
 191
 192	/*
 193	 * ext4_ext_truncate() doesn't reserve any slop when it
 194	 * restarts journal transactions; therefore there may not be
 195	 * enough credits left in the handle to remove the inode from
 196	 * the orphan list and set the dtime field.
 197	 */
 198	if (!ext4_handle_has_enough_credits(handle, 3)) {
 199		err = ext4_journal_extend(handle, 3);
 200		if (err > 0)
 201			err = ext4_journal_restart(handle, 3);
 202		if (err != 0) {
 203			ext4_warning(inode->i_sb,
 204				     "couldn't extend journal (err %d)", err);
 205		stop_handle:
 206			ext4_journal_stop(handle);
 207			ext4_orphan_del(NULL, inode);
 
 208			goto no_delete;
 209		}
 210	}
 211
 212	/*
 213	 * Kill off the orphan record which ext4_truncate created.
 214	 * AKPM: I think this can be inside the above `if'.
 215	 * Note that ext4_orphan_del() has to be able to cope with the
 216	 * deletion of a non-existent orphan - this is because we don't
 217	 * know if ext4_truncate() actually created an orphan record.
 218	 * (Well, we could do this if we need to, but heck - it works)
 219	 */
 220	ext4_orphan_del(handle, inode);
 221	EXT4_I(inode)->i_dtime	= get_seconds();
 222
 223	/*
 224	 * One subtle ordering requirement: if anything has gone wrong
 225	 * (transaction abort, IO errors, whatever), then we can still
 226	 * do these next steps (the fs will already have been marked as
 227	 * having errors), but we can't free the inode if the mark_dirty
 228	 * fails.
 229	 */
 230	if (ext4_mark_inode_dirty(handle, inode))
 231		/* If that failed, just do the required in-core inode clear. */
 232		ext4_clear_inode(inode);
 233	else
 234		ext4_free_inode(handle, inode);
 235	ext4_journal_stop(handle);
 
 236	return;
 237no_delete:
 238	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 239}
 240
 241#ifdef CONFIG_QUOTA
 242qsize_t *ext4_get_reserved_space(struct inode *inode)
 243{
 244	return &EXT4_I(inode)->i_reserved_quota;
 245}
 246#endif
 247
 248/*
 249 * Calculate the number of metadata blocks need to reserve
 250 * to allocate a block located at @lblock
 251 */
 252static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
 253{
 254	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 255		return ext4_ext_calc_metadata_amount(inode, lblock);
 256
 257	return ext4_ind_calc_metadata_amount(inode, lblock);
 258}
 259
 260/*
 261 * Called with i_data_sem down, which is important since we can call
 262 * ext4_discard_preallocations() from here.
 263 */
 264void ext4_da_update_reserve_space(struct inode *inode,
 265					int used, int quota_claim)
 266{
 267	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 268	struct ext4_inode_info *ei = EXT4_I(inode);
 269
 270	spin_lock(&ei->i_block_reservation_lock);
 271	trace_ext4_da_update_reserve_space(inode, used);
 272	if (unlikely(used > ei->i_reserved_data_blocks)) {
 273		ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
 274			 "with only %d reserved data blocks\n",
 275			 __func__, inode->i_ino, used,
 276			 ei->i_reserved_data_blocks);
 277		WARN_ON(1);
 278		used = ei->i_reserved_data_blocks;
 279	}
 280
 281	/* Update per-inode reservations */
 282	ei->i_reserved_data_blocks -= used;
 283	ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
 284	percpu_counter_sub(&sbi->s_dirtyblocks_counter,
 285			   used + ei->i_allocated_meta_blocks);
 286	ei->i_allocated_meta_blocks = 0;
 287
 288	if (ei->i_reserved_data_blocks == 0) {
 289		/*
 290		 * We can release all of the reserved metadata blocks
 291		 * only when we have written all of the delayed
 292		 * allocation blocks.
 293		 */
 294		percpu_counter_sub(&sbi->s_dirtyblocks_counter,
 295				   ei->i_reserved_meta_blocks);
 296		ei->i_reserved_meta_blocks = 0;
 297		ei->i_da_metadata_calc_len = 0;
 298	}
 299	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 300
 301	/* Update quota subsystem for data blocks */
 302	if (quota_claim)
 303		dquot_claim_block(inode, used);
 304	else {
 305		/*
 306		 * We did fallocate with an offset that is already delayed
 307		 * allocated. So on delayed allocated writeback we should
 308		 * not re-claim the quota for fallocated blocks.
 309		 */
 310		dquot_release_reservation_block(inode, used);
 311	}
 312
 313	/*
 314	 * If we have done all the pending block allocations and if
 315	 * there aren't any writers on the inode, we can discard the
 316	 * inode's preallocations.
 317	 */
 318	if ((ei->i_reserved_data_blocks == 0) &&
 319	    (atomic_read(&inode->i_writecount) == 0))
 320		ext4_discard_preallocations(inode);
 321}
 322
 323static int __check_block_validity(struct inode *inode, const char *func,
 324				unsigned int line,
 325				struct ext4_map_blocks *map)
 326{
 327	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
 328				   map->m_len)) {
 329		ext4_error_inode(inode, func, line, map->m_pblk,
 330				 "lblock %lu mapped to illegal pblock "
 331				 "(length %d)", (unsigned long) map->m_lblk,
 332				 map->m_len);
 333		return -EIO;
 334	}
 335	return 0;
 336}
 337
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 338#define check_block_validity(inode, map)	\
 339	__check_block_validity((inode), __func__, __LINE__, (map))
 340
 341/*
 342 * Return the number of contiguous dirty pages in a given inode
 343 * starting at page frame idx.
 344 */
 345static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
 346				    unsigned int max_pages)
 347{
 348	struct address_space *mapping = inode->i_mapping;
 349	pgoff_t	index;
 350	struct pagevec pvec;
 351	pgoff_t num = 0;
 352	int i, nr_pages, done = 0;
 353
 354	if (max_pages == 0)
 355		return 0;
 356	pagevec_init(&pvec, 0);
 357	while (!done) {
 358		index = idx;
 359		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 360					      PAGECACHE_TAG_DIRTY,
 361					      (pgoff_t)PAGEVEC_SIZE);
 362		if (nr_pages == 0)
 363			break;
 364		for (i = 0; i < nr_pages; i++) {
 365			struct page *page = pvec.pages[i];
 366			struct buffer_head *bh, *head;
 
 
 
 
 367
 368			lock_page(page);
 369			if (unlikely(page->mapping != mapping) ||
 370			    !PageDirty(page) ||
 371			    PageWriteback(page) ||
 372			    page->index != idx) {
 373				done = 1;
 374				unlock_page(page);
 375				break;
 376			}
 377			if (page_has_buffers(page)) {
 378				bh = head = page_buffers(page);
 379				do {
 380					if (!buffer_delay(bh) &&
 381					    !buffer_unwritten(bh))
 382						done = 1;
 383					bh = bh->b_this_page;
 384				} while (!done && (bh != head));
 385			}
 386			unlock_page(page);
 387			if (done)
 388				break;
 389			idx++;
 390			num++;
 391			if (num >= max_pages) {
 392				done = 1;
 393				break;
 394			}
 395		}
 396		pagevec_release(&pvec);
 397	}
 398	return num;
 399}
 
 400
 401/*
 402 * The ext4_map_blocks() function tries to look up the requested blocks,
 403 * and returns if the blocks are already mapped.
 404 *
 405 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 406 * and store the allocated blocks in the result buffer head and mark it
 407 * mapped.
 408 *
 409 * If file type is extents based, it will call ext4_ext_map_blocks(),
 410 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 411 * based files
 412 *
 413 * On success, it returns the number of blocks being mapped or allocate.
 414 * if create==0 and the blocks are pre-allocated and uninitialized block,
 415 * the result buffer head is unmapped. If the create ==1, it will make sure
 416 * the buffer head is mapped.
 417 *
 418 * It returns 0 if plain look up failed (blocks have not been allocated), in
 419 * that casem, buffer head is unmapped
 
 420 *
 421 * It returns the error in case of allocation failure.
 422 */
 423int ext4_map_blocks(handle_t *handle, struct inode *inode,
 424		    struct ext4_map_blocks *map, int flags)
 425{
 
 426	int retval;
 
 
 
 
 
 
 427
 428	map->m_flags = 0;
 429	ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
 430		  "logical block %lu\n", inode->i_ino, flags, map->m_len,
 431		  (unsigned long) map->m_lblk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 432	/*
 433	 * Try to see if we can get the block without requesting a new
 434	 * file system block.
 435	 */
 436	down_read((&EXT4_I(inode)->i_data_sem));
 437	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 438		retval = ext4_ext_map_blocks(handle, inode, map, 0);
 
 439	} else {
 440		retval = ext4_ind_map_blocks(handle, inode, map, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 441	}
 442	up_read((&EXT4_I(inode)->i_data_sem));
 443
 
 444	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 445		int ret = check_block_validity(inode, map);
 446		if (ret != 0)
 447			return ret;
 448	}
 449
 450	/* If it is only a block(s) look up */
 451	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 452		return retval;
 453
 454	/*
 455	 * Returns if the blocks have already allocated
 456	 *
 457	 * Note that if blocks have been preallocated
 458	 * ext4_ext_get_block() returns th create = 0
 459	 * with buffer head unmapped.
 460	 */
 461	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 462		return retval;
 
 
 
 
 
 
 463
 464	/*
 465	 * When we call get_blocks without the create flag, the
 466	 * BH_Unwritten flag could have gotten set if the blocks
 467	 * requested were part of a uninitialized extent.  We need to
 468	 * clear this flag now that we are committed to convert all or
 469	 * part of the uninitialized extent to be an initialized
 470	 * extent.  This is because we need to avoid the combination
 471	 * of BH_Unwritten and BH_Mapped flags being simultaneously
 472	 * set on the buffer_head.
 473	 */
 474	map->m_flags &= ~EXT4_MAP_UNWRITTEN;
 475
 476	/*
 477	 * New blocks allocate and/or writing to uninitialized extent
 478	 * will possibly result in updating i_data, so we take
 479	 * the write lock of i_data_sem, and call get_blocks()
 480	 * with create == 1 flag.
 481	 */
 482	down_write((&EXT4_I(inode)->i_data_sem));
 483
 484	/*
 485	 * if the caller is from delayed allocation writeout path
 486	 * we have already reserved fs blocks for allocation
 487	 * let the underlying get_block() function know to
 488	 * avoid double accounting
 489	 */
 490	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 491		ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 492	/*
 493	 * We need to check for EXT4 here because migrate
 494	 * could have changed the inode type in between
 495	 */
 496	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 497		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 498	} else {
 499		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 500
 501		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 502			/*
 503			 * We allocated new blocks which will result in
 504			 * i_data's format changing.  Force the migrate
 505			 * to fail by clearing migrate flags
 506			 */
 507			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 508		}
 509
 510		/*
 511		 * Update reserved blocks/metadata blocks after successful
 512		 * block allocation which had been deferred till now. We don't
 513		 * support fallocate for non extent files. So we can update
 514		 * reserve space here.
 515		 */
 516		if ((retval > 0) &&
 517			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 518			ext4_da_update_reserve_space(inode, retval, 1);
 519	}
 520	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 521		ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
 522
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 523	up_write((&EXT4_I(inode)->i_data_sem));
 524	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 525		int ret = check_block_validity(inode, map);
 526		if (ret != 0)
 527			return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 528	}
 529	return retval;
 530}
 531
 532/* Maximum number of blocks we map for direct IO at once. */
 533#define DIO_MAX_BLOCKS 4096
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 534
 535static int _ext4_get_block(struct inode *inode, sector_t iblock,
 536			   struct buffer_head *bh, int flags)
 537{
 538	handle_t *handle = ext4_journal_current_handle();
 539	struct ext4_map_blocks map;
 540	int ret = 0, started = 0;
 541	int dio_credits;
 
 
 542
 543	map.m_lblk = iblock;
 544	map.m_len = bh->b_size >> inode->i_blkbits;
 545
 546	if (flags && !handle) {
 547		/* Direct IO write... */
 548		if (map.m_len > DIO_MAX_BLOCKS)
 549			map.m_len = DIO_MAX_BLOCKS;
 550		dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
 551		handle = ext4_journal_start(inode, dio_credits);
 552		if (IS_ERR(handle)) {
 553			ret = PTR_ERR(handle);
 554			return ret;
 555		}
 556		started = 1;
 557	}
 558
 559	ret = ext4_map_blocks(handle, inode, &map, flags);
 560	if (ret > 0) {
 561		map_bh(bh, inode->i_sb, map.m_pblk);
 562		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
 563		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 564		ret = 0;
 
 
 
 565	}
 566	if (started)
 567		ext4_journal_stop(handle);
 568	return ret;
 569}
 570
 571int ext4_get_block(struct inode *inode, sector_t iblock,
 572		   struct buffer_head *bh, int create)
 573{
 574	return _ext4_get_block(inode, iblock, bh,
 575			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 576}
 577
 578/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 579 * `handle' can be NULL if create is zero
 580 */
 581struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 582				ext4_lblk_t block, int create, int *errp)
 583{
 584	struct ext4_map_blocks map;
 585	struct buffer_head *bh;
 586	int fatal = 0, err;
 
 587
 588	J_ASSERT(handle != NULL || create == 0);
 589
 590	map.m_lblk = block;
 591	map.m_len = 1;
 592	err = ext4_map_blocks(handle, inode, &map,
 593			      create ? EXT4_GET_BLOCKS_CREATE : 0);
 594
 
 
 595	if (err < 0)
 596		*errp = err;
 597	if (err <= 0)
 598		return NULL;
 599	*errp = 0;
 600
 601	bh = sb_getblk(inode->i_sb, map.m_pblk);
 602	if (!bh) {
 603		*errp = -EIO;
 604		return NULL;
 605	}
 606	if (map.m_flags & EXT4_MAP_NEW) {
 607		J_ASSERT(create != 0);
 608		J_ASSERT(handle != NULL);
 609
 610		/*
 611		 * Now that we do not always journal data, we should
 612		 * keep in mind whether this should always journal the
 613		 * new buffer as metadata.  For now, regular file
 614		 * writes use ext4_get_block instead, so it's not a
 615		 * problem.
 616		 */
 617		lock_buffer(bh);
 618		BUFFER_TRACE(bh, "call get_create_access");
 619		fatal = ext4_journal_get_create_access(handle, bh);
 620		if (!fatal && !buffer_uptodate(bh)) {
 
 
 
 
 621			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 622			set_buffer_uptodate(bh);
 623		}
 624		unlock_buffer(bh);
 625		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 626		err = ext4_handle_dirty_metadata(handle, inode, bh);
 627		if (!fatal)
 628			fatal = err;
 629	} else {
 630		BUFFER_TRACE(bh, "not a new buffer");
 631	}
 632	if (fatal) {
 633		*errp = fatal;
 634		brelse(bh);
 635		bh = NULL;
 636	}
 637	return bh;
 
 
 
 638}
 639
 640struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 641			       ext4_lblk_t block, int create, int *err)
 642{
 643	struct buffer_head *bh;
 644
 645	bh = ext4_getblk(handle, inode, block, create, err);
 646	if (!bh)
 647		return bh;
 648	if (buffer_uptodate(bh))
 649		return bh;
 650	ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
 651	wait_on_buffer(bh);
 652	if (buffer_uptodate(bh))
 653		return bh;
 654	put_bh(bh);
 655	*err = -EIO;
 656	return NULL;
 657}
 658
 659static int walk_page_buffers(handle_t *handle,
 660			     struct buffer_head *head,
 661			     unsigned from,
 662			     unsigned to,
 663			     int *partial,
 664			     int (*fn)(handle_t *handle,
 665				       struct buffer_head *bh))
 666{
 667	struct buffer_head *bh;
 668	unsigned block_start, block_end;
 669	unsigned blocksize = head->b_size;
 670	int err, ret = 0;
 671	struct buffer_head *next;
 672
 673	for (bh = head, block_start = 0;
 674	     ret == 0 && (bh != head || !block_start);
 675	     block_start = block_end, bh = next) {
 676		next = bh->b_this_page;
 677		block_end = block_start + blocksize;
 678		if (block_end <= from || block_start >= to) {
 679			if (partial && !buffer_uptodate(bh))
 680				*partial = 1;
 681			continue;
 682		}
 683		err = (*fn)(handle, bh);
 684		if (!ret)
 685			ret = err;
 686	}
 687	return ret;
 688}
 689
 690/*
 691 * To preserve ordering, it is essential that the hole instantiation and
 692 * the data write be encapsulated in a single transaction.  We cannot
 693 * close off a transaction and start a new one between the ext4_get_block()
 694 * and the commit_write().  So doing the jbd2_journal_start at the start of
 695 * prepare_write() is the right place.
 696 *
 697 * Also, this function can nest inside ext4_writepage() ->
 698 * block_write_full_page(). In that case, we *know* that ext4_writepage()
 699 * has generated enough buffer credits to do the whole page.  So we won't
 700 * block on the journal in that case, which is good, because the caller may
 701 * be PF_MEMALLOC.
 702 *
 703 * By accident, ext4 can be reentered when a transaction is open via
 704 * quota file writes.  If we were to commit the transaction while thus
 705 * reentered, there can be a deadlock - we would be holding a quota
 706 * lock, and the commit would never complete if another thread had a
 707 * transaction open and was blocking on the quota lock - a ranking
 708 * violation.
 709 *
 710 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
 711 * will _not_ run commit under these circumstances because handle->h_ref
 712 * is elevated.  We'll still have enough credits for the tiny quotafile
 713 * write.
 714 */
 715static int do_journal_get_write_access(handle_t *handle,
 716				       struct buffer_head *bh)
 717{
 718	int dirty = buffer_dirty(bh);
 719	int ret;
 720
 721	if (!buffer_mapped(bh) || buffer_freed(bh))
 722		return 0;
 723	/*
 724	 * __block_write_begin() could have dirtied some buffers. Clean
 725	 * the dirty bit as jbd2_journal_get_write_access() could complain
 726	 * otherwise about fs integrity issues. Setting of the dirty bit
 727	 * by __block_write_begin() isn't a real problem here as we clear
 728	 * the bit before releasing a page lock and thus writeback cannot
 729	 * ever write the buffer.
 730	 */
 731	if (dirty)
 732		clear_buffer_dirty(bh);
 
 733	ret = ext4_journal_get_write_access(handle, bh);
 734	if (!ret && dirty)
 735		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
 736	return ret;
 737}
 738
 739static int ext4_get_block_write(struct inode *inode, sector_t iblock,
 740		   struct buffer_head *bh_result, int create);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 741static int ext4_write_begin(struct file *file, struct address_space *mapping,
 742			    loff_t pos, unsigned len, unsigned flags,
 743			    struct page **pagep, void **fsdata)
 744{
 745	struct inode *inode = mapping->host;
 746	int ret, needed_blocks;
 747	handle_t *handle;
 748	int retries = 0;
 749	struct page *page;
 750	pgoff_t index;
 751	unsigned from, to;
 752
 753	trace_ext4_write_begin(inode, pos, len, flags);
 754	/*
 755	 * Reserve one block more for addition to orphan list in case
 756	 * we allocate blocks but write fails for some reason
 757	 */
 758	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
 759	index = pos >> PAGE_CACHE_SHIFT;
 760	from = pos & (PAGE_CACHE_SIZE - 1);
 761	to = from + len;
 762
 763retry:
 764	handle = ext4_journal_start(inode, needed_blocks);
 765	if (IS_ERR(handle)) {
 766		ret = PTR_ERR(handle);
 767		goto out;
 
 
 768	}
 769
 770	/* We cannot recurse into the filesystem as the transaction is already
 771	 * started */
 772	flags |= AOP_FLAG_NOFS;
 773
 
 
 
 
 774	page = grab_cache_page_write_begin(mapping, index, flags);
 775	if (!page) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 776		ext4_journal_stop(handle);
 777		ret = -ENOMEM;
 778		goto out;
 779	}
 780	*pagep = page;
 
 781
 
 
 
 
 
 
 
 
 782	if (ext4_should_dioread_nolock(inode))
 783		ret = __block_write_begin(page, pos, len, ext4_get_block_write);
 
 784	else
 785		ret = __block_write_begin(page, pos, len, ext4_get_block);
 786
 787	if (!ret && ext4_should_journal_data(inode)) {
 788		ret = walk_page_buffers(handle, page_buffers(page),
 789				from, to, NULL, do_journal_get_write_access);
 
 790	}
 791
 792	if (ret) {
 793		unlock_page(page);
 794		page_cache_release(page);
 795		/*
 796		 * __block_write_begin may have instantiated a few blocks
 797		 * outside i_size.  Trim these off again. Don't need
 798		 * i_size_read because we hold i_mutex.
 799		 *
 800		 * Add inode to orphan list in case we crash before
 801		 * truncate finishes
 802		 */
 803		if (pos + len > inode->i_size && ext4_can_truncate(inode))
 804			ext4_orphan_add(handle, inode);
 805
 806		ext4_journal_stop(handle);
 807		if (pos + len > inode->i_size) {
 808			ext4_truncate_failed_write(inode);
 809			/*
 810			 * If truncate failed early the inode might
 811			 * still be on the orphan list; we need to
 812			 * make sure the inode is removed from the
 813			 * orphan list in that case.
 814			 */
 815			if (inode->i_nlink)
 816				ext4_orphan_del(NULL, inode);
 817		}
 818	}
 819
 820	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 821		goto retry;
 822out:
 
 
 
 
 823	return ret;
 824}
 825
 826/* For write_end() in data=journal mode */
 827static int write_end_fn(handle_t *handle, struct buffer_head *bh)
 828{
 
 829	if (!buffer_mapped(bh) || buffer_freed(bh))
 830		return 0;
 831	set_buffer_uptodate(bh);
 832	return ext4_handle_dirty_metadata(handle, NULL, bh);
 
 
 
 833}
 834
 835static int ext4_generic_write_end(struct file *file,
 836				  struct address_space *mapping,
 837				  loff_t pos, unsigned len, unsigned copied,
 838				  struct page *page, void *fsdata)
 
 
 
 
 
 
 
 839{
 840	int i_size_changed = 0;
 841	struct inode *inode = mapping->host;
 842	handle_t *handle = ext4_journal_current_handle();
 
 
 
 
 843
 844	copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 845
 
 
 
 
 
 
 
 
 
 
 
 846	/*
 847	 * No need to use i_size_read() here, the i_size
 848	 * cannot change under us because we hold i_mutex.
 849	 *
 850	 * But it's important to update i_size while still holding page lock:
 851	 * page writeout could otherwise come in and zero beyond i_size.
 852	 */
 853	if (pos + copied > inode->i_size) {
 854		i_size_write(inode, pos + copied);
 855		i_size_changed = 1;
 856	}
 857
 858	if (pos + copied >  EXT4_I(inode)->i_disksize) {
 859		/* We need to mark inode dirty even if
 860		 * new_i_size is less that inode->i_size
 861		 * bu greater than i_disksize.(hint delalloc)
 862		 */
 863		ext4_update_i_disksize(inode, (pos + copied));
 864		i_size_changed = 1;
 865	}
 866	unlock_page(page);
 867	page_cache_release(page);
 868
 
 
 869	/*
 870	 * Don't mark the inode dirty under page lock. First, it unnecessarily
 871	 * makes the holding time of page lock longer. Second, it forces lock
 872	 * ordering of page lock and transaction start for journaling
 873	 * filesystems.
 874	 */
 875	if (i_size_changed)
 876		ext4_mark_inode_dirty(handle, inode);
 877
 878	return copied;
 879}
 880
 881/*
 882 * We need to pick up the new inode size which generic_commit_write gave us
 883 * `file' can be NULL - eg, when called from page_symlink().
 884 *
 885 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
 886 * buffers are managed internally.
 887 */
 888static int ext4_ordered_write_end(struct file *file,
 889				  struct address_space *mapping,
 890				  loff_t pos, unsigned len, unsigned copied,
 891				  struct page *page, void *fsdata)
 892{
 893	handle_t *handle = ext4_journal_current_handle();
 894	struct inode *inode = mapping->host;
 895	int ret = 0, ret2;
 896
 897	trace_ext4_ordered_write_end(inode, pos, len, copied);
 898	ret = ext4_jbd2_file_inode(handle, inode);
 899
 900	if (ret == 0) {
 901		ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
 902							page, fsdata);
 903		copied = ret2;
 904		if (pos + len > inode->i_size && ext4_can_truncate(inode))
 905			/* if we have allocated more blocks and copied
 906			 * less. We will have blocks allocated outside
 907			 * inode->i_size. So truncate them
 908			 */
 909			ext4_orphan_add(handle, inode);
 910		if (ret2 < 0)
 911			ret = ret2;
 912	}
 913	ret2 = ext4_journal_stop(handle);
 914	if (!ret)
 915		ret = ret2;
 916
 917	if (pos + len > inode->i_size) {
 918		ext4_truncate_failed_write(inode);
 919		/*
 920		 * If truncate failed early the inode might still be
 921		 * on the orphan list; we need to make sure the inode
 922		 * is removed from the orphan list in that case.
 923		 */
 924		if (inode->i_nlink)
 925			ext4_orphan_del(NULL, inode);
 926	}
 927
 928
 929	return ret ? ret : copied;
 930}
 931
 932static int ext4_writeback_write_end(struct file *file,
 933				    struct address_space *mapping,
 934				    loff_t pos, unsigned len, unsigned copied,
 935				    struct page *page, void *fsdata)
 936{
 937	handle_t *handle = ext4_journal_current_handle();
 938	struct inode *inode = mapping->host;
 939	int ret = 0, ret2;
 940
 941	trace_ext4_writeback_write_end(inode, pos, len, copied);
 942	ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
 943							page, fsdata);
 944	copied = ret2;
 945	if (pos + len > inode->i_size && ext4_can_truncate(inode))
 946		/* if we have allocated more blocks and copied
 947		 * less. We will have blocks allocated outside
 948		 * inode->i_size. So truncate them
 949		 */
 950		ext4_orphan_add(handle, inode);
 951
 952	if (ret2 < 0)
 953		ret = ret2;
 954
 955	ret2 = ext4_journal_stop(handle);
 956	if (!ret)
 957		ret = ret2;
 958
 959	if (pos + len > inode->i_size) {
 960		ext4_truncate_failed_write(inode);
 961		/*
 962		 * If truncate failed early the inode might still be
 963		 * on the orphan list; we need to make sure the inode
 964		 * is removed from the orphan list in that case.
 965		 */
 966		if (inode->i_nlink)
 967			ext4_orphan_del(NULL, inode);
 968	}
 969
 970	return ret ? ret : copied;
 971}
 972
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 973static int ext4_journalled_write_end(struct file *file,
 974				     struct address_space *mapping,
 975				     loff_t pos, unsigned len, unsigned copied,
 976				     struct page *page, void *fsdata)
 977{
 978	handle_t *handle = ext4_journal_current_handle();
 979	struct inode *inode = mapping->host;
 
 980	int ret = 0, ret2;
 981	int partial = 0;
 982	unsigned from, to;
 983	loff_t new_i_size;
 984
 985	trace_ext4_journalled_write_end(inode, pos, len, copied);
 986	from = pos & (PAGE_CACHE_SIZE - 1);
 987	to = from + len;
 988
 989	BUG_ON(!ext4_handle_valid(handle));
 990
 991	if (copied < len) {
 992		if (!PageUptodate(page))
 993			copied = 0;
 994		page_zero_new_buffers(page, from+copied, to);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 995	}
 996
 997	ret = walk_page_buffers(handle, page_buffers(page), from,
 998				to, &partial, write_end_fn);
 999	if (!partial)
1000		SetPageUptodate(page);
1001	new_i_size = pos + copied;
1002	if (new_i_size > inode->i_size)
1003		i_size_write(inode, pos+copied);
1004	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1005	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1006	if (new_i_size > EXT4_I(inode)->i_disksize) {
1007		ext4_update_i_disksize(inode, new_i_size);
 
 
 
 
 
1008		ret2 = ext4_mark_inode_dirty(handle, inode);
1009		if (!ret)
1010			ret = ret2;
1011	}
1012
1013	unlock_page(page);
1014	page_cache_release(page);
1015	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1016		/* if we have allocated more blocks and copied
1017		 * less. We will have blocks allocated outside
1018		 * inode->i_size. So truncate them
1019		 */
1020		ext4_orphan_add(handle, inode);
1021
 
1022	ret2 = ext4_journal_stop(handle);
1023	if (!ret)
1024		ret = ret2;
1025	if (pos + len > inode->i_size) {
1026		ext4_truncate_failed_write(inode);
1027		/*
1028		 * If truncate failed early the inode might still be
1029		 * on the orphan list; we need to make sure the inode
1030		 * is removed from the orphan list in that case.
1031		 */
1032		if (inode->i_nlink)
1033			ext4_orphan_del(NULL, inode);
1034	}
1035
1036	return ret ? ret : copied;
1037}
1038
1039/*
1040 * Reserve a single block located at lblock
1041 */
1042static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1043{
1044	int retries = 0;
1045	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1046	struct ext4_inode_info *ei = EXT4_I(inode);
1047	unsigned long md_needed;
1048	int ret;
1049
1050	/*
1051	 * recalculate the amount of metadata blocks to reserve
1052	 * in order to allocate nrblocks
1053	 * worse case is one extent per block
1054	 */
1055repeat:
1056	spin_lock(&ei->i_block_reservation_lock);
1057	md_needed = ext4_calc_metadata_amount(inode, lblock);
1058	trace_ext4_da_reserve_space(inode, md_needed);
1059	spin_unlock(&ei->i_block_reservation_lock);
1060
1061	/*
1062	 * We will charge metadata quota at writeout time; this saves
1063	 * us from metadata over-estimation, though we may go over by
1064	 * a small amount in the end.  Here we just reserve for data.
1065	 */
1066	ret = dquot_reserve_block(inode, 1);
1067	if (ret)
1068		return ret;
1069	/*
1070	 * We do still charge estimated metadata to the sb though;
1071	 * we cannot afford to run out of free blocks.
1072	 */
1073	if (ext4_claim_free_blocks(sbi, md_needed + 1, 0)) {
1074		dquot_release_reservation_block(inode, 1);
1075		if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1076			yield();
1077			goto repeat;
1078		}
1079		return -ENOSPC;
1080	}
1081	spin_lock(&ei->i_block_reservation_lock);
1082	ei->i_reserved_data_blocks++;
1083	ei->i_reserved_meta_blocks += md_needed;
1084	spin_unlock(&ei->i_block_reservation_lock);
1085
1086	return 0;       /* success */
1087}
1088
1089static void ext4_da_release_space(struct inode *inode, int to_free)
1090{
1091	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092	struct ext4_inode_info *ei = EXT4_I(inode);
1093
1094	if (!to_free)
1095		return;		/* Nothing to release, exit */
1096
1097	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1098
1099	trace_ext4_da_release_space(inode, to_free);
1100	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1101		/*
1102		 * if there aren't enough reserved blocks, then the
1103		 * counter is messed up somewhere.  Since this
1104		 * function is called from invalidate page, it's
1105		 * harmless to return without any action.
1106		 */
1107		ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1108			 "ino %lu, to_free %d with only %d reserved "
1109			 "data blocks\n", inode->i_ino, to_free,
1110			 ei->i_reserved_data_blocks);
1111		WARN_ON(1);
1112		to_free = ei->i_reserved_data_blocks;
1113	}
1114	ei->i_reserved_data_blocks -= to_free;
1115
1116	if (ei->i_reserved_data_blocks == 0) {
1117		/*
1118		 * We can release all of the reserved metadata blocks
1119		 * only when we have written all of the delayed
1120		 * allocation blocks.
1121		 */
1122		percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1123				   ei->i_reserved_meta_blocks);
1124		ei->i_reserved_meta_blocks = 0;
1125		ei->i_da_metadata_calc_len = 0;
1126	}
1127
1128	/* update fs dirty data blocks counter */
1129	percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1130
1131	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1132
1133	dquot_release_reservation_block(inode, to_free);
1134}
1135
1136static void ext4_da_page_release_reservation(struct page *page,
1137					     unsigned long offset)
 
1138{
1139	int to_release = 0;
1140	struct buffer_head *head, *bh;
1141	unsigned int curr_off = 0;
 
 
 
 
 
 
 
1142
1143	head = page_buffers(page);
1144	bh = head;
1145	do {
1146		unsigned int next_off = curr_off + bh->b_size;
1147
 
 
 
1148		if ((offset <= curr_off) && (buffer_delay(bh))) {
1149			to_release++;
 
1150			clear_buffer_delay(bh);
 
 
 
 
 
 
 
1151		}
1152		curr_off = next_off;
1153	} while ((bh = bh->b_this_page) != head);
1154	ext4_da_release_space(page->mapping->host, to_release);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1155}
1156
1157/*
1158 * Delayed allocation stuff
1159 */
1160
1161/*
1162 * mpage_da_submit_io - walks through extent of pages and try to write
1163 * them with writepage() call back
1164 *
1165 * @mpd->inode: inode
1166 * @mpd->first_page: first page of the extent
1167 * @mpd->next_page: page after the last page of the extent
1168 *
1169 * By the time mpage_da_submit_io() is called we expect all blocks
1170 * to be allocated. this may be wrong if allocation failed.
1171 *
1172 * As pages are already locked by write_cache_pages(), we can't use it
1173 */
1174static int mpage_da_submit_io(struct mpage_da_data *mpd,
1175			      struct ext4_map_blocks *map)
1176{
1177	struct pagevec pvec;
1178	unsigned long index, end;
1179	int ret = 0, err, nr_pages, i;
1180	struct inode *inode = mpd->inode;
1181	struct address_space *mapping = inode->i_mapping;
1182	loff_t size = i_size_read(inode);
1183	unsigned int len, block_start;
1184	struct buffer_head *bh, *page_bufs = NULL;
1185	int journal_data = ext4_should_journal_data(inode);
1186	sector_t pblock = 0, cur_logical = 0;
1187	struct ext4_io_submit io_submit;
1188
1189	BUG_ON(mpd->next_page <= mpd->first_page);
1190	memset(&io_submit, 0, sizeof(io_submit));
1191	/*
1192	 * We need to start from the first_page to the next_page - 1
1193	 * to make sure we also write the mapped dirty buffer_heads.
1194	 * If we look at mpd->b_blocknr we would only be looking
1195	 * at the currently mapped buffer_heads.
1196	 */
1197	index = mpd->first_page;
1198	end = mpd->next_page - 1;
1199
1200	pagevec_init(&pvec, 0);
1201	while (index <= end) {
1202		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1203		if (nr_pages == 0)
1204			break;
1205		for (i = 0; i < nr_pages; i++) {
1206			int commit_write = 0, skip_page = 0;
1207			struct page *page = pvec.pages[i];
1208
1209			index = page->index;
1210			if (index > end)
1211				break;
1212
1213			if (index == size >> PAGE_CACHE_SHIFT)
1214				len = size & ~PAGE_CACHE_MASK;
1215			else
1216				len = PAGE_CACHE_SIZE;
1217			if (map) {
1218				cur_logical = index << (PAGE_CACHE_SHIFT -
1219							inode->i_blkbits);
1220				pblock = map->m_pblk + (cur_logical -
1221							map->m_lblk);
1222			}
1223			index++;
1224
1225			BUG_ON(!PageLocked(page));
1226			BUG_ON(PageWriteback(page));
1227
1228			/*
1229			 * If the page does not have buffers (for
1230			 * whatever reason), try to create them using
1231			 * __block_write_begin.  If this fails,
1232			 * skip the page and move on.
1233			 */
1234			if (!page_has_buffers(page)) {
1235				if (__block_write_begin(page, 0, len,
1236						noalloc_get_block_write)) {
1237				skip_page:
1238					unlock_page(page);
1239					continue;
1240				}
1241				commit_write = 1;
1242			}
1243
1244			bh = page_bufs = page_buffers(page);
1245			block_start = 0;
1246			do {
1247				if (!bh)
1248					goto skip_page;
1249				if (map && (cur_logical >= map->m_lblk) &&
1250				    (cur_logical <= (map->m_lblk +
1251						     (map->m_len - 1)))) {
1252					if (buffer_delay(bh)) {
1253						clear_buffer_delay(bh);
1254						bh->b_blocknr = pblock;
1255					}
1256					if (buffer_unwritten(bh) ||
1257					    buffer_mapped(bh))
1258						BUG_ON(bh->b_blocknr != pblock);
1259					if (map->m_flags & EXT4_MAP_UNINIT)
1260						set_buffer_uninit(bh);
1261					clear_buffer_unwritten(bh);
1262				}
1263
1264				/* skip page if block allocation undone */
1265				if (buffer_delay(bh) || buffer_unwritten(bh))
1266					skip_page = 1;
1267				bh = bh->b_this_page;
1268				block_start += bh->b_size;
1269				cur_logical++;
1270				pblock++;
1271			} while (bh != page_bufs);
1272
1273			if (skip_page)
1274				goto skip_page;
1275
1276			if (commit_write)
1277				/* mark the buffer_heads as dirty & uptodate */
1278				block_commit_write(page, 0, len);
1279
1280			clear_page_dirty_for_io(page);
1281			/*
1282			 * Delalloc doesn't support data journalling,
1283			 * but eventually maybe we'll lift this
1284			 * restriction.
1285			 */
1286			if (unlikely(journal_data && PageChecked(page)))
1287				err = __ext4_journalled_writepage(page, len);
1288			else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
1289				err = ext4_bio_write_page(&io_submit, page,
1290							  len, mpd->wbc);
1291			else if (buffer_uninit(page_bufs)) {
1292				ext4_set_bh_endio(page_bufs, inode);
1293				err = block_write_full_page_endio(page,
1294					noalloc_get_block_write,
1295					mpd->wbc, ext4_end_io_buffer_write);
1296			} else
1297				err = block_write_full_page(page,
1298					noalloc_get_block_write, mpd->wbc);
1299
1300			if (!err)
1301				mpd->pages_written++;
1302			/*
1303			 * In error case, we have to continue because
1304			 * remaining pages are still locked
1305			 */
1306			if (ret == 0)
1307				ret = err;
1308		}
1309		pagevec_release(&pvec);
1310	}
1311	ext4_io_submit(&io_submit);
1312	return ret;
1313}
1314
1315static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
 
1316{
1317	int nr_pages, i;
1318	pgoff_t index, end;
1319	struct pagevec pvec;
1320	struct inode *inode = mpd->inode;
1321	struct address_space *mapping = inode->i_mapping;
1322
 
 
 
 
1323	index = mpd->first_page;
1324	end   = mpd->next_page - 1;
 
 
 
 
 
 
 
 
1325	while (index <= end) {
1326		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1327		if (nr_pages == 0)
1328			break;
1329		for (i = 0; i < nr_pages; i++) {
1330			struct page *page = pvec.pages[i];
1331			if (page->index > end)
1332				break;
1333			BUG_ON(!PageLocked(page));
1334			BUG_ON(PageWriteback(page));
1335			block_invalidatepage(page, 0);
1336			ClearPageUptodate(page);
 
 
 
 
1337			unlock_page(page);
1338		}
1339		index = pvec.pages[nr_pages - 1]->index + 1;
1340		pagevec_release(&pvec);
1341	}
1342	return;
1343}
1344
1345static void ext4_print_free_blocks(struct inode *inode)
1346{
1347	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1348	printk(KERN_CRIT "Total free blocks count %lld\n",
1349	       ext4_count_free_blocks(inode->i_sb));
1350	printk(KERN_CRIT "Free/Dirty block details\n");
1351	printk(KERN_CRIT "free_blocks=%lld\n",
1352	       (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
1353	printk(KERN_CRIT "dirty_blocks=%lld\n",
1354	       (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1355	printk(KERN_CRIT "Block reservation details\n");
1356	printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
1357	       EXT4_I(inode)->i_reserved_data_blocks);
1358	printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
1359	       EXT4_I(inode)->i_reserved_meta_blocks);
 
 
 
 
1360	return;
1361}
1362
 
 
 
 
 
1363/*
1364 * mpage_da_map_and_submit - go through given space, map them
1365 *       if necessary, and then submit them for I/O
1366 *
1367 * @mpd - bh describing space
1368 *
1369 * The function skips space we know is already mapped to disk blocks.
1370 *
1371 */
1372static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
1373{
1374	int err, blks, get_blocks_flags;
1375	struct ext4_map_blocks map, *mapp = NULL;
1376	sector_t next = mpd->b_blocknr;
1377	unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1378	loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1379	handle_t *handle = NULL;
1380
1381	/*
1382	 * If the blocks are mapped already, or we couldn't accumulate
1383	 * any blocks, then proceed immediately to the submission stage.
1384	 */
1385	if ((mpd->b_size == 0) ||
1386	    ((mpd->b_state  & (1 << BH_Mapped)) &&
1387	     !(mpd->b_state & (1 << BH_Delay)) &&
1388	     !(mpd->b_state & (1 << BH_Unwritten))))
1389		goto submit_io;
1390
1391	handle = ext4_journal_current_handle();
1392	BUG_ON(!handle);
1393
1394	/*
1395	 * Call ext4_map_blocks() to allocate any delayed allocation
1396	 * blocks, or to convert an uninitialized extent to be
1397	 * initialized (in the case where we have written into
1398	 * one or more preallocated blocks).
1399	 *
1400	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1401	 * indicate that we are on the delayed allocation path.  This
1402	 * affects functions in many different parts of the allocation
1403	 * call path.  This flag exists primarily because we don't
1404	 * want to change *many* call functions, so ext4_map_blocks()
1405	 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1406	 * inode's allocation semaphore is taken.
1407	 *
1408	 * If the blocks in questions were delalloc blocks, set
1409	 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1410	 * variables are updated after the blocks have been allocated.
1411	 */
1412	map.m_lblk = next;
1413	map.m_len = max_blocks;
1414	get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
1415	if (ext4_should_dioread_nolock(mpd->inode))
1416		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
1417	if (mpd->b_state & (1 << BH_Delay))
1418		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1419
1420	blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
1421	if (blks < 0) {
1422		struct super_block *sb = mpd->inode->i_sb;
1423
1424		err = blks;
1425		/*
1426		 * If get block returns EAGAIN or ENOSPC and there
1427		 * appears to be free blocks we will just let
1428		 * mpage_da_submit_io() unlock all of the pages.
1429		 */
1430		if (err == -EAGAIN)
1431			goto submit_io;
1432
1433		if (err == -ENOSPC &&
1434		    ext4_count_free_blocks(sb)) {
1435			mpd->retval = err;
1436			goto submit_io;
 
 
1437		}
1438
1439		/*
1440		 * get block failure will cause us to loop in
1441		 * writepages, because a_ops->writepage won't be able
1442		 * to make progress. The page will be redirtied by
1443		 * writepage and writepages will again try to write
1444		 * the same.
1445		 */
1446		if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1447			ext4_msg(sb, KERN_CRIT,
1448				 "delayed block allocation failed for inode %lu "
1449				 "at logical offset %llu with max blocks %zd "
1450				 "with error %d", mpd->inode->i_ino,
1451				 (unsigned long long) next,
1452				 mpd->b_size >> mpd->inode->i_blkbits, err);
1453			ext4_msg(sb, KERN_CRIT,
1454				"This should not happen!! Data will be lost\n");
1455			if (err == -ENOSPC)
1456				ext4_print_free_blocks(mpd->inode);
1457		}
1458		/* invalidate all the pages */
1459		ext4_da_block_invalidatepages(mpd);
1460
1461		/* Mark this page range as having been completed */
1462		mpd->io_done = 1;
1463		return;
1464	}
1465	BUG_ON(blks == 0);
1466
1467	mapp = &map;
1468	if (map.m_flags & EXT4_MAP_NEW) {
1469		struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1470		int i;
1471
1472		for (i = 0; i < map.m_len; i++)
1473			unmap_underlying_metadata(bdev, map.m_pblk + i);
1474	}
 
 
 
 
 
 
 
 
1475
1476	if (ext4_should_order_data(mpd->inode)) {
1477		err = ext4_jbd2_file_inode(handle, mpd->inode);
1478		if (err)
1479			/* This only happens if the journal is aborted */
1480			return;
1481	}
1482
1483	/*
1484	 * Update on-disk size along with block allocation.
 
1485	 */
1486	disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1487	if (disksize > i_size_read(mpd->inode))
1488		disksize = i_size_read(mpd->inode);
1489	if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1490		ext4_update_i_disksize(mpd->inode, disksize);
1491		err = ext4_mark_inode_dirty(handle, mpd->inode);
1492		if (err)
1493			ext4_error(mpd->inode->i_sb,
1494				   "Failed to mark inode %lu dirty",
1495				   mpd->inode->i_ino);
1496	}
1497
1498submit_io:
1499	mpage_da_submit_io(mpd, mapp);
1500	mpd->io_done = 1;
1501}
1502
1503#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1504		(1 << BH_Delay) | (1 << BH_Unwritten))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1505
1506/*
1507 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1508 *
1509 * @mpd->lbh - extent of blocks
1510 * @logical - logical number of the block in the file
1511 * @bh - bh of the block (used to access block's state)
1512 *
1513 * the function is used to collect contig. blocks in same state
1514 */
1515static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1516				   sector_t logical, size_t b_size,
1517				   unsigned long b_state)
1518{
1519	sector_t next;
1520	int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
1521
1522	/*
1523	 * XXX Don't go larger than mballoc is willing to allocate
1524	 * This is a stopgap solution.  We eventually need to fold
1525	 * mpage_da_submit_io() into this function and then call
1526	 * ext4_map_blocks() multiple times in a loop
1527	 */
1528	if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1529		goto flush_it;
1530
1531	/* check if thereserved journal credits might overflow */
1532	if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
1533		if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1534			/*
1535			 * With non-extent format we are limited by the journal
1536			 * credit available.  Total credit needed to insert
1537			 * nrblocks contiguous blocks is dependent on the
1538			 * nrblocks.  So limit nrblocks.
1539			 */
1540			goto flush_it;
1541		} else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1542				EXT4_MAX_TRANS_DATA) {
1543			/*
1544			 * Adding the new buffer_head would make it cross the
1545			 * allowed limit for which we have journal credit
1546			 * reserved. So limit the new bh->b_size
1547			 */
1548			b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1549						mpd->inode->i_blkbits;
1550			/* we will do mpage_da_submit_io in the next loop */
1551		}
1552	}
1553	/*
1554	 * First block in the extent
1555	 */
1556	if (mpd->b_size == 0) {
1557		mpd->b_blocknr = logical;
1558		mpd->b_size = b_size;
1559		mpd->b_state = b_state & BH_FLAGS;
1560		return;
1561	}
1562
1563	next = mpd->b_blocknr + nrblocks;
1564	/*
1565	 * Can we merge the block to our big extent?
1566	 */
1567	if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1568		mpd->b_size += b_size;
1569		return;
1570	}
1571
1572flush_it:
1573	/*
1574	 * We couldn't merge the block to our extent, so we
1575	 * need to flush current  extent and start new one
1576	 */
1577	mpage_da_map_and_submit(mpd);
1578	return;
1579}
1580
1581static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1582{
1583	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1584}
1585
1586/*
1587 * This is a special get_blocks_t callback which is used by
1588 * ext4_da_write_begin().  It will either return mapped block or
1589 * reserve space for a single block.
1590 *
1591 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1592 * We also have b_blocknr = -1 and b_bdev initialized properly
1593 *
1594 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1595 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1596 * initialized properly.
1597 */
1598static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1599				  struct buffer_head *bh, int create)
1600{
1601	struct ext4_map_blocks map;
1602	int ret = 0;
1603	sector_t invalid_block = ~((sector_t) 0xffff);
1604
1605	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1606		invalid_block = ~0;
1607
1608	BUG_ON(create == 0);
1609	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1610
1611	map.m_lblk = iblock;
1612	map.m_len = 1;
1613
1614	/*
1615	 * first, we need to know whether the block is allocated already
1616	 * preallocated blocks are unmapped but should treated
1617	 * the same as allocated blocks.
1618	 */
1619	ret = ext4_map_blocks(NULL, inode, &map, 0);
1620	if (ret < 0)
1621		return ret;
1622	if (ret == 0) {
1623		if (buffer_delay(bh))
1624			return 0; /* Not sure this could or should happen */
1625		/*
1626		 * XXX: __block_write_begin() unmaps passed block, is it OK?
1627		 */
1628		ret = ext4_da_reserve_space(inode, iblock);
1629		if (ret)
1630			/* not enough space to reserve */
1631			return ret;
1632
1633		map_bh(bh, inode->i_sb, invalid_block);
1634		set_buffer_new(bh);
1635		set_buffer_delay(bh);
1636		return 0;
1637	}
1638
1639	map_bh(bh, inode->i_sb, map.m_pblk);
1640	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1641
1642	if (buffer_unwritten(bh)) {
1643		/* A delayed write to unwritten bh should be marked
1644		 * new and mapped.  Mapped ensures that we don't do
1645		 * get_block multiple times when we write to the same
1646		 * offset and new ensures that we do proper zero out
1647		 * for partial write.
1648		 */
1649		set_buffer_new(bh);
1650		set_buffer_mapped(bh);
1651	}
1652	return 0;
1653}
1654
1655/*
1656 * This function is used as a standard get_block_t calback function
1657 * when there is no desire to allocate any blocks.  It is used as a
1658 * callback function for block_write_begin() and block_write_full_page().
1659 * These functions should only try to map a single block at a time.
1660 *
1661 * Since this function doesn't do block allocations even if the caller
1662 * requests it by passing in create=1, it is critically important that
1663 * any caller checks to make sure that any buffer heads are returned
1664 * by this function are either all already mapped or marked for
1665 * delayed allocation before calling  block_write_full_page().  Otherwise,
1666 * b_blocknr could be left unitialized, and the page write functions will
1667 * be taken by surprise.
1668 */
1669static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
1670				   struct buffer_head *bh_result, int create)
1671{
1672	BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1673	return _ext4_get_block(inode, iblock, bh_result, 0);
1674}
1675
1676static int bget_one(handle_t *handle, struct buffer_head *bh)
1677{
1678	get_bh(bh);
1679	return 0;
1680}
1681
1682static int bput_one(handle_t *handle, struct buffer_head *bh)
1683{
1684	put_bh(bh);
1685	return 0;
1686}
1687
1688static int __ext4_journalled_writepage(struct page *page,
1689				       unsigned int len)
1690{
1691	struct address_space *mapping = page->mapping;
1692	struct inode *inode = mapping->host;
1693	struct buffer_head *page_bufs;
1694	handle_t *handle = NULL;
1695	int ret = 0;
1696	int err;
 
1697
1698	ClearPageChecked(page);
1699	page_bufs = page_buffers(page);
1700	BUG_ON(!page_bufs);
1701	walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
1702	/* As soon as we unlock the page, it can go away, but we have
1703	 * references to buffers so we are safe */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1704	unlock_page(page);
1705
1706	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
 
1707	if (IS_ERR(handle)) {
1708		ret = PTR_ERR(handle);
 
 
 
 
 
 
 
 
 
 
 
1709		goto out;
1710	}
1711
1712	BUG_ON(!ext4_handle_valid(handle));
 
 
1713
1714	ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1715				do_journal_get_write_access);
1716
1717	err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
1718				write_end_fn);
 
 
 
 
 
1719	if (ret == 0)
1720		ret = err;
1721	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1722	err = ext4_journal_stop(handle);
1723	if (!ret)
1724		ret = err;
1725
1726	walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
 
 
1727	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1728out:
 
 
 
1729	return ret;
1730}
1731
1732static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
1733static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
1734
1735/*
1736 * Note that we don't need to start a transaction unless we're journaling data
1737 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1738 * need to file the inode to the transaction's list in ordered mode because if
1739 * we are writing back data added by write(), the inode is already there and if
1740 * we are writing back data modified via mmap(), no one guarantees in which
1741 * transaction the data will hit the disk. In case we are journaling data, we
1742 * cannot start transaction directly because transaction start ranks above page
1743 * lock so we have to do some magic.
1744 *
1745 * This function can get called via...
1746 *   - ext4_da_writepages after taking page lock (have journal handle)
1747 *   - journal_submit_inode_data_buffers (no journal handle)
1748 *   - shrink_page_list via pdflush (no journal handle)
1749 *   - grab_page_cache when doing write_begin (have journal handle)
1750 *
1751 * We don't do any block allocation in this function. If we have page with
1752 * multiple blocks we need to write those buffer_heads that are mapped. This
1753 * is important for mmaped based write. So if we do with blocksize 1K
1754 * truncate(f, 1024);
1755 * a = mmap(f, 0, 4096);
1756 * a[0] = 'a';
1757 * truncate(f, 4096);
1758 * we have in the page first buffer_head mapped via page_mkwrite call back
1759 * but other bufer_heads would be unmapped but dirty(dirty done via the
1760 * do_wp_page). So writepage should write the first block. If we modify
1761 * the mmap area beyond 1024 we will again get a page_fault and the
1762 * page_mkwrite callback will do the block allocation and mark the
1763 * buffer_heads mapped.
1764 *
1765 * We redirty the page if we have any buffer_heads that is either delay or
1766 * unwritten in the page.
1767 *
1768 * We can get recursively called as show below.
1769 *
1770 *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1771 *		ext4_writepage()
1772 *
1773 * But since we don't do any block allocation we should not deadlock.
1774 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1775 */
1776static int ext4_writepage(struct page *page,
1777			  struct writeback_control *wbc)
1778{
1779	int ret = 0, commit_write = 0;
1780	loff_t size;
1781	unsigned int len;
1782	struct buffer_head *page_bufs = NULL;
1783	struct inode *inode = page->mapping->host;
 
 
1784
1785	trace_ext4_writepage(page);
1786	size = i_size_read(inode);
1787	if (page->index == size >> PAGE_CACHE_SHIFT)
1788		len = size & ~PAGE_CACHE_MASK;
1789	else
1790		len = PAGE_CACHE_SIZE;
1791
 
1792	/*
1793	 * If the page does not have buffers (for whatever reason),
1794	 * try to create them using __block_write_begin.  If this
1795	 * fails, redirty the page and move on.
1796	 */
1797	if (!page_has_buffers(page)) {
1798		if (__block_write_begin(page, 0, len,
1799					noalloc_get_block_write)) {
1800		redirty_page:
1801			redirty_page_for_writepage(wbc, page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1802			unlock_page(page);
1803			return 0;
1804		}
1805		commit_write = 1;
1806	}
1807	page_bufs = page_buffers(page);
1808	if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1809			      ext4_bh_delay_or_unwritten)) {
1810		/*
1811		 * We don't want to do block allocation, so redirty
1812		 * the page and return.  We may reach here when we do
1813		 * a journal commit via journal_submit_inode_data_buffers.
1814		 * We can also reach here via shrink_page_list
1815		 */
1816		goto redirty_page;
1817	}
1818	if (commit_write)
1819		/* now mark the buffer_heads as dirty and uptodate */
1820		block_commit_write(page, 0, len);
1821
1822	if (PageChecked(page) && ext4_should_journal_data(inode))
1823		/*
1824		 * It's mmapped pagecache.  Add buffers and journal it.  There
1825		 * doesn't seem much point in redirtying the page here.
1826		 */
1827		return __ext4_journalled_writepage(page, len);
1828
1829	if (buffer_uninit(page_bufs)) {
1830		ext4_set_bh_endio(page_bufs, inode);
1831		ret = block_write_full_page_endio(page, noalloc_get_block_write,
1832					    wbc, ext4_end_io_buffer_write);
1833	} else
1834		ret = block_write_full_page(page, noalloc_get_block_write,
1835					    wbc);
1836
 
 
 
1837	return ret;
1838}
1839
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1840/*
1841 * This is called via ext4_da_writepages() to
1842 * calculate the total number of credits to reserve to fit
1843 * a single extent allocation into a single transaction,
1844 * ext4_da_writpeages() will loop calling this before
1845 * the block allocation.
1846 */
 
1847
1848static int ext4_da_writepages_trans_blocks(struct inode *inode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1849{
1850	int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
1851
1852	/*
1853	 * With non-extent format the journal credit needed to
1854	 * insert nrblocks contiguous block is dependent on
1855	 * number of contiguous block. So we will limit
1856	 * number of contiguous block to a sane value
1857	 */
1858	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
1859	    (max_blocks > EXT4_MAX_TRANS_DATA))
1860		max_blocks = EXT4_MAX_TRANS_DATA;
 
 
 
 
 
 
 
 
 
 
 
1861
1862	return ext4_chunk_trans_blocks(inode, max_blocks);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1863}
1864
1865/*
1866 * write_cache_pages_da - walk the list of dirty pages of the given
1867 * address space and accumulate pages that need writing, and call
1868 * mpage_da_map_and_submit to map a single contiguous memory region
1869 * and then write them.
 
 
 
 
 
 
 
 
1870 */
1871static int write_cache_pages_da(struct address_space *mapping,
1872				struct writeback_control *wbc,
1873				struct mpage_da_data *mpd,
1874				pgoff_t *done_index)
1875{
1876	struct buffer_head	*bh, *head;
1877	struct inode		*inode = mapping->host;
1878	struct pagevec		pvec;
1879	unsigned int		nr_pages;
1880	sector_t		logical;
1881	pgoff_t			index, end;
1882	long			nr_to_write = wbc->nr_to_write;
1883	int			i, tag, ret = 0;
 
 
 
 
 
 
1884
1885	memset(mpd, 0, sizeof(struct mpage_da_data));
1886	mpd->wbc = wbc;
1887	mpd->inode = inode;
1888	pagevec_init(&pvec, 0);
1889	index = wbc->range_start >> PAGE_CACHE_SHIFT;
1890	end = wbc->range_end >> PAGE_CACHE_SHIFT;
 
 
 
 
 
1891
1892	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1893		tag = PAGECACHE_TAG_TOWRITE;
1894	else
1895		tag = PAGECACHE_TAG_DIRTY;
1896
1897	*done_index = index;
 
 
1898	while (index <= end) {
1899		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1900			      min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1901		if (nr_pages == 0)
1902			return 0;
1903
1904		for (i = 0; i < nr_pages; i++) {
1905			struct page *page = pvec.pages[i];
1906
1907			/*
1908			 * At this point, the page may be truncated or
1909			 * invalidated (changing page->mapping to NULL), or
1910			 * even swizzled back from swapper_space to tmpfs file
1911			 * mapping. However, page->index will not change
1912			 * because we have a reference on the page.
1913			 */
1914			if (page->index > end)
1915				goto out;
1916
1917			*done_index = page->index + 1;
1918
1919			/*
1920			 * If we can't merge this page, and we have
1921			 * accumulated an contiguous region, write it
 
 
 
 
1922			 */
1923			if ((mpd->next_page != page->index) &&
1924			    (mpd->next_page != mpd->first_page)) {
1925				mpage_da_map_and_submit(mpd);
1926				goto ret_extent_tail;
1927			}
1928
1929			lock_page(page);
 
 
1930
 
1931			/*
1932			 * If the page is no longer dirty, or its
1933			 * mapping no longer corresponds to inode we
1934			 * are writing (which means it has been
1935			 * truncated or invalidated), or the page is
1936			 * already under writeback and we are not
1937			 * doing a data integrity writeback, skip the page
1938			 */
1939			if (!PageDirty(page) ||
1940			    (PageWriteback(page) &&
1941			     (wbc->sync_mode == WB_SYNC_NONE)) ||
1942			    unlikely(page->mapping != mapping)) {
1943				unlock_page(page);
1944				continue;
1945			}
1946
1947			wait_on_page_writeback(page);
1948			BUG_ON(PageWriteback(page));
1949
1950			if (mpd->next_page != page->index)
1951				mpd->first_page = page->index;
1952			mpd->next_page = page->index + 1;
1953			logical = (sector_t) page->index <<
1954				(PAGE_CACHE_SHIFT - inode->i_blkbits);
1955
1956			if (!page_has_buffers(page)) {
1957				mpage_add_bh_to_extent(mpd, logical,
1958						       PAGE_CACHE_SIZE,
1959						       (1 << BH_Dirty) | (1 << BH_Uptodate));
1960				if (mpd->io_done)
1961					goto ret_extent_tail;
1962			} else {
1963				/*
1964				 * Page with regular buffer heads,
1965				 * just add all dirty ones
1966				 */
1967				head = page_buffers(page);
1968				bh = head;
1969				do {
1970					BUG_ON(buffer_locked(bh));
1971					/*
1972					 * We need to try to allocate
1973					 * unmapped blocks in the same page.
1974					 * Otherwise we won't make progress
1975					 * with the page in ext4_writepage
1976					 */
1977					if (ext4_bh_delay_or_unwritten(NULL, bh)) {
1978						mpage_add_bh_to_extent(mpd, logical,
1979								       bh->b_size,
1980								       bh->b_state);
1981						if (mpd->io_done)
1982							goto ret_extent_tail;
1983					} else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
1984						/*
1985						 * mapped dirty buffer. We need
1986						 * to update the b_state
1987						 * because we look at b_state
1988						 * in mpage_da_map_blocks.  We
1989						 * don't update b_size because
1990						 * if we find an unmapped
1991						 * buffer_head later we need to
1992						 * use the b_state flag of that
1993						 * buffer_head.
1994						 */
1995						if (mpd->b_size == 0)
1996							mpd->b_state = bh->b_state & BH_FLAGS;
1997					}
1998					logical++;
1999				} while ((bh = bh->b_this_page) != head);
2000			}
2001
2002			if (nr_to_write > 0) {
2003				nr_to_write--;
2004				if (nr_to_write == 0 &&
2005				    wbc->sync_mode == WB_SYNC_NONE)
2006					/*
2007					 * We stop writing back only if we are
2008					 * not doing integrity sync. In case of
2009					 * integrity sync we have to keep going
2010					 * because someone may be concurrently
2011					 * dirtying pages, and we might have
2012					 * synced a lot of newly appeared dirty
2013					 * pages, but have not synced all of the
2014					 * old dirty pages.
2015					 */
2016					goto out;
2017			}
2018		}
2019		pagevec_release(&pvec);
2020		cond_resched();
2021	}
2022	return 0;
2023ret_extent_tail:
2024	ret = MPAGE_DA_EXTENT_TAIL;
2025out:
2026	pagevec_release(&pvec);
2027	cond_resched();
2028	return ret;
2029}
2030
 
 
 
 
 
 
 
 
2031
2032static int ext4_da_writepages(struct address_space *mapping,
2033			      struct writeback_control *wbc)
2034{
2035	pgoff_t	index;
 
2036	int range_whole = 0;
 
2037	handle_t *handle = NULL;
2038	struct mpage_da_data mpd;
2039	struct inode *inode = mapping->host;
2040	int pages_written = 0;
2041	unsigned int max_pages;
2042	int range_cyclic, cycled = 1, io_done = 0;
2043	int needed_blocks, ret = 0;
2044	long desired_nr_to_write, nr_to_writebump = 0;
2045	loff_t range_start = wbc->range_start;
2046	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2047	pgoff_t done_index = 0;
2048	pgoff_t end;
2049
2050	trace_ext4_da_writepages(inode, wbc);
 
 
 
 
 
 
 
 
2051
2052	/*
2053	 * No pages to write? This is mainly a kludge to avoid starting
2054	 * a transaction for special inodes like journal inode on last iput()
2055	 * because that could violate lock ordering on umount
2056	 */
2057	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2058		return 0;
 
 
 
 
 
 
 
 
 
2059
2060	/*
2061	 * If the filesystem has aborted, it is read-only, so return
2062	 * right away instead of dumping stack traces later on that
2063	 * will obscure the real source of the problem.  We test
2064	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2065	 * the latter could be true if the filesystem is mounted
2066	 * read-only, and in that case, ext4_da_writepages should
2067	 * *never* be called, so if that ever happens, we would want
2068	 * the stack trace.
2069	 */
2070	if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2071		return -EROFS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2072
2073	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2074		range_whole = 1;
2075
2076	range_cyclic = wbc->range_cyclic;
2077	if (wbc->range_cyclic) {
2078		index = mapping->writeback_index;
2079		if (index)
2080			cycled = 0;
2081		wbc->range_start = index << PAGE_CACHE_SHIFT;
2082		wbc->range_end  = LLONG_MAX;
2083		wbc->range_cyclic = 0;
2084		end = -1;
2085	} else {
2086		index = wbc->range_start >> PAGE_CACHE_SHIFT;
2087		end = wbc->range_end >> PAGE_CACHE_SHIFT;
2088	}
2089
2090	/*
2091	 * This works around two forms of stupidity.  The first is in
2092	 * the writeback code, which caps the maximum number of pages
2093	 * written to be 1024 pages.  This is wrong on multiple
2094	 * levels; different architectues have a different page size,
2095	 * which changes the maximum amount of data which gets
2096	 * written.  Secondly, 4 megabytes is way too small.  XFS
2097	 * forces this value to be 16 megabytes by multiplying
2098	 * nr_to_write parameter by four, and then relies on its
2099	 * allocator to allocate larger extents to make them
2100	 * contiguous.  Unfortunately this brings us to the second
2101	 * stupidity, which is that ext4's mballoc code only allocates
2102	 * at most 2048 blocks.  So we force contiguous writes up to
2103	 * the number of dirty blocks in the inode, or
2104	 * sbi->max_writeback_mb_bump whichever is smaller.
2105	 */
2106	max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2107	if (!range_cyclic && range_whole) {
2108		if (wbc->nr_to_write == LONG_MAX)
2109			desired_nr_to_write = wbc->nr_to_write;
2110		else
2111			desired_nr_to_write = wbc->nr_to_write * 8;
2112	} else
2113		desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2114							   max_pages);
2115	if (desired_nr_to_write > max_pages)
2116		desired_nr_to_write = max_pages;
2117
2118	if (wbc->nr_to_write < desired_nr_to_write) {
2119		nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2120		wbc->nr_to_write = desired_nr_to_write;
2121	}
2122
 
 
 
2123retry:
2124	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2125		tag_pages_for_writeback(mapping, index, end);
2126
2127	while (!ret && wbc->nr_to_write > 0) {
 
 
 
 
 
 
 
2128
2129		/*
2130		 * we  insert one extent at a time. So we need
2131		 * credit needed for single extent allocation.
2132		 * journalled mode is currently not supported
2133		 * by delalloc
 
2134		 */
2135		BUG_ON(ext4_should_journal_data(inode));
2136		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2137
2138		/* start a new transaction*/
2139		handle = ext4_journal_start(inode, needed_blocks);
 
2140		if (IS_ERR(handle)) {
2141			ret = PTR_ERR(handle);
2142			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2143			       "%ld pages, ino %lu; err %d", __func__,
2144				wbc->nr_to_write, inode->i_ino, ret);
2145			goto out_writepages;
 
 
2146		}
2147
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2148		/*
2149		 * Now call write_cache_pages_da() to find the next
2150		 * contiguous region of logical blocks that need
2151		 * blocks to be allocated by ext4 and submit them.
2152		 */
2153		ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
2154		/*
2155		 * If we have a contiguous extent of pages and we
2156		 * haven't done the I/O yet, map the blocks and submit
2157		 * them for I/O.
2158		 */
2159		if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2160			mpage_da_map_and_submit(&mpd);
2161			ret = MPAGE_DA_EXTENT_TAIL;
2162		}
2163		trace_ext4_da_write_pages(inode, &mpd);
2164		wbc->nr_to_write -= mpd.pages_written;
2165
2166		ext4_journal_stop(handle);
 
 
 
 
 
 
 
 
 
 
 
 
2167
2168		if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2169			/* commit the transaction which would
 
2170			 * free blocks released in the transaction
2171			 * and try again
2172			 */
2173			jbd2_journal_force_commit_nested(sbi->s_journal);
2174			ret = 0;
2175		} else if (ret == MPAGE_DA_EXTENT_TAIL) {
2176			/*
2177			 * got one extent now try with
2178			 * rest of the pages
2179			 */
2180			pages_written += mpd.pages_written;
2181			ret = 0;
2182			io_done = 1;
2183		} else if (wbc->nr_to_write)
2184			/*
2185			 * There is no more writeout needed
2186			 * or we requested for a noblocking writeout
2187			 * and we found the device congested
2188			 */
2189			break;
2190	}
2191	if (!io_done && !cycled) {
 
2192		cycled = 1;
2193		index = 0;
2194		wbc->range_start = index << PAGE_CACHE_SHIFT;
2195		wbc->range_end  = mapping->writeback_index - 1;
2196		goto retry;
2197	}
2198
2199	/* Update index */
2200	wbc->range_cyclic = range_cyclic;
2201	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2202		/*
2203		 * set the writeback_index so that range_cyclic
2204		 * mode will write it back later
2205		 */
2206		mapping->writeback_index = done_index;
2207
2208out_writepages:
2209	wbc->nr_to_write -= nr_to_writebump;
2210	wbc->range_start = range_start;
2211	trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2212	return ret;
2213}
2214
2215#define FALL_BACK_TO_NONDELALLOC 1
2216static int ext4_nonda_switch(struct super_block *sb)
2217{
2218	s64 free_blocks, dirty_blocks;
2219	struct ext4_sb_info *sbi = EXT4_SB(sb);
2220
2221	/*
2222	 * switch to non delalloc mode if we are running low
2223	 * on free block. The free block accounting via percpu
2224	 * counters can get slightly wrong with percpu_counter_batch getting
2225	 * accumulated on each CPU without updating global counters
2226	 * Delalloc need an accurate free block accounting. So switch
2227	 * to non delalloc when we are near to error range.
2228	 */
2229	free_blocks  = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2230	dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2231	if (2 * free_blocks < 3 * dirty_blocks ||
2232		free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
 
 
 
 
 
 
 
 
2233		/*
2234		 * free block count is less than 150% of dirty blocks
2235		 * or free blocks is less than watermark
2236		 */
2237		return 1;
2238	}
2239	/*
2240	 * Even if we don't switch but are nearing capacity,
2241	 * start pushing delalloc when 1/2 of free blocks are dirty.
2242	 */
2243	if (free_blocks < 2 * dirty_blocks)
2244		writeback_inodes_sb_if_idle(sb);
2245
2246	return 0;
2247}
2248
 
 
 
 
 
 
 
 
 
 
 
 
 
2249static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2250			       loff_t pos, unsigned len, unsigned flags,
2251			       struct page **pagep, void **fsdata)
2252{
2253	int ret, retries = 0;
2254	struct page *page;
2255	pgoff_t index;
2256	struct inode *inode = mapping->host;
2257	handle_t *handle;
2258
2259	index = pos >> PAGE_CACHE_SHIFT;
2260
2261	if (ext4_nonda_switch(inode->i_sb)) {
 
2262		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2263		return ext4_write_begin(file, mapping, pos,
2264					len, flags, pagep, fsdata);
2265	}
2266	*fsdata = (void *)0;
2267	trace_ext4_da_write_begin(inode, pos, len, flags);
2268retry:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2269	/*
2270	 * With delayed allocation, we don't log the i_disksize update
2271	 * if there is delayed block allocation. But we still need
2272	 * to journalling the i_disksize update if writes to the end
2273	 * of file which has an already mapped buffer.
2274	 */
2275	handle = ext4_journal_start(inode, 1);
 
 
2276	if (IS_ERR(handle)) {
2277		ret = PTR_ERR(handle);
2278		goto out;
2279	}
2280	/* We cannot recurse into the filesystem as the transaction is already
2281	 * started */
2282	flags |= AOP_FLAG_NOFS;
2283
2284	page = grab_cache_page_write_begin(mapping, index, flags);
2285	if (!page) {
 
 
 
2286		ext4_journal_stop(handle);
2287		ret = -ENOMEM;
2288		goto out;
2289	}
2290	*pagep = page;
 
2291
 
 
 
 
2292	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
 
2293	if (ret < 0) {
2294		unlock_page(page);
2295		ext4_journal_stop(handle);
2296		page_cache_release(page);
2297		/*
2298		 * block_write_begin may have instantiated a few blocks
2299		 * outside i_size.  Trim these off again. Don't need
2300		 * i_size_read because we hold i_mutex.
2301		 */
2302		if (pos + len > inode->i_size)
2303			ext4_truncate_failed_write(inode);
 
 
 
 
 
 
 
2304	}
2305
2306	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2307		goto retry;
2308out:
2309	return ret;
2310}
2311
2312/*
2313 * Check if we should update i_disksize
2314 * when write to the end of file but not require block allocation
2315 */
2316static int ext4_da_should_update_i_disksize(struct page *page,
2317					    unsigned long offset)
2318{
2319	struct buffer_head *bh;
2320	struct inode *inode = page->mapping->host;
2321	unsigned int idx;
2322	int i;
2323
2324	bh = page_buffers(page);
2325	idx = offset >> inode->i_blkbits;
2326
2327	for (i = 0; i < idx; i++)
2328		bh = bh->b_this_page;
2329
2330	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2331		return 0;
2332	return 1;
2333}
2334
2335static int ext4_da_write_end(struct file *file,
2336			     struct address_space *mapping,
2337			     loff_t pos, unsigned len, unsigned copied,
2338			     struct page *page, void *fsdata)
2339{
2340	struct inode *inode = mapping->host;
2341	int ret = 0, ret2;
2342	handle_t *handle = ext4_journal_current_handle();
2343	loff_t new_i_size;
2344	unsigned long start, end;
2345	int write_mode = (int)(unsigned long)fsdata;
2346
2347	if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2348		if (ext4_should_order_data(inode)) {
2349			return ext4_ordered_write_end(file, mapping, pos,
2350					len, copied, page, fsdata);
2351		} else if (ext4_should_writeback_data(inode)) {
2352			return ext4_writeback_write_end(file, mapping, pos,
2353					len, copied, page, fsdata);
2354		} else {
2355			BUG();
2356		}
2357	}
2358
2359	trace_ext4_da_write_end(inode, pos, len, copied);
2360	start = pos & (PAGE_CACHE_SIZE - 1);
2361	end = start + copied - 1;
2362
2363	/*
2364	 * generic_write_end() will run mark_inode_dirty() if i_size
2365	 * changes.  So let's piggyback the i_disksize mark_inode_dirty
2366	 * into that.
2367	 */
2368
2369	new_i_size = pos + copied;
2370	if (new_i_size > EXT4_I(inode)->i_disksize) {
2371		if (ext4_da_should_update_i_disksize(page, end)) {
2372			down_write(&EXT4_I(inode)->i_data_sem);
2373			if (new_i_size > EXT4_I(inode)->i_disksize) {
2374				/*
2375				 * Updating i_disksize when extending file
2376				 * without needing block allocation
2377				 */
2378				if (ext4_should_order_data(inode))
2379					ret = ext4_jbd2_file_inode(handle,
2380								   inode);
2381
2382				EXT4_I(inode)->i_disksize = new_i_size;
2383			}
2384			up_write(&EXT4_I(inode)->i_data_sem);
2385			/* We need to mark inode dirty even if
2386			 * new_i_size is less that inode->i_size
2387			 * bu greater than i_disksize.(hint delalloc)
2388			 */
2389			ext4_mark_inode_dirty(handle, inode);
2390		}
2391	}
2392	ret2 = generic_write_end(file, mapping, pos, len, copied,
 
 
 
 
 
 
 
2393							page, fsdata);
 
2394	copied = ret2;
2395	if (ret2 < 0)
2396		ret = ret2;
2397	ret2 = ext4_journal_stop(handle);
2398	if (!ret)
2399		ret = ret2;
2400
2401	return ret ? ret : copied;
2402}
2403
2404static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
 
2405{
2406	/*
2407	 * Drop reserved blocks
2408	 */
2409	BUG_ON(!PageLocked(page));
2410	if (!page_has_buffers(page))
2411		goto out;
2412
2413	ext4_da_page_release_reservation(page, offset);
2414
2415out:
2416	ext4_invalidatepage(page, offset);
2417
2418	return;
2419}
2420
2421/*
2422 * Force all delayed allocation blocks to be allocated for a given inode.
2423 */
2424int ext4_alloc_da_blocks(struct inode *inode)
2425{
2426	trace_ext4_alloc_da_blocks(inode);
2427
2428	if (!EXT4_I(inode)->i_reserved_data_blocks &&
2429	    !EXT4_I(inode)->i_reserved_meta_blocks)
2430		return 0;
2431
2432	/*
2433	 * We do something simple for now.  The filemap_flush() will
2434	 * also start triggering a write of the data blocks, which is
2435	 * not strictly speaking necessary (and for users of
2436	 * laptop_mode, not even desirable).  However, to do otherwise
2437	 * would require replicating code paths in:
2438	 *
2439	 * ext4_da_writepages() ->
2440	 *    write_cache_pages() ---> (via passed in callback function)
2441	 *        __mpage_da_writepage() -->
2442	 *           mpage_add_bh_to_extent()
2443	 *           mpage_da_map_blocks()
2444	 *
2445	 * The problem is that write_cache_pages(), located in
2446	 * mm/page-writeback.c, marks pages clean in preparation for
2447	 * doing I/O, which is not desirable if we're not planning on
2448	 * doing I/O at all.
2449	 *
2450	 * We could call write_cache_pages(), and then redirty all of
2451	 * the pages by calling redirty_page_for_writepage() but that
2452	 * would be ugly in the extreme.  So instead we would need to
2453	 * replicate parts of the code in the above functions,
2454	 * simplifying them because we wouldn't actually intend to
2455	 * write out the pages, but rather only collect contiguous
2456	 * logical block extents, call the multi-block allocator, and
2457	 * then update the buffer heads with the block allocations.
2458	 *
2459	 * For now, though, we'll cheat by calling filemap_flush(),
2460	 * which will map the blocks, and start the I/O, but not
2461	 * actually wait for the I/O to complete.
2462	 */
2463	return filemap_flush(inode->i_mapping);
2464}
2465
2466/*
2467 * bmap() is special.  It gets used by applications such as lilo and by
2468 * the swapper to find the on-disk block of a specific piece of data.
2469 *
2470 * Naturally, this is dangerous if the block concerned is still in the
2471 * journal.  If somebody makes a swapfile on an ext4 data-journaling
2472 * filesystem and enables swap, then they may get a nasty shock when the
2473 * data getting swapped to that swapfile suddenly gets overwritten by
2474 * the original zero's written out previously to the journal and
2475 * awaiting writeback in the kernel's buffer cache.
2476 *
2477 * So, if we see any bmap calls here on a modified, data-journaled file,
2478 * take extra steps to flush any blocks which might be in the cache.
2479 */
2480static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2481{
2482	struct inode *inode = mapping->host;
2483	journal_t *journal;
2484	int err;
2485
 
 
 
 
 
 
2486	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2487			test_opt(inode->i_sb, DELALLOC)) {
2488		/*
2489		 * With delalloc we want to sync the file
2490		 * so that we can make sure we allocate
2491		 * blocks for file
2492		 */
2493		filemap_write_and_wait(mapping);
2494	}
2495
2496	if (EXT4_JOURNAL(inode) &&
2497	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
2498		/*
2499		 * This is a REALLY heavyweight approach, but the use of
2500		 * bmap on dirty files is expected to be extremely rare:
2501		 * only if we run lilo or swapon on a freshly made file
2502		 * do we expect this to happen.
2503		 *
2504		 * (bmap requires CAP_SYS_RAWIO so this does not
2505		 * represent an unprivileged user DOS attack --- we'd be
2506		 * in trouble if mortal users could trigger this path at
2507		 * will.)
2508		 *
2509		 * NB. EXT4_STATE_JDATA is not set on files other than
2510		 * regular files.  If somebody wants to bmap a directory
2511		 * or symlink and gets confused because the buffer
2512		 * hasn't yet been flushed to disk, they deserve
2513		 * everything they get.
2514		 */
2515
2516		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
2517		journal = EXT4_JOURNAL(inode);
2518		jbd2_journal_lock_updates(journal);
2519		err = jbd2_journal_flush(journal);
2520		jbd2_journal_unlock_updates(journal);
2521
2522		if (err)
2523			return 0;
2524	}
2525
2526	return generic_block_bmap(mapping, block, ext4_get_block);
2527}
2528
2529static int ext4_readpage(struct file *file, struct page *page)
2530{
 
 
 
2531	trace_ext4_readpage(page);
2532	return mpage_readpage(page, ext4_get_block);
 
 
 
 
 
 
 
2533}
2534
2535static int
2536ext4_readpages(struct file *file, struct address_space *mapping,
2537		struct list_head *pages, unsigned nr_pages)
2538{
2539	return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
 
 
 
 
 
 
2540}
2541
2542static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
 
2543{
2544	struct buffer_head *head, *bh;
2545	unsigned int curr_off = 0;
2546
2547	if (!page_has_buffers(page))
2548		return;
2549	head = bh = page_buffers(page);
2550	do {
2551		if (offset <= curr_off && test_clear_buffer_uninit(bh)
2552					&& bh->b_private) {
2553			ext4_free_io_end(bh->b_private);
2554			bh->b_private = NULL;
2555			bh->b_end_io = NULL;
2556		}
2557		curr_off = curr_off + bh->b_size;
2558		bh = bh->b_this_page;
2559	} while (bh != head);
2560}
2561
2562static void ext4_invalidatepage(struct page *page, unsigned long offset)
 
 
2563{
2564	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2565
2566	trace_ext4_invalidatepage(page, offset);
2567
2568	/*
2569	 * free any io_end structure allocated for buffers to be discarded
2570	 */
2571	if (ext4_should_dioread_nolock(page->mapping->host))
2572		ext4_invalidatepage_free_endio(page, offset);
2573	/*
2574	 * If it's a full truncate we just forget about the pending dirtying
2575	 */
2576	if (offset == 0)
2577		ClearPageChecked(page);
2578
2579	if (journal)
2580		jbd2_journal_invalidatepage(journal, page, offset);
2581	else
2582		block_invalidatepage(page, offset);
 
 
 
 
 
2583}
2584
2585static int ext4_releasepage(struct page *page, gfp_t wait)
2586{
2587	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2588
2589	trace_ext4_releasepage(page);
2590
2591	WARN_ON(PageChecked(page));
2592	if (!page_has_buffers(page))
2593		return 0;
2594	if (journal)
2595		return jbd2_journal_try_to_free_buffers(journal, page, wait);
2596	else
2597		return try_to_free_buffers(page);
2598}
2599
2600/*
2601 * ext4_get_block used when preparing for a DIO write or buffer write.
2602 * We allocate an uinitialized extent if blocks haven't been allocated.
2603 * The extent will be converted to initialized after the IO is complete.
2604 */
2605static int ext4_get_block_write(struct inode *inode, sector_t iblock,
2606		   struct buffer_head *bh_result, int create)
2607{
2608	ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2609		   inode->i_ino, create);
2610	return _ext4_get_block(inode, iblock, bh_result,
2611			       EXT4_GET_BLOCKS_IO_CREATE_EXT);
2612}
2613
2614static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
2615			    ssize_t size, void *private, int ret,
2616			    bool is_async)
2617{
2618	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
2619        ext4_io_end_t *io_end = iocb->private;
2620	struct workqueue_struct *wq;
2621	unsigned long flags;
2622	struct ext4_inode_info *ei;
2623
2624	/* if not async direct IO or dio with 0 bytes write, just return */
2625	if (!io_end || !size)
2626		goto out;
2627
2628	ext_debug("ext4_end_io_dio(): io_end 0x%p"
2629		  "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2630 		  iocb->private, io_end->inode->i_ino, iocb, offset,
2631		  size);
2632
2633	/* if not aio dio with unwritten extents, just free io and return */
2634	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2635		ext4_free_io_end(io_end);
2636		iocb->private = NULL;
2637out:
2638		if (is_async)
2639			aio_complete(iocb, ret, 0);
2640		inode_dio_done(inode);
2641		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2642	}
2643
2644	io_end->offset = offset;
2645	io_end->size = size;
2646	if (is_async) {
2647		io_end->iocb = iocb;
2648		io_end->result = ret;
2649	}
2650	wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
2651
2652	/* Add the io_end to per-inode completed aio dio list*/
2653	ei = EXT4_I(io_end->inode);
2654	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
2655	list_add_tail(&io_end->list, &ei->i_completed_io_list);
2656	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
2657
2658	/* queue the work to convert unwritten extents to written */
2659	queue_work(wq, &io_end->work);
2660	iocb->private = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
2661
2662	/* XXX: probably should move into the real I/O completion handler */
2663	inode_dio_done(inode);
 
2664}
2665
2666static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
 
2667{
2668	ext4_io_end_t *io_end = bh->b_private;
2669	struct workqueue_struct *wq;
2670	struct inode *inode;
2671	unsigned long flags;
2672
2673	if (!test_clear_buffer_uninit(bh) || !io_end)
2674		goto out;
2675
2676	if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
2677		printk("sb umounted, discard end_io request for inode %lu\n",
2678			io_end->inode->i_ino);
2679		ext4_free_io_end(io_end);
2680		goto out;
2681	}
 
 
 
 
 
 
 
 
2682
 
 
 
 
 
2683	/*
2684	 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2685	 * but being more careful is always safe for the future change.
2686	 */
2687	inode = io_end->inode;
2688	if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
2689		io_end->flag |= EXT4_IO_END_UNWRITTEN;
2690		atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
2691	}
2692
2693	/* Add the io_end to per-inode completed io list*/
2694	spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
2695	list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
2696	spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
2697
2698	wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
2699	/* queue the work to convert unwritten extents to written */
2700	queue_work(wq, &io_end->work);
2701out:
2702	bh->b_private = NULL;
2703	bh->b_end_io = NULL;
2704	clear_buffer_uninit(bh);
2705	end_buffer_async_write(bh, uptodate);
2706}
2707
2708static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
2709{
2710	ext4_io_end_t *io_end;
2711	struct page *page = bh->b_page;
2712	loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
2713	size_t size = bh->b_size;
2714
2715retry:
2716	io_end = ext4_init_io_end(inode, GFP_ATOMIC);
2717	if (!io_end) {
2718		pr_warn_ratelimited("%s: allocation fail\n", __func__);
2719		schedule();
2720		goto retry;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2721	}
2722	io_end->offset = offset;
2723	io_end->size = size;
2724	/*
2725	 * We need to hold a reference to the page to make sure it
2726	 * doesn't get evicted before ext4_end_io_work() has a chance
2727	 * to convert the extent from written to unwritten.
2728	 */
2729	io_end->page = page;
2730	get_page(io_end->page);
2731
2732	bh->b_private = io_end;
2733	bh->b_end_io = ext4_end_io_buffer_write;
2734	return 0;
2735}
2736
2737/*
2738 * For ext4 extent files, ext4 will do direct-io write to holes,
 
 
2739 * preallocated extents, and those write extend the file, no need to
2740 * fall back to buffered IO.
2741 *
2742 * For holes, we fallocate those blocks, mark them as uninitialized
2743 * If those blocks were preallocated, we mark sure they are splited, but
2744 * still keep the range to write as uninitialized.
2745 *
2746 * The unwrritten extents will be converted to written when DIO is completed.
2747 * For async direct IO, since the IO may still pending when return, we
2748 * set up an end_io call back function, which will do the conversion
2749 * when async direct IO completed.
2750 *
2751 * If the O_DIRECT write will extend the file then add this inode to the
2752 * orphan list.  So recovery will truncate it back to the original size
2753 * if the machine crashes during the write.
2754 *
2755 */
2756static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
2757			      const struct iovec *iov, loff_t offset,
2758			      unsigned long nr_segs)
2759{
2760	struct file *file = iocb->ki_filp;
2761	struct inode *inode = file->f_mapping->host;
 
2762	ssize_t ret;
2763	size_t count = iov_length(iov, nr_segs);
2764
 
 
 
2765	loff_t final_size = offset + count;
2766	if (rw == WRITE && final_size <= inode->i_size) {
2767		/*
2768 		 * We could direct write to holes and fallocate.
2769		 *
2770 		 * Allocated blocks to fill the hole are marked as uninitialized
2771 		 * to prevent parallel buffered read to expose the stale data
2772 		 * before DIO complete the data IO.
2773		 *
2774 		 * As to previously fallocated extents, ext4 get_block
2775 		 * will just simply mark the buffer mapped but still
2776 		 * keep the extents uninitialized.
2777 		 *
2778		 * for non AIO case, we will convert those unwritten extents
2779		 * to written after return back from blockdev_direct_IO.
2780		 *
2781		 * for async DIO, the conversion needs to be defered when
2782		 * the IO is completed. The ext4 end_io callback function
2783		 * will be called to take care of the conversion work.
2784		 * Here for async case, we allocate an io_end structure to
2785		 * hook to the iocb.
2786 		 */
2787		iocb->private = NULL;
2788		EXT4_I(inode)->cur_aio_dio = NULL;
2789		if (!is_sync_kiocb(iocb)) {
2790			iocb->private = ext4_init_io_end(inode, GFP_NOFS);
2791			if (!iocb->private)
2792				return -ENOMEM;
2793			/*
2794			 * we save the io structure for current async
2795			 * direct IO, so that later ext4_map_blocks()
2796			 * could flag the io structure whether there
2797			 * is a unwritten extents needs to be converted
2798			 * when IO is completed.
2799			 */
2800			EXT4_I(inode)->cur_aio_dio = iocb->private;
2801		}
 
 
 
 
2802
2803		ret = __blockdev_direct_IO(rw, iocb, inode,
2804					 inode->i_sb->s_bdev, iov,
2805					 offset, nr_segs,
2806					 ext4_get_block_write,
2807					 ext4_end_io_dio,
2808					 NULL,
2809					 DIO_LOCKING | DIO_SKIP_HOLES);
2810		if (iocb->private)
2811			EXT4_I(inode)->cur_aio_dio = NULL;
2812		/*
2813		 * The io_end structure takes a reference to the inode,
2814		 * that structure needs to be destroyed and the
2815		 * reference to the inode need to be dropped, when IO is
2816		 * complete, even with 0 byte write, or failed.
2817		 *
2818		 * In the successful AIO DIO case, the io_end structure will be
2819		 * desctroyed and the reference to the inode will be dropped
2820		 * after the end_io call back function is called.
2821		 *
2822		 * In the case there is 0 byte write, or error case, since
2823		 * VFS direct IO won't invoke the end_io call back function,
2824		 * we need to free the end_io structure here.
2825		 */
2826		if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
2827			ext4_free_io_end(iocb->private);
2828			iocb->private = NULL;
2829		} else if (ret > 0 && ext4_test_inode_state(inode,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2830						EXT4_STATE_DIO_UNWRITTEN)) {
2831			int err;
2832			/*
2833			 * for non AIO case, since the IO is already
2834			 * completed, we could do the conversion right here
2835			 */
2836			err = ext4_convert_unwritten_extents(inode,
2837							     offset, ret);
2838			if (err < 0)
2839				ret = err;
2840			ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2841		}
2842		return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2843	}
 
 
 
 
 
 
 
 
 
 
2844
2845	/* for write the the end of file case, we fall back to old way */
2846	return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
 
 
 
 
 
 
 
 
 
 
 
 
 
2847}
2848
2849static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2850			      const struct iovec *iov, loff_t offset,
2851			      unsigned long nr_segs)
2852{
2853	struct file *file = iocb->ki_filp;
2854	struct inode *inode = file->f_mapping->host;
 
 
2855	ssize_t ret;
2856
2857	trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
2858	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
2859		ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2860	else
2861		ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
2862	trace_ext4_direct_IO_exit(inode, offset,
2863				iov_length(iov, nr_segs), rw, ret);
2864	return ret;
2865}
2866
2867/*
2868 * Pages can be marked dirty completely asynchronously from ext4's journalling
2869 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
2870 * much here because ->set_page_dirty is called under VFS locks.  The page is
2871 * not necessarily locked.
2872 *
2873 * We cannot just dirty the page and leave attached buffers clean, because the
2874 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
2875 * or jbddirty because all the journalling code will explode.
2876 *
2877 * So what we do is to mark the page "pending dirty" and next time writepage
2878 * is called, propagate that into the buffers appropriately.
2879 */
2880static int ext4_journalled_set_page_dirty(struct page *page)
2881{
2882	SetPageChecked(page);
2883	return __set_page_dirty_nobuffers(page);
2884}
2885
2886static const struct address_space_operations ext4_ordered_aops = {
2887	.readpage		= ext4_readpage,
2888	.readpages		= ext4_readpages,
2889	.writepage		= ext4_writepage,
2890	.write_begin		= ext4_write_begin,
2891	.write_end		= ext4_ordered_write_end,
2892	.bmap			= ext4_bmap,
2893	.invalidatepage		= ext4_invalidatepage,
2894	.releasepage		= ext4_releasepage,
2895	.direct_IO		= ext4_direct_IO,
2896	.migratepage		= buffer_migrate_page,
2897	.is_partially_uptodate  = block_is_partially_uptodate,
2898	.error_remove_page	= generic_error_remove_page,
2899};
2900
2901static const struct address_space_operations ext4_writeback_aops = {
2902	.readpage		= ext4_readpage,
2903	.readpages		= ext4_readpages,
2904	.writepage		= ext4_writepage,
 
2905	.write_begin		= ext4_write_begin,
2906	.write_end		= ext4_writeback_write_end,
 
2907	.bmap			= ext4_bmap,
2908	.invalidatepage		= ext4_invalidatepage,
2909	.releasepage		= ext4_releasepage,
2910	.direct_IO		= ext4_direct_IO,
2911	.migratepage		= buffer_migrate_page,
2912	.is_partially_uptodate  = block_is_partially_uptodate,
2913	.error_remove_page	= generic_error_remove_page,
2914};
2915
2916static const struct address_space_operations ext4_journalled_aops = {
2917	.readpage		= ext4_readpage,
2918	.readpages		= ext4_readpages,
2919	.writepage		= ext4_writepage,
 
2920	.write_begin		= ext4_write_begin,
2921	.write_end		= ext4_journalled_write_end,
2922	.set_page_dirty		= ext4_journalled_set_page_dirty,
2923	.bmap			= ext4_bmap,
2924	.invalidatepage		= ext4_invalidatepage,
2925	.releasepage		= ext4_releasepage,
 
2926	.is_partially_uptodate  = block_is_partially_uptodate,
2927	.error_remove_page	= generic_error_remove_page,
2928};
2929
2930static const struct address_space_operations ext4_da_aops = {
2931	.readpage		= ext4_readpage,
2932	.readpages		= ext4_readpages,
2933	.writepage		= ext4_writepage,
2934	.writepages		= ext4_da_writepages,
2935	.write_begin		= ext4_da_write_begin,
2936	.write_end		= ext4_da_write_end,
 
2937	.bmap			= ext4_bmap,
2938	.invalidatepage		= ext4_da_invalidatepage,
2939	.releasepage		= ext4_releasepage,
2940	.direct_IO		= ext4_direct_IO,
2941	.migratepage		= buffer_migrate_page,
2942	.is_partially_uptodate  = block_is_partially_uptodate,
2943	.error_remove_page	= generic_error_remove_page,
2944};
2945
2946void ext4_set_aops(struct inode *inode)
2947{
2948	if (ext4_should_order_data(inode) &&
2949		test_opt(inode->i_sb, DELALLOC))
2950		inode->i_mapping->a_ops = &ext4_da_aops;
2951	else if (ext4_should_order_data(inode))
2952		inode->i_mapping->a_ops = &ext4_ordered_aops;
2953	else if (ext4_should_writeback_data(inode) &&
2954		 test_opt(inode->i_sb, DELALLOC))
 
 
 
 
2955		inode->i_mapping->a_ops = &ext4_da_aops;
2956	else if (ext4_should_writeback_data(inode))
2957		inode->i_mapping->a_ops = &ext4_writeback_aops;
2958	else
2959		inode->i_mapping->a_ops = &ext4_journalled_aops;
2960}
2961
2962/*
2963 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2964 * up to the end of the block which corresponds to `from'.
2965 * This required during truncate. We need to physically zero the tail end
2966 * of that block so it doesn't yield old data if the file is later grown.
2967 */
2968int ext4_block_truncate_page(handle_t *handle,
2969		struct address_space *mapping, loff_t from)
2970{
2971	unsigned offset = from & (PAGE_CACHE_SIZE-1);
2972	unsigned length;
2973	unsigned blocksize;
2974	struct inode *inode = mapping->host;
2975
2976	blocksize = inode->i_sb->s_blocksize;
2977	length = blocksize - (offset & (blocksize - 1));
2978
2979	return ext4_block_zero_page_range(handle, mapping, from, length);
2980}
2981
2982/*
2983 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
2984 * starting from file offset 'from'.  The range to be zero'd must
2985 * be contained with in one block.  If the specified range exceeds
2986 * the end of the block it will be shortened to end of the block
2987 * that cooresponds to 'from'
2988 */
2989int ext4_block_zero_page_range(handle_t *handle,
2990		struct address_space *mapping, loff_t from, loff_t length)
2991{
2992	ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2993	unsigned offset = from & (PAGE_CACHE_SIZE-1);
2994	unsigned blocksize, max, pos;
2995	ext4_lblk_t iblock;
2996	struct inode *inode = mapping->host;
2997	struct buffer_head *bh;
2998	struct page *page;
2999	int err = 0;
3000
3001	page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3002				   mapping_gfp_mask(mapping) & ~__GFP_FS);
3003	if (!page)
3004		return -EINVAL;
3005
3006	blocksize = inode->i_sb->s_blocksize;
3007	max = blocksize - (offset & (blocksize - 1));
3008
3009	/*
3010	 * correct length if it does not fall between
3011	 * 'from' and the end of the block
3012	 */
3013	if (length > max || length < 0)
3014		length = max;
3015
3016	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3017
3018	if (!page_has_buffers(page))
3019		create_empty_buffers(page, blocksize, 0);
3020
3021	/* Find the buffer that contains "offset" */
3022	bh = page_buffers(page);
3023	pos = blocksize;
3024	while (offset >= pos) {
3025		bh = bh->b_this_page;
3026		iblock++;
3027		pos += blocksize;
3028	}
3029
3030	err = 0;
3031	if (buffer_freed(bh)) {
3032		BUFFER_TRACE(bh, "freed: skip");
3033		goto unlock;
3034	}
3035
3036	if (!buffer_mapped(bh)) {
3037		BUFFER_TRACE(bh, "unmapped");
3038		ext4_get_block(inode, iblock, bh, 0);
3039		/* unmapped? It's a hole - nothing to do */
3040		if (!buffer_mapped(bh)) {
3041			BUFFER_TRACE(bh, "still unmapped");
3042			goto unlock;
3043		}
3044	}
3045
3046	/* Ok, it's mapped. Make sure it's up-to-date */
3047	if (PageUptodate(page))
3048		set_buffer_uptodate(bh);
3049
3050	if (!buffer_uptodate(bh)) {
3051		err = -EIO;
3052		ll_rw_block(READ, 1, &bh);
3053		wait_on_buffer(bh);
3054		/* Uhhuh. Read error. Complain and punt. */
3055		if (!buffer_uptodate(bh))
3056			goto unlock;
 
 
 
 
 
 
 
 
3057	}
3058
3059	if (ext4_should_journal_data(inode)) {
3060		BUFFER_TRACE(bh, "get write access");
3061		err = ext4_journal_get_write_access(handle, bh);
3062		if (err)
3063			goto unlock;
3064	}
3065
3066	zero_user(page, offset, length);
3067
3068	BUFFER_TRACE(bh, "zeroed end of block");
3069
3070	err = 0;
3071	if (ext4_should_journal_data(inode)) {
3072		err = ext4_handle_dirty_metadata(handle, inode, bh);
3073	} else {
3074		if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
3075			err = ext4_jbd2_file_inode(handle, inode);
3076		mark_buffer_dirty(bh);
 
 
3077	}
3078
3079unlock:
3080	unlock_page(page);
3081	page_cache_release(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3082	return err;
3083}
3084
3085int ext4_can_truncate(struct inode *inode)
3086{
3087	if (S_ISREG(inode->i_mode))
3088		return 1;
3089	if (S_ISDIR(inode->i_mode))
3090		return 1;
3091	if (S_ISLNK(inode->i_mode))
3092		return !ext4_inode_is_fast_symlink(inode);
3093	return 0;
3094}
3095
3096/*
3097 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3098 * associated with the given offset and length
3099 *
3100 * @inode:  File inode
3101 * @offset: The offset where the hole will begin
3102 * @len:    The length of the hole
3103 *
3104 * Returns: 0 on sucess or negative on failure
3105 */
3106
3107int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3108{
3109	struct inode *inode = file->f_path.dentry->d_inode;
 
 
 
 
 
 
 
3110	if (!S_ISREG(inode->i_mode))
3111		return -ENOTSUPP;
 
 
3112
3113	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3114		/* TODO: Add support for non extent hole punching */
3115		return -ENOTSUPP;
 
 
 
 
 
 
3116	}
3117
3118	return ext4_ext_punch_hole(file, offset, length);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3119}
3120
3121/*
3122 * ext4_truncate()
3123 *
3124 * We block out ext4_get_block() block instantiations across the entire
3125 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3126 * simultaneously on behalf of the same inode.
3127 *
3128 * As we work through the truncate and commmit bits of it to the journal there
3129 * is one core, guiding principle: the file's tree must always be consistent on
3130 * disk.  We must be able to restart the truncate after a crash.
3131 *
3132 * The file's tree may be transiently inconsistent in memory (although it
3133 * probably isn't), but whenever we close off and commit a journal transaction,
3134 * the contents of (the filesystem + the journal) must be consistent and
3135 * restartable.  It's pretty simple, really: bottom up, right to left (although
3136 * left-to-right works OK too).
3137 *
3138 * Note that at recovery time, journal replay occurs *before* the restart of
3139 * truncate against the orphan inode list.
3140 *
3141 * The committed inode has the new, desired i_size (which is the same as
3142 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
3143 * that this inode's truncate did not complete and it will again call
3144 * ext4_truncate() to have another go.  So there will be instantiated blocks
3145 * to the right of the truncation point in a crashed ext4 filesystem.  But
3146 * that's fine - as long as they are linked from the inode, the post-crash
3147 * ext4_truncate() run will find them and release them.
3148 */
3149void ext4_truncate(struct inode *inode)
3150{
 
 
 
 
 
 
 
 
 
 
 
 
 
3151	trace_ext4_truncate_enter(inode);
3152
3153	if (!ext4_can_truncate(inode))
3154		return;
3155
3156	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3157
3158	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3159		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
3160
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3161	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3162		ext4_ext_truncate(inode);
3163	else
3164		ext4_ind_truncate(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3165
3166	trace_ext4_truncate_exit(inode);
 
3167}
3168
3169/*
3170 * ext4_get_inode_loc returns with an extra refcount against the inode's
3171 * underlying buffer_head on success. If 'in_mem' is true, we have all
3172 * data in memory that is needed to recreate the on-disk version of this
3173 * inode.
3174 */
3175static int __ext4_get_inode_loc(struct inode *inode,
3176				struct ext4_iloc *iloc, int in_mem)
3177{
3178	struct ext4_group_desc	*gdp;
3179	struct buffer_head	*bh;
3180	struct super_block	*sb = inode->i_sb;
3181	ext4_fsblk_t		block;
3182	int			inodes_per_block, inode_offset;
3183
3184	iloc->bh = NULL;
3185	if (!ext4_valid_inum(sb, inode->i_ino))
3186		return -EIO;
3187
3188	iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3189	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3190	if (!gdp)
3191		return -EIO;
3192
3193	/*
3194	 * Figure out the offset within the block group inode table
3195	 */
3196	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
3197	inode_offset = ((inode->i_ino - 1) %
3198			EXT4_INODES_PER_GROUP(sb));
3199	block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3200	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3201
3202	bh = sb_getblk(sb, block);
3203	if (!bh) {
3204		EXT4_ERROR_INODE_BLOCK(inode, block,
3205				       "unable to read itable block");
3206		return -EIO;
3207	}
3208	if (!buffer_uptodate(bh)) {
3209		lock_buffer(bh);
3210
3211		/*
3212		 * If the buffer has the write error flag, we have failed
3213		 * to write out another inode in the same block.  In this
3214		 * case, we don't have to read the block because we may
3215		 * read the old inode data successfully.
3216		 */
3217		if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3218			set_buffer_uptodate(bh);
3219
3220		if (buffer_uptodate(bh)) {
3221			/* someone brought it uptodate while we waited */
3222			unlock_buffer(bh);
3223			goto has_buffer;
3224		}
3225
3226		/*
3227		 * If we have all information of the inode in memory and this
3228		 * is the only valid inode in the block, we need not read the
3229		 * block.
3230		 */
3231		if (in_mem) {
3232			struct buffer_head *bitmap_bh;
3233			int i, start;
3234
3235			start = inode_offset & ~(inodes_per_block - 1);
3236
3237			/* Is the inode bitmap in cache? */
3238			bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3239			if (!bitmap_bh)
3240				goto make_io;
3241
3242			/*
3243			 * If the inode bitmap isn't in cache then the
3244			 * optimisation may end up performing two reads instead
3245			 * of one, so skip it.
3246			 */
3247			if (!buffer_uptodate(bitmap_bh)) {
3248				brelse(bitmap_bh);
3249				goto make_io;
3250			}
3251			for (i = start; i < start + inodes_per_block; i++) {
3252				if (i == inode_offset)
3253					continue;
3254				if (ext4_test_bit(i, bitmap_bh->b_data))
3255					break;
3256			}
3257			brelse(bitmap_bh);
3258			if (i == start + inodes_per_block) {
3259				/* all other inodes are free, so skip I/O */
3260				memset(bh->b_data, 0, bh->b_size);
3261				set_buffer_uptodate(bh);
3262				unlock_buffer(bh);
3263				goto has_buffer;
3264			}
3265		}
3266
3267make_io:
3268		/*
3269		 * If we need to do any I/O, try to pre-readahead extra
3270		 * blocks from the inode table.
3271		 */
3272		if (EXT4_SB(sb)->s_inode_readahead_blks) {
3273			ext4_fsblk_t b, end, table;
3274			unsigned num;
 
3275
3276			table = ext4_inode_table(sb, gdp);
3277			/* s_inode_readahead_blks is always a power of 2 */
3278			b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3279			if (table > b)
3280				b = table;
3281			end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3282			num = EXT4_INODES_PER_GROUP(sb);
3283			if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3284				       EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3285				num -= ext4_itable_unused_count(sb, gdp);
3286			table += num / inodes_per_block;
3287			if (end > table)
3288				end = table;
3289			while (b <= end)
3290				sb_breadahead(sb, b++);
3291		}
3292
3293		/*
3294		 * There are other valid inodes in the buffer, this inode
3295		 * has in-inode xattrs, or we don't have this inode in memory.
3296		 * Read the block from disk.
3297		 */
3298		trace_ext4_load_inode(inode);
3299		get_bh(bh);
3300		bh->b_end_io = end_buffer_read_sync;
3301		submit_bh(READ | REQ_META | REQ_PRIO, bh);
3302		wait_on_buffer(bh);
3303		if (!buffer_uptodate(bh)) {
3304			EXT4_ERROR_INODE_BLOCK(inode, block,
3305					       "unable to read itable block");
3306			brelse(bh);
3307			return -EIO;
3308		}
3309	}
3310has_buffer:
3311	iloc->bh = bh;
3312	return 0;
3313}
3314
3315int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3316{
3317	/* We have all inode data except xattrs in memory here. */
3318	return __ext4_get_inode_loc(inode, iloc,
3319		!ext4_test_inode_state(inode, EXT4_STATE_XATTR));
3320}
3321
3322void ext4_set_inode_flags(struct inode *inode)
3323{
3324	unsigned int flags = EXT4_I(inode)->i_flags;
 
3325
3326	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3327	if (flags & EXT4_SYNC_FL)
3328		inode->i_flags |= S_SYNC;
3329	if (flags & EXT4_APPEND_FL)
3330		inode->i_flags |= S_APPEND;
3331	if (flags & EXT4_IMMUTABLE_FL)
3332		inode->i_flags |= S_IMMUTABLE;
3333	if (flags & EXT4_NOATIME_FL)
3334		inode->i_flags |= S_NOATIME;
3335	if (flags & EXT4_DIRSYNC_FL)
3336		inode->i_flags |= S_DIRSYNC;
 
 
 
 
 
 
3337}
3338
3339/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3340void ext4_get_inode_flags(struct ext4_inode_info *ei)
3341{
3342	unsigned int vfs_fl;
3343	unsigned long old_fl, new_fl;
3344
3345	do {
3346		vfs_fl = ei->vfs_inode.i_flags;
3347		old_fl = ei->i_flags;
3348		new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3349				EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3350				EXT4_DIRSYNC_FL);
3351		if (vfs_fl & S_SYNC)
3352			new_fl |= EXT4_SYNC_FL;
3353		if (vfs_fl & S_APPEND)
3354			new_fl |= EXT4_APPEND_FL;
3355		if (vfs_fl & S_IMMUTABLE)
3356			new_fl |= EXT4_IMMUTABLE_FL;
3357		if (vfs_fl & S_NOATIME)
3358			new_fl |= EXT4_NOATIME_FL;
3359		if (vfs_fl & S_DIRSYNC)
3360			new_fl |= EXT4_DIRSYNC_FL;
3361	} while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
3362}
3363
3364static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3365				  struct ext4_inode_info *ei)
3366{
3367	blkcnt_t i_blocks ;
3368	struct inode *inode = &(ei->vfs_inode);
3369	struct super_block *sb = inode->i_sb;
3370
3371	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3372				EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3373		/* we are using combined 48 bit field */
3374		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3375					le32_to_cpu(raw_inode->i_blocks_lo);
3376		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
3377			/* i_blocks represent file system block size */
3378			return i_blocks  << (inode->i_blkbits - 9);
3379		} else {
3380			return i_blocks;
3381		}
3382	} else {
3383		return le32_to_cpu(raw_inode->i_blocks_lo);
3384	}
3385}
3386
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3387struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3388{
3389	struct ext4_iloc iloc;
3390	struct ext4_inode *raw_inode;
3391	struct ext4_inode_info *ei;
3392	struct inode *inode;
3393	journal_t *journal = EXT4_SB(sb)->s_journal;
3394	long ret;
 
3395	int block;
 
 
 
3396
3397	inode = iget_locked(sb, ino);
3398	if (!inode)
3399		return ERR_PTR(-ENOMEM);
3400	if (!(inode->i_state & I_NEW))
3401		return inode;
3402
3403	ei = EXT4_I(inode);
3404	iloc.bh = NULL;
3405
3406	ret = __ext4_get_inode_loc(inode, &iloc, 0);
3407	if (ret < 0)
3408		goto bad_inode;
3409	raw_inode = ext4_raw_inode(&iloc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3410	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3411	inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3412	inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
 
 
 
 
 
 
 
3413	if (!(test_opt(inode->i_sb, NO_UID32))) {
3414		inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3415		inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3416	}
3417	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
 
 
 
3418
3419	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
 
3420	ei->i_dir_start_lookup = 0;
3421	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3422	/* We now have enough fields to check if the inode was active or not.
3423	 * This is needed because nfsd might try to access dead inodes
3424	 * the test is that same one that e2fsck uses
3425	 * NeilBrown 1999oct15
3426	 */
3427	if (inode->i_nlink == 0) {
3428		if (inode->i_mode == 0 ||
3429		    !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
 
3430			/* this inode is deleted */
3431			ret = -ESTALE;
3432			goto bad_inode;
3433		}
3434		/* The only unlinked inodes we let through here have
3435		 * valid i_mode and are being read by the orphan
3436		 * recovery code: that's fine, we're about to complete
3437		 * the process of deleting those. */
 
 
3438	}
3439	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3440	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3441	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3442	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
3443		ei->i_file_acl |=
3444			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3445	inode->i_size = ext4_isize(raw_inode);
 
 
 
 
 
3446	ei->i_disksize = inode->i_size;
3447#ifdef CONFIG_QUOTA
3448	ei->i_reserved_quota = 0;
3449#endif
3450	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3451	ei->i_block_group = iloc.block_group;
3452	ei->i_last_alloc_group = ~0;
3453	/*
3454	 * NOTE! The in-memory inode i_data array is in little-endian order
3455	 * even on big-endian machines: we do NOT byteswap the block numbers!
3456	 */
3457	for (block = 0; block < EXT4_N_BLOCKS; block++)
3458		ei->i_data[block] = raw_inode->i_block[block];
3459	INIT_LIST_HEAD(&ei->i_orphan);
3460
3461	/*
3462	 * Set transaction id's of transactions that have to be committed
3463	 * to finish f[data]sync. We set them to currently running transaction
3464	 * as we cannot be sure that the inode or some of its metadata isn't
3465	 * part of the transaction - the inode could have been reclaimed and
3466	 * now it is reread from disk.
3467	 */
3468	if (journal) {
3469		transaction_t *transaction;
3470		tid_t tid;
3471
3472		read_lock(&journal->j_state_lock);
3473		if (journal->j_running_transaction)
3474			transaction = journal->j_running_transaction;
3475		else
3476			transaction = journal->j_committing_transaction;
3477		if (transaction)
3478			tid = transaction->t_tid;
3479		else
3480			tid = journal->j_commit_sequence;
3481		read_unlock(&journal->j_state_lock);
3482		ei->i_sync_tid = tid;
3483		ei->i_datasync_tid = tid;
3484	}
3485
3486	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3487		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3488		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3489		    EXT4_INODE_SIZE(inode->i_sb)) {
3490			ret = -EIO;
3491			goto bad_inode;
3492		}
3493		if (ei->i_extra_isize == 0) {
3494			/* The extra space is currently unused. Use it. */
 
3495			ei->i_extra_isize = sizeof(struct ext4_inode) -
3496					    EXT4_GOOD_OLD_INODE_SIZE;
3497		} else {
3498			__le32 *magic = (void *)raw_inode +
3499					EXT4_GOOD_OLD_INODE_SIZE +
3500					ei->i_extra_isize;
3501			if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3502				ext4_set_inode_state(inode, EXT4_STATE_XATTR);
3503		}
3504	} else
3505		ei->i_extra_isize = 0;
3506
3507	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3508	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3509	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3510	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3511
3512	inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3513	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3514		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3515			inode->i_version |=
3516			(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
 
 
3517	}
3518
3519	ret = 0;
3520	if (ei->i_file_acl &&
3521	    !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
3522		EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
3523				 ei->i_file_acl);
3524		ret = -EIO;
3525		goto bad_inode;
3526	} else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3527		if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3528		    (S_ISLNK(inode->i_mode) &&
3529		     !ext4_inode_is_fast_symlink(inode)))
3530			/* Validate extent which is part of inode */
3531			ret = ext4_ext_check_inode(inode);
3532	} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3533		   (S_ISLNK(inode->i_mode) &&
3534		    !ext4_inode_is_fast_symlink(inode))) {
3535		/* Validate block references which are part of inode */
3536		ret = ext4_ind_check_inode(inode);
 
 
3537	}
3538	if (ret)
3539		goto bad_inode;
3540
3541	if (S_ISREG(inode->i_mode)) {
3542		inode->i_op = &ext4_file_inode_operations;
3543		inode->i_fop = &ext4_file_operations;
3544		ext4_set_aops(inode);
3545	} else if (S_ISDIR(inode->i_mode)) {
3546		inode->i_op = &ext4_dir_inode_operations;
3547		inode->i_fop = &ext4_dir_operations;
3548	} else if (S_ISLNK(inode->i_mode)) {
3549		if (ext4_inode_is_fast_symlink(inode)) {
 
 
 
 
3550			inode->i_op = &ext4_fast_symlink_inode_operations;
3551			nd_terminate_link(ei->i_data, inode->i_size,
3552				sizeof(ei->i_data) - 1);
3553		} else {
3554			inode->i_op = &ext4_symlink_inode_operations;
3555			ext4_set_aops(inode);
3556		}
 
3557	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
3558	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
3559		inode->i_op = &ext4_special_inode_operations;
3560		if (raw_inode->i_block[0])
3561			init_special_inode(inode, inode->i_mode,
3562			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3563		else
3564			init_special_inode(inode, inode->i_mode,
3565			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
 
 
3566	} else {
3567		ret = -EIO;
3568		EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
3569		goto bad_inode;
3570	}
3571	brelse(iloc.bh);
3572	ext4_set_inode_flags(inode);
3573	unlock_new_inode(inode);
3574	return inode;
3575
3576bad_inode:
3577	brelse(iloc.bh);
3578	iget_failed(inode);
3579	return ERR_PTR(ret);
3580}
3581
 
 
 
 
 
 
 
3582static int ext4_inode_blocks_set(handle_t *handle,
3583				struct ext4_inode *raw_inode,
3584				struct ext4_inode_info *ei)
3585{
3586	struct inode *inode = &(ei->vfs_inode);
3587	u64 i_blocks = inode->i_blocks;
3588	struct super_block *sb = inode->i_sb;
3589
3590	if (i_blocks <= ~0U) {
3591		/*
3592		 * i_blocks can be represnted in a 32 bit variable
3593		 * as multiple of 512 bytes
3594		 */
3595		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3596		raw_inode->i_blocks_high = 0;
3597		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3598		return 0;
3599	}
3600	if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
3601		return -EFBIG;
3602
3603	if (i_blocks <= 0xffffffffffffULL) {
3604		/*
3605		 * i_blocks can be represented in a 48 bit variable
3606		 * as multiple of 512 bytes
3607		 */
3608		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3609		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3610		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3611	} else {
3612		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
3613		/* i_block is stored in file system block size */
3614		i_blocks = i_blocks >> (inode->i_blkbits - 9);
3615		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
3616		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3617	}
3618	return 0;
3619}
3620
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3621/*
3622 * Post the struct inode info into an on-disk inode location in the
3623 * buffer-cache.  This gobbles the caller's reference to the
3624 * buffer_head in the inode location struct.
3625 *
3626 * The caller must have write access to iloc->bh.
3627 */
3628static int ext4_do_update_inode(handle_t *handle,
3629				struct inode *inode,
3630				struct ext4_iloc *iloc)
3631{
3632	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3633	struct ext4_inode_info *ei = EXT4_I(inode);
3634	struct buffer_head *bh = iloc->bh;
 
3635	int err = 0, rc, block;
 
 
 
 
 
 
3636
3637	/* For fields not not tracking in the in-memory inode,
3638	 * initialise them to zero for new inodes. */
3639	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
3640		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3641
3642	ext4_get_inode_flags(ei);
3643	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 
 
 
3644	if (!(test_opt(inode->i_sb, NO_UID32))) {
3645		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3646		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3647/*
3648 * Fix up interoperability with old kernels. Otherwise, old inodes get
3649 * re-used with the upper 16 bits of the uid/gid intact
3650 */
3651		if (!ei->i_dtime) {
3652			raw_inode->i_uid_high =
3653				cpu_to_le16(high_16_bits(inode->i_uid));
3654			raw_inode->i_gid_high =
3655				cpu_to_le16(high_16_bits(inode->i_gid));
3656		} else {
3657			raw_inode->i_uid_high = 0;
3658			raw_inode->i_gid_high = 0;
 
 
 
 
 
3659		}
3660	} else {
3661		raw_inode->i_uid_low =
3662			cpu_to_le16(fs_high2lowuid(inode->i_uid));
3663		raw_inode->i_gid_low =
3664			cpu_to_le16(fs_high2lowgid(inode->i_gid));
3665		raw_inode->i_uid_high = 0;
3666		raw_inode->i_gid_high = 0;
3667	}
3668	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3669
3670	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3671	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3672	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3673	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3674
3675	if (ext4_inode_blocks_set(handle, raw_inode, ei))
 
 
3676		goto out_brelse;
 
3677	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3678	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
3679	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3680	    cpu_to_le32(EXT4_OS_HURD))
3681		raw_inode->i_file_acl_high =
3682			cpu_to_le16(ei->i_file_acl >> 32);
3683	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
3684	ext4_isize_set(raw_inode, ei->i_disksize);
 
 
 
3685	if (ei->i_disksize > 0x7fffffffULL) {
3686		struct super_block *sb = inode->i_sb;
3687		if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
3688				EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
3689				EXT4_SB(sb)->s_es->s_rev_level ==
3690				cpu_to_le32(EXT4_GOOD_OLD_REV)) {
3691			/* If this is the first large file
3692			 * created, add a flag to the superblock.
3693			 */
3694			err = ext4_journal_get_write_access(handle,
3695					EXT4_SB(sb)->s_sbh);
3696			if (err)
3697				goto out_brelse;
3698			ext4_update_dynamic_rev(sb);
3699			EXT4_SET_RO_COMPAT_FEATURE(sb,
3700					EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
3701			sb->s_dirt = 1;
3702			ext4_handle_sync(handle);
3703			err = ext4_handle_dirty_metadata(handle, NULL,
3704					EXT4_SB(sb)->s_sbh);
3705		}
3706	}
3707	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3708	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3709		if (old_valid_dev(inode->i_rdev)) {
3710			raw_inode->i_block[0] =
3711				cpu_to_le32(old_encode_dev(inode->i_rdev));
3712			raw_inode->i_block[1] = 0;
3713		} else {
3714			raw_inode->i_block[0] = 0;
3715			raw_inode->i_block[1] =
3716				cpu_to_le32(new_encode_dev(inode->i_rdev));
3717			raw_inode->i_block[2] = 0;
3718		}
3719	} else
3720		for (block = 0; block < EXT4_N_BLOCKS; block++)
3721			raw_inode->i_block[block] = ei->i_data[block];
 
3722
3723	raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
3724	if (ei->i_extra_isize) {
3725		if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3726			raw_inode->i_version_hi =
3727			cpu_to_le32(inode->i_version >> 32);
3728		raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
 
 
 
3729	}
3730
 
 
 
 
 
 
 
 
 
 
 
 
 
3731	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3732	rc = ext4_handle_dirty_metadata(handle, NULL, bh);
3733	if (!err)
3734		err = rc;
3735	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
3736
3737	ext4_update_inode_fsync_trans(handle, inode, 0);
 
 
 
 
 
 
 
 
 
3738out_brelse:
3739	brelse(bh);
3740	ext4_std_error(inode->i_sb, err);
3741	return err;
3742}
3743
3744/*
3745 * ext4_write_inode()
3746 *
3747 * We are called from a few places:
3748 *
3749 * - Within generic_file_write() for O_SYNC files.
3750 *   Here, there will be no transaction running. We wait for any running
3751 *   trasnaction to commit.
3752 *
3753 * - Within sys_sync(), kupdate and such.
3754 *   We wait on commit, if tol to.
3755 *
3756 * - Within prune_icache() (PF_MEMALLOC == true)
3757 *   Here we simply return.  We can't afford to block kswapd on the
3758 *   journal commit.
3759 *
3760 * In all cases it is actually safe for us to return without doing anything,
3761 * because the inode has been copied into a raw inode buffer in
3762 * ext4_mark_inode_dirty().  This is a correctness thing for O_SYNC and for
3763 * knfsd.
3764 *
3765 * Note that we are absolutely dependent upon all inode dirtiers doing the
3766 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3767 * which we are interested.
3768 *
3769 * It would be a bug for them to not do this.  The code:
3770 *
3771 *	mark_inode_dirty(inode)
3772 *	stuff();
3773 *	inode->i_size = expr;
3774 *
3775 * is in error because a kswapd-driven write_inode() could occur while
3776 * `stuff()' is running, and the new i_size will be lost.  Plus the inode
3777 * will no longer be on the superblock's dirty inode list.
3778 */
3779int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
3780{
3781	int err;
3782
3783	if (current->flags & PF_MEMALLOC)
3784		return 0;
3785
3786	if (EXT4_SB(inode->i_sb)->s_journal) {
3787		if (ext4_journal_current_handle()) {
3788			jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3789			dump_stack();
3790			return -EIO;
3791		}
3792
3793		if (wbc->sync_mode != WB_SYNC_ALL)
 
 
 
 
 
3794			return 0;
3795
3796		err = ext4_force_commit(inode->i_sb);
3797	} else {
3798		struct ext4_iloc iloc;
3799
3800		err = __ext4_get_inode_loc(inode, &iloc, 0);
3801		if (err)
3802			return err;
3803		if (wbc->sync_mode == WB_SYNC_ALL)
 
 
 
 
3804			sync_dirty_buffer(iloc.bh);
3805		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
3806			EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
3807					 "IO error syncing inode");
3808			err = -EIO;
3809		}
3810		brelse(iloc.bh);
3811	}
3812	return err;
3813}
3814
3815/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3816 * ext4_setattr()
3817 *
3818 * Called from notify_change.
3819 *
3820 * We want to trap VFS attempts to truncate the file as soon as
3821 * possible.  In particular, we want to make sure that when the VFS
3822 * shrinks i_size, we put the inode on the orphan list and modify
3823 * i_disksize immediately, so that during the subsequent flushing of
3824 * dirty pages and freeing of disk blocks, we can guarantee that any
3825 * commit will leave the blocks being flushed in an unused state on
3826 * disk.  (On recovery, the inode will get truncated and the blocks will
3827 * be freed, so we have a strong guarantee that no future commit will
3828 * leave these blocks visible to the user.)
3829 *
3830 * Another thing we have to assure is that if we are in ordered mode
3831 * and inode is still attached to the committing transaction, we must
3832 * we start writeout of all the dirty pages which are being truncated.
3833 * This way we are sure that all the data written in the previous
3834 * transaction are already on disk (truncate waits for pages under
3835 * writeback).
3836 *
3837 * Called with inode->i_mutex down.
3838 */
3839int ext4_setattr(struct dentry *dentry, struct iattr *attr)
3840{
3841	struct inode *inode = dentry->d_inode;
3842	int error, rc = 0;
3843	int orphan = 0;
3844	const unsigned int ia_valid = attr->ia_valid;
3845
3846	error = inode_change_ok(inode, attr);
3847	if (error)
3848		return error;
3849
3850	if (is_quota_modification(inode, attr))
3851		dquot_initialize(inode);
3852	if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3853		(ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
 
 
 
3854		handle_t *handle;
3855
3856		/* (user+group)*(old+new) structure, inode write (sb,
3857		 * inode block, ? - but truncate inode update has it) */
3858		handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3859					EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
 
3860		if (IS_ERR(handle)) {
3861			error = PTR_ERR(handle);
3862			goto err_out;
3863		}
3864		error = dquot_transfer(inode, attr);
3865		if (error) {
3866			ext4_journal_stop(handle);
3867			return error;
3868		}
3869		/* Update corresponding info in inode so that everything is in
3870		 * one transaction */
3871		if (attr->ia_valid & ATTR_UID)
3872			inode->i_uid = attr->ia_uid;
3873		if (attr->ia_valid & ATTR_GID)
3874			inode->i_gid = attr->ia_gid;
3875		error = ext4_mark_inode_dirty(handle, inode);
3876		ext4_journal_stop(handle);
3877	}
3878
3879	if (attr->ia_valid & ATTR_SIZE) {
3880		inode_dio_wait(inode);
 
 
3881
3882		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3883			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3884
3885			if (attr->ia_size > sbi->s_bitmap_maxbytes)
3886				return -EFBIG;
3887		}
3888	}
3889
3890	if (S_ISREG(inode->i_mode) &&
3891	    attr->ia_valid & ATTR_SIZE &&
3892	    (attr->ia_size < inode->i_size)) {
3893		handle_t *handle;
3894
3895		handle = ext4_journal_start(inode, 3);
3896		if (IS_ERR(handle)) {
3897			error = PTR_ERR(handle);
3898			goto err_out;
3899		}
3900		if (ext4_handle_valid(handle)) {
3901			error = ext4_orphan_add(handle, inode);
3902			orphan = 1;
3903		}
3904		EXT4_I(inode)->i_disksize = attr->ia_size;
3905		rc = ext4_mark_inode_dirty(handle, inode);
3906		if (!error)
3907			error = rc;
3908		ext4_journal_stop(handle);
3909
3910		if (ext4_should_order_data(inode)) {
 
3911			error = ext4_begin_ordered_truncate(inode,
3912							    attr->ia_size);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3913			if (error) {
3914				/* Do as much error cleanup as possible */
3915				handle = ext4_journal_start(inode, 3);
3916				if (IS_ERR(handle)) {
3917					ext4_orphan_del(NULL, inode);
3918					goto err_out;
3919				}
3920				ext4_orphan_del(handle, inode);
3921				orphan = 0;
3922				ext4_journal_stop(handle);
3923				goto err_out;
3924			}
3925		}
3926	}
 
3927
3928	if (attr->ia_valid & ATTR_SIZE) {
3929		if (attr->ia_size != i_size_read(inode)) {
3930			truncate_setsize(inode, attr->ia_size);
3931			ext4_truncate(inode);
3932		} else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3933			ext4_truncate(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3934	}
3935
3936	if (!rc) {
3937		setattr_copy(inode, attr);
3938		mark_inode_dirty(inode);
3939	}
3940
3941	/*
3942	 * If the call to ext4_truncate failed to get a transaction handle at
3943	 * all, we need to clean up the in-core orphan list manually.
3944	 */
3945	if (orphan && inode->i_nlink)
3946		ext4_orphan_del(NULL, inode);
3947
3948	if (!rc && (ia_valid & ATTR_MODE))
3949		rc = ext4_acl_chmod(inode);
3950
3951err_out:
3952	ext4_std_error(inode->i_sb, error);
3953	if (!error)
3954		error = rc;
3955	return error;
3956}
3957
3958int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
3959		 struct kstat *stat)
3960{
3961	struct inode *inode;
3962	unsigned long delalloc_blocks;
3963
3964	inode = dentry->d_inode;
3965	generic_fillattr(inode, stat);
3966
3967	/*
 
 
 
 
 
 
 
 
 
3968	 * We can't update i_blocks if the block allocation is delayed
3969	 * otherwise in the case of system crash before the real block
3970	 * allocation is done, we will have i_blocks inconsistent with
3971	 * on-disk file blocks.
3972	 * We always keep i_blocks updated together with real
3973	 * allocation. But to not confuse with user, stat
3974	 * will return the blocks that include the delayed allocation
3975	 * blocks for this file.
3976	 */
3977	delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
3978
3979	stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
3980	return 0;
3981}
3982
3983static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
3984{
3985	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
3986		return ext4_ind_trans_blocks(inode, nrblocks, chunk);
3987	return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
3988}
3989
3990/*
3991 * Account for index blocks, block groups bitmaps and block group
3992 * descriptor blocks if modify datablocks and index blocks
3993 * worse case, the indexs blocks spread over different block groups
3994 *
3995 * If datablocks are discontiguous, they are possible to spread over
3996 * different block groups too. If they are contiuguous, with flexbg,
3997 * they could still across block group boundary.
3998 *
3999 * Also account for superblock, inode, quota and xattr blocks
4000 */
4001static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 
4002{
4003	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4004	int gdpblocks;
4005	int idxblocks;
4006	int ret = 0;
4007
4008	/*
4009	 * How many index blocks need to touch to modify nrblocks?
4010	 * The "Chunk" flag indicating whether the nrblocks is
4011	 * physically contiguous on disk
4012	 *
4013	 * For Direct IO and fallocate, they calls get_block to allocate
4014	 * one single extent at a time, so they could set the "Chunk" flag
4015	 */
4016	idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4017
4018	ret = idxblocks;
4019
4020	/*
4021	 * Now let's see how many group bitmaps and group descriptors need
4022	 * to account
4023	 */
4024	groups = idxblocks;
4025	if (chunk)
4026		groups += 1;
4027	else
4028		groups += nrblocks;
4029
4030	gdpblocks = groups;
4031	if (groups > ngroups)
4032		groups = ngroups;
4033	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4034		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4035
4036	/* bitmaps and block group descriptor blocks */
4037	ret += groups + gdpblocks;
4038
4039	/* Blocks for super block, inode, quota and xattr blocks */
4040	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4041
4042	return ret;
4043}
4044
4045/*
4046 * Calculate the total number of credits to reserve to fit
4047 * the modification of a single pages into a single transaction,
4048 * which may include multiple chunks of block allocations.
4049 *
4050 * This could be called via ext4_write_begin()
4051 *
4052 * We need to consider the worse case, when
4053 * one new block per extent.
4054 */
4055int ext4_writepage_trans_blocks(struct inode *inode)
4056{
4057	int bpp = ext4_journal_blocks_per_page(inode);
4058	int ret;
4059
4060	ret = ext4_meta_trans_blocks(inode, bpp, 0);
4061
4062	/* Account for data blocks for journalled mode */
4063	if (ext4_should_journal_data(inode))
4064		ret += bpp;
4065	return ret;
4066}
4067
4068/*
4069 * Calculate the journal credits for a chunk of data modification.
4070 *
4071 * This is called from DIO, fallocate or whoever calling
4072 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4073 *
4074 * journal buffers for data blocks are not included here, as DIO
4075 * and fallocate do no need to journal data buffers.
4076 */
4077int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4078{
4079	return ext4_meta_trans_blocks(inode, nrblocks, 1);
4080}
4081
4082/*
4083 * The caller must have previously called ext4_reserve_inode_write().
4084 * Give this, we know that the caller already has write access to iloc->bh.
4085 */
4086int ext4_mark_iloc_dirty(handle_t *handle,
4087			 struct inode *inode, struct ext4_iloc *iloc)
4088{
4089	int err = 0;
4090
4091	if (test_opt(inode->i_sb, I_VERSION))
4092		inode_inc_iversion(inode);
4093
4094	/* the do_update_inode consumes one bh->b_count */
4095	get_bh(iloc->bh);
4096
4097	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4098	err = ext4_do_update_inode(handle, inode, iloc);
4099	put_bh(iloc->bh);
4100	return err;
4101}
4102
4103/*
4104 * On success, We end up with an outstanding reference count against
4105 * iloc->bh.  This _must_ be cleaned up later.
4106 */
4107
4108int
4109ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4110			 struct ext4_iloc *iloc)
4111{
4112	int err;
4113
4114	err = ext4_get_inode_loc(inode, iloc);
4115	if (!err) {
4116		BUFFER_TRACE(iloc->bh, "get_write_access");
4117		err = ext4_journal_get_write_access(handle, iloc->bh);
4118		if (err) {
4119			brelse(iloc->bh);
4120			iloc->bh = NULL;
4121		}
4122	}
4123	ext4_std_error(inode->i_sb, err);
4124	return err;
4125}
4126
4127/*
4128 * Expand an inode by new_extra_isize bytes.
4129 * Returns 0 on success or negative error number on failure.
4130 */
4131static int ext4_expand_extra_isize(struct inode *inode,
4132				   unsigned int new_extra_isize,
4133				   struct ext4_iloc iloc,
4134				   handle_t *handle)
4135{
4136	struct ext4_inode *raw_inode;
4137	struct ext4_xattr_ibody_header *header;
4138
4139	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4140		return 0;
4141
4142	raw_inode = ext4_raw_inode(&iloc);
4143
4144	header = IHDR(inode, raw_inode);
4145
4146	/* No extended attributes present */
4147	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4148	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4149		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4150			new_extra_isize);
4151		EXT4_I(inode)->i_extra_isize = new_extra_isize;
4152		return 0;
4153	}
4154
4155	/* try to expand with EAs present */
4156	return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4157					  raw_inode, handle);
4158}
4159
4160/*
4161 * What we do here is to mark the in-core inode as clean with respect to inode
4162 * dirtiness (it may still be data-dirty).
4163 * This means that the in-core inode may be reaped by prune_icache
4164 * without having to perform any I/O.  This is a very good thing,
4165 * because *any* task may call prune_icache - even ones which
4166 * have a transaction open against a different journal.
4167 *
4168 * Is this cheating?  Not really.  Sure, we haven't written the
4169 * inode out, but prune_icache isn't a user-visible syncing function.
4170 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4171 * we start and wait on commits.
4172 *
4173 * Is this efficient/effective?  Well, we're being nice to the system
4174 * by cleaning up our inodes proactively so they can be reaped
4175 * without I/O.  But we are potentially leaving up to five seconds'
4176 * worth of inodes floating about which prune_icache wants us to
4177 * write out.  One way to fix that would be to get prune_icache()
4178 * to do a write_super() to free up some memory.  It has the desired
4179 * effect.
4180 */
4181int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4182{
4183	struct ext4_iloc iloc;
4184	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4185	static unsigned int mnt_count;
4186	int err, ret;
4187
4188	might_sleep();
4189	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
4190	err = ext4_reserve_inode_write(handle, inode, &iloc);
4191	if (ext4_handle_valid(handle) &&
4192	    EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
 
4193	    !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
4194		/*
4195		 * We need extra buffer credits since we may write into EA block
 
 
4196		 * with this same handle. If journal_extend fails, then it will
4197		 * only result in a minor loss of functionality for that inode.
4198		 * If this is felt to be critical, then e2fsck should be run to
4199		 * force a large enough s_min_extra_isize.
4200		 */
4201		if ((jbd2_journal_extend(handle,
4202			     EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
 
4203			ret = ext4_expand_extra_isize(inode,
4204						      sbi->s_want_extra_isize,
4205						      iloc, handle);
4206			if (ret) {
4207				ext4_set_inode_state(inode,
4208						     EXT4_STATE_NO_EXPAND);
4209				if (mnt_count !=
4210					le16_to_cpu(sbi->s_es->s_mnt_count)) {
4211					ext4_warning(inode->i_sb,
4212					"Unable to expand inode %lu. Delete"
4213					" some EAs or run e2fsck.",
4214					inode->i_ino);
4215					mnt_count =
4216					  le16_to_cpu(sbi->s_es->s_mnt_count);
4217				}
4218			}
4219		}
4220	}
4221	if (!err)
4222		err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4223	return err;
4224}
4225
4226/*
4227 * ext4_dirty_inode() is called from __mark_inode_dirty()
4228 *
4229 * We're really interested in the case where a file is being extended.
4230 * i_size has been changed by generic_commit_write() and we thus need
4231 * to include the updated inode in the current transaction.
4232 *
4233 * Also, dquot_alloc_block() will always dirty the inode when blocks
4234 * are allocated to the file.
4235 *
4236 * If the inode is marked synchronous, we don't honour that here - doing
4237 * so would cause a commit on atime updates, which we don't bother doing.
4238 * We handle synchronous inodes at the highest possible level.
 
 
 
 
4239 */
4240void ext4_dirty_inode(struct inode *inode, int flags)
4241{
4242	handle_t *handle;
4243
4244	handle = ext4_journal_start(inode, 2);
 
 
4245	if (IS_ERR(handle))
4246		goto out;
4247
4248	ext4_mark_inode_dirty(handle, inode);
4249
4250	ext4_journal_stop(handle);
4251out:
4252	return;
4253}
4254
4255#if 0
4256/*
4257 * Bind an inode's backing buffer_head into this transaction, to prevent
4258 * it from being flushed to disk early.  Unlike
4259 * ext4_reserve_inode_write, this leaves behind no bh reference and
4260 * returns no iloc structure, so the caller needs to repeat the iloc
4261 * lookup to mark the inode dirty later.
4262 */
4263static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4264{
4265	struct ext4_iloc iloc;
4266
4267	int err = 0;
4268	if (handle) {
4269		err = ext4_get_inode_loc(inode, &iloc);
4270		if (!err) {
4271			BUFFER_TRACE(iloc.bh, "get_write_access");
4272			err = jbd2_journal_get_write_access(handle, iloc.bh);
4273			if (!err)
4274				err = ext4_handle_dirty_metadata(handle,
4275								 NULL,
4276								 iloc.bh);
4277			brelse(iloc.bh);
4278		}
4279	}
4280	ext4_std_error(inode->i_sb, err);
4281	return err;
4282}
4283#endif
4284
4285int ext4_change_inode_journal_flag(struct inode *inode, int val)
4286{
4287	journal_t *journal;
4288	handle_t *handle;
4289	int err;
 
4290
4291	/*
4292	 * We have to be very careful here: changing a data block's
4293	 * journaling status dynamically is dangerous.  If we write a
4294	 * data block to the journal, change the status and then delete
4295	 * that block, we risk forgetting to revoke the old log record
4296	 * from the journal and so a subsequent replay can corrupt data.
4297	 * So, first we make sure that the journal is empty and that
4298	 * nobody is changing anything.
4299	 */
4300
4301	journal = EXT4_JOURNAL(inode);
4302	if (!journal)
4303		return 0;
4304	if (is_journal_aborted(journal))
4305		return -EROFS;
4306
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4307	jbd2_journal_lock_updates(journal);
4308	jbd2_journal_flush(journal);
4309
4310	/*
4311	 * OK, there are no updates running now, and all cached data is
4312	 * synced to disk.  We are now in a completely consistent state
4313	 * which doesn't have anything in the journal, and we know that
4314	 * no filesystem updates are running, so it is safe to modify
4315	 * the inode's in-core data-journaling state flag now.
4316	 */
4317
4318	if (val)
4319		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
4320	else
 
 
 
 
 
 
 
4321		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
 
4322	ext4_set_aops(inode);
 
 
 
 
 
4323
4324	jbd2_journal_unlock_updates(journal);
 
 
 
 
 
4325
4326	/* Finally we can mark the inode as dirty. */
4327
4328	handle = ext4_journal_start(inode, 1);
4329	if (IS_ERR(handle))
4330		return PTR_ERR(handle);
4331
4332	err = ext4_mark_inode_dirty(handle, inode);
4333	ext4_handle_sync(handle);
4334	ext4_journal_stop(handle);
4335	ext4_std_error(inode->i_sb, err);
4336
4337	return err;
4338}
4339
4340static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4341{
4342	return !buffer_mapped(bh);
4343}
4344
4345int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4346{
4347	struct page *page = vmf->page;
4348	loff_t size;
4349	unsigned long len;
4350	int ret;
4351	struct file *file = vma->vm_file;
4352	struct inode *inode = file->f_path.dentry->d_inode;
4353	struct address_space *mapping = inode->i_mapping;
4354	handle_t *handle;
4355	get_block_t *get_block;
4356	int retries = 0;
4357
4358	/*
4359	 * This check is racy but catches the common case. We rely on
4360	 * __block_page_mkwrite() to do a reliable check.
4361	 */
4362	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
4363	/* Delalloc case is easy... */
4364	if (test_opt(inode->i_sb, DELALLOC) &&
4365	    !ext4_should_journal_data(inode) &&
4366	    !ext4_nonda_switch(inode->i_sb)) {
4367		do {
4368			ret = __block_page_mkwrite(vma, vmf,
4369						   ext4_da_get_block_prep);
4370		} while (ret == -ENOSPC &&
4371		       ext4_should_retry_alloc(inode->i_sb, &retries));
4372		goto out_ret;
4373	}
4374
4375	lock_page(page);
4376	size = i_size_read(inode);
4377	/* Page got truncated from under us? */
4378	if (page->mapping != mapping || page_offset(page) > size) {
4379		unlock_page(page);
4380		ret = VM_FAULT_NOPAGE;
4381		goto out;
4382	}
4383
4384	if (page->index == size >> PAGE_CACHE_SHIFT)
4385		len = size & ~PAGE_CACHE_MASK;
4386	else
4387		len = PAGE_CACHE_SIZE;
4388	/*
4389	 * Return if we have all the buffers mapped. This avoids the need to do
4390	 * journal_start/journal_stop which can block and take a long time
4391	 */
4392	if (page_has_buffers(page)) {
4393		if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4394					ext4_bh_unmapped)) {
 
4395			/* Wait so that we don't change page under IO */
4396			wait_on_page_writeback(page);
4397			ret = VM_FAULT_LOCKED;
4398			goto out;
4399		}
4400	}
4401	unlock_page(page);
4402	/* OK, we need to fill the hole... */
4403	if (ext4_should_dioread_nolock(inode))
4404		get_block = ext4_get_block_write;
4405	else
4406		get_block = ext4_get_block;
4407retry_alloc:
4408	handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
 
4409	if (IS_ERR(handle)) {
4410		ret = VM_FAULT_SIGBUS;
4411		goto out;
4412	}
4413	ret = __block_page_mkwrite(vma, vmf, get_block);
4414	if (!ret && ext4_should_journal_data(inode)) {
4415		if (walk_page_buffers(handle, page_buffers(page), 0,
4416			  PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4417			unlock_page(page);
4418			ret = VM_FAULT_SIGBUS;
 
4419			goto out;
4420		}
4421		ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4422	}
4423	ext4_journal_stop(handle);
4424	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4425		goto retry_alloc;
4426out_ret:
4427	ret = block_page_mkwrite_return(ret);
4428out:
 
 
4429	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4430}
v4.10.11
   1/*
   2 *  linux/fs/ext4/inode.c
   3 *
   4 * Copyright (C) 1992, 1993, 1994, 1995
   5 * Remy Card (card@masi.ibp.fr)
   6 * Laboratoire MASI - Institut Blaise Pascal
   7 * Universite Pierre et Marie Curie (Paris VI)
   8 *
   9 *  from
  10 *
  11 *  linux/fs/minix/inode.c
  12 *
  13 *  Copyright (C) 1991, 1992  Linus Torvalds
  14 *
  15 *  64-bit file support on 64-bit platforms by Jakub Jelinek
  16 *	(jj@sunsite.ms.mff.cuni.cz)
  17 *
  18 *  Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
  19 */
  20
 
  21#include <linux/fs.h>
  22#include <linux/time.h>
 
  23#include <linux/highuid.h>
  24#include <linux/pagemap.h>
  25#include <linux/dax.h>
  26#include <linux/quotaops.h>
  27#include <linux/string.h>
  28#include <linux/buffer_head.h>
  29#include <linux/writeback.h>
  30#include <linux/pagevec.h>
  31#include <linux/mpage.h>
  32#include <linux/namei.h>
  33#include <linux/uio.h>
  34#include <linux/bio.h>
  35#include <linux/workqueue.h>
  36#include <linux/kernel.h>
  37#include <linux/printk.h>
  38#include <linux/slab.h>
  39#include <linux/bitops.h>
  40#include <linux/iomap.h>
  41
  42#include "ext4_jbd2.h"
  43#include "xattr.h"
  44#include "acl.h"
 
  45#include "truncate.h"
  46
  47#include <trace/events/ext4.h>
  48
  49#define MPAGE_DA_EXTENT_TAIL 0x01
  50
  51static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
  52			      struct ext4_inode_info *ei)
  53{
  54	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
  55	__u32 csum;
  56	__u16 dummy_csum = 0;
  57	int offset = offsetof(struct ext4_inode, i_checksum_lo);
  58	unsigned int csum_size = sizeof(dummy_csum);
  59
  60	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
  61	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
  62	offset += csum_size;
  63	csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  64			   EXT4_GOOD_OLD_INODE_SIZE - offset);
  65
  66	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
  67		offset = offsetof(struct ext4_inode, i_checksum_hi);
  68		csum = ext4_chksum(sbi, csum, (__u8 *)raw +
  69				   EXT4_GOOD_OLD_INODE_SIZE,
  70				   offset - EXT4_GOOD_OLD_INODE_SIZE);
  71		if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
  72			csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
  73					   csum_size);
  74			offset += csum_size;
  75		}
  76		csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
  77				   EXT4_INODE_SIZE(inode->i_sb) - offset);
  78	}
  79
  80	return csum;
  81}
  82
  83static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
  84				  struct ext4_inode_info *ei)
  85{
  86	__u32 provided, calculated;
  87
  88	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
  89	    cpu_to_le32(EXT4_OS_LINUX) ||
  90	    !ext4_has_metadata_csum(inode->i_sb))
  91		return 1;
  92
  93	provided = le16_to_cpu(raw->i_checksum_lo);
  94	calculated = ext4_inode_csum(inode, raw, ei);
  95	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
  96	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
  97		provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
  98	else
  99		calculated &= 0xFFFF;
 100
 101	return provided == calculated;
 102}
 103
 104static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
 105				struct ext4_inode_info *ei)
 106{
 107	__u32 csum;
 108
 109	if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
 110	    cpu_to_le32(EXT4_OS_LINUX) ||
 111	    !ext4_has_metadata_csum(inode->i_sb))
 112		return;
 113
 114	csum = ext4_inode_csum(inode, raw, ei);
 115	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
 116	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
 117	    EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
 118		raw->i_checksum_hi = cpu_to_le16(csum >> 16);
 119}
 120
 121static inline int ext4_begin_ordered_truncate(struct inode *inode,
 122					      loff_t new_size)
 123{
 124	trace_ext4_begin_ordered_truncate(inode, new_size);
 125	/*
 126	 * If jinode is zero, then we never opened the file for
 127	 * writing, so there's no need to call
 128	 * jbd2_journal_begin_ordered_truncate() since there's no
 129	 * outstanding writes we need to flush.
 130	 */
 131	if (!EXT4_I(inode)->jinode)
 132		return 0;
 133	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
 134						   EXT4_I(inode)->jinode,
 135						   new_size);
 136}
 137
 138static void ext4_invalidatepage(struct page *page, unsigned int offset,
 139				unsigned int length);
 
 
 
 140static int __ext4_journalled_writepage(struct page *page, unsigned int len);
 141static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
 142static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 143				  int pextents);
 144
 145/*
 146 * Test whether an inode is a fast symlink.
 147 */
 148int ext4_inode_is_fast_symlink(struct inode *inode)
 149{
 150        int ea_blocks = EXT4_I(inode)->i_file_acl ?
 151		EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
 152
 153	if (ext4_has_inline_data(inode))
 154		return 0;
 155
 156	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
 157}
 158
 159/*
 160 * Restart the transaction associated with *handle.  This does a commit,
 161 * so before we call here everything must be consistently dirtied against
 162 * this transaction.
 163 */
 164int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
 165				 int nblocks)
 166{
 167	int ret;
 168
 169	/*
 170	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
 171	 * moment, get_block can be called only for blocks inside i_size since
 172	 * page cache has been already dropped and writes are blocked by
 173	 * i_mutex. So we can safely drop the i_data_sem here.
 174	 */
 175	BUG_ON(EXT4_JOURNAL(inode) == NULL);
 176	jbd_debug(2, "restarting handle %p\n", handle);
 177	up_write(&EXT4_I(inode)->i_data_sem);
 178	ret = ext4_journal_restart(handle, nblocks);
 179	down_write(&EXT4_I(inode)->i_data_sem);
 180	ext4_discard_preallocations(inode);
 181
 182	return ret;
 183}
 184
 185/*
 186 * Called at the last iput() if i_nlink is zero.
 187 */
 188void ext4_evict_inode(struct inode *inode)
 189{
 190	handle_t *handle;
 191	int err;
 192
 193	trace_ext4_evict_inode(inode);
 194
 
 
 195	if (inode->i_nlink) {
 196		/*
 197		 * When journalling data dirty buffers are tracked only in the
 198		 * journal. So although mm thinks everything is clean and
 199		 * ready for reaping the inode might still have some pages to
 200		 * write in the running transaction or waiting to be
 201		 * checkpointed. Thus calling jbd2_journal_invalidatepage()
 202		 * (via truncate_inode_pages()) to discard these buffers can
 203		 * cause data loss. Also even if we did not discard these
 204		 * buffers, we would have no way to find them after the inode
 205		 * is reaped and thus user could see stale data if he tries to
 206		 * read them before the transaction is checkpointed. So be
 207		 * careful and force everything to disk here... We use
 208		 * ei->i_datasync_tid to store the newest transaction
 209		 * containing inode's data.
 210		 *
 211		 * Note that directories do not have this problem because they
 212		 * don't use page cache.
 213		 */
 214		if (inode->i_ino != EXT4_JOURNAL_INO &&
 215		    ext4_should_journal_data(inode) &&
 216		    (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
 217			journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 218			tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
 219
 220			jbd2_complete_transaction(journal, commit_tid);
 
 221			filemap_write_and_wait(&inode->i_data);
 222		}
 223		truncate_inode_pages_final(&inode->i_data);
 224
 225		goto no_delete;
 226	}
 227
 228	if (is_bad_inode(inode))
 229		goto no_delete;
 230	dquot_initialize(inode);
 231
 232	if (ext4_should_order_data(inode))
 233		ext4_begin_ordered_truncate(inode, 0);
 234	truncate_inode_pages_final(&inode->i_data);
 
 
 
 235
 236	/*
 237	 * Protect us against freezing - iput() caller didn't have to have any
 238	 * protection against it
 239	 */
 240	sb_start_intwrite(inode->i_sb);
 241	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
 242				    ext4_blocks_for_truncate(inode)+3);
 243	if (IS_ERR(handle)) {
 244		ext4_std_error(inode->i_sb, PTR_ERR(handle));
 245		/*
 246		 * If we're going to skip the normal cleanup, we still need to
 247		 * make sure that the in-core orphan linked list is properly
 248		 * cleaned up.
 249		 */
 250		ext4_orphan_del(NULL, inode);
 251		sb_end_intwrite(inode->i_sb);
 252		goto no_delete;
 253	}
 254
 255	if (IS_SYNC(inode))
 256		ext4_handle_sync(handle);
 257	inode->i_size = 0;
 258	err = ext4_mark_inode_dirty(handle, inode);
 259	if (err) {
 260		ext4_warning(inode->i_sb,
 261			     "couldn't mark inode dirty (err %d)", err);
 262		goto stop_handle;
 263	}
 264	if (inode->i_blocks) {
 265		err = ext4_truncate(inode);
 266		if (err) {
 267			ext4_error(inode->i_sb,
 268				   "couldn't truncate inode %lu (err %d)",
 269				   inode->i_ino, err);
 270			goto stop_handle;
 271		}
 272	}
 273
 274	/*
 275	 * ext4_ext_truncate() doesn't reserve any slop when it
 276	 * restarts journal transactions; therefore there may not be
 277	 * enough credits left in the handle to remove the inode from
 278	 * the orphan list and set the dtime field.
 279	 */
 280	if (!ext4_handle_has_enough_credits(handle, 3)) {
 281		err = ext4_journal_extend(handle, 3);
 282		if (err > 0)
 283			err = ext4_journal_restart(handle, 3);
 284		if (err != 0) {
 285			ext4_warning(inode->i_sb,
 286				     "couldn't extend journal (err %d)", err);
 287		stop_handle:
 288			ext4_journal_stop(handle);
 289			ext4_orphan_del(NULL, inode);
 290			sb_end_intwrite(inode->i_sb);
 291			goto no_delete;
 292		}
 293	}
 294
 295	/*
 296	 * Kill off the orphan record which ext4_truncate created.
 297	 * AKPM: I think this can be inside the above `if'.
 298	 * Note that ext4_orphan_del() has to be able to cope with the
 299	 * deletion of a non-existent orphan - this is because we don't
 300	 * know if ext4_truncate() actually created an orphan record.
 301	 * (Well, we could do this if we need to, but heck - it works)
 302	 */
 303	ext4_orphan_del(handle, inode);
 304	EXT4_I(inode)->i_dtime	= get_seconds();
 305
 306	/*
 307	 * One subtle ordering requirement: if anything has gone wrong
 308	 * (transaction abort, IO errors, whatever), then we can still
 309	 * do these next steps (the fs will already have been marked as
 310	 * having errors), but we can't free the inode if the mark_dirty
 311	 * fails.
 312	 */
 313	if (ext4_mark_inode_dirty(handle, inode))
 314		/* If that failed, just do the required in-core inode clear. */
 315		ext4_clear_inode(inode);
 316	else
 317		ext4_free_inode(handle, inode);
 318	ext4_journal_stop(handle);
 319	sb_end_intwrite(inode->i_sb);
 320	return;
 321no_delete:
 322	ext4_clear_inode(inode);	/* We must guarantee clearing of inode... */
 323}
 324
 325#ifdef CONFIG_QUOTA
 326qsize_t *ext4_get_reserved_space(struct inode *inode)
 327{
 328	return &EXT4_I(inode)->i_reserved_quota;
 329}
 330#endif
 331
 332/*
 
 
 
 
 
 
 
 
 
 
 
 
 333 * Called with i_data_sem down, which is important since we can call
 334 * ext4_discard_preallocations() from here.
 335 */
 336void ext4_da_update_reserve_space(struct inode *inode,
 337					int used, int quota_claim)
 338{
 339	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 340	struct ext4_inode_info *ei = EXT4_I(inode);
 341
 342	spin_lock(&ei->i_block_reservation_lock);
 343	trace_ext4_da_update_reserve_space(inode, used, quota_claim);
 344	if (unlikely(used > ei->i_reserved_data_blocks)) {
 345		ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
 346			 "with only %d reserved data blocks",
 347			 __func__, inode->i_ino, used,
 348			 ei->i_reserved_data_blocks);
 349		WARN_ON(1);
 350		used = ei->i_reserved_data_blocks;
 351	}
 352
 353	/* Update per-inode reservations */
 354	ei->i_reserved_data_blocks -= used;
 355	percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
 356
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 357	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 358
 359	/* Update quota subsystem for data blocks */
 360	if (quota_claim)
 361		dquot_claim_block(inode, EXT4_C2B(sbi, used));
 362	else {
 363		/*
 364		 * We did fallocate with an offset that is already delayed
 365		 * allocated. So on delayed allocated writeback we should
 366		 * not re-claim the quota for fallocated blocks.
 367		 */
 368		dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
 369	}
 370
 371	/*
 372	 * If we have done all the pending block allocations and if
 373	 * there aren't any writers on the inode, we can discard the
 374	 * inode's preallocations.
 375	 */
 376	if ((ei->i_reserved_data_blocks == 0) &&
 377	    (atomic_read(&inode->i_writecount) == 0))
 378		ext4_discard_preallocations(inode);
 379}
 380
 381static int __check_block_validity(struct inode *inode, const char *func,
 382				unsigned int line,
 383				struct ext4_map_blocks *map)
 384{
 385	if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
 386				   map->m_len)) {
 387		ext4_error_inode(inode, func, line, map->m_pblk,
 388				 "lblock %lu mapped to illegal pblock "
 389				 "(length %d)", (unsigned long) map->m_lblk,
 390				 map->m_len);
 391		return -EFSCORRUPTED;
 392	}
 393	return 0;
 394}
 395
 396int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
 397		       ext4_lblk_t len)
 398{
 399	int ret;
 400
 401	if (ext4_encrypted_inode(inode))
 402		return fscrypt_zeroout_range(inode, lblk, pblk, len);
 403
 404	ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
 405	if (ret > 0)
 406		ret = 0;
 407
 408	return ret;
 409}
 410
 411#define check_block_validity(inode, map)	\
 412	__check_block_validity((inode), __func__, __LINE__, (map))
 413
 414#ifdef ES_AGGRESSIVE_TEST
 415static void ext4_map_blocks_es_recheck(handle_t *handle,
 416				       struct inode *inode,
 417				       struct ext4_map_blocks *es_map,
 418				       struct ext4_map_blocks *map,
 419				       int flags)
 420{
 421	int retval;
 
 
 
 
 422
 423	map->m_flags = 0;
 424	/*
 425	 * There is a race window that the result is not the same.
 426	 * e.g. xfstests #223 when dioread_nolock enables.  The reason
 427	 * is that we lookup a block mapping in extent status tree with
 428	 * out taking i_data_sem.  So at the time the unwritten extent
 429	 * could be converted.
 430	 */
 431	down_read(&EXT4_I(inode)->i_data_sem);
 432	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 433		retval = ext4_ext_map_blocks(handle, inode, map, flags &
 434					     EXT4_GET_BLOCKS_KEEP_SIZE);
 435	} else {
 436		retval = ext4_ind_map_blocks(handle, inode, map, flags &
 437					     EXT4_GET_BLOCKS_KEEP_SIZE);
 438	}
 439	up_read((&EXT4_I(inode)->i_data_sem));
 440
 441	/*
 442	 * We don't check m_len because extent will be collpased in status
 443	 * tree.  So the m_len might not equal.
 444	 */
 445	if (es_map->m_lblk != map->m_lblk ||
 446	    es_map->m_flags != map->m_flags ||
 447	    es_map->m_pblk != map->m_pblk) {
 448		printk("ES cache assertion failed for inode: %lu "
 449		       "es_cached ex [%d/%d/%llu/%x] != "
 450		       "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
 451		       inode->i_ino, es_map->m_lblk, es_map->m_len,
 452		       es_map->m_pblk, es_map->m_flags, map->m_lblk,
 453		       map->m_len, map->m_pblk, map->m_flags,
 454		       retval, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 455	}
 
 456}
 457#endif /* ES_AGGRESSIVE_TEST */
 458
 459/*
 460 * The ext4_map_blocks() function tries to look up the requested blocks,
 461 * and returns if the blocks are already mapped.
 462 *
 463 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
 464 * and store the allocated blocks in the result buffer head and mark it
 465 * mapped.
 466 *
 467 * If file type is extents based, it will call ext4_ext_map_blocks(),
 468 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
 469 * based files
 470 *
 471 * On success, it returns the number of blocks being mapped or allocated.  if
 472 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
 473 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
 
 474 *
 475 * It returns 0 if plain look up failed (blocks have not been allocated), in
 476 * that case, @map is returned as unmapped but we still do fill map->m_len to
 477 * indicate the length of a hole starting at map->m_lblk.
 478 *
 479 * It returns the error in case of allocation failure.
 480 */
 481int ext4_map_blocks(handle_t *handle, struct inode *inode,
 482		    struct ext4_map_blocks *map, int flags)
 483{
 484	struct extent_status es;
 485	int retval;
 486	int ret = 0;
 487#ifdef ES_AGGRESSIVE_TEST
 488	struct ext4_map_blocks orig_map;
 489
 490	memcpy(&orig_map, map, sizeof(*map));
 491#endif
 492
 493	map->m_flags = 0;
 494	ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
 495		  "logical block %lu\n", inode->i_ino, flags, map->m_len,
 496		  (unsigned long) map->m_lblk);
 497
 498	/*
 499	 * ext4_map_blocks returns an int, and m_len is an unsigned int
 500	 */
 501	if (unlikely(map->m_len > INT_MAX))
 502		map->m_len = INT_MAX;
 503
 504	/* We can handle the block number less than EXT_MAX_BLOCKS */
 505	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
 506		return -EFSCORRUPTED;
 507
 508	/* Lookup extent status tree firstly */
 509	if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
 510		if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
 511			map->m_pblk = ext4_es_pblock(&es) +
 512					map->m_lblk - es.es_lblk;
 513			map->m_flags |= ext4_es_is_written(&es) ?
 514					EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
 515			retval = es.es_len - (map->m_lblk - es.es_lblk);
 516			if (retval > map->m_len)
 517				retval = map->m_len;
 518			map->m_len = retval;
 519		} else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
 520			map->m_pblk = 0;
 521			retval = es.es_len - (map->m_lblk - es.es_lblk);
 522			if (retval > map->m_len)
 523				retval = map->m_len;
 524			map->m_len = retval;
 525			retval = 0;
 526		} else {
 527			BUG_ON(1);
 528		}
 529#ifdef ES_AGGRESSIVE_TEST
 530		ext4_map_blocks_es_recheck(handle, inode, map,
 531					   &orig_map, flags);
 532#endif
 533		goto found;
 534	}
 535
 536	/*
 537	 * Try to see if we can get the block without requesting a new
 538	 * file system block.
 539	 */
 540	down_read(&EXT4_I(inode)->i_data_sem);
 541	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 542		retval = ext4_ext_map_blocks(handle, inode, map, flags &
 543					     EXT4_GET_BLOCKS_KEEP_SIZE);
 544	} else {
 545		retval = ext4_ind_map_blocks(handle, inode, map, flags &
 546					     EXT4_GET_BLOCKS_KEEP_SIZE);
 547	}
 548	if (retval > 0) {
 549		unsigned int status;
 550
 551		if (unlikely(retval != map->m_len)) {
 552			ext4_warning(inode->i_sb,
 553				     "ES len assertion failed for inode "
 554				     "%lu: retval %d != map->m_len %d",
 555				     inode->i_ino, retval, map->m_len);
 556			WARN_ON(1);
 557		}
 558
 559		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 560				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 561		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 562		    !(status & EXTENT_STATUS_WRITTEN) &&
 563		    ext4_find_delalloc_range(inode, map->m_lblk,
 564					     map->m_lblk + map->m_len - 1))
 565			status |= EXTENT_STATUS_DELAYED;
 566		ret = ext4_es_insert_extent(inode, map->m_lblk,
 567					    map->m_len, map->m_pblk, status);
 568		if (ret < 0)
 569			retval = ret;
 570	}
 571	up_read((&EXT4_I(inode)->i_data_sem));
 572
 573found:
 574	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 575		ret = check_block_validity(inode, map);
 576		if (ret != 0)
 577			return ret;
 578	}
 579
 580	/* If it is only a block(s) look up */
 581	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
 582		return retval;
 583
 584	/*
 585	 * Returns if the blocks have already allocated
 586	 *
 587	 * Note that if blocks have been preallocated
 588	 * ext4_ext_get_block() returns the create = 0
 589	 * with buffer head unmapped.
 590	 */
 591	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
 592		/*
 593		 * If we need to convert extent to unwritten
 594		 * we continue and do the actual work in
 595		 * ext4_ext_map_blocks()
 596		 */
 597		if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
 598			return retval;
 599
 600	/*
 601	 * Here we clear m_flags because after allocating an new extent,
 602	 * it will be set again.
 
 
 
 
 
 
 603	 */
 604	map->m_flags &= ~EXT4_MAP_FLAGS;
 605
 606	/*
 607	 * New blocks allocate and/or writing to unwritten extent
 608	 * will possibly result in updating i_data, so we take
 609	 * the write lock of i_data_sem, and call get_block()
 610	 * with create == 1 flag.
 611	 */
 612	down_write(&EXT4_I(inode)->i_data_sem);
 613
 614	/*
 
 
 
 
 
 
 
 
 615	 * We need to check for EXT4 here because migrate
 616	 * could have changed the inode type in between
 617	 */
 618	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
 619		retval = ext4_ext_map_blocks(handle, inode, map, flags);
 620	} else {
 621		retval = ext4_ind_map_blocks(handle, inode, map, flags);
 622
 623		if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
 624			/*
 625			 * We allocated new blocks which will result in
 626			 * i_data's format changing.  Force the migrate
 627			 * to fail by clearing migrate flags
 628			 */
 629			ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 630		}
 631
 632		/*
 633		 * Update reserved blocks/metadata blocks after successful
 634		 * block allocation which had been deferred till now. We don't
 635		 * support fallocate for non extent files. So we can update
 636		 * reserve space here.
 637		 */
 638		if ((retval > 0) &&
 639			(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
 640			ext4_da_update_reserve_space(inode, retval, 1);
 641	}
 
 
 642
 643	if (retval > 0) {
 644		unsigned int status;
 645
 646		if (unlikely(retval != map->m_len)) {
 647			ext4_warning(inode->i_sb,
 648				     "ES len assertion failed for inode "
 649				     "%lu: retval %d != map->m_len %d",
 650				     inode->i_ino, retval, map->m_len);
 651			WARN_ON(1);
 652		}
 653
 654		/*
 655		 * We have to zeroout blocks before inserting them into extent
 656		 * status tree. Otherwise someone could look them up there and
 657		 * use them before they are really zeroed. We also have to
 658		 * unmap metadata before zeroing as otherwise writeback can
 659		 * overwrite zeros with stale data from block device.
 660		 */
 661		if (flags & EXT4_GET_BLOCKS_ZERO &&
 662		    map->m_flags & EXT4_MAP_MAPPED &&
 663		    map->m_flags & EXT4_MAP_NEW) {
 664			clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
 665					   map->m_len);
 666			ret = ext4_issue_zeroout(inode, map->m_lblk,
 667						 map->m_pblk, map->m_len);
 668			if (ret) {
 669				retval = ret;
 670				goto out_sem;
 671			}
 672		}
 673
 674		/*
 675		 * If the extent has been zeroed out, we don't need to update
 676		 * extent status tree.
 677		 */
 678		if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
 679		    ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
 680			if (ext4_es_is_written(&es))
 681				goto out_sem;
 682		}
 683		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
 684				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
 685		if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
 686		    !(status & EXTENT_STATUS_WRITTEN) &&
 687		    ext4_find_delalloc_range(inode, map->m_lblk,
 688					     map->m_lblk + map->m_len - 1))
 689			status |= EXTENT_STATUS_DELAYED;
 690		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
 691					    map->m_pblk, status);
 692		if (ret < 0) {
 693			retval = ret;
 694			goto out_sem;
 695		}
 696	}
 697
 698out_sem:
 699	up_write((&EXT4_I(inode)->i_data_sem));
 700	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
 701		ret = check_block_validity(inode, map);
 702		if (ret != 0)
 703			return ret;
 704
 705		/*
 706		 * Inodes with freshly allocated blocks where contents will be
 707		 * visible after transaction commit must be on transaction's
 708		 * ordered data list.
 709		 */
 710		if (map->m_flags & EXT4_MAP_NEW &&
 711		    !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
 712		    !(flags & EXT4_GET_BLOCKS_ZERO) &&
 713		    !IS_NOQUOTA(inode) &&
 714		    ext4_should_order_data(inode)) {
 715			if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
 716				ret = ext4_jbd2_inode_add_wait(handle, inode);
 717			else
 718				ret = ext4_jbd2_inode_add_write(handle, inode);
 719			if (ret)
 720				return ret;
 721		}
 722	}
 723	return retval;
 724}
 725
 726/*
 727 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
 728 * we have to be careful as someone else may be manipulating b_state as well.
 729 */
 730static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
 731{
 732	unsigned long old_state;
 733	unsigned long new_state;
 734
 735	flags &= EXT4_MAP_FLAGS;
 736
 737	/* Dummy buffer_head? Set non-atomically. */
 738	if (!bh->b_page) {
 739		bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
 740		return;
 741	}
 742	/*
 743	 * Someone else may be modifying b_state. Be careful! This is ugly but
 744	 * once we get rid of using bh as a container for mapping information
 745	 * to pass to / from get_block functions, this can go away.
 746	 */
 747	do {
 748		old_state = READ_ONCE(bh->b_state);
 749		new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
 750	} while (unlikely(
 751		 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
 752}
 753
 754static int _ext4_get_block(struct inode *inode, sector_t iblock,
 755			   struct buffer_head *bh, int flags)
 756{
 
 757	struct ext4_map_blocks map;
 758	int ret = 0;
 759
 760	if (ext4_has_inline_data(inode))
 761		return -ERANGE;
 762
 763	map.m_lblk = iblock;
 764	map.m_len = bh->b_size >> inode->i_blkbits;
 765
 766	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
 767			      flags);
 
 
 
 
 
 
 
 
 
 
 
 
 768	if (ret > 0) {
 769		map_bh(bh, inode->i_sb, map.m_pblk);
 770		ext4_update_bh_state(bh, map.m_flags);
 771		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 772		ret = 0;
 773	} else if (ret == 0) {
 774		/* hole case, need to fill in bh->b_size */
 775		bh->b_size = inode->i_sb->s_blocksize * map.m_len;
 776	}
 
 
 777	return ret;
 778}
 779
 780int ext4_get_block(struct inode *inode, sector_t iblock,
 781		   struct buffer_head *bh, int create)
 782{
 783	return _ext4_get_block(inode, iblock, bh,
 784			       create ? EXT4_GET_BLOCKS_CREATE : 0);
 785}
 786
 787/*
 788 * Get block function used when preparing for buffered write if we require
 789 * creating an unwritten extent if blocks haven't been allocated.  The extent
 790 * will be converted to written after the IO is complete.
 791 */
 792int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
 793			     struct buffer_head *bh_result, int create)
 794{
 795	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
 796		   inode->i_ino, create);
 797	return _ext4_get_block(inode, iblock, bh_result,
 798			       EXT4_GET_BLOCKS_IO_CREATE_EXT);
 799}
 800
 801/* Maximum number of blocks we map for direct IO at once. */
 802#define DIO_MAX_BLOCKS 4096
 803
 804/*
 805 * Get blocks function for the cases that need to start a transaction -
 806 * generally difference cases of direct IO and DAX IO. It also handles retries
 807 * in case of ENOSPC.
 808 */
 809static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
 810				struct buffer_head *bh_result, int flags)
 811{
 812	int dio_credits;
 813	handle_t *handle;
 814	int retries = 0;
 815	int ret;
 816
 817	/* Trim mapping request to maximum we can map at once for DIO */
 818	if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
 819		bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
 820	dio_credits = ext4_chunk_trans_blocks(inode,
 821				      bh_result->b_size >> inode->i_blkbits);
 822retry:
 823	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
 824	if (IS_ERR(handle))
 825		return PTR_ERR(handle);
 826
 827	ret = _ext4_get_block(inode, iblock, bh_result, flags);
 828	ext4_journal_stop(handle);
 829
 830	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 831		goto retry;
 832	return ret;
 833}
 834
 835/* Get block function for DIO reads and writes to inodes without extents */
 836int ext4_dio_get_block(struct inode *inode, sector_t iblock,
 837		       struct buffer_head *bh, int create)
 838{
 839	/* We don't expect handle for direct IO */
 840	WARN_ON_ONCE(ext4_journal_current_handle());
 841
 842	if (!create)
 843		return _ext4_get_block(inode, iblock, bh, 0);
 844	return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
 845}
 846
 847/*
 848 * Get block function for AIO DIO writes when we create unwritten extent if
 849 * blocks are not allocated yet. The extent will be converted to written
 850 * after IO is complete.
 851 */
 852static int ext4_dio_get_block_unwritten_async(struct inode *inode,
 853		sector_t iblock, struct buffer_head *bh_result,	int create)
 854{
 855	int ret;
 856
 857	/* We don't expect handle for direct IO */
 858	WARN_ON_ONCE(ext4_journal_current_handle());
 859
 860	ret = ext4_get_block_trans(inode, iblock, bh_result,
 861				   EXT4_GET_BLOCKS_IO_CREATE_EXT);
 862
 863	/*
 864	 * When doing DIO using unwritten extents, we need io_end to convert
 865	 * unwritten extents to written on IO completion. We allocate io_end
 866	 * once we spot unwritten extent and store it in b_private. Generic
 867	 * DIO code keeps b_private set and furthermore passes the value to
 868	 * our completion callback in 'private' argument.
 869	 */
 870	if (!ret && buffer_unwritten(bh_result)) {
 871		if (!bh_result->b_private) {
 872			ext4_io_end_t *io_end;
 873
 874			io_end = ext4_init_io_end(inode, GFP_KERNEL);
 875			if (!io_end)
 876				return -ENOMEM;
 877			bh_result->b_private = io_end;
 878			ext4_set_io_unwritten_flag(inode, io_end);
 879		}
 880		set_buffer_defer_completion(bh_result);
 881	}
 882
 883	return ret;
 884}
 885
 886/*
 887 * Get block function for non-AIO DIO writes when we create unwritten extent if
 888 * blocks are not allocated yet. The extent will be converted to written
 889 * after IO is complete from ext4_ext_direct_IO() function.
 890 */
 891static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
 892		sector_t iblock, struct buffer_head *bh_result,	int create)
 893{
 894	int ret;
 895
 896	/* We don't expect handle for direct IO */
 897	WARN_ON_ONCE(ext4_journal_current_handle());
 898
 899	ret = ext4_get_block_trans(inode, iblock, bh_result,
 900				   EXT4_GET_BLOCKS_IO_CREATE_EXT);
 901
 902	/*
 903	 * Mark inode as having pending DIO writes to unwritten extents.
 904	 * ext4_ext_direct_IO() checks this flag and converts extents to
 905	 * written.
 906	 */
 907	if (!ret && buffer_unwritten(bh_result))
 908		ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
 909
 910	return ret;
 911}
 912
 913static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
 914		   struct buffer_head *bh_result, int create)
 915{
 916	int ret;
 917
 918	ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
 919		   inode->i_ino, create);
 920	/* We don't expect handle for direct IO */
 921	WARN_ON_ONCE(ext4_journal_current_handle());
 922
 923	ret = _ext4_get_block(inode, iblock, bh_result, 0);
 924	/*
 925	 * Blocks should have been preallocated! ext4_file_write_iter() checks
 926	 * that.
 927	 */
 928	WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
 929
 930	return ret;
 931}
 932
 933
 934/*
 935 * `handle' can be NULL if create is zero
 936 */
 937struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 938				ext4_lblk_t block, int map_flags)
 939{
 940	struct ext4_map_blocks map;
 941	struct buffer_head *bh;
 942	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
 943	int err;
 944
 945	J_ASSERT(handle != NULL || create == 0);
 946
 947	map.m_lblk = block;
 948	map.m_len = 1;
 949	err = ext4_map_blocks(handle, inode, &map, map_flags);
 
 950
 951	if (err == 0)
 952		return create ? ERR_PTR(-ENOSPC) : NULL;
 953	if (err < 0)
 954		return ERR_PTR(err);
 
 
 
 955
 956	bh = sb_getblk(inode->i_sb, map.m_pblk);
 957	if (unlikely(!bh))
 958		return ERR_PTR(-ENOMEM);
 
 
 959	if (map.m_flags & EXT4_MAP_NEW) {
 960		J_ASSERT(create != 0);
 961		J_ASSERT(handle != NULL);
 962
 963		/*
 964		 * Now that we do not always journal data, we should
 965		 * keep in mind whether this should always journal the
 966		 * new buffer as metadata.  For now, regular file
 967		 * writes use ext4_get_block instead, so it's not a
 968		 * problem.
 969		 */
 970		lock_buffer(bh);
 971		BUFFER_TRACE(bh, "call get_create_access");
 972		err = ext4_journal_get_create_access(handle, bh);
 973		if (unlikely(err)) {
 974			unlock_buffer(bh);
 975			goto errout;
 976		}
 977		if (!buffer_uptodate(bh)) {
 978			memset(bh->b_data, 0, inode->i_sb->s_blocksize);
 979			set_buffer_uptodate(bh);
 980		}
 981		unlock_buffer(bh);
 982		BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
 983		err = ext4_handle_dirty_metadata(handle, inode, bh);
 984		if (unlikely(err))
 985			goto errout;
 986	} else
 987		BUFFER_TRACE(bh, "not a new buffer");
 
 
 
 
 
 
 988	return bh;
 989errout:
 990	brelse(bh);
 991	return ERR_PTR(err);
 992}
 993
 994struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
 995			       ext4_lblk_t block, int map_flags)
 996{
 997	struct buffer_head *bh;
 998
 999	bh = ext4_getblk(handle, inode, block, map_flags);
1000	if (IS_ERR(bh))
1001		return bh;
1002	if (!bh || buffer_uptodate(bh))
1003		return bh;
1004	ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1005	wait_on_buffer(bh);
1006	if (buffer_uptodate(bh))
1007		return bh;
1008	put_bh(bh);
1009	return ERR_PTR(-EIO);
 
1010}
1011
1012int ext4_walk_page_buffers(handle_t *handle,
1013			   struct buffer_head *head,
1014			   unsigned from,
1015			   unsigned to,
1016			   int *partial,
1017			   int (*fn)(handle_t *handle,
1018				     struct buffer_head *bh))
1019{
1020	struct buffer_head *bh;
1021	unsigned block_start, block_end;
1022	unsigned blocksize = head->b_size;
1023	int err, ret = 0;
1024	struct buffer_head *next;
1025
1026	for (bh = head, block_start = 0;
1027	     ret == 0 && (bh != head || !block_start);
1028	     block_start = block_end, bh = next) {
1029		next = bh->b_this_page;
1030		block_end = block_start + blocksize;
1031		if (block_end <= from || block_start >= to) {
1032			if (partial && !buffer_uptodate(bh))
1033				*partial = 1;
1034			continue;
1035		}
1036		err = (*fn)(handle, bh);
1037		if (!ret)
1038			ret = err;
1039	}
1040	return ret;
1041}
1042
1043/*
1044 * To preserve ordering, it is essential that the hole instantiation and
1045 * the data write be encapsulated in a single transaction.  We cannot
1046 * close off a transaction and start a new one between the ext4_get_block()
1047 * and the commit_write().  So doing the jbd2_journal_start at the start of
1048 * prepare_write() is the right place.
1049 *
1050 * Also, this function can nest inside ext4_writepage().  In that case, we
1051 * *know* that ext4_writepage() has generated enough buffer credits to do the
1052 * whole page.  So we won't block on the journal in that case, which is good,
1053 * because the caller may be PF_MEMALLOC.
 
1054 *
1055 * By accident, ext4 can be reentered when a transaction is open via
1056 * quota file writes.  If we were to commit the transaction while thus
1057 * reentered, there can be a deadlock - we would be holding a quota
1058 * lock, and the commit would never complete if another thread had a
1059 * transaction open and was blocking on the quota lock - a ranking
1060 * violation.
1061 *
1062 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1063 * will _not_ run commit under these circumstances because handle->h_ref
1064 * is elevated.  We'll still have enough credits for the tiny quotafile
1065 * write.
1066 */
1067int do_journal_get_write_access(handle_t *handle,
1068				struct buffer_head *bh)
1069{
1070	int dirty = buffer_dirty(bh);
1071	int ret;
1072
1073	if (!buffer_mapped(bh) || buffer_freed(bh))
1074		return 0;
1075	/*
1076	 * __block_write_begin() could have dirtied some buffers. Clean
1077	 * the dirty bit as jbd2_journal_get_write_access() could complain
1078	 * otherwise about fs integrity issues. Setting of the dirty bit
1079	 * by __block_write_begin() isn't a real problem here as we clear
1080	 * the bit before releasing a page lock and thus writeback cannot
1081	 * ever write the buffer.
1082	 */
1083	if (dirty)
1084		clear_buffer_dirty(bh);
1085	BUFFER_TRACE(bh, "get write access");
1086	ret = ext4_journal_get_write_access(handle, bh);
1087	if (!ret && dirty)
1088		ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1089	return ret;
1090}
1091
1092#ifdef CONFIG_EXT4_FS_ENCRYPTION
1093static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1094				  get_block_t *get_block)
1095{
1096	unsigned from = pos & (PAGE_SIZE - 1);
1097	unsigned to = from + len;
1098	struct inode *inode = page->mapping->host;
1099	unsigned block_start, block_end;
1100	sector_t block;
1101	int err = 0;
1102	unsigned blocksize = inode->i_sb->s_blocksize;
1103	unsigned bbits;
1104	struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1105	bool decrypt = false;
1106
1107	BUG_ON(!PageLocked(page));
1108	BUG_ON(from > PAGE_SIZE);
1109	BUG_ON(to > PAGE_SIZE);
1110	BUG_ON(from > to);
1111
1112	if (!page_has_buffers(page))
1113		create_empty_buffers(page, blocksize, 0);
1114	head = page_buffers(page);
1115	bbits = ilog2(blocksize);
1116	block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1117
1118	for (bh = head, block_start = 0; bh != head || !block_start;
1119	    block++, block_start = block_end, bh = bh->b_this_page) {
1120		block_end = block_start + blocksize;
1121		if (block_end <= from || block_start >= to) {
1122			if (PageUptodate(page)) {
1123				if (!buffer_uptodate(bh))
1124					set_buffer_uptodate(bh);
1125			}
1126			continue;
1127		}
1128		if (buffer_new(bh))
1129			clear_buffer_new(bh);
1130		if (!buffer_mapped(bh)) {
1131			WARN_ON(bh->b_size != blocksize);
1132			err = get_block(inode, block, bh, 1);
1133			if (err)
1134				break;
1135			if (buffer_new(bh)) {
1136				clean_bdev_bh_alias(bh);
1137				if (PageUptodate(page)) {
1138					clear_buffer_new(bh);
1139					set_buffer_uptodate(bh);
1140					mark_buffer_dirty(bh);
1141					continue;
1142				}
1143				if (block_end > to || block_start < from)
1144					zero_user_segments(page, to, block_end,
1145							   block_start, from);
1146				continue;
1147			}
1148		}
1149		if (PageUptodate(page)) {
1150			if (!buffer_uptodate(bh))
1151				set_buffer_uptodate(bh);
1152			continue;
1153		}
1154		if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1155		    !buffer_unwritten(bh) &&
1156		    (block_start < from || block_end > to)) {
1157			ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1158			*wait_bh++ = bh;
1159			decrypt = ext4_encrypted_inode(inode) &&
1160				S_ISREG(inode->i_mode);
1161		}
1162	}
1163	/*
1164	 * If we issued read requests, let them complete.
1165	 */
1166	while (wait_bh > wait) {
1167		wait_on_buffer(*--wait_bh);
1168		if (!buffer_uptodate(*wait_bh))
1169			err = -EIO;
1170	}
1171	if (unlikely(err))
1172		page_zero_new_buffers(page, from, to);
1173	else if (decrypt)
1174		err = fscrypt_decrypt_page(page->mapping->host, page,
1175				PAGE_SIZE, 0, page->index);
1176	return err;
1177}
1178#endif
1179
1180static int ext4_write_begin(struct file *file, struct address_space *mapping,
1181			    loff_t pos, unsigned len, unsigned flags,
1182			    struct page **pagep, void **fsdata)
1183{
1184	struct inode *inode = mapping->host;
1185	int ret, needed_blocks;
1186	handle_t *handle;
1187	int retries = 0;
1188	struct page *page;
1189	pgoff_t index;
1190	unsigned from, to;
1191
1192	trace_ext4_write_begin(inode, pos, len, flags);
1193	/*
1194	 * Reserve one block more for addition to orphan list in case
1195	 * we allocate blocks but write fails for some reason
1196	 */
1197	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1198	index = pos >> PAGE_SHIFT;
1199	from = pos & (PAGE_SIZE - 1);
1200	to = from + len;
1201
1202	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1203		ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1204						    flags, pagep);
1205		if (ret < 0)
1206			return ret;
1207		if (ret == 1)
1208			return 0;
1209	}
1210
1211	/*
1212	 * grab_cache_page_write_begin() can take a long time if the
1213	 * system is thrashing due to memory pressure, or if the page
1214	 * is being written back.  So grab it first before we start
1215	 * the transaction handle.  This also allows us to allocate
1216	 * the page (if needed) without using GFP_NOFS.
1217	 */
1218retry_grab:
1219	page = grab_cache_page_write_begin(mapping, index, flags);
1220	if (!page)
1221		return -ENOMEM;
1222	unlock_page(page);
1223
1224retry_journal:
1225	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1226	if (IS_ERR(handle)) {
1227		put_page(page);
1228		return PTR_ERR(handle);
1229	}
1230
1231	lock_page(page);
1232	if (page->mapping != mapping) {
1233		/* The page got truncated from under us */
1234		unlock_page(page);
1235		put_page(page);
1236		ext4_journal_stop(handle);
1237		goto retry_grab;
 
1238	}
1239	/* In case writeback began while the page was unlocked */
1240	wait_for_stable_page(page);
1241
1242#ifdef CONFIG_EXT4_FS_ENCRYPTION
1243	if (ext4_should_dioread_nolock(inode))
1244		ret = ext4_block_write_begin(page, pos, len,
1245					     ext4_get_block_unwritten);
1246	else
1247		ret = ext4_block_write_begin(page, pos, len,
1248					     ext4_get_block);
1249#else
1250	if (ext4_should_dioread_nolock(inode))
1251		ret = __block_write_begin(page, pos, len,
1252					  ext4_get_block_unwritten);
1253	else
1254		ret = __block_write_begin(page, pos, len, ext4_get_block);
1255#endif
1256	if (!ret && ext4_should_journal_data(inode)) {
1257		ret = ext4_walk_page_buffers(handle, page_buffers(page),
1258					     from, to, NULL,
1259					     do_journal_get_write_access);
1260	}
1261
1262	if (ret) {
1263		unlock_page(page);
 
1264		/*
1265		 * __block_write_begin may have instantiated a few blocks
1266		 * outside i_size.  Trim these off again. Don't need
1267		 * i_size_read because we hold i_mutex.
1268		 *
1269		 * Add inode to orphan list in case we crash before
1270		 * truncate finishes
1271		 */
1272		if (pos + len > inode->i_size && ext4_can_truncate(inode))
1273			ext4_orphan_add(handle, inode);
1274
1275		ext4_journal_stop(handle);
1276		if (pos + len > inode->i_size) {
1277			ext4_truncate_failed_write(inode);
1278			/*
1279			 * If truncate failed early the inode might
1280			 * still be on the orphan list; we need to
1281			 * make sure the inode is removed from the
1282			 * orphan list in that case.
1283			 */
1284			if (inode->i_nlink)
1285				ext4_orphan_del(NULL, inode);
1286		}
 
1287
1288		if (ret == -ENOSPC &&
1289		    ext4_should_retry_alloc(inode->i_sb, &retries))
1290			goto retry_journal;
1291		put_page(page);
1292		return ret;
1293	}
1294	*pagep = page;
1295	return ret;
1296}
1297
1298/* For write_end() in data=journal mode */
1299static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1300{
1301	int ret;
1302	if (!buffer_mapped(bh) || buffer_freed(bh))
1303		return 0;
1304	set_buffer_uptodate(bh);
1305	ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1306	clear_buffer_meta(bh);
1307	clear_buffer_prio(bh);
1308	return ret;
1309}
1310
1311/*
1312 * We need to pick up the new inode size which generic_commit_write gave us
1313 * `file' can be NULL - eg, when called from page_symlink().
1314 *
1315 * ext4 never places buffers on inode->i_mapping->private_list.  metadata
1316 * buffers are managed internally.
1317 */
1318static int ext4_write_end(struct file *file,
1319			  struct address_space *mapping,
1320			  loff_t pos, unsigned len, unsigned copied,
1321			  struct page *page, void *fsdata)
1322{
 
 
1323	handle_t *handle = ext4_journal_current_handle();
1324	struct inode *inode = mapping->host;
1325	loff_t old_size = inode->i_size;
1326	int ret = 0, ret2;
1327	int i_size_changed = 0;
1328
1329	trace_ext4_write_end(inode, pos, len, copied);
1330	if (ext4_has_inline_data(inode)) {
1331		ret = ext4_write_inline_data_end(inode, pos, len,
1332						 copied, page);
1333		if (ret < 0) {
1334			unlock_page(page);
1335			put_page(page);
1336			goto errout;
1337		}
1338		copied = ret;
1339	} else
1340		copied = block_write_end(file, mapping, pos,
1341					 len, copied, page, fsdata);
1342	/*
1343	 * it's important to update i_size while still holding page lock:
 
 
 
1344	 * page writeout could otherwise come in and zero beyond i_size.
1345	 */
1346	i_size_changed = ext4_update_inode_size(inode, pos + copied);
 
 
 
 
 
 
 
 
 
 
 
 
1347	unlock_page(page);
1348	put_page(page);
1349
1350	if (old_size < pos)
1351		pagecache_isize_extended(inode, old_size, pos);
1352	/*
1353	 * Don't mark the inode dirty under page lock. First, it unnecessarily
1354	 * makes the holding time of page lock longer. Second, it forces lock
1355	 * ordering of page lock and transaction start for journaling
1356	 * filesystems.
1357	 */
1358	if (i_size_changed)
1359		ext4_mark_inode_dirty(handle, inode);
1360
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1361	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1362		/* if we have allocated more blocks and copied
1363		 * less. We will have blocks allocated outside
1364		 * inode->i_size. So truncate them
1365		 */
1366		ext4_orphan_add(handle, inode);
1367errout:
 
 
 
1368	ret2 = ext4_journal_stop(handle);
1369	if (!ret)
1370		ret = ret2;
1371
1372	if (pos + len > inode->i_size) {
1373		ext4_truncate_failed_write(inode);
1374		/*
1375		 * If truncate failed early the inode might still be
1376		 * on the orphan list; we need to make sure the inode
1377		 * is removed from the orphan list in that case.
1378		 */
1379		if (inode->i_nlink)
1380			ext4_orphan_del(NULL, inode);
1381	}
1382
1383	return ret ? ret : copied;
1384}
1385
1386/*
1387 * This is a private version of page_zero_new_buffers() which doesn't
1388 * set the buffer to be dirty, since in data=journalled mode we need
1389 * to call ext4_handle_dirty_metadata() instead.
1390 */
1391static void ext4_journalled_zero_new_buffers(handle_t *handle,
1392					    struct page *page,
1393					    unsigned from, unsigned to)
1394{
1395	unsigned int block_start = 0, block_end;
1396	struct buffer_head *head, *bh;
1397
1398	bh = head = page_buffers(page);
1399	do {
1400		block_end = block_start + bh->b_size;
1401		if (buffer_new(bh)) {
1402			if (block_end > from && block_start < to) {
1403				if (!PageUptodate(page)) {
1404					unsigned start, size;
1405
1406					start = max(from, block_start);
1407					size = min(to, block_end) - start;
1408
1409					zero_user(page, start, size);
1410					write_end_fn(handle, bh);
1411				}
1412				clear_buffer_new(bh);
1413			}
1414		}
1415		block_start = block_end;
1416		bh = bh->b_this_page;
1417	} while (bh != head);
1418}
1419
1420static int ext4_journalled_write_end(struct file *file,
1421				     struct address_space *mapping,
1422				     loff_t pos, unsigned len, unsigned copied,
1423				     struct page *page, void *fsdata)
1424{
1425	handle_t *handle = ext4_journal_current_handle();
1426	struct inode *inode = mapping->host;
1427	loff_t old_size = inode->i_size;
1428	int ret = 0, ret2;
1429	int partial = 0;
1430	unsigned from, to;
1431	int size_changed = 0;
1432
1433	trace_ext4_journalled_write_end(inode, pos, len, copied);
1434	from = pos & (PAGE_SIZE - 1);
1435	to = from + len;
1436
1437	BUG_ON(!ext4_handle_valid(handle));
1438
1439	if (ext4_has_inline_data(inode)) {
1440		ret = ext4_write_inline_data_end(inode, pos, len,
1441						 copied, page);
1442		if (ret < 0) {
1443			unlock_page(page);
1444			put_page(page);
1445			goto errout;
1446		}
1447		copied = ret;
1448	} else if (unlikely(copied < len) && !PageUptodate(page)) {
1449		copied = 0;
1450		ext4_journalled_zero_new_buffers(handle, page, from, to);
1451	} else {
1452		if (unlikely(copied < len))
1453			ext4_journalled_zero_new_buffers(handle, page,
1454							 from + copied, to);
1455		ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1456					     from + copied, &partial,
1457					     write_end_fn);
1458		if (!partial)
1459			SetPageUptodate(page);
1460	}
1461	size_changed = ext4_update_inode_size(inode, pos + copied);
 
 
 
 
 
 
 
1462	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1463	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1464	unlock_page(page);
1465	put_page(page);
1466
1467	if (old_size < pos)
1468		pagecache_isize_extended(inode, old_size, pos);
1469
1470	if (size_changed) {
1471		ret2 = ext4_mark_inode_dirty(handle, inode);
1472		if (!ret)
1473			ret = ret2;
1474	}
1475
 
 
1476	if (pos + len > inode->i_size && ext4_can_truncate(inode))
1477		/* if we have allocated more blocks and copied
1478		 * less. We will have blocks allocated outside
1479		 * inode->i_size. So truncate them
1480		 */
1481		ext4_orphan_add(handle, inode);
1482
1483errout:
1484	ret2 = ext4_journal_stop(handle);
1485	if (!ret)
1486		ret = ret2;
1487	if (pos + len > inode->i_size) {
1488		ext4_truncate_failed_write(inode);
1489		/*
1490		 * If truncate failed early the inode might still be
1491		 * on the orphan list; we need to make sure the inode
1492		 * is removed from the orphan list in that case.
1493		 */
1494		if (inode->i_nlink)
1495			ext4_orphan_del(NULL, inode);
1496	}
1497
1498	return ret ? ret : copied;
1499}
1500
1501/*
1502 * Reserve space for a single cluster
1503 */
1504static int ext4_da_reserve_space(struct inode *inode)
1505{
 
1506	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1507	struct ext4_inode_info *ei = EXT4_I(inode);
 
1508	int ret;
1509
1510	/*
 
 
 
 
 
 
 
 
 
 
 
1511	 * We will charge metadata quota at writeout time; this saves
1512	 * us from metadata over-estimation, though we may go over by
1513	 * a small amount in the end.  Here we just reserve for data.
1514	 */
1515	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1516	if (ret)
1517		return ret;
1518
1519	spin_lock(&ei->i_block_reservation_lock);
1520	if (ext4_claim_free_clusters(sbi, 1, 0)) {
1521		spin_unlock(&ei->i_block_reservation_lock);
1522		dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
 
 
 
 
 
1523		return -ENOSPC;
1524	}
 
1525	ei->i_reserved_data_blocks++;
1526	trace_ext4_da_reserve_space(inode);
1527	spin_unlock(&ei->i_block_reservation_lock);
1528
1529	return 0;       /* success */
1530}
1531
1532static void ext4_da_release_space(struct inode *inode, int to_free)
1533{
1534	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1535	struct ext4_inode_info *ei = EXT4_I(inode);
1536
1537	if (!to_free)
1538		return;		/* Nothing to release, exit */
1539
1540	spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1541
1542	trace_ext4_da_release_space(inode, to_free);
1543	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1544		/*
1545		 * if there aren't enough reserved blocks, then the
1546		 * counter is messed up somewhere.  Since this
1547		 * function is called from invalidate page, it's
1548		 * harmless to return without any action.
1549		 */
1550		ext4_warning(inode->i_sb, "ext4_da_release_space: "
1551			 "ino %lu, to_free %d with only %d reserved "
1552			 "data blocks", inode->i_ino, to_free,
1553			 ei->i_reserved_data_blocks);
1554		WARN_ON(1);
1555		to_free = ei->i_reserved_data_blocks;
1556	}
1557	ei->i_reserved_data_blocks -= to_free;
1558
 
 
 
 
 
 
 
 
 
 
 
 
1559	/* update fs dirty data blocks counter */
1560	percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1561
1562	spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1563
1564	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1565}
1566
1567static void ext4_da_page_release_reservation(struct page *page,
1568					     unsigned int offset,
1569					     unsigned int length)
1570{
1571	int to_release = 0, contiguous_blks = 0;
1572	struct buffer_head *head, *bh;
1573	unsigned int curr_off = 0;
1574	struct inode *inode = page->mapping->host;
1575	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1576	unsigned int stop = offset + length;
1577	int num_clusters;
1578	ext4_fsblk_t lblk;
1579
1580	BUG_ON(stop > PAGE_SIZE || stop < length);
1581
1582	head = page_buffers(page);
1583	bh = head;
1584	do {
1585		unsigned int next_off = curr_off + bh->b_size;
1586
1587		if (next_off > stop)
1588			break;
1589
1590		if ((offset <= curr_off) && (buffer_delay(bh))) {
1591			to_release++;
1592			contiguous_blks++;
1593			clear_buffer_delay(bh);
1594		} else if (contiguous_blks) {
1595			lblk = page->index <<
1596			       (PAGE_SHIFT - inode->i_blkbits);
1597			lblk += (curr_off >> inode->i_blkbits) -
1598				contiguous_blks;
1599			ext4_es_remove_extent(inode, lblk, contiguous_blks);
1600			contiguous_blks = 0;
1601		}
1602		curr_off = next_off;
1603	} while ((bh = bh->b_this_page) != head);
1604
1605	if (contiguous_blks) {
1606		lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1607		lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1608		ext4_es_remove_extent(inode, lblk, contiguous_blks);
1609	}
1610
1611	/* If we have released all the blocks belonging to a cluster, then we
1612	 * need to release the reserved space for that cluster. */
1613	num_clusters = EXT4_NUM_B2C(sbi, to_release);
1614	while (num_clusters > 0) {
1615		lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1616			((num_clusters - 1) << sbi->s_cluster_bits);
1617		if (sbi->s_cluster_ratio == 1 ||
1618		    !ext4_find_delalloc_cluster(inode, lblk))
1619			ext4_da_release_space(inode, 1);
1620
1621		num_clusters--;
1622	}
1623}
1624
1625/*
1626 * Delayed allocation stuff
1627 */
1628
1629struct mpage_da_data {
1630	struct inode *inode;
1631	struct writeback_control *wbc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1632
1633	pgoff_t first_page;	/* The first page to write */
1634	pgoff_t next_page;	/* Current page to examine */
1635	pgoff_t last_page;	/* Last page to examine */
1636	/*
1637	 * Extent to map - this can be after first_page because that can be
1638	 * fully mapped. We somewhat abuse m_flags to store whether the extent
1639	 * is delalloc or unwritten.
1640	 */
1641	struct ext4_map_blocks map;
1642	struct ext4_io_submit io_submit;	/* IO submission data */
1643};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1644
1645static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1646				       bool invalidate)
1647{
1648	int nr_pages, i;
1649	pgoff_t index, end;
1650	struct pagevec pvec;
1651	struct inode *inode = mpd->inode;
1652	struct address_space *mapping = inode->i_mapping;
1653
1654	/* This is necessary when next_page == 0. */
1655	if (mpd->first_page >= mpd->next_page)
1656		return;
1657
1658	index = mpd->first_page;
1659	end   = mpd->next_page - 1;
1660	if (invalidate) {
1661		ext4_lblk_t start, last;
1662		start = index << (PAGE_SHIFT - inode->i_blkbits);
1663		last = end << (PAGE_SHIFT - inode->i_blkbits);
1664		ext4_es_remove_extent(inode, start, last - start + 1);
1665	}
1666
1667	pagevec_init(&pvec, 0);
1668	while (index <= end) {
1669		nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1670		if (nr_pages == 0)
1671			break;
1672		for (i = 0; i < nr_pages; i++) {
1673			struct page *page = pvec.pages[i];
1674			if (page->index > end)
1675				break;
1676			BUG_ON(!PageLocked(page));
1677			BUG_ON(PageWriteback(page));
1678			if (invalidate) {
1679				if (page_mapped(page))
1680					clear_page_dirty_for_io(page);
1681				block_invalidatepage(page, 0, PAGE_SIZE);
1682				ClearPageUptodate(page);
1683			}
1684			unlock_page(page);
1685		}
1686		index = pvec.pages[nr_pages - 1]->index + 1;
1687		pagevec_release(&pvec);
1688	}
 
1689}
1690
1691static void ext4_print_free_blocks(struct inode *inode)
1692{
1693	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1694	struct super_block *sb = inode->i_sb;
1695	struct ext4_inode_info *ei = EXT4_I(inode);
1696
1697	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1698	       EXT4_C2B(EXT4_SB(inode->i_sb),
1699			ext4_count_free_clusters(sb)));
1700	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1701	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1702	       (long long) EXT4_C2B(EXT4_SB(sb),
1703		percpu_counter_sum(&sbi->s_freeclusters_counter)));
1704	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1705	       (long long) EXT4_C2B(EXT4_SB(sb),
1706		percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1707	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1708	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1709		 ei->i_reserved_data_blocks);
1710	return;
1711}
1712
1713static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1714{
1715	return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1716}
1717
1718/*
1719 * This function is grabs code from the very beginning of
1720 * ext4_map_blocks, but assumes that the caller is from delayed write
1721 * time. This function looks up the requested blocks and sets the
1722 * buffer delay bit under the protection of i_data_sem.
1723 */
1724static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1725			      struct ext4_map_blocks *map,
1726			      struct buffer_head *bh)
 
1727{
1728	struct extent_status es;
1729	int retval;
1730	sector_t invalid_block = ~((sector_t) 0xffff);
1731#ifdef ES_AGGRESSIVE_TEST
1732	struct ext4_map_blocks orig_map;
 
1733
1734	memcpy(&orig_map, map, sizeof(*map));
1735#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1736
1737	if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1738		invalid_block = ~0;
 
1739
1740	map->m_flags = 0;
1741	ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1742		  "logical block %lu\n", inode->i_ino, map->m_len,
1743		  (unsigned long) map->m_lblk);
 
 
 
 
1744
1745	/* Lookup extent status tree firstly */
1746	if (ext4_es_lookup_extent(inode, iblock, &es)) {
1747		if (ext4_es_is_hole(&es)) {
1748			retval = 0;
1749			down_read(&EXT4_I(inode)->i_data_sem);
1750			goto add_delayed;
1751		}
1752
1753		/*
1754		 * Delayed extent could be allocated by fallocate.
1755		 * So we need to check it.
 
 
 
1756		 */
1757		if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1758			map_bh(bh, inode->i_sb, invalid_block);
1759			set_buffer_new(bh);
1760			set_buffer_delay(bh);
1761			return 0;
 
 
 
 
 
 
1762		}
 
 
 
 
 
 
 
 
 
 
 
 
 
1763
1764		map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1765		retval = es.es_len - (iblock - es.es_lblk);
1766		if (retval > map->m_len)
1767			retval = map->m_len;
1768		map->m_len = retval;
1769		if (ext4_es_is_written(&es))
1770			map->m_flags |= EXT4_MAP_MAPPED;
1771		else if (ext4_es_is_unwritten(&es))
1772			map->m_flags |= EXT4_MAP_UNWRITTEN;
1773		else
1774			BUG_ON(1);
1775
1776#ifdef ES_AGGRESSIVE_TEST
1777		ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1778#endif
1779		return retval;
 
1780	}
1781
1782	/*
1783	 * Try to see if we can get the block without requesting a new
1784	 * file system block.
1785	 */
1786	down_read(&EXT4_I(inode)->i_data_sem);
1787	if (ext4_has_inline_data(inode))
1788		retval = 0;
1789	else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1790		retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1791	else
1792		retval = ext4_ind_map_blocks(NULL, inode, map, 0);
 
 
 
 
 
 
 
 
 
1793
1794add_delayed:
1795	if (retval == 0) {
1796		int ret;
1797		/*
1798		 * XXX: __block_prepare_write() unmaps passed block,
1799		 * is it OK?
1800		 */
1801		/*
1802		 * If the block was allocated from previously allocated cluster,
1803		 * then we don't need to reserve it again. However we still need
1804		 * to reserve metadata for every block we're going to write.
1805		 */
1806		if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1807		    !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1808			ret = ext4_da_reserve_space(inode);
1809			if (ret) {
1810				/* not enough space to reserve */
1811				retval = ret;
1812				goto out_unlock;
1813			}
1814		}
1815
1816		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1817					    ~0, EXTENT_STATUS_DELAYED);
1818		if (ret) {
1819			retval = ret;
1820			goto out_unlock;
1821		}
 
 
 
 
 
 
 
 
 
1822
1823		map_bh(bh, inode->i_sb, invalid_block);
1824		set_buffer_new(bh);
1825		set_buffer_delay(bh);
1826	} else if (retval > 0) {
1827		int ret;
1828		unsigned int status;
 
 
1829
1830		if (unlikely(retval != map->m_len)) {
1831			ext4_warning(inode->i_sb,
1832				     "ES len assertion failed for inode "
1833				     "%lu: retval %d != map->m_len %d",
1834				     inode->i_ino, retval, map->m_len);
1835			WARN_ON(1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1836		}
 
 
 
 
 
 
 
 
 
 
1837
1838		status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1839				EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1840		ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1841					    map->m_pblk, status);
1842		if (ret != 0)
1843			retval = ret;
 
1844	}
1845
1846out_unlock:
1847	up_read((&EXT4_I(inode)->i_data_sem));
 
 
 
 
 
 
1848
1849	return retval;
 
 
1850}
1851
1852/*
1853 * This is a special get_block_t callback which is used by
1854 * ext4_da_write_begin().  It will either return mapped block or
1855 * reserve space for a single block.
1856 *
1857 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1858 * We also have b_blocknr = -1 and b_bdev initialized properly
1859 *
1860 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1861 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1862 * initialized properly.
1863 */
1864int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1865			   struct buffer_head *bh, int create)
1866{
1867	struct ext4_map_blocks map;
1868	int ret = 0;
 
 
 
 
1869
1870	BUG_ON(create == 0);
1871	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1872
1873	map.m_lblk = iblock;
1874	map.m_len = 1;
1875
1876	/*
1877	 * first, we need to know whether the block is allocated already
1878	 * preallocated blocks are unmapped but should treated
1879	 * the same as allocated blocks.
1880	 */
1881	ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1882	if (ret <= 0)
1883		return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1884
1885	map_bh(bh, inode->i_sb, map.m_pblk);
1886	ext4_update_bh_state(bh, map.m_flags);
1887
1888	if (buffer_unwritten(bh)) {
1889		/* A delayed write to unwritten bh should be marked
1890		 * new and mapped.  Mapped ensures that we don't do
1891		 * get_block multiple times when we write to the same
1892		 * offset and new ensures that we do proper zero out
1893		 * for partial write.
1894		 */
1895		set_buffer_new(bh);
1896		set_buffer_mapped(bh);
1897	}
1898	return 0;
1899}
1900
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1901static int bget_one(handle_t *handle, struct buffer_head *bh)
1902{
1903	get_bh(bh);
1904	return 0;
1905}
1906
1907static int bput_one(handle_t *handle, struct buffer_head *bh)
1908{
1909	put_bh(bh);
1910	return 0;
1911}
1912
1913static int __ext4_journalled_writepage(struct page *page,
1914				       unsigned int len)
1915{
1916	struct address_space *mapping = page->mapping;
1917	struct inode *inode = mapping->host;
1918	struct buffer_head *page_bufs = NULL;
1919	handle_t *handle = NULL;
1920	int ret = 0, err = 0;
1921	int inline_data = ext4_has_inline_data(inode);
1922	struct buffer_head *inode_bh = NULL;
1923
1924	ClearPageChecked(page);
1925
1926	if (inline_data) {
1927		BUG_ON(page->index != 0);
1928		BUG_ON(len > ext4_get_max_inline_size(inode));
1929		inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1930		if (inode_bh == NULL)
1931			goto out;
1932	} else {
1933		page_bufs = page_buffers(page);
1934		if (!page_bufs) {
1935			BUG();
1936			goto out;
1937		}
1938		ext4_walk_page_buffers(handle, page_bufs, 0, len,
1939				       NULL, bget_one);
1940	}
1941	/*
1942	 * We need to release the page lock before we start the
1943	 * journal, so grab a reference so the page won't disappear
1944	 * out from under us.
1945	 */
1946	get_page(page);
1947	unlock_page(page);
1948
1949	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1950				    ext4_writepage_trans_blocks(inode));
1951	if (IS_ERR(handle)) {
1952		ret = PTR_ERR(handle);
1953		put_page(page);
1954		goto out_no_pagelock;
1955	}
1956	BUG_ON(!ext4_handle_valid(handle));
1957
1958	lock_page(page);
1959	put_page(page);
1960	if (page->mapping != mapping) {
1961		/* The page got truncated from under us */
1962		ext4_journal_stop(handle);
1963		ret = 0;
1964		goto out;
1965	}
1966
1967	if (inline_data) {
1968		BUFFER_TRACE(inode_bh, "get write access");
1969		ret = ext4_journal_get_write_access(handle, inode_bh);
1970
1971		err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
 
1972
1973	} else {
1974		ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1975					     do_journal_get_write_access);
1976
1977		err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1978					     write_end_fn);
1979	}
1980	if (ret == 0)
1981		ret = err;
1982	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1983	err = ext4_journal_stop(handle);
1984	if (!ret)
1985		ret = err;
1986
1987	if (!ext4_has_inline_data(inode))
1988		ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1989				       NULL, bput_one);
1990	ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1991out:
1992	unlock_page(page);
1993out_no_pagelock:
1994	brelse(inode_bh);
1995	return ret;
1996}
1997
 
 
 
1998/*
1999 * Note that we don't need to start a transaction unless we're journaling data
2000 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2001 * need to file the inode to the transaction's list in ordered mode because if
2002 * we are writing back data added by write(), the inode is already there and if
2003 * we are writing back data modified via mmap(), no one guarantees in which
2004 * transaction the data will hit the disk. In case we are journaling data, we
2005 * cannot start transaction directly because transaction start ranks above page
2006 * lock so we have to do some magic.
2007 *
2008 * This function can get called via...
2009 *   - ext4_writepages after taking page lock (have journal handle)
2010 *   - journal_submit_inode_data_buffers (no journal handle)
2011 *   - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2012 *   - grab_page_cache when doing write_begin (have journal handle)
2013 *
2014 * We don't do any block allocation in this function. If we have page with
2015 * multiple blocks we need to write those buffer_heads that are mapped. This
2016 * is important for mmaped based write. So if we do with blocksize 1K
2017 * truncate(f, 1024);
2018 * a = mmap(f, 0, 4096);
2019 * a[0] = 'a';
2020 * truncate(f, 4096);
2021 * we have in the page first buffer_head mapped via page_mkwrite call back
2022 * but other buffer_heads would be unmapped but dirty (dirty done via the
2023 * do_wp_page). So writepage should write the first block. If we modify
2024 * the mmap area beyond 1024 we will again get a page_fault and the
2025 * page_mkwrite callback will do the block allocation and mark the
2026 * buffer_heads mapped.
2027 *
2028 * We redirty the page if we have any buffer_heads that is either delay or
2029 * unwritten in the page.
2030 *
2031 * We can get recursively called as show below.
2032 *
2033 *	ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2034 *		ext4_writepage()
2035 *
2036 * But since we don't do any block allocation we should not deadlock.
2037 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2038 */
2039static int ext4_writepage(struct page *page,
2040			  struct writeback_control *wbc)
2041{
2042	int ret = 0;
2043	loff_t size;
2044	unsigned int len;
2045	struct buffer_head *page_bufs = NULL;
2046	struct inode *inode = page->mapping->host;
2047	struct ext4_io_submit io_submit;
2048	bool keep_towrite = false;
2049
2050	trace_ext4_writepage(page);
2051	size = i_size_read(inode);
2052	if (page->index == size >> PAGE_SHIFT)
2053		len = size & ~PAGE_MASK;
2054	else
2055		len = PAGE_SIZE;
2056
2057	page_bufs = page_buffers(page);
2058	/*
2059	 * We cannot do block allocation or other extent handling in this
2060	 * function. If there are buffers needing that, we have to redirty
2061	 * the page. But we may reach here when we do a journal commit via
2062	 * journal_submit_inode_data_buffers() and in that case we must write
2063	 * allocated buffers to achieve data=ordered mode guarantees.
2064	 *
2065	 * Also, if there is only one buffer per page (the fs block
2066	 * size == the page size), if one buffer needs block
2067	 * allocation or needs to modify the extent tree to clear the
2068	 * unwritten flag, we know that the page can't be written at
2069	 * all, so we might as well refuse the write immediately.
2070	 * Unfortunately if the block size != page size, we can't as
2071	 * easily detect this case using ext4_walk_page_buffers(), but
2072	 * for the extremely common case, this is an optimization that
2073	 * skips a useless round trip through ext4_bio_write_page().
2074	 */
2075	if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2076				   ext4_bh_delay_or_unwritten)) {
2077		redirty_page_for_writepage(wbc, page);
2078		if ((current->flags & PF_MEMALLOC) ||
2079		    (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2080			/*
2081			 * For memory cleaning there's no point in writing only
2082			 * some buffers. So just bail out. Warn if we came here
2083			 * from direct reclaim.
2084			 */
2085			WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2086							== PF_MEMALLOC);
2087			unlock_page(page);
2088			return 0;
2089		}
2090		keep_towrite = true;
2091	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2092
2093	if (PageChecked(page) && ext4_should_journal_data(inode))
2094		/*
2095		 * It's mmapped pagecache.  Add buffers and journal it.  There
2096		 * doesn't seem much point in redirtying the page here.
2097		 */
2098		return __ext4_journalled_writepage(page, len);
2099
2100	ext4_io_submit_init(&io_submit, wbc);
2101	io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2102	if (!io_submit.io_end) {
2103		redirty_page_for_writepage(wbc, page);
2104		unlock_page(page);
2105		return -ENOMEM;
2106	}
2107	ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2108	ext4_io_submit(&io_submit);
2109	/* Drop io_end reference we got from init */
2110	ext4_put_io_end_defer(io_submit.io_end);
2111	return ret;
2112}
2113
2114static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2115{
2116	int len;
2117	loff_t size = i_size_read(mpd->inode);
2118	int err;
2119
2120	BUG_ON(page->index != mpd->first_page);
2121	if (page->index == size >> PAGE_SHIFT)
2122		len = size & ~PAGE_MASK;
2123	else
2124		len = PAGE_SIZE;
2125	clear_page_dirty_for_io(page);
2126	err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2127	if (!err)
2128		mpd->wbc->nr_to_write--;
2129	mpd->first_page++;
2130
2131	return err;
2132}
2133
2134#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2135
2136/*
2137 * mballoc gives us at most this number of blocks...
2138 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2139 * The rest of mballoc seems to handle chunks up to full group size.
 
 
2140 */
2141#define MAX_WRITEPAGES_EXTENT_LEN 2048
2142
2143/*
2144 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2145 *
2146 * @mpd - extent of blocks
2147 * @lblk - logical number of the block in the file
2148 * @bh - buffer head we want to add to the extent
2149 *
2150 * The function is used to collect contig. blocks in the same state. If the
2151 * buffer doesn't require mapping for writeback and we haven't started the
2152 * extent of buffers to map yet, the function returns 'true' immediately - the
2153 * caller can write the buffer right away. Otherwise the function returns true
2154 * if the block has been added to the extent, false if the block couldn't be
2155 * added.
2156 */
2157static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2158				   struct buffer_head *bh)
2159{
2160	struct ext4_map_blocks *map = &mpd->map;
2161
2162	/* Buffer that doesn't need mapping for writeback? */
2163	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2164	    (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2165		/* So far no extent to map => we write the buffer right away */
2166		if (map->m_len == 0)
2167			return true;
2168		return false;
2169	}
2170
2171	/* First block in the extent? */
2172	if (map->m_len == 0) {
2173		map->m_lblk = lblk;
2174		map->m_len = 1;
2175		map->m_flags = bh->b_state & BH_FLAGS;
2176		return true;
2177	}
2178
2179	/* Don't go larger than mballoc is willing to allocate */
2180	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2181		return false;
2182
2183	/* Can we merge the block to our big extent? */
2184	if (lblk == map->m_lblk + map->m_len &&
2185	    (bh->b_state & BH_FLAGS) == map->m_flags) {
2186		map->m_len++;
2187		return true;
2188	}
2189	return false;
2190}
2191
2192/*
2193 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2194 *
2195 * @mpd - extent of blocks for mapping
2196 * @head - the first buffer in the page
2197 * @bh - buffer we should start processing from
2198 * @lblk - logical number of the block in the file corresponding to @bh
2199 *
2200 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2201 * the page for IO if all buffers in this page were mapped and there's no
2202 * accumulated extent of buffers to map or add buffers in the page to the
2203 * extent of buffers to map. The function returns 1 if the caller can continue
2204 * by processing the next page, 0 if it should stop adding buffers to the
2205 * extent to map because we cannot extend it anymore. It can also return value
2206 * < 0 in case of error during IO submission.
2207 */
2208static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2209				   struct buffer_head *head,
2210				   struct buffer_head *bh,
2211				   ext4_lblk_t lblk)
2212{
2213	struct inode *inode = mpd->inode;
2214	int err;
2215	ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2216							>> inode->i_blkbits;
2217
2218	do {
2219		BUG_ON(buffer_locked(bh));
2220
2221		if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2222			/* Found extent to map? */
2223			if (mpd->map.m_len)
2224				return 0;
2225			/* Everything mapped so far and we hit EOF */
2226			break;
2227		}
2228	} while (lblk++, (bh = bh->b_this_page) != head);
2229	/* So far everything mapped? Submit the page for IO. */
2230	if (mpd->map.m_len == 0) {
2231		err = mpage_submit_page(mpd, head->b_page);
2232		if (err < 0)
2233			return err;
2234	}
2235	return lblk < blocks;
2236}
2237
2238/*
2239 * mpage_map_buffers - update buffers corresponding to changed extent and
2240 *		       submit fully mapped pages for IO
2241 *
2242 * @mpd - description of extent to map, on return next extent to map
2243 *
2244 * Scan buffers corresponding to changed extent (we expect corresponding pages
2245 * to be already locked) and update buffer state according to new extent state.
2246 * We map delalloc buffers to their physical location, clear unwritten bits,
2247 * and mark buffers as uninit when we perform writes to unwritten extents
2248 * and do extent conversion after IO is finished. If the last page is not fully
2249 * mapped, we update @map to the next extent in the last page that needs
2250 * mapping. Otherwise we submit the page for IO.
2251 */
2252static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
 
 
 
2253{
2254	struct pagevec pvec;
2255	int nr_pages, i;
2256	struct inode *inode = mpd->inode;
2257	struct buffer_head *head, *bh;
2258	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2259	pgoff_t start, end;
2260	ext4_lblk_t lblk;
2261	sector_t pblock;
2262	int err;
2263
2264	start = mpd->map.m_lblk >> bpp_bits;
2265	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2266	lblk = start << bpp_bits;
2267	pblock = mpd->map.m_pblk;
2268
 
 
 
2269	pagevec_init(&pvec, 0);
2270	while (start <= end) {
2271		nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2272					  PAGEVEC_SIZE);
2273		if (nr_pages == 0)
2274			break;
2275		for (i = 0; i < nr_pages; i++) {
2276			struct page *page = pvec.pages[i];
2277
2278			if (page->index > end)
2279				break;
2280			/* Up to 'end' pages must be contiguous */
2281			BUG_ON(page->index != start);
2282			bh = head = page_buffers(page);
2283			do {
2284				if (lblk < mpd->map.m_lblk)
2285					continue;
2286				if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2287					/*
2288					 * Buffer after end of mapped extent.
2289					 * Find next buffer in the page to map.
2290					 */
2291					mpd->map.m_len = 0;
2292					mpd->map.m_flags = 0;
2293					/*
2294					 * FIXME: If dioread_nolock supports
2295					 * blocksize < pagesize, we need to make
2296					 * sure we add size mapped so far to
2297					 * io_end->size as the following call
2298					 * can submit the page for IO.
2299					 */
2300					err = mpage_process_page_bufs(mpd, head,
2301								      bh, lblk);
2302					pagevec_release(&pvec);
2303					if (err > 0)
2304						err = 0;
2305					return err;
2306				}
2307				if (buffer_delay(bh)) {
2308					clear_buffer_delay(bh);
2309					bh->b_blocknr = pblock++;
2310				}
2311				clear_buffer_unwritten(bh);
2312			} while (lblk++, (bh = bh->b_this_page) != head);
2313
2314			/*
2315			 * FIXME: This is going to break if dioread_nolock
2316			 * supports blocksize < pagesize as we will try to
2317			 * convert potentially unmapped parts of inode.
2318			 */
2319			mpd->io_submit.io_end->size += PAGE_SIZE;
2320			/* Page fully mapped - let IO run! */
2321			err = mpage_submit_page(mpd, page);
2322			if (err < 0) {
2323				pagevec_release(&pvec);
2324				return err;
2325			}
2326			start++;
2327		}
2328		pagevec_release(&pvec);
2329	}
2330	/* Extent fully mapped and matches with page boundary. We are done. */
2331	mpd->map.m_len = 0;
2332	mpd->map.m_flags = 0;
2333	return 0;
2334}
2335
2336static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2337{
2338	struct inode *inode = mpd->inode;
2339	struct ext4_map_blocks *map = &mpd->map;
2340	int get_blocks_flags;
2341	int err, dioread_nolock;
2342
2343	trace_ext4_da_write_pages_extent(inode, map);
2344	/*
2345	 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2346	 * to convert an unwritten extent to be initialized (in the case
2347	 * where we have written into one or more preallocated blocks).  It is
2348	 * possible that we're going to need more metadata blocks than
2349	 * previously reserved. However we must not fail because we're in
2350	 * writeback and there is nothing we can do about it so it might result
2351	 * in data loss.  So use reserved blocks to allocate metadata if
2352	 * possible.
2353	 *
2354	 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2355	 * the blocks in question are delalloc blocks.  This indicates
2356	 * that the blocks and quotas has already been checked when
2357	 * the data was copied into the page cache.
2358	 */
2359	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2360			   EXT4_GET_BLOCKS_METADATA_NOFAIL |
2361			   EXT4_GET_BLOCKS_IO_SUBMIT;
2362	dioread_nolock = ext4_should_dioread_nolock(inode);
2363	if (dioread_nolock)
2364		get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2365	if (map->m_flags & (1 << BH_Delay))
2366		get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2367
2368	err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2369	if (err < 0)
2370		return err;
2371	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2372		if (!mpd->io_submit.io_end->handle &&
2373		    ext4_handle_valid(handle)) {
2374			mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2375			handle->h_rsv_handle = NULL;
2376		}
2377		ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2378	}
2379
2380	BUG_ON(map->m_len == 0);
2381	if (map->m_flags & EXT4_MAP_NEW) {
2382		clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2383				   map->m_len);
2384	}
2385	return 0;
2386}
2387
2388/*
2389 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2390 *				 mpd->len and submit pages underlying it for IO
2391 *
2392 * @handle - handle for journal operations
2393 * @mpd - extent to map
2394 * @give_up_on_write - we set this to true iff there is a fatal error and there
2395 *                     is no hope of writing the data. The caller should discard
2396 *                     dirty pages to avoid infinite loops.
2397 *
2398 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2399 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2400 * them to initialized or split the described range from larger unwritten
2401 * extent. Note that we need not map all the described range since allocation
2402 * can return less blocks or the range is covered by more unwritten extents. We
2403 * cannot map more because we are limited by reserved transaction credits. On
2404 * the other hand we always make sure that the last touched page is fully
2405 * mapped so that it can be written out (and thus forward progress is
2406 * guaranteed). After mapping we submit all mapped pages for IO.
2407 */
2408static int mpage_map_and_submit_extent(handle_t *handle,
2409				       struct mpage_da_data *mpd,
2410				       bool *give_up_on_write)
2411{
2412	struct inode *inode = mpd->inode;
2413	struct ext4_map_blocks *map = &mpd->map;
2414	int err;
2415	loff_t disksize;
2416	int progress = 0;
2417
2418	mpd->io_submit.io_end->offset =
2419				((loff_t)map->m_lblk) << inode->i_blkbits;
2420	do {
2421		err = mpage_map_one_extent(handle, mpd);
2422		if (err < 0) {
2423			struct super_block *sb = inode->i_sb;
2424
2425			if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2426				goto invalidate_dirty_pages;
2427			/*
2428			 * Let the uper layers retry transient errors.
2429			 * In the case of ENOSPC, if ext4_count_free_blocks()
2430			 * is non-zero, a commit should free up blocks.
2431			 */
2432			if ((err == -ENOMEM) ||
2433			    (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2434				if (progress)
2435					goto update_disksize;
2436				return err;
2437			}
2438			ext4_msg(sb, KERN_CRIT,
2439				 "Delayed block allocation failed for "
2440				 "inode %lu at logical offset %llu with"
2441				 " max blocks %u with error %d",
2442				 inode->i_ino,
2443				 (unsigned long long)map->m_lblk,
2444				 (unsigned)map->m_len, -err);
2445			ext4_msg(sb, KERN_CRIT,
2446				 "This should not happen!! Data will "
2447				 "be lost\n");
2448			if (err == -ENOSPC)
2449				ext4_print_free_blocks(inode);
2450		invalidate_dirty_pages:
2451			*give_up_on_write = true;
2452			return err;
2453		}
2454		progress = 1;
2455		/*
2456		 * Update buffer state, submit mapped pages, and get us new
2457		 * extent to map
2458		 */
2459		err = mpage_map_and_submit_buffers(mpd);
2460		if (err < 0)
2461			goto update_disksize;
2462	} while (map->m_len);
2463
2464update_disksize:
2465	/*
2466	 * Update on-disk size after IO is submitted.  Races with
2467	 * truncate are avoided by checking i_size under i_data_sem.
2468	 */
2469	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2470	if (disksize > EXT4_I(inode)->i_disksize) {
2471		int err2;
2472		loff_t i_size;
2473
2474		down_write(&EXT4_I(inode)->i_data_sem);
2475		i_size = i_size_read(inode);
2476		if (disksize > i_size)
2477			disksize = i_size;
2478		if (disksize > EXT4_I(inode)->i_disksize)
2479			EXT4_I(inode)->i_disksize = disksize;
2480		err2 = ext4_mark_inode_dirty(handle, inode);
2481		up_write(&EXT4_I(inode)->i_data_sem);
2482		if (err2)
2483			ext4_error(inode->i_sb,
2484				   "Failed to mark inode %lu dirty",
2485				   inode->i_ino);
2486		if (!err)
2487			err = err2;
2488	}
2489	return err;
2490}
2491
2492/*
2493 * Calculate the total number of credits to reserve for one writepages
2494 * iteration. This is called from ext4_writepages(). We map an extent of
2495 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2496 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2497 * bpp - 1 blocks in bpp different extents.
2498 */
2499static int ext4_da_writepages_trans_blocks(struct inode *inode)
2500{
2501	int bpp = ext4_journal_blocks_per_page(inode);
2502
2503	return ext4_meta_trans_blocks(inode,
2504				MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2505}
2506
2507/*
2508 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2509 * 				 and underlying extent to map
2510 *
2511 * @mpd - where to look for pages
2512 *
2513 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2514 * IO immediately. When we find a page which isn't mapped we start accumulating
2515 * extent of buffers underlying these pages that needs mapping (formed by
2516 * either delayed or unwritten buffers). We also lock the pages containing
2517 * these buffers. The extent found is returned in @mpd structure (starting at
2518 * mpd->lblk with length mpd->len blocks).
2519 *
2520 * Note that this function can attach bios to one io_end structure which are
2521 * neither logically nor physically contiguous. Although it may seem as an
2522 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2523 * case as we need to track IO to all buffers underlying a page in one io_end.
2524 */
2525static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2526{
2527	struct address_space *mapping = mpd->inode->i_mapping;
2528	struct pagevec pvec;
2529	unsigned int nr_pages;
2530	long left = mpd->wbc->nr_to_write;
2531	pgoff_t index = mpd->first_page;
2532	pgoff_t end = mpd->last_page;
2533	int tag;
2534	int i, err = 0;
2535	int blkbits = mpd->inode->i_blkbits;
2536	ext4_lblk_t lblk;
2537	struct buffer_head *head;
2538
2539	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2540		tag = PAGECACHE_TAG_TOWRITE;
2541	else
2542		tag = PAGECACHE_TAG_DIRTY;
2543
2544	pagevec_init(&pvec, 0);
2545	mpd->map.m_len = 0;
2546	mpd->next_page = index;
2547	while (index <= end) {
2548		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2549			      min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2550		if (nr_pages == 0)
2551			goto out;
2552
2553		for (i = 0; i < nr_pages; i++) {
2554			struct page *page = pvec.pages[i];
2555
2556			/*
2557			 * At this point, the page may be truncated or
2558			 * invalidated (changing page->mapping to NULL), or
2559			 * even swizzled back from swapper_space to tmpfs file
2560			 * mapping. However, page->index will not change
2561			 * because we have a reference on the page.
2562			 */
2563			if (page->index > end)
2564				goto out;
2565
 
 
2566			/*
2567			 * Accumulated enough dirty pages? This doesn't apply
2568			 * to WB_SYNC_ALL mode. For integrity sync we have to
2569			 * keep going because someone may be concurrently
2570			 * dirtying pages, and we might have synced a lot of
2571			 * newly appeared dirty pages, but have not synced all
2572			 * of the old dirty pages.
2573			 */
2574			if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2575				goto out;
 
 
 
2576
2577			/* If we can't merge this page, we are done. */
2578			if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2579				goto out;
2580
2581			lock_page(page);
2582			/*
2583			 * If the page is no longer dirty, or its mapping no
2584			 * longer corresponds to inode we are writing (which
2585			 * means it has been truncated or invalidated), or the
2586			 * page is already under writeback and we are not doing
2587			 * a data integrity writeback, skip the page
 
2588			 */
2589			if (!PageDirty(page) ||
2590			    (PageWriteback(page) &&
2591			     (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2592			    unlikely(page->mapping != mapping)) {
2593				unlock_page(page);
2594				continue;
2595			}
2596
2597			wait_on_page_writeback(page);
2598			BUG_ON(PageWriteback(page));
2599
2600			if (mpd->map.m_len == 0)
2601				mpd->first_page = page->index;
2602			mpd->next_page = page->index + 1;
2603			/* Add all dirty buffers to mpd */
2604			lblk = ((ext4_lblk_t)page->index) <<
2605				(PAGE_SHIFT - blkbits);
2606			head = page_buffers(page);
2607			err = mpage_process_page_bufs(mpd, head, head, lblk);
2608			if (err <= 0)
2609				goto out;
2610			err = 0;
2611			left--;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2612		}
2613		pagevec_release(&pvec);
2614		cond_resched();
2615	}
2616	return 0;
 
 
2617out:
2618	pagevec_release(&pvec);
2619	return err;
 
2620}
2621
2622static int __writepage(struct page *page, struct writeback_control *wbc,
2623		       void *data)
2624{
2625	struct address_space *mapping = data;
2626	int ret = ext4_writepage(page, wbc);
2627	mapping_set_error(mapping, ret);
2628	return ret;
2629}
2630
2631static int ext4_writepages(struct address_space *mapping,
2632			   struct writeback_control *wbc)
2633{
2634	pgoff_t	writeback_index = 0;
2635	long nr_to_write = wbc->nr_to_write;
2636	int range_whole = 0;
2637	int cycled = 1;
2638	handle_t *handle = NULL;
2639	struct mpage_da_data mpd;
2640	struct inode *inode = mapping->host;
2641	int needed_blocks, rsv_blocks = 0, ret = 0;
 
 
 
 
 
2642	struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2643	bool done;
2644	struct blk_plug plug;
2645	bool give_up_on_write = false;
2646
2647	percpu_down_read(&sbi->s_journal_flag_rwsem);
2648	trace_ext4_writepages(inode, wbc);
2649
2650	if (dax_mapping(mapping)) {
2651		ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2652						  wbc);
2653		goto out_writepages;
2654	}
2655
2656	/*
2657	 * No pages to write? This is mainly a kludge to avoid starting
2658	 * a transaction for special inodes like journal inode on last iput()
2659	 * because that could violate lock ordering on umount
2660	 */
2661	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2662		goto out_writepages;
2663
2664	if (ext4_should_journal_data(inode)) {
2665		struct blk_plug plug;
2666
2667		blk_start_plug(&plug);
2668		ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2669		blk_finish_plug(&plug);
2670		goto out_writepages;
2671	}
2672
2673	/*
2674	 * If the filesystem has aborted, it is read-only, so return
2675	 * right away instead of dumping stack traces later on that
2676	 * will obscure the real source of the problem.  We test
2677	 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2678	 * the latter could be true if the filesystem is mounted
2679	 * read-only, and in that case, ext4_writepages should
2680	 * *never* be called, so if that ever happens, we would want
2681	 * the stack trace.
2682	 */
2683	if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2684		ret = -EROFS;
2685		goto out_writepages;
2686	}
2687
2688	if (ext4_should_dioread_nolock(inode)) {
2689		/*
2690		 * We may need to convert up to one extent per block in
2691		 * the page and we may dirty the inode.
2692		 */
2693		rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits);
2694	}
2695
2696	/*
2697	 * If we have inline data and arrive here, it means that
2698	 * we will soon create the block for the 1st page, so
2699	 * we'd better clear the inline data here.
2700	 */
2701	if (ext4_has_inline_data(inode)) {
2702		/* Just inode will be modified... */
2703		handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2704		if (IS_ERR(handle)) {
2705			ret = PTR_ERR(handle);
2706			goto out_writepages;
2707		}
2708		BUG_ON(ext4_test_inode_state(inode,
2709				EXT4_STATE_MAY_INLINE_DATA));
2710		ext4_destroy_inline_data(handle, inode);
2711		ext4_journal_stop(handle);
2712	}
2713
2714	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2715		range_whole = 1;
2716
 
2717	if (wbc->range_cyclic) {
2718		writeback_index = mapping->writeback_index;
2719		if (writeback_index)
2720			cycled = 0;
2721		mpd.first_page = writeback_index;
2722		mpd.last_page = -1;
 
 
2723	} else {
2724		mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2725		mpd.last_page = wbc->range_end >> PAGE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2726	}
2727
2728	mpd.inode = inode;
2729	mpd.wbc = wbc;
2730	ext4_io_submit_init(&mpd.io_submit, wbc);
2731retry:
2732	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2733		tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2734	done = false;
2735	blk_start_plug(&plug);
2736	while (!done && mpd.first_page <= mpd.last_page) {
2737		/* For each extent of pages we use new io_end */
2738		mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2739		if (!mpd.io_submit.io_end) {
2740			ret = -ENOMEM;
2741			break;
2742		}
2743
2744		/*
2745		 * We have two constraints: We find one extent to map and we
2746		 * must always write out whole page (makes a difference when
2747		 * blocksize < pagesize) so that we don't block on IO when we
2748		 * try to write out the rest of the page. Journalled mode is
2749		 * not supported by delalloc.
2750		 */
2751		BUG_ON(ext4_should_journal_data(inode));
2752		needed_blocks = ext4_da_writepages_trans_blocks(inode);
2753
2754		/* start a new transaction */
2755		handle = ext4_journal_start_with_reserve(inode,
2756				EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2757		if (IS_ERR(handle)) {
2758			ret = PTR_ERR(handle);
2759			ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2760			       "%ld pages, ino %lu; err %d", __func__,
2761				wbc->nr_to_write, inode->i_ino, ret);
2762			/* Release allocated io_end */
2763			ext4_put_io_end(mpd.io_submit.io_end);
2764			break;
2765		}
2766
2767		trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2768		ret = mpage_prepare_extent_to_map(&mpd);
2769		if (!ret) {
2770			if (mpd.map.m_len)
2771				ret = mpage_map_and_submit_extent(handle, &mpd,
2772					&give_up_on_write);
2773			else {
2774				/*
2775				 * We scanned the whole range (or exhausted
2776				 * nr_to_write), submitted what was mapped and
2777				 * didn't find anything needing mapping. We are
2778				 * done.
2779				 */
2780				done = true;
2781			}
2782		}
2783		/*
2784		 * Caution: If the handle is synchronous,
2785		 * ext4_journal_stop() can wait for transaction commit
2786		 * to finish which may depend on writeback of pages to
2787		 * complete or on page lock to be released.  In that
2788		 * case, we have to wait until after after we have
2789		 * submitted all the IO, released page locks we hold,
2790		 * and dropped io_end reference (for extent conversion
2791		 * to be able to complete) before stopping the handle.
2792		 */
2793		if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2794			ext4_journal_stop(handle);
2795			handle = NULL;
 
2796		}
2797		/* Submit prepared bio */
2798		ext4_io_submit(&mpd.io_submit);
2799		/* Unlock pages we didn't use */
2800		mpage_release_unused_pages(&mpd, give_up_on_write);
2801		/*
2802		 * Drop our io_end reference we got from init. We have
2803		 * to be careful and use deferred io_end finishing if
2804		 * we are still holding the transaction as we can
2805		 * release the last reference to io_end which may end
2806		 * up doing unwritten extent conversion.
2807		 */
2808		if (handle) {
2809			ext4_put_io_end_defer(mpd.io_submit.io_end);
2810			ext4_journal_stop(handle);
2811		} else
2812			ext4_put_io_end(mpd.io_submit.io_end);
2813
2814		if (ret == -ENOSPC && sbi->s_journal) {
2815			/*
2816			 * Commit the transaction which would
2817			 * free blocks released in the transaction
2818			 * and try again
2819			 */
2820			jbd2_journal_force_commit_nested(sbi->s_journal);
2821			ret = 0;
2822			continue;
2823		}
2824		/* Fatal error - ENOMEM, EIO... */
2825		if (ret)
 
 
 
 
 
 
 
 
 
 
2826			break;
2827	}
2828	blk_finish_plug(&plug);
2829	if (!ret && !cycled && wbc->nr_to_write > 0) {
2830		cycled = 1;
2831		mpd.last_page = writeback_index - 1;
2832		mpd.first_page = 0;
 
2833		goto retry;
2834	}
2835
2836	/* Update index */
 
2837	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2838		/*
2839		 * Set the writeback_index so that range_cyclic
2840		 * mode will write it back later
2841		 */
2842		mapping->writeback_index = mpd.first_page;
2843
2844out_writepages:
2845	trace_ext4_writepages_result(inode, wbc, ret,
2846				     nr_to_write - wbc->nr_to_write);
2847	percpu_up_read(&sbi->s_journal_flag_rwsem);
2848	return ret;
2849}
2850
 
2851static int ext4_nonda_switch(struct super_block *sb)
2852{
2853	s64 free_clusters, dirty_clusters;
2854	struct ext4_sb_info *sbi = EXT4_SB(sb);
2855
2856	/*
2857	 * switch to non delalloc mode if we are running low
2858	 * on free block. The free block accounting via percpu
2859	 * counters can get slightly wrong with percpu_counter_batch getting
2860	 * accumulated on each CPU without updating global counters
2861	 * Delalloc need an accurate free block accounting. So switch
2862	 * to non delalloc when we are near to error range.
2863	 */
2864	free_clusters =
2865		percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2866	dirty_clusters =
2867		percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2868	/*
2869	 * Start pushing delalloc when 1/2 of free blocks are dirty.
2870	 */
2871	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2872		try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2873
2874	if (2 * free_clusters < 3 * dirty_clusters ||
2875	    free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2876		/*
2877		 * free block count is less than 150% of dirty blocks
2878		 * or free blocks is less than watermark
2879		 */
2880		return 1;
2881	}
 
 
 
 
 
 
 
2882	return 0;
2883}
2884
2885/* We always reserve for an inode update; the superblock could be there too */
2886static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2887{
2888	if (likely(ext4_has_feature_large_file(inode->i_sb)))
2889		return 1;
2890
2891	if (pos + len <= 0x7fffffffULL)
2892		return 1;
2893
2894	/* We might need to update the superblock to set LARGE_FILE */
2895	return 2;
2896}
2897
2898static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2899			       loff_t pos, unsigned len, unsigned flags,
2900			       struct page **pagep, void **fsdata)
2901{
2902	int ret, retries = 0;
2903	struct page *page;
2904	pgoff_t index;
2905	struct inode *inode = mapping->host;
2906	handle_t *handle;
2907
2908	index = pos >> PAGE_SHIFT;
2909
2910	if (ext4_nonda_switch(inode->i_sb) ||
2911	    S_ISLNK(inode->i_mode)) {
2912		*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2913		return ext4_write_begin(file, mapping, pos,
2914					len, flags, pagep, fsdata);
2915	}
2916	*fsdata = (void *)0;
2917	trace_ext4_da_write_begin(inode, pos, len, flags);
2918
2919	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2920		ret = ext4_da_write_inline_data_begin(mapping, inode,
2921						      pos, len, flags,
2922						      pagep, fsdata);
2923		if (ret < 0)
2924			return ret;
2925		if (ret == 1)
2926			return 0;
2927	}
2928
2929	/*
2930	 * grab_cache_page_write_begin() can take a long time if the
2931	 * system is thrashing due to memory pressure, or if the page
2932	 * is being written back.  So grab it first before we start
2933	 * the transaction handle.  This also allows us to allocate
2934	 * the page (if needed) without using GFP_NOFS.
2935	 */
2936retry_grab:
2937	page = grab_cache_page_write_begin(mapping, index, flags);
2938	if (!page)
2939		return -ENOMEM;
2940	unlock_page(page);
2941
2942	/*
2943	 * With delayed allocation, we don't log the i_disksize update
2944	 * if there is delayed block allocation. But we still need
2945	 * to journalling the i_disksize update if writes to the end
2946	 * of file which has an already mapped buffer.
2947	 */
2948retry_journal:
2949	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2950				ext4_da_write_credits(inode, pos, len));
2951	if (IS_ERR(handle)) {
2952		put_page(page);
2953		return PTR_ERR(handle);
2954	}
 
 
 
2955
2956	lock_page(page);
2957	if (page->mapping != mapping) {
2958		/* The page got truncated from under us */
2959		unlock_page(page);
2960		put_page(page);
2961		ext4_journal_stop(handle);
2962		goto retry_grab;
 
2963	}
2964	/* In case writeback began while the page was unlocked */
2965	wait_for_stable_page(page);
2966
2967#ifdef CONFIG_EXT4_FS_ENCRYPTION
2968	ret = ext4_block_write_begin(page, pos, len,
2969				     ext4_da_get_block_prep);
2970#else
2971	ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
2972#endif
2973	if (ret < 0) {
2974		unlock_page(page);
2975		ext4_journal_stop(handle);
 
2976		/*
2977		 * block_write_begin may have instantiated a few blocks
2978		 * outside i_size.  Trim these off again. Don't need
2979		 * i_size_read because we hold i_mutex.
2980		 */
2981		if (pos + len > inode->i_size)
2982			ext4_truncate_failed_write(inode);
2983
2984		if (ret == -ENOSPC &&
2985		    ext4_should_retry_alloc(inode->i_sb, &retries))
2986			goto retry_journal;
2987
2988		put_page(page);
2989		return ret;
2990	}
2991
2992	*pagep = page;
 
 
2993	return ret;
2994}
2995
2996/*
2997 * Check if we should update i_disksize
2998 * when write to the end of file but not require block allocation
2999 */
3000static int ext4_da_should_update_i_disksize(struct page *page,
3001					    unsigned long offset)
3002{
3003	struct buffer_head *bh;
3004	struct inode *inode = page->mapping->host;
3005	unsigned int idx;
3006	int i;
3007
3008	bh = page_buffers(page);
3009	idx = offset >> inode->i_blkbits;
3010
3011	for (i = 0; i < idx; i++)
3012		bh = bh->b_this_page;
3013
3014	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3015		return 0;
3016	return 1;
3017}
3018
3019static int ext4_da_write_end(struct file *file,
3020			     struct address_space *mapping,
3021			     loff_t pos, unsigned len, unsigned copied,
3022			     struct page *page, void *fsdata)
3023{
3024	struct inode *inode = mapping->host;
3025	int ret = 0, ret2;
3026	handle_t *handle = ext4_journal_current_handle();
3027	loff_t new_i_size;
3028	unsigned long start, end;
3029	int write_mode = (int)(unsigned long)fsdata;
3030
3031	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3032		return ext4_write_end(file, mapping, pos,
3033				      len, copied, page, fsdata);
 
 
 
 
 
 
 
 
3034
3035	trace_ext4_da_write_end(inode, pos, len, copied);
3036	start = pos & (PAGE_SIZE - 1);
3037	end = start + copied - 1;
3038
3039	/*
3040	 * generic_write_end() will run mark_inode_dirty() if i_size
3041	 * changes.  So let's piggyback the i_disksize mark_inode_dirty
3042	 * into that.
3043	 */
 
3044	new_i_size = pos + copied;
3045	if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3046		if (ext4_has_inline_data(inode) ||
3047		    ext4_da_should_update_i_disksize(page, end)) {
3048			ext4_update_i_disksize(inode, new_i_size);
 
 
 
 
 
 
 
 
 
 
 
3049			/* We need to mark inode dirty even if
3050			 * new_i_size is less that inode->i_size
3051			 * bu greater than i_disksize.(hint delalloc)
3052			 */
3053			ext4_mark_inode_dirty(handle, inode);
3054		}
3055	}
3056
3057	if (write_mode != CONVERT_INLINE_DATA &&
3058	    ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3059	    ext4_has_inline_data(inode))
3060		ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3061						     page);
3062	else
3063		ret2 = generic_write_end(file, mapping, pos, len, copied,
3064							page, fsdata);
3065
3066	copied = ret2;
3067	if (ret2 < 0)
3068		ret = ret2;
3069	ret2 = ext4_journal_stop(handle);
3070	if (!ret)
3071		ret = ret2;
3072
3073	return ret ? ret : copied;
3074}
3075
3076static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3077				   unsigned int length)
3078{
3079	/*
3080	 * Drop reserved blocks
3081	 */
3082	BUG_ON(!PageLocked(page));
3083	if (!page_has_buffers(page))
3084		goto out;
3085
3086	ext4_da_page_release_reservation(page, offset, length);
3087
3088out:
3089	ext4_invalidatepage(page, offset, length);
3090
3091	return;
3092}
3093
3094/*
3095 * Force all delayed allocation blocks to be allocated for a given inode.
3096 */
3097int ext4_alloc_da_blocks(struct inode *inode)
3098{
3099	trace_ext4_alloc_da_blocks(inode);
3100
3101	if (!EXT4_I(inode)->i_reserved_data_blocks)
 
3102		return 0;
3103
3104	/*
3105	 * We do something simple for now.  The filemap_flush() will
3106	 * also start triggering a write of the data blocks, which is
3107	 * not strictly speaking necessary (and for users of
3108	 * laptop_mode, not even desirable).  However, to do otherwise
3109	 * would require replicating code paths in:
3110	 *
3111	 * ext4_writepages() ->
3112	 *    write_cache_pages() ---> (via passed in callback function)
3113	 *        __mpage_da_writepage() -->
3114	 *           mpage_add_bh_to_extent()
3115	 *           mpage_da_map_blocks()
3116	 *
3117	 * The problem is that write_cache_pages(), located in
3118	 * mm/page-writeback.c, marks pages clean in preparation for
3119	 * doing I/O, which is not desirable if we're not planning on
3120	 * doing I/O at all.
3121	 *
3122	 * We could call write_cache_pages(), and then redirty all of
3123	 * the pages by calling redirty_page_for_writepage() but that
3124	 * would be ugly in the extreme.  So instead we would need to
3125	 * replicate parts of the code in the above functions,
3126	 * simplifying them because we wouldn't actually intend to
3127	 * write out the pages, but rather only collect contiguous
3128	 * logical block extents, call the multi-block allocator, and
3129	 * then update the buffer heads with the block allocations.
3130	 *
3131	 * For now, though, we'll cheat by calling filemap_flush(),
3132	 * which will map the blocks, and start the I/O, but not
3133	 * actually wait for the I/O to complete.
3134	 */
3135	return filemap_flush(inode->i_mapping);
3136}
3137
3138/*
3139 * bmap() is special.  It gets used by applications such as lilo and by
3140 * the swapper to find the on-disk block of a specific piece of data.
3141 *
3142 * Naturally, this is dangerous if the block concerned is still in the
3143 * journal.  If somebody makes a swapfile on an ext4 data-journaling
3144 * filesystem and enables swap, then they may get a nasty shock when the
3145 * data getting swapped to that swapfile suddenly gets overwritten by
3146 * the original zero's written out previously to the journal and
3147 * awaiting writeback in the kernel's buffer cache.
3148 *
3149 * So, if we see any bmap calls here on a modified, data-journaled file,
3150 * take extra steps to flush any blocks which might be in the cache.
3151 */
3152static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3153{
3154	struct inode *inode = mapping->host;
3155	journal_t *journal;
3156	int err;
3157
3158	/*
3159	 * We can get here for an inline file via the FIBMAP ioctl
3160	 */
3161	if (ext4_has_inline_data(inode))
3162		return 0;
3163
3164	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3165			test_opt(inode->i_sb, DELALLOC)) {
3166		/*
3167		 * With delalloc we want to sync the file
3168		 * so that we can make sure we allocate
3169		 * blocks for file
3170		 */
3171		filemap_write_and_wait(mapping);
3172	}
3173
3174	if (EXT4_JOURNAL(inode) &&
3175	    ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3176		/*
3177		 * This is a REALLY heavyweight approach, but the use of
3178		 * bmap on dirty files is expected to be extremely rare:
3179		 * only if we run lilo or swapon on a freshly made file
3180		 * do we expect this to happen.
3181		 *
3182		 * (bmap requires CAP_SYS_RAWIO so this does not
3183		 * represent an unprivileged user DOS attack --- we'd be
3184		 * in trouble if mortal users could trigger this path at
3185		 * will.)
3186		 *
3187		 * NB. EXT4_STATE_JDATA is not set on files other than
3188		 * regular files.  If somebody wants to bmap a directory
3189		 * or symlink and gets confused because the buffer
3190		 * hasn't yet been flushed to disk, they deserve
3191		 * everything they get.
3192		 */
3193
3194		ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3195		journal = EXT4_JOURNAL(inode);
3196		jbd2_journal_lock_updates(journal);
3197		err = jbd2_journal_flush(journal);
3198		jbd2_journal_unlock_updates(journal);
3199
3200		if (err)
3201			return 0;
3202	}
3203
3204	return generic_block_bmap(mapping, block, ext4_get_block);
3205}
3206
3207static int ext4_readpage(struct file *file, struct page *page)
3208{
3209	int ret = -EAGAIN;
3210	struct inode *inode = page->mapping->host;
3211
3212	trace_ext4_readpage(page);
3213
3214	if (ext4_has_inline_data(inode))
3215		ret = ext4_readpage_inline(inode, page);
3216
3217	if (ret == -EAGAIN)
3218		return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3219
3220	return ret;
3221}
3222
3223static int
3224ext4_readpages(struct file *file, struct address_space *mapping,
3225		struct list_head *pages, unsigned nr_pages)
3226{
3227	struct inode *inode = mapping->host;
3228
3229	/* If the file has inline data, no need to do readpages. */
3230	if (ext4_has_inline_data(inode))
3231		return 0;
3232
3233	return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3234}
3235
3236static void ext4_invalidatepage(struct page *page, unsigned int offset,
3237				unsigned int length)
3238{
3239	trace_ext4_invalidatepage(page, offset, length);
 
3240
3241	/* No journalling happens on data buffers when this function is used */
3242	WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3243
3244	block_invalidatepage(page, offset, length);
 
 
 
 
 
 
 
 
 
3245}
3246
3247static int __ext4_journalled_invalidatepage(struct page *page,
3248					    unsigned int offset,
3249					    unsigned int length)
3250{
3251	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3252
3253	trace_ext4_journalled_invalidatepage(page, offset, length);
3254
3255	/*
 
 
 
 
 
3256	 * If it's a full truncate we just forget about the pending dirtying
3257	 */
3258	if (offset == 0 && length == PAGE_SIZE)
3259		ClearPageChecked(page);
3260
3261	return jbd2_journal_invalidatepage(journal, page, offset, length);
3262}
3263
3264/* Wrapper for aops... */
3265static void ext4_journalled_invalidatepage(struct page *page,
3266					   unsigned int offset,
3267					   unsigned int length)
3268{
3269	WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3270}
3271
3272static int ext4_releasepage(struct page *page, gfp_t wait)
3273{
3274	journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3275
3276	trace_ext4_releasepage(page);
3277
3278	/* Page has dirty journalled data -> cannot release */
3279	if (PageChecked(page))
3280		return 0;
3281	if (journal)
3282		return jbd2_journal_try_to_free_buffers(journal, page, wait);
3283	else
3284		return try_to_free_buffers(page);
3285}
3286
3287#ifdef CONFIG_FS_DAX
3288static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3289			    unsigned flags, struct iomap *iomap)
3290{
3291	unsigned int blkbits = inode->i_blkbits;
3292	unsigned long first_block = offset >> blkbits;
3293	unsigned long last_block = (offset + length - 1) >> blkbits;
3294	struct ext4_map_blocks map;
3295	int ret;
 
 
 
 
3296
3297	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3298		return -ERANGE;
 
 
 
 
 
 
 
3299
3300	map.m_lblk = first_block;
3301	map.m_len = last_block - first_block + 1;
 
3302
3303	if (!(flags & IOMAP_WRITE)) {
3304		ret = ext4_map_blocks(NULL, inode, &map, 0);
3305	} else {
3306		int dio_credits;
3307		handle_t *handle;
3308		int retries = 0;
3309
3310		/* Trim mapping request to maximum we can map at once for DIO */
3311		if (map.m_len > DIO_MAX_BLOCKS)
3312			map.m_len = DIO_MAX_BLOCKS;
3313		dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3314retry:
3315		/*
3316		 * Either we allocate blocks and then we don't get unwritten
3317		 * extent so we have reserved enough credits, or the blocks
3318		 * are already allocated and unwritten and in that case
3319		 * extent conversion fits in the credits as well.
3320		 */
3321		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3322					    dio_credits);
3323		if (IS_ERR(handle))
3324			return PTR_ERR(handle);
3325
3326		ret = ext4_map_blocks(handle, inode, &map,
3327				      EXT4_GET_BLOCKS_CREATE_ZERO);
3328		if (ret < 0) {
3329			ext4_journal_stop(handle);
3330			if (ret == -ENOSPC &&
3331			    ext4_should_retry_alloc(inode->i_sb, &retries))
3332				goto retry;
3333			return ret;
3334		}
3335
3336		/*
3337		 * If we added blocks beyond i_size, we need to make sure they
3338		 * will get truncated if we crash before updating i_size in
3339		 * ext4_iomap_end(). For faults we don't need to do that (and
3340		 * even cannot because for orphan list operations inode_lock is
3341		 * required) - if we happen to instantiate block beyond i_size,
3342		 * it is because we race with truncate which has already added
3343		 * the inode to the orphan list.
3344		 */
3345		if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3346		    (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3347			int err;
3348
3349			err = ext4_orphan_add(handle, inode);
3350			if (err < 0) {
3351				ext4_journal_stop(handle);
3352				return err;
3353			}
3354		}
3355		ext4_journal_stop(handle);
3356	}
3357
3358	iomap->flags = 0;
3359	iomap->bdev = inode->i_sb->s_bdev;
3360	iomap->offset = first_block << blkbits;
 
 
 
 
 
 
 
 
 
 
3361
3362	if (ret == 0) {
3363		iomap->type = IOMAP_HOLE;
3364		iomap->blkno = IOMAP_NULL_BLOCK;
3365		iomap->length = (u64)map.m_len << blkbits;
3366	} else {
3367		if (map.m_flags & EXT4_MAP_MAPPED) {
3368			iomap->type = IOMAP_MAPPED;
3369		} else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3370			iomap->type = IOMAP_UNWRITTEN;
3371		} else {
3372			WARN_ON_ONCE(1);
3373			return -EIO;
3374		}
3375		iomap->blkno = (sector_t)map.m_pblk << (blkbits - 9);
3376		iomap->length = (u64)map.m_len << blkbits;
3377	}
3378
3379	if (map.m_flags & EXT4_MAP_NEW)
3380		iomap->flags |= IOMAP_F_NEW;
3381	return 0;
3382}
3383
3384static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3385			  ssize_t written, unsigned flags, struct iomap *iomap)
3386{
3387	int ret = 0;
3388	handle_t *handle;
3389	int blkbits = inode->i_blkbits;
3390	bool truncate = false;
3391
3392	if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3393		return 0;
3394
3395	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3396	if (IS_ERR(handle)) {
3397		ret = PTR_ERR(handle);
3398		goto orphan_del;
 
3399	}
3400	if (ext4_update_inode_size(inode, offset + written))
3401		ext4_mark_inode_dirty(handle, inode);
3402	/*
3403	 * We may need to truncate allocated but not written blocks beyond EOF.
3404	 */
3405	if (iomap->offset + iomap->length > 
3406	    ALIGN(inode->i_size, 1 << blkbits)) {
3407		ext4_lblk_t written_blk, end_blk;
3408
3409		written_blk = (offset + written) >> blkbits;
3410		end_blk = (offset + length) >> blkbits;
3411		if (written_blk < end_blk && ext4_can_truncate(inode))
3412			truncate = true;
3413	}
3414	/*
3415	 * Remove inode from orphan list if we were extending a inode and
3416	 * everything went fine.
3417	 */
3418	if (!truncate && inode->i_nlink &&
3419	    !list_empty(&EXT4_I(inode)->i_orphan))
3420		ext4_orphan_del(handle, inode);
3421	ext4_journal_stop(handle);
3422	if (truncate) {
3423		ext4_truncate_failed_write(inode);
3424orphan_del:
3425		/*
3426		 * If truncate failed early the inode might still be on the
3427		 * orphan list; we need to make sure the inode is removed from
3428		 * the orphan list in that case.
3429		 */
3430		if (inode->i_nlink)
3431			ext4_orphan_del(NULL, inode);
3432	}
3433	return ret;
 
 
 
3434}
3435
3436struct iomap_ops ext4_iomap_ops = {
3437	.iomap_begin		= ext4_iomap_begin,
3438	.iomap_end		= ext4_iomap_end,
3439};
3440
3441#endif
3442
3443static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3444			    ssize_t size, void *private)
3445{
3446        ext4_io_end_t *io_end = private;
3447
3448	/* if not async direct IO just return */
3449	if (!io_end)
3450		return 0;
3451
3452	ext_debug("ext4_end_io_dio(): io_end 0x%p "
3453		  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3454		  io_end, io_end->inode->i_ino, iocb, offset, size);
3455
3456	/*
3457	 * Error during AIO DIO. We cannot convert unwritten extents as the
3458	 * data was not written. Just clear the unwritten flag and drop io_end.
3459	 */
3460	if (size <= 0) {
3461		ext4_clear_io_unwritten_flag(io_end);
3462		size = 0;
3463	}
3464	io_end->offset = offset;
3465	io_end->size = size;
3466	ext4_put_io_end(io_end);
 
 
 
 
 
 
3467
 
 
3468	return 0;
3469}
3470
3471/*
3472 * Handling of direct IO writes.
3473 *
3474 * For ext4 extent files, ext4 will do direct-io write even to holes,
3475 * preallocated extents, and those write extend the file, no need to
3476 * fall back to buffered IO.
3477 *
3478 * For holes, we fallocate those blocks, mark them as unwritten
3479 * If those blocks were preallocated, we mark sure they are split, but
3480 * still keep the range to write as unwritten.
3481 *
3482 * The unwritten extents will be converted to written when DIO is completed.
3483 * For async direct IO, since the IO may still pending when return, we
3484 * set up an end_io call back function, which will do the conversion
3485 * when async direct IO completed.
3486 *
3487 * If the O_DIRECT write will extend the file then add this inode to the
3488 * orphan list.  So recovery will truncate it back to the original size
3489 * if the machine crashes during the write.
3490 *
3491 */
3492static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
 
 
3493{
3494	struct file *file = iocb->ki_filp;
3495	struct inode *inode = file->f_mapping->host;
3496	struct ext4_inode_info *ei = EXT4_I(inode);
3497	ssize_t ret;
3498	loff_t offset = iocb->ki_pos;
3499	size_t count = iov_iter_count(iter);
3500	int overwrite = 0;
3501	get_block_t *get_block_func = NULL;
3502	int dio_flags = 0;
3503	loff_t final_size = offset + count;
3504	int orphan = 0;
3505	handle_t *handle;
3506
3507	if (final_size > inode->i_size) {
3508		/* Credits for sb + inode write */
3509		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3510		if (IS_ERR(handle)) {
3511			ret = PTR_ERR(handle);
3512			goto out;
3513		}
3514		ret = ext4_orphan_add(handle, inode);
3515		if (ret) {
3516			ext4_journal_stop(handle);
3517			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3518		}
3519		orphan = 1;
3520		ei->i_disksize = inode->i_size;
3521		ext4_journal_stop(handle);
3522	}
3523
3524	BUG_ON(iocb->private == NULL);
3525
3526	/*
3527	 * Make all waiters for direct IO properly wait also for extent
3528	 * conversion. This also disallows race between truncate() and
3529	 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3530	 */
3531	inode_dio_begin(inode);
3532
3533	/* If we do a overwrite dio, i_mutex locking can be released */
3534	overwrite = *((int *)iocb->private);
3535
3536	if (overwrite)
3537		inode_unlock(inode);
3538
3539	/*
3540	 * For extent mapped files we could direct write to holes and fallocate.
3541	 *
3542	 * Allocated blocks to fill the hole are marked as unwritten to prevent
3543	 * parallel buffered read to expose the stale data before DIO complete
3544	 * the data IO.
3545	 *
3546	 * As to previously fallocated extents, ext4 get_block will just simply
3547	 * mark the buffer mapped but still keep the extents unwritten.
3548	 *
3549	 * For non AIO case, we will convert those unwritten extents to written
3550	 * after return back from blockdev_direct_IO. That way we save us from
3551	 * allocating io_end structure and also the overhead of offloading
3552	 * the extent convertion to a workqueue.
3553	 *
3554	 * For async DIO, the conversion needs to be deferred when the
3555	 * IO is completed. The ext4 end_io callback function will be
3556	 * called to take care of the conversion work.  Here for async
3557	 * case, we allocate an io_end structure to hook to the iocb.
3558	 */
3559	iocb->private = NULL;
3560	if (overwrite)
3561		get_block_func = ext4_dio_get_block_overwrite;
3562	else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3563		   round_down(offset, 1 << inode->i_blkbits) >= inode->i_size) {
3564		get_block_func = ext4_dio_get_block;
3565		dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3566	} else if (is_sync_kiocb(iocb)) {
3567		get_block_func = ext4_dio_get_block_unwritten_sync;
3568		dio_flags = DIO_LOCKING;
3569	} else {
3570		get_block_func = ext4_dio_get_block_unwritten_async;
3571		dio_flags = DIO_LOCKING;
3572	}
3573#ifdef CONFIG_EXT4_FS_ENCRYPTION
3574	BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3575#endif
3576	ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3577				   get_block_func, ext4_end_io_dio, NULL,
3578				   dio_flags);
3579
3580	if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3581						EXT4_STATE_DIO_UNWRITTEN)) {
3582		int err;
3583		/*
3584		 * for non AIO case, since the IO is already
3585		 * completed, we could do the conversion right here
3586		 */
3587		err = ext4_convert_unwritten_extents(NULL, inode,
3588						     offset, ret);
3589		if (err < 0)
3590			ret = err;
3591		ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3592	}
3593
3594	inode_dio_end(inode);
3595	/* take i_mutex locking again if we do a ovewrite dio */
3596	if (overwrite)
3597		inode_lock(inode);
3598
3599	if (ret < 0 && final_size > inode->i_size)
3600		ext4_truncate_failed_write(inode);
3601
3602	/* Handle extending of i_size after direct IO write */
3603	if (orphan) {
3604		int err;
3605
3606		/* Credits for sb + inode write */
3607		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3608		if (IS_ERR(handle)) {
3609			/* This is really bad luck. We've written the data
3610			 * but cannot extend i_size. Bail out and pretend
3611			 * the write failed... */
3612			ret = PTR_ERR(handle);
3613			if (inode->i_nlink)
3614				ext4_orphan_del(NULL, inode);
3615
3616			goto out;
3617		}
3618		if (inode->i_nlink)
3619			ext4_orphan_del(handle, inode);
3620		if (ret > 0) {
3621			loff_t end = offset + ret;
3622			if (end > inode->i_size) {
3623				ei->i_disksize = end;
3624				i_size_write(inode, end);
3625				/*
3626				 * We're going to return a positive `ret'
3627				 * here due to non-zero-length I/O, so there's
3628				 * no way of reporting error returns from
3629				 * ext4_mark_inode_dirty() to userspace.  So
3630				 * ignore it.
3631				 */
3632				ext4_mark_inode_dirty(handle, inode);
3633			}
3634		}
3635		err = ext4_journal_stop(handle);
3636		if (ret == 0)
3637			ret = err;
3638	}
3639out:
3640	return ret;
3641}
3642
3643static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3644{
3645	struct address_space *mapping = iocb->ki_filp->f_mapping;
3646	struct inode *inode = mapping->host;
3647	size_t count = iov_iter_count(iter);
3648	ssize_t ret;
3649
3650	/*
3651	 * Shared inode_lock is enough for us - it protects against concurrent
3652	 * writes & truncates and since we take care of writing back page cache,
3653	 * we are protected against page writeback as well.
3654	 */
3655	inode_lock_shared(inode);
3656	ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3657					   iocb->ki_pos + count);
3658	if (ret)
3659		goto out_unlock;
3660	ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3661				   iter, ext4_dio_get_block, NULL, NULL, 0);
3662out_unlock:
3663	inode_unlock_shared(inode);
3664	return ret;
3665}
3666
3667static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 
 
3668{
3669	struct file *file = iocb->ki_filp;
3670	struct inode *inode = file->f_mapping->host;
3671	size_t count = iov_iter_count(iter);
3672	loff_t offset = iocb->ki_pos;
3673	ssize_t ret;
3674
3675#ifdef CONFIG_EXT4_FS_ENCRYPTION
3676	if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3677		return 0;
3678#endif
3679
3680	/*
3681	 * If we are doing data journalling we don't support O_DIRECT
3682	 */
3683	if (ext4_should_journal_data(inode))
3684		return 0;
3685
3686	/* Let buffer I/O handle the inline data case. */
3687	if (ext4_has_inline_data(inode))
3688		return 0;
3689
3690	/* DAX uses iomap path now */
3691	if (WARN_ON_ONCE(IS_DAX(inode)))
3692		return 0;
3693
3694	trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3695	if (iov_iter_rw(iter) == READ)
3696		ret = ext4_direct_IO_read(iocb, iter);
3697	else
3698		ret = ext4_direct_IO_write(iocb, iter);
3699	trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
 
3700	return ret;
3701}
3702
3703/*
3704 * Pages can be marked dirty completely asynchronously from ext4's journalling
3705 * activity.  By filemap_sync_pte(), try_to_unmap_one(), etc.  We cannot do
3706 * much here because ->set_page_dirty is called under VFS locks.  The page is
3707 * not necessarily locked.
3708 *
3709 * We cannot just dirty the page and leave attached buffers clean, because the
3710 * buffers' dirty state is "definitive".  We cannot just set the buffers dirty
3711 * or jbddirty because all the journalling code will explode.
3712 *
3713 * So what we do is to mark the page "pending dirty" and next time writepage
3714 * is called, propagate that into the buffers appropriately.
3715 */
3716static int ext4_journalled_set_page_dirty(struct page *page)
3717{
3718	SetPageChecked(page);
3719	return __set_page_dirty_nobuffers(page);
3720}
3721
3722static int ext4_set_page_dirty(struct page *page)
3723{
3724	WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3725	WARN_ON_ONCE(!page_has_buffers(page));
3726	return __set_page_dirty_buffers(page);
3727}
 
 
 
 
 
 
 
 
3728
3729static const struct address_space_operations ext4_aops = {
3730	.readpage		= ext4_readpage,
3731	.readpages		= ext4_readpages,
3732	.writepage		= ext4_writepage,
3733	.writepages		= ext4_writepages,
3734	.write_begin		= ext4_write_begin,
3735	.write_end		= ext4_write_end,
3736	.set_page_dirty		= ext4_set_page_dirty,
3737	.bmap			= ext4_bmap,
3738	.invalidatepage		= ext4_invalidatepage,
3739	.releasepage		= ext4_releasepage,
3740	.direct_IO		= ext4_direct_IO,
3741	.migratepage		= buffer_migrate_page,
3742	.is_partially_uptodate  = block_is_partially_uptodate,
3743	.error_remove_page	= generic_error_remove_page,
3744};
3745
3746static const struct address_space_operations ext4_journalled_aops = {
3747	.readpage		= ext4_readpage,
3748	.readpages		= ext4_readpages,
3749	.writepage		= ext4_writepage,
3750	.writepages		= ext4_writepages,
3751	.write_begin		= ext4_write_begin,
3752	.write_end		= ext4_journalled_write_end,
3753	.set_page_dirty		= ext4_journalled_set_page_dirty,
3754	.bmap			= ext4_bmap,
3755	.invalidatepage		= ext4_journalled_invalidatepage,
3756	.releasepage		= ext4_releasepage,
3757	.direct_IO		= ext4_direct_IO,
3758	.is_partially_uptodate  = block_is_partially_uptodate,
3759	.error_remove_page	= generic_error_remove_page,
3760};
3761
3762static const struct address_space_operations ext4_da_aops = {
3763	.readpage		= ext4_readpage,
3764	.readpages		= ext4_readpages,
3765	.writepage		= ext4_writepage,
3766	.writepages		= ext4_writepages,
3767	.write_begin		= ext4_da_write_begin,
3768	.write_end		= ext4_da_write_end,
3769	.set_page_dirty		= ext4_set_page_dirty,
3770	.bmap			= ext4_bmap,
3771	.invalidatepage		= ext4_da_invalidatepage,
3772	.releasepage		= ext4_releasepage,
3773	.direct_IO		= ext4_direct_IO,
3774	.migratepage		= buffer_migrate_page,
3775	.is_partially_uptodate  = block_is_partially_uptodate,
3776	.error_remove_page	= generic_error_remove_page,
3777};
3778
3779void ext4_set_aops(struct inode *inode)
3780{
3781	switch (ext4_inode_journal_mode(inode)) {
3782	case EXT4_INODE_ORDERED_DATA_MODE:
3783	case EXT4_INODE_WRITEBACK_DATA_MODE:
3784		break;
3785	case EXT4_INODE_JOURNAL_DATA_MODE:
3786		inode->i_mapping->a_ops = &ext4_journalled_aops;
3787		return;
3788	default:
3789		BUG();
3790	}
3791	if (test_opt(inode->i_sb, DELALLOC))
3792		inode->i_mapping->a_ops = &ext4_da_aops;
 
 
3793	else
3794		inode->i_mapping->a_ops = &ext4_aops;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3795}
3796
3797static int __ext4_block_zero_page_range(handle_t *handle,
 
 
 
 
 
 
 
3798		struct address_space *mapping, loff_t from, loff_t length)
3799{
3800	ext4_fsblk_t index = from >> PAGE_SHIFT;
3801	unsigned offset = from & (PAGE_SIZE-1);
3802	unsigned blocksize, pos;
3803	ext4_lblk_t iblock;
3804	struct inode *inode = mapping->host;
3805	struct buffer_head *bh;
3806	struct page *page;
3807	int err = 0;
3808
3809	page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3810				   mapping_gfp_constraint(mapping, ~__GFP_FS));
3811	if (!page)
3812		return -ENOMEM;
3813
3814	blocksize = inode->i_sb->s_blocksize;
 
3815
3816	iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
 
 
 
 
 
 
 
3817
3818	if (!page_has_buffers(page))
3819		create_empty_buffers(page, blocksize, 0);
3820
3821	/* Find the buffer that contains "offset" */
3822	bh = page_buffers(page);
3823	pos = blocksize;
3824	while (offset >= pos) {
3825		bh = bh->b_this_page;
3826		iblock++;
3827		pos += blocksize;
3828	}
 
 
3829	if (buffer_freed(bh)) {
3830		BUFFER_TRACE(bh, "freed: skip");
3831		goto unlock;
3832	}
 
3833	if (!buffer_mapped(bh)) {
3834		BUFFER_TRACE(bh, "unmapped");
3835		ext4_get_block(inode, iblock, bh, 0);
3836		/* unmapped? It's a hole - nothing to do */
3837		if (!buffer_mapped(bh)) {
3838			BUFFER_TRACE(bh, "still unmapped");
3839			goto unlock;
3840		}
3841	}
3842
3843	/* Ok, it's mapped. Make sure it's up-to-date */
3844	if (PageUptodate(page))
3845		set_buffer_uptodate(bh);
3846
3847	if (!buffer_uptodate(bh)) {
3848		err = -EIO;
3849		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3850		wait_on_buffer(bh);
3851		/* Uhhuh. Read error. Complain and punt. */
3852		if (!buffer_uptodate(bh))
3853			goto unlock;
3854		if (S_ISREG(inode->i_mode) &&
3855		    ext4_encrypted_inode(inode)) {
3856			/* We expect the key to be set. */
3857			BUG_ON(!fscrypt_has_encryption_key(inode));
3858			BUG_ON(blocksize != PAGE_SIZE);
3859			WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
3860						page, PAGE_SIZE, 0, page->index));
3861		}
3862	}
 
3863	if (ext4_should_journal_data(inode)) {
3864		BUFFER_TRACE(bh, "get write access");
3865		err = ext4_journal_get_write_access(handle, bh);
3866		if (err)
3867			goto unlock;
3868	}
 
3869	zero_user(page, offset, length);
 
3870	BUFFER_TRACE(bh, "zeroed end of block");
3871
 
3872	if (ext4_should_journal_data(inode)) {
3873		err = ext4_handle_dirty_metadata(handle, inode, bh);
3874	} else {
3875		err = 0;
 
3876		mark_buffer_dirty(bh);
3877		if (ext4_should_order_data(inode))
3878			err = ext4_jbd2_inode_add_write(handle, inode);
3879	}
3880
3881unlock:
3882	unlock_page(page);
3883	put_page(page);
3884	return err;
3885}
3886
3887/*
3888 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3889 * starting from file offset 'from'.  The range to be zero'd must
3890 * be contained with in one block.  If the specified range exceeds
3891 * the end of the block it will be shortened to end of the block
3892 * that cooresponds to 'from'
3893 */
3894static int ext4_block_zero_page_range(handle_t *handle,
3895		struct address_space *mapping, loff_t from, loff_t length)
3896{
3897	struct inode *inode = mapping->host;
3898	unsigned offset = from & (PAGE_SIZE-1);
3899	unsigned blocksize = inode->i_sb->s_blocksize;
3900	unsigned max = blocksize - (offset & (blocksize - 1));
3901
3902	/*
3903	 * correct length if it does not fall between
3904	 * 'from' and the end of the block
3905	 */
3906	if (length > max || length < 0)
3907		length = max;
3908
3909	if (IS_DAX(inode)) {
3910		return iomap_zero_range(inode, from, length, NULL,
3911					&ext4_iomap_ops);
3912	}
3913	return __ext4_block_zero_page_range(handle, mapping, from, length);
3914}
3915
3916/*
3917 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3918 * up to the end of the block which corresponds to `from'.
3919 * This required during truncate. We need to physically zero the tail end
3920 * of that block so it doesn't yield old data if the file is later grown.
3921 */
3922static int ext4_block_truncate_page(handle_t *handle,
3923		struct address_space *mapping, loff_t from)
3924{
3925	unsigned offset = from & (PAGE_SIZE-1);
3926	unsigned length;
3927	unsigned blocksize;
3928	struct inode *inode = mapping->host;
3929
3930	/* If we are processing an encrypted inode during orphan list handling */
3931	if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
3932		return 0;
3933
3934	blocksize = inode->i_sb->s_blocksize;
3935	length = blocksize - (offset & (blocksize - 1));
3936
3937	return ext4_block_zero_page_range(handle, mapping, from, length);
3938}
3939
3940int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3941			     loff_t lstart, loff_t length)
3942{
3943	struct super_block *sb = inode->i_sb;
3944	struct address_space *mapping = inode->i_mapping;
3945	unsigned partial_start, partial_end;
3946	ext4_fsblk_t start, end;
3947	loff_t byte_end = (lstart + length - 1);
3948	int err = 0;
3949
3950	partial_start = lstart & (sb->s_blocksize - 1);
3951	partial_end = byte_end & (sb->s_blocksize - 1);
3952
3953	start = lstart >> sb->s_blocksize_bits;
3954	end = byte_end >> sb->s_blocksize_bits;
3955
3956	/* Handle partial zero within the single block */
3957	if (start == end &&
3958	    (partial_start || (partial_end != sb->s_blocksize - 1))) {
3959		err = ext4_block_zero_page_range(handle, mapping,
3960						 lstart, length);
3961		return err;
3962	}
3963	/* Handle partial zero out on the start of the range */
3964	if (partial_start) {
3965		err = ext4_block_zero_page_range(handle, mapping,
3966						 lstart, sb->s_blocksize);
3967		if (err)
3968			return err;
3969	}
3970	/* Handle partial zero out on the end of the range */
3971	if (partial_end != sb->s_blocksize - 1)
3972		err = ext4_block_zero_page_range(handle, mapping,
3973						 byte_end - partial_end,
3974						 partial_end + 1);
3975	return err;
3976}
3977
3978int ext4_can_truncate(struct inode *inode)
3979{
3980	if (S_ISREG(inode->i_mode))
3981		return 1;
3982	if (S_ISDIR(inode->i_mode))
3983		return 1;
3984	if (S_ISLNK(inode->i_mode))
3985		return !ext4_inode_is_fast_symlink(inode);
3986	return 0;
3987}
3988
3989/*
3990 * We have to make sure i_disksize gets properly updated before we truncate
3991 * page cache due to hole punching or zero range. Otherwise i_disksize update
3992 * can get lost as it may have been postponed to submission of writeback but
3993 * that will never happen after we truncate page cache.
3994 */
3995int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3996				      loff_t len)
3997{
3998	handle_t *handle;
3999	loff_t size = i_size_read(inode);
4000
4001	WARN_ON(!inode_is_locked(inode));
4002	if (offset > size || offset + len < size)
4003		return 0;
4004
4005	if (EXT4_I(inode)->i_disksize >= size)
4006		return 0;
4007
4008	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4009	if (IS_ERR(handle))
4010		return PTR_ERR(handle);
4011	ext4_update_i_disksize(inode, size);
4012	ext4_mark_inode_dirty(handle, inode);
4013	ext4_journal_stop(handle);
4014
4015	return 0;
4016}
4017
4018/*
4019 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4020 * associated with the given offset and length
4021 *
4022 * @inode:  File inode
4023 * @offset: The offset where the hole will begin
4024 * @len:    The length of the hole
4025 *
4026 * Returns: 0 on success or negative on failure
4027 */
4028
4029int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4030{
4031	struct super_block *sb = inode->i_sb;
4032	ext4_lblk_t first_block, stop_block;
4033	struct address_space *mapping = inode->i_mapping;
4034	loff_t first_block_offset, last_block_offset;
4035	handle_t *handle;
4036	unsigned int credits;
4037	int ret = 0;
4038
4039	if (!S_ISREG(inode->i_mode))
4040		return -EOPNOTSUPP;
4041
4042	trace_ext4_punch_hole(inode, offset, length, 0);
4043
4044	/*
4045	 * Write out all dirty pages to avoid race conditions
4046	 * Then release them.
4047	 */
4048	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4049		ret = filemap_write_and_wait_range(mapping, offset,
4050						   offset + length - 1);
4051		if (ret)
4052			return ret;
4053	}
4054
4055	inode_lock(inode);
4056
4057	/* No need to punch hole beyond i_size */
4058	if (offset >= inode->i_size)
4059		goto out_mutex;
4060
4061	/*
4062	 * If the hole extends beyond i_size, set the hole
4063	 * to end after the page that contains i_size
4064	 */
4065	if (offset + length > inode->i_size) {
4066		length = inode->i_size +
4067		   PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4068		   offset;
4069	}
4070
4071	if (offset & (sb->s_blocksize - 1) ||
4072	    (offset + length) & (sb->s_blocksize - 1)) {
4073		/*
4074		 * Attach jinode to inode for jbd2 if we do any zeroing of
4075		 * partial block
4076		 */
4077		ret = ext4_inode_attach_jinode(inode);
4078		if (ret < 0)
4079			goto out_mutex;
4080
4081	}
4082
4083	/* Wait all existing dio workers, newcomers will block on i_mutex */
4084	ext4_inode_block_unlocked_dio(inode);
4085	inode_dio_wait(inode);
4086
4087	/*
4088	 * Prevent page faults from reinstantiating pages we have released from
4089	 * page cache.
4090	 */
4091	down_write(&EXT4_I(inode)->i_mmap_sem);
4092	first_block_offset = round_up(offset, sb->s_blocksize);
4093	last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4094
4095	/* Now release the pages and zero block aligned part of pages*/
4096	if (last_block_offset > first_block_offset) {
4097		ret = ext4_update_disksize_before_punch(inode, offset, length);
4098		if (ret)
4099			goto out_dio;
4100		truncate_pagecache_range(inode, first_block_offset,
4101					 last_block_offset);
4102	}
4103
4104	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4105		credits = ext4_writepage_trans_blocks(inode);
4106	else
4107		credits = ext4_blocks_for_truncate(inode);
4108	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4109	if (IS_ERR(handle)) {
4110		ret = PTR_ERR(handle);
4111		ext4_std_error(sb, ret);
4112		goto out_dio;
4113	}
4114
4115	ret = ext4_zero_partial_blocks(handle, inode, offset,
4116				       length);
4117	if (ret)
4118		goto out_stop;
4119
4120	first_block = (offset + sb->s_blocksize - 1) >>
4121		EXT4_BLOCK_SIZE_BITS(sb);
4122	stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4123
4124	/* If there are no blocks to remove, return now */
4125	if (first_block >= stop_block)
4126		goto out_stop;
4127
4128	down_write(&EXT4_I(inode)->i_data_sem);
4129	ext4_discard_preallocations(inode);
4130
4131	ret = ext4_es_remove_extent(inode, first_block,
4132				    stop_block - first_block);
4133	if (ret) {
4134		up_write(&EXT4_I(inode)->i_data_sem);
4135		goto out_stop;
4136	}
4137
4138	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4139		ret = ext4_ext_remove_space(inode, first_block,
4140					    stop_block - 1);
4141	else
4142		ret = ext4_ind_remove_space(handle, inode, first_block,
4143					    stop_block);
4144
4145	up_write(&EXT4_I(inode)->i_data_sem);
4146	if (IS_SYNC(inode))
4147		ext4_handle_sync(handle);
4148
4149	inode->i_mtime = inode->i_ctime = current_time(inode);
4150	ext4_mark_inode_dirty(handle, inode);
4151out_stop:
4152	ext4_journal_stop(handle);
4153out_dio:
4154	up_write(&EXT4_I(inode)->i_mmap_sem);
4155	ext4_inode_resume_unlocked_dio(inode);
4156out_mutex:
4157	inode_unlock(inode);
4158	return ret;
4159}
4160
4161int ext4_inode_attach_jinode(struct inode *inode)
4162{
4163	struct ext4_inode_info *ei = EXT4_I(inode);
4164	struct jbd2_inode *jinode;
4165
4166	if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4167		return 0;
4168
4169	jinode = jbd2_alloc_inode(GFP_KERNEL);
4170	spin_lock(&inode->i_lock);
4171	if (!ei->jinode) {
4172		if (!jinode) {
4173			spin_unlock(&inode->i_lock);
4174			return -ENOMEM;
4175		}
4176		ei->jinode = jinode;
4177		jbd2_journal_init_jbd_inode(ei->jinode, inode);
4178		jinode = NULL;
4179	}
4180	spin_unlock(&inode->i_lock);
4181	if (unlikely(jinode != NULL))
4182		jbd2_free_inode(jinode);
4183	return 0;
4184}
4185
4186/*
4187 * ext4_truncate()
4188 *
4189 * We block out ext4_get_block() block instantiations across the entire
4190 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4191 * simultaneously on behalf of the same inode.
4192 *
4193 * As we work through the truncate and commit bits of it to the journal there
4194 * is one core, guiding principle: the file's tree must always be consistent on
4195 * disk.  We must be able to restart the truncate after a crash.
4196 *
4197 * The file's tree may be transiently inconsistent in memory (although it
4198 * probably isn't), but whenever we close off and commit a journal transaction,
4199 * the contents of (the filesystem + the journal) must be consistent and
4200 * restartable.  It's pretty simple, really: bottom up, right to left (although
4201 * left-to-right works OK too).
4202 *
4203 * Note that at recovery time, journal replay occurs *before* the restart of
4204 * truncate against the orphan inode list.
4205 *
4206 * The committed inode has the new, desired i_size (which is the same as
4207 * i_disksize in this case).  After a crash, ext4_orphan_cleanup() will see
4208 * that this inode's truncate did not complete and it will again call
4209 * ext4_truncate() to have another go.  So there will be instantiated blocks
4210 * to the right of the truncation point in a crashed ext4 filesystem.  But
4211 * that's fine - as long as they are linked from the inode, the post-crash
4212 * ext4_truncate() run will find them and release them.
4213 */
4214int ext4_truncate(struct inode *inode)
4215{
4216	struct ext4_inode_info *ei = EXT4_I(inode);
4217	unsigned int credits;
4218	int err = 0;
4219	handle_t *handle;
4220	struct address_space *mapping = inode->i_mapping;
4221
4222	/*
4223	 * There is a possibility that we're either freeing the inode
4224	 * or it's a completely new inode. In those cases we might not
4225	 * have i_mutex locked because it's not necessary.
4226	 */
4227	if (!(inode->i_state & (I_NEW|I_FREEING)))
4228		WARN_ON(!inode_is_locked(inode));
4229	trace_ext4_truncate_enter(inode);
4230
4231	if (!ext4_can_truncate(inode))
4232		return 0;
4233
4234	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4235
4236	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4237		ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4238
4239	if (ext4_has_inline_data(inode)) {
4240		int has_inline = 1;
4241
4242		ext4_inline_data_truncate(inode, &has_inline);
4243		if (has_inline)
4244			return 0;
4245	}
4246
4247	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4248	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4249		if (ext4_inode_attach_jinode(inode) < 0)
4250			return 0;
4251	}
4252
4253	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4254		credits = ext4_writepage_trans_blocks(inode);
4255	else
4256		credits = ext4_blocks_for_truncate(inode);
4257
4258	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4259	if (IS_ERR(handle))
4260		return PTR_ERR(handle);
4261
4262	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4263		ext4_block_truncate_page(handle, mapping, inode->i_size);
4264
4265	/*
4266	 * We add the inode to the orphan list, so that if this
4267	 * truncate spans multiple transactions, and we crash, we will
4268	 * resume the truncate when the filesystem recovers.  It also
4269	 * marks the inode dirty, to catch the new size.
4270	 *
4271	 * Implication: the file must always be in a sane, consistent
4272	 * truncatable state while each transaction commits.
4273	 */
4274	err = ext4_orphan_add(handle, inode);
4275	if (err)
4276		goto out_stop;
4277
4278	down_write(&EXT4_I(inode)->i_data_sem);
4279
4280	ext4_discard_preallocations(inode);
4281
4282	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4283		err = ext4_ext_truncate(handle, inode);
4284	else
4285		ext4_ind_truncate(handle, inode);
4286
4287	up_write(&ei->i_data_sem);
4288	if (err)
4289		goto out_stop;
4290
4291	if (IS_SYNC(inode))
4292		ext4_handle_sync(handle);
4293
4294out_stop:
4295	/*
4296	 * If this was a simple ftruncate() and the file will remain alive,
4297	 * then we need to clear up the orphan record which we created above.
4298	 * However, if this was a real unlink then we were called by
4299	 * ext4_evict_inode(), and we allow that function to clean up the
4300	 * orphan info for us.
4301	 */
4302	if (inode->i_nlink)
4303		ext4_orphan_del(handle, inode);
4304
4305	inode->i_mtime = inode->i_ctime = current_time(inode);
4306	ext4_mark_inode_dirty(handle, inode);
4307	ext4_journal_stop(handle);
4308
4309	trace_ext4_truncate_exit(inode);
4310	return err;
4311}
4312
4313/*
4314 * ext4_get_inode_loc returns with an extra refcount against the inode's
4315 * underlying buffer_head on success. If 'in_mem' is true, we have all
4316 * data in memory that is needed to recreate the on-disk version of this
4317 * inode.
4318 */
4319static int __ext4_get_inode_loc(struct inode *inode,
4320				struct ext4_iloc *iloc, int in_mem)
4321{
4322	struct ext4_group_desc	*gdp;
4323	struct buffer_head	*bh;
4324	struct super_block	*sb = inode->i_sb;
4325	ext4_fsblk_t		block;
4326	int			inodes_per_block, inode_offset;
4327
4328	iloc->bh = NULL;
4329	if (!ext4_valid_inum(sb, inode->i_ino))
4330		return -EFSCORRUPTED;
4331
4332	iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4333	gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4334	if (!gdp)
4335		return -EIO;
4336
4337	/*
4338	 * Figure out the offset within the block group inode table
4339	 */
4340	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4341	inode_offset = ((inode->i_ino - 1) %
4342			EXT4_INODES_PER_GROUP(sb));
4343	block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4344	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4345
4346	bh = sb_getblk(sb, block);
4347	if (unlikely(!bh))
4348		return -ENOMEM;
 
 
 
4349	if (!buffer_uptodate(bh)) {
4350		lock_buffer(bh);
4351
4352		/*
4353		 * If the buffer has the write error flag, we have failed
4354		 * to write out another inode in the same block.  In this
4355		 * case, we don't have to read the block because we may
4356		 * read the old inode data successfully.
4357		 */
4358		if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4359			set_buffer_uptodate(bh);
4360
4361		if (buffer_uptodate(bh)) {
4362			/* someone brought it uptodate while we waited */
4363			unlock_buffer(bh);
4364			goto has_buffer;
4365		}
4366
4367		/*
4368		 * If we have all information of the inode in memory and this
4369		 * is the only valid inode in the block, we need not read the
4370		 * block.
4371		 */
4372		if (in_mem) {
4373			struct buffer_head *bitmap_bh;
4374			int i, start;
4375
4376			start = inode_offset & ~(inodes_per_block - 1);
4377
4378			/* Is the inode bitmap in cache? */
4379			bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4380			if (unlikely(!bitmap_bh))
4381				goto make_io;
4382
4383			/*
4384			 * If the inode bitmap isn't in cache then the
4385			 * optimisation may end up performing two reads instead
4386			 * of one, so skip it.
4387			 */
4388			if (!buffer_uptodate(bitmap_bh)) {
4389				brelse(bitmap_bh);
4390				goto make_io;
4391			}
4392			for (i = start; i < start + inodes_per_block; i++) {
4393				if (i == inode_offset)
4394					continue;
4395				if (ext4_test_bit(i, bitmap_bh->b_data))
4396					break;
4397			}
4398			brelse(bitmap_bh);
4399			if (i == start + inodes_per_block) {
4400				/* all other inodes are free, so skip I/O */
4401				memset(bh->b_data, 0, bh->b_size);
4402				set_buffer_uptodate(bh);
4403				unlock_buffer(bh);
4404				goto has_buffer;
4405			}
4406		}
4407
4408make_io:
4409		/*
4410		 * If we need to do any I/O, try to pre-readahead extra
4411		 * blocks from the inode table.
4412		 */
4413		if (EXT4_SB(sb)->s_inode_readahead_blks) {
4414			ext4_fsblk_t b, end, table;
4415			unsigned num;
4416			__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4417
4418			table = ext4_inode_table(sb, gdp);
4419			/* s_inode_readahead_blks is always a power of 2 */
4420			b = block & ~((ext4_fsblk_t) ra_blks - 1);
4421			if (table > b)
4422				b = table;
4423			end = b + ra_blks;
4424			num = EXT4_INODES_PER_GROUP(sb);
4425			if (ext4_has_group_desc_csum(sb))
 
4426				num -= ext4_itable_unused_count(sb, gdp);
4427			table += num / inodes_per_block;
4428			if (end > table)
4429				end = table;
4430			while (b <= end)
4431				sb_breadahead(sb, b++);
4432		}
4433
4434		/*
4435		 * There are other valid inodes in the buffer, this inode
4436		 * has in-inode xattrs, or we don't have this inode in memory.
4437		 * Read the block from disk.
4438		 */
4439		trace_ext4_load_inode(inode);
4440		get_bh(bh);
4441		bh->b_end_io = end_buffer_read_sync;
4442		submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4443		wait_on_buffer(bh);
4444		if (!buffer_uptodate(bh)) {
4445			EXT4_ERROR_INODE_BLOCK(inode, block,
4446					       "unable to read itable block");
4447			brelse(bh);
4448			return -EIO;
4449		}
4450	}
4451has_buffer:
4452	iloc->bh = bh;
4453	return 0;
4454}
4455
4456int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4457{
4458	/* We have all inode data except xattrs in memory here. */
4459	return __ext4_get_inode_loc(inode, iloc,
4460		!ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4461}
4462
4463void ext4_set_inode_flags(struct inode *inode)
4464{
4465	unsigned int flags = EXT4_I(inode)->i_flags;
4466	unsigned int new_fl = 0;
4467
 
4468	if (flags & EXT4_SYNC_FL)
4469		new_fl |= S_SYNC;
4470	if (flags & EXT4_APPEND_FL)
4471		new_fl |= S_APPEND;
4472	if (flags & EXT4_IMMUTABLE_FL)
4473		new_fl |= S_IMMUTABLE;
4474	if (flags & EXT4_NOATIME_FL)
4475		new_fl |= S_NOATIME;
4476	if (flags & EXT4_DIRSYNC_FL)
4477		new_fl |= S_DIRSYNC;
4478	if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) &&
4479	    !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) &&
4480	    !ext4_encrypted_inode(inode))
4481		new_fl |= S_DAX;
4482	inode_set_flags(inode, new_fl,
4483			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4484}
4485
4486/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4487void ext4_get_inode_flags(struct ext4_inode_info *ei)
4488{
4489	unsigned int vfs_fl;
4490	unsigned long old_fl, new_fl;
4491
4492	do {
4493		vfs_fl = ei->vfs_inode.i_flags;
4494		old_fl = ei->i_flags;
4495		new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4496				EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4497				EXT4_DIRSYNC_FL);
4498		if (vfs_fl & S_SYNC)
4499			new_fl |= EXT4_SYNC_FL;
4500		if (vfs_fl & S_APPEND)
4501			new_fl |= EXT4_APPEND_FL;
4502		if (vfs_fl & S_IMMUTABLE)
4503			new_fl |= EXT4_IMMUTABLE_FL;
4504		if (vfs_fl & S_NOATIME)
4505			new_fl |= EXT4_NOATIME_FL;
4506		if (vfs_fl & S_DIRSYNC)
4507			new_fl |= EXT4_DIRSYNC_FL;
4508	} while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4509}
4510
4511static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4512				  struct ext4_inode_info *ei)
4513{
4514	blkcnt_t i_blocks ;
4515	struct inode *inode = &(ei->vfs_inode);
4516	struct super_block *sb = inode->i_sb;
4517
4518	if (ext4_has_feature_huge_file(sb)) {
 
4519		/* we are using combined 48 bit field */
4520		i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4521					le32_to_cpu(raw_inode->i_blocks_lo);
4522		if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4523			/* i_blocks represent file system block size */
4524			return i_blocks  << (inode->i_blkbits - 9);
4525		} else {
4526			return i_blocks;
4527		}
4528	} else {
4529		return le32_to_cpu(raw_inode->i_blocks_lo);
4530	}
4531}
4532
4533static inline void ext4_iget_extra_inode(struct inode *inode,
4534					 struct ext4_inode *raw_inode,
4535					 struct ext4_inode_info *ei)
4536{
4537	__le32 *magic = (void *)raw_inode +
4538			EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4539	if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4540	    EXT4_INODE_SIZE(inode->i_sb) &&
4541	    *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4542		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4543		ext4_find_inline_data_nolock(inode);
4544	} else
4545		EXT4_I(inode)->i_inline_off = 0;
4546}
4547
4548int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4549{
4550	if (!ext4_has_feature_project(inode->i_sb))
4551		return -EOPNOTSUPP;
4552	*projid = EXT4_I(inode)->i_projid;
4553	return 0;
4554}
4555
4556struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4557{
4558	struct ext4_iloc iloc;
4559	struct ext4_inode *raw_inode;
4560	struct ext4_inode_info *ei;
4561	struct inode *inode;
4562	journal_t *journal = EXT4_SB(sb)->s_journal;
4563	long ret;
4564	loff_t size;
4565	int block;
4566	uid_t i_uid;
4567	gid_t i_gid;
4568	projid_t i_projid;
4569
4570	inode = iget_locked(sb, ino);
4571	if (!inode)
4572		return ERR_PTR(-ENOMEM);
4573	if (!(inode->i_state & I_NEW))
4574		return inode;
4575
4576	ei = EXT4_I(inode);
4577	iloc.bh = NULL;
4578
4579	ret = __ext4_get_inode_loc(inode, &iloc, 0);
4580	if (ret < 0)
4581		goto bad_inode;
4582	raw_inode = ext4_raw_inode(&iloc);
4583
4584	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4585		ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4586		if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4587			EXT4_INODE_SIZE(inode->i_sb) ||
4588		    (ei->i_extra_isize & 3)) {
4589			EXT4_ERROR_INODE(inode,
4590					 "bad extra_isize %u (inode size %u)",
4591					 ei->i_extra_isize,
4592					 EXT4_INODE_SIZE(inode->i_sb));
4593			ret = -EFSCORRUPTED;
4594			goto bad_inode;
4595		}
4596	} else
4597		ei->i_extra_isize = 0;
4598
4599	/* Precompute checksum seed for inode metadata */
4600	if (ext4_has_metadata_csum(sb)) {
4601		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4602		__u32 csum;
4603		__le32 inum = cpu_to_le32(inode->i_ino);
4604		__le32 gen = raw_inode->i_generation;
4605		csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4606				   sizeof(inum));
4607		ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4608					      sizeof(gen));
4609	}
4610
4611	if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4612		EXT4_ERROR_INODE(inode, "checksum invalid");
4613		ret = -EFSBADCRC;
4614		goto bad_inode;
4615	}
4616
4617	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4618	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4619	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4620	if (ext4_has_feature_project(sb) &&
4621	    EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4622	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4623		i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4624	else
4625		i_projid = EXT4_DEF_PROJID;
4626
4627	if (!(test_opt(inode->i_sb, NO_UID32))) {
4628		i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4629		i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4630	}
4631	i_uid_write(inode, i_uid);
4632	i_gid_write(inode, i_gid);
4633	ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4634	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4635
4636	ext4_clear_state_flags(ei);	/* Only relevant on 32-bit archs */
4637	ei->i_inline_off = 0;
4638	ei->i_dir_start_lookup = 0;
4639	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4640	/* We now have enough fields to check if the inode was active or not.
4641	 * This is needed because nfsd might try to access dead inodes
4642	 * the test is that same one that e2fsck uses
4643	 * NeilBrown 1999oct15
4644	 */
4645	if (inode->i_nlink == 0) {
4646		if ((inode->i_mode == 0 ||
4647		     !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4648		    ino != EXT4_BOOT_LOADER_INO) {
4649			/* this inode is deleted */
4650			ret = -ESTALE;
4651			goto bad_inode;
4652		}
4653		/* The only unlinked inodes we let through here have
4654		 * valid i_mode and are being read by the orphan
4655		 * recovery code: that's fine, we're about to complete
4656		 * the process of deleting those.
4657		 * OR it is the EXT4_BOOT_LOADER_INO which is
4658		 * not initialized on a new filesystem. */
4659	}
4660	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4661	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4662	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4663	if (ext4_has_feature_64bit(sb))
4664		ei->i_file_acl |=
4665			((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4666	inode->i_size = ext4_isize(raw_inode);
4667	if ((size = i_size_read(inode)) < 0) {
4668		EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
4669		ret = -EFSCORRUPTED;
4670		goto bad_inode;
4671	}
4672	ei->i_disksize = inode->i_size;
4673#ifdef CONFIG_QUOTA
4674	ei->i_reserved_quota = 0;
4675#endif
4676	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4677	ei->i_block_group = iloc.block_group;
4678	ei->i_last_alloc_group = ~0;
4679	/*
4680	 * NOTE! The in-memory inode i_data array is in little-endian order
4681	 * even on big-endian machines: we do NOT byteswap the block numbers!
4682	 */
4683	for (block = 0; block < EXT4_N_BLOCKS; block++)
4684		ei->i_data[block] = raw_inode->i_block[block];
4685	INIT_LIST_HEAD(&ei->i_orphan);
4686
4687	/*
4688	 * Set transaction id's of transactions that have to be committed
4689	 * to finish f[data]sync. We set them to currently running transaction
4690	 * as we cannot be sure that the inode or some of its metadata isn't
4691	 * part of the transaction - the inode could have been reclaimed and
4692	 * now it is reread from disk.
4693	 */
4694	if (journal) {
4695		transaction_t *transaction;
4696		tid_t tid;
4697
4698		read_lock(&journal->j_state_lock);
4699		if (journal->j_running_transaction)
4700			transaction = journal->j_running_transaction;
4701		else
4702			transaction = journal->j_committing_transaction;
4703		if (transaction)
4704			tid = transaction->t_tid;
4705		else
4706			tid = journal->j_commit_sequence;
4707		read_unlock(&journal->j_state_lock);
4708		ei->i_sync_tid = tid;
4709		ei->i_datasync_tid = tid;
4710	}
4711
4712	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
 
 
 
 
 
 
4713		if (ei->i_extra_isize == 0) {
4714			/* The extra space is currently unused. Use it. */
4715			BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4716			ei->i_extra_isize = sizeof(struct ext4_inode) -
4717					    EXT4_GOOD_OLD_INODE_SIZE;
4718		} else {
4719			ext4_iget_extra_inode(inode, raw_inode, ei);
 
 
 
 
4720		}
4721	}
 
4722
4723	EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4724	EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4725	EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4726	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4727
4728	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4729		inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4730		if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4731			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4732				inode->i_version |=
4733		    (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4734		}
4735	}
4736
4737	ret = 0;
4738	if (ei->i_file_acl &&
4739	    !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4740		EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4741				 ei->i_file_acl);
4742		ret = -EFSCORRUPTED;
4743		goto bad_inode;
4744	} else if (!ext4_has_inline_data(inode)) {
4745		if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4746			if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4747			    (S_ISLNK(inode->i_mode) &&
4748			     !ext4_inode_is_fast_symlink(inode))))
4749				/* Validate extent which is part of inode */
4750				ret = ext4_ext_check_inode(inode);
4751		} else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4752			   (S_ISLNK(inode->i_mode) &&
4753			    !ext4_inode_is_fast_symlink(inode))) {
4754			/* Validate block references which are part of inode */
4755			ret = ext4_ind_check_inode(inode);
4756		}
4757	}
4758	if (ret)
4759		goto bad_inode;
4760
4761	if (S_ISREG(inode->i_mode)) {
4762		inode->i_op = &ext4_file_inode_operations;
4763		inode->i_fop = &ext4_file_operations;
4764		ext4_set_aops(inode);
4765	} else if (S_ISDIR(inode->i_mode)) {
4766		inode->i_op = &ext4_dir_inode_operations;
4767		inode->i_fop = &ext4_dir_operations;
4768	} else if (S_ISLNK(inode->i_mode)) {
4769		if (ext4_encrypted_inode(inode)) {
4770			inode->i_op = &ext4_encrypted_symlink_inode_operations;
4771			ext4_set_aops(inode);
4772		} else if (ext4_inode_is_fast_symlink(inode)) {
4773			inode->i_link = (char *)ei->i_data;
4774			inode->i_op = &ext4_fast_symlink_inode_operations;
4775			nd_terminate_link(ei->i_data, inode->i_size,
4776				sizeof(ei->i_data) - 1);
4777		} else {
4778			inode->i_op = &ext4_symlink_inode_operations;
4779			ext4_set_aops(inode);
4780		}
4781		inode_nohighmem(inode);
4782	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4783	      S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4784		inode->i_op = &ext4_special_inode_operations;
4785		if (raw_inode->i_block[0])
4786			init_special_inode(inode, inode->i_mode,
4787			   old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4788		else
4789			init_special_inode(inode, inode->i_mode,
4790			   new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4791	} else if (ino == EXT4_BOOT_LOADER_INO) {
4792		make_bad_inode(inode);
4793	} else {
4794		ret = -EFSCORRUPTED;
4795		EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4796		goto bad_inode;
4797	}
4798	brelse(iloc.bh);
4799	ext4_set_inode_flags(inode);
4800	unlock_new_inode(inode);
4801	return inode;
4802
4803bad_inode:
4804	brelse(iloc.bh);
4805	iget_failed(inode);
4806	return ERR_PTR(ret);
4807}
4808
4809struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4810{
4811	if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4812		return ERR_PTR(-EFSCORRUPTED);
4813	return ext4_iget(sb, ino);
4814}
4815
4816static int ext4_inode_blocks_set(handle_t *handle,
4817				struct ext4_inode *raw_inode,
4818				struct ext4_inode_info *ei)
4819{
4820	struct inode *inode = &(ei->vfs_inode);
4821	u64 i_blocks = inode->i_blocks;
4822	struct super_block *sb = inode->i_sb;
4823
4824	if (i_blocks <= ~0U) {
4825		/*
4826		 * i_blocks can be represented in a 32 bit variable
4827		 * as multiple of 512 bytes
4828		 */
4829		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4830		raw_inode->i_blocks_high = 0;
4831		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4832		return 0;
4833	}
4834	if (!ext4_has_feature_huge_file(sb))
4835		return -EFBIG;
4836
4837	if (i_blocks <= 0xffffffffffffULL) {
4838		/*
4839		 * i_blocks can be represented in a 48 bit variable
4840		 * as multiple of 512 bytes
4841		 */
4842		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4843		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4844		ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4845	} else {
4846		ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4847		/* i_block is stored in file system block size */
4848		i_blocks = i_blocks >> (inode->i_blkbits - 9);
4849		raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
4850		raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4851	}
4852	return 0;
4853}
4854
4855struct other_inode {
4856	unsigned long		orig_ino;
4857	struct ext4_inode	*raw_inode;
4858};
4859
4860static int other_inode_match(struct inode * inode, unsigned long ino,
4861			     void *data)
4862{
4863	struct other_inode *oi = (struct other_inode *) data;
4864
4865	if ((inode->i_ino != ino) ||
4866	    (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4867			       I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4868	    ((inode->i_state & I_DIRTY_TIME) == 0))
4869		return 0;
4870	spin_lock(&inode->i_lock);
4871	if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4872				I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4873	    (inode->i_state & I_DIRTY_TIME)) {
4874		struct ext4_inode_info	*ei = EXT4_I(inode);
4875
4876		inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4877		spin_unlock(&inode->i_lock);
4878
4879		spin_lock(&ei->i_raw_lock);
4880		EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4881		EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4882		EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4883		ext4_inode_csum_set(inode, oi->raw_inode, ei);
4884		spin_unlock(&ei->i_raw_lock);
4885		trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4886		return -1;
4887	}
4888	spin_unlock(&inode->i_lock);
4889	return -1;
4890}
4891
4892/*
4893 * Opportunistically update the other time fields for other inodes in
4894 * the same inode table block.
4895 */
4896static void ext4_update_other_inodes_time(struct super_block *sb,
4897					  unsigned long orig_ino, char *buf)
4898{
4899	struct other_inode oi;
4900	unsigned long ino;
4901	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4902	int inode_size = EXT4_INODE_SIZE(sb);
4903
4904	oi.orig_ino = orig_ino;
4905	/*
4906	 * Calculate the first inode in the inode table block.  Inode
4907	 * numbers are one-based.  That is, the first inode in a block
4908	 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4909	 */
4910	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4911	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4912		if (ino == orig_ino)
4913			continue;
4914		oi.raw_inode = (struct ext4_inode *) buf;
4915		(void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4916	}
4917}
4918
4919/*
4920 * Post the struct inode info into an on-disk inode location in the
4921 * buffer-cache.  This gobbles the caller's reference to the
4922 * buffer_head in the inode location struct.
4923 *
4924 * The caller must have write access to iloc->bh.
4925 */
4926static int ext4_do_update_inode(handle_t *handle,
4927				struct inode *inode,
4928				struct ext4_iloc *iloc)
4929{
4930	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4931	struct ext4_inode_info *ei = EXT4_I(inode);
4932	struct buffer_head *bh = iloc->bh;
4933	struct super_block *sb = inode->i_sb;
4934	int err = 0, rc, block;
4935	int need_datasync = 0, set_large_file = 0;
4936	uid_t i_uid;
4937	gid_t i_gid;
4938	projid_t i_projid;
4939
4940	spin_lock(&ei->i_raw_lock);
4941
4942	/* For fields not tracked in the in-memory inode,
4943	 * initialise them to zero for new inodes. */
4944	if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4945		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4946
4947	ext4_get_inode_flags(ei);
4948	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4949	i_uid = i_uid_read(inode);
4950	i_gid = i_gid_read(inode);
4951	i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4952	if (!(test_opt(inode->i_sb, NO_UID32))) {
4953		raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4954		raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4955/*
4956 * Fix up interoperability with old kernels. Otherwise, old inodes get
4957 * re-used with the upper 16 bits of the uid/gid intact
4958 */
4959		if (ei->i_dtime && list_empty(&ei->i_orphan)) {
 
 
 
 
 
4960			raw_inode->i_uid_high = 0;
4961			raw_inode->i_gid_high = 0;
4962		} else {
4963			raw_inode->i_uid_high =
4964				cpu_to_le16(high_16_bits(i_uid));
4965			raw_inode->i_gid_high =
4966				cpu_to_le16(high_16_bits(i_gid));
4967		}
4968	} else {
4969		raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4970		raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
 
 
4971		raw_inode->i_uid_high = 0;
4972		raw_inode->i_gid_high = 0;
4973	}
4974	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4975
4976	EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4977	EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4978	EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4979	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4980
4981	err = ext4_inode_blocks_set(handle, raw_inode, ei);
4982	if (err) {
4983		spin_unlock(&ei->i_raw_lock);
4984		goto out_brelse;
4985	}
4986	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4987	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4988	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
 
4989		raw_inode->i_file_acl_high =
4990			cpu_to_le16(ei->i_file_acl >> 32);
4991	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4992	if (ei->i_disksize != ext4_isize(raw_inode)) {
4993		ext4_isize_set(raw_inode, ei->i_disksize);
4994		need_datasync = 1;
4995	}
4996	if (ei->i_disksize > 0x7fffffffULL) {
4997		if (!ext4_has_feature_large_file(sb) ||
 
 
4998				EXT4_SB(sb)->s_es->s_rev_level ==
4999		    cpu_to_le32(EXT4_GOOD_OLD_REV))
5000			set_large_file = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5001	}
5002	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5003	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5004		if (old_valid_dev(inode->i_rdev)) {
5005			raw_inode->i_block[0] =
5006				cpu_to_le32(old_encode_dev(inode->i_rdev));
5007			raw_inode->i_block[1] = 0;
5008		} else {
5009			raw_inode->i_block[0] = 0;
5010			raw_inode->i_block[1] =
5011				cpu_to_le32(new_encode_dev(inode->i_rdev));
5012			raw_inode->i_block[2] = 0;
5013		}
5014	} else if (!ext4_has_inline_data(inode)) {
5015		for (block = 0; block < EXT4_N_BLOCKS; block++)
5016			raw_inode->i_block[block] = ei->i_data[block];
5017	}
5018
5019	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5020		raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5021		if (ei->i_extra_isize) {
5022			if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5023				raw_inode->i_version_hi =
5024					cpu_to_le32(inode->i_version >> 32);
5025			raw_inode->i_extra_isize =
5026				cpu_to_le16(ei->i_extra_isize);
5027		}
5028	}
5029
5030	BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5031	       i_projid != EXT4_DEF_PROJID);
5032
5033	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5034	    EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5035		raw_inode->i_projid = cpu_to_le32(i_projid);
5036
5037	ext4_inode_csum_set(inode, raw_inode, ei);
5038	spin_unlock(&ei->i_raw_lock);
5039	if (inode->i_sb->s_flags & MS_LAZYTIME)
5040		ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5041					      bh->b_data);
5042
5043	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5044	rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5045	if (!err)
5046		err = rc;
5047	ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5048	if (set_large_file) {
5049		BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5050		err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5051		if (err)
5052			goto out_brelse;
5053		ext4_update_dynamic_rev(sb);
5054		ext4_set_feature_large_file(sb);
5055		ext4_handle_sync(handle);
5056		err = ext4_handle_dirty_super(handle, sb);
5057	}
5058	ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5059out_brelse:
5060	brelse(bh);
5061	ext4_std_error(inode->i_sb, err);
5062	return err;
5063}
5064
5065/*
5066 * ext4_write_inode()
5067 *
5068 * We are called from a few places:
5069 *
5070 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5071 *   Here, there will be no transaction running. We wait for any running
5072 *   transaction to commit.
5073 *
5074 * - Within flush work (sys_sync(), kupdate and such).
5075 *   We wait on commit, if told to.
5076 *
5077 * - Within iput_final() -> write_inode_now()
5078 *   We wait on commit, if told to.
 
5079 *
5080 * In all cases it is actually safe for us to return without doing anything,
5081 * because the inode has been copied into a raw inode buffer in
5082 * ext4_mark_inode_dirty().  This is a correctness thing for WB_SYNC_ALL
5083 * writeback.
5084 *
5085 * Note that we are absolutely dependent upon all inode dirtiers doing the
5086 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5087 * which we are interested.
5088 *
5089 * It would be a bug for them to not do this.  The code:
5090 *
5091 *	mark_inode_dirty(inode)
5092 *	stuff();
5093 *	inode->i_size = expr;
5094 *
5095 * is in error because write_inode() could occur while `stuff()' is running,
5096 * and the new i_size will be lost.  Plus the inode will no longer be on the
5097 * superblock's dirty inode list.
5098 */
5099int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5100{
5101	int err;
5102
5103	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5104		return 0;
5105
5106	if (EXT4_SB(inode->i_sb)->s_journal) {
5107		if (ext4_journal_current_handle()) {
5108			jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5109			dump_stack();
5110			return -EIO;
5111		}
5112
5113		/*
5114		 * No need to force transaction in WB_SYNC_NONE mode. Also
5115		 * ext4_sync_fs() will force the commit after everything is
5116		 * written.
5117		 */
5118		if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5119			return 0;
5120
5121		err = ext4_force_commit(inode->i_sb);
5122	} else {
5123		struct ext4_iloc iloc;
5124
5125		err = __ext4_get_inode_loc(inode, &iloc, 0);
5126		if (err)
5127			return err;
5128		/*
5129		 * sync(2) will flush the whole buffer cache. No need to do
5130		 * it here separately for each inode.
5131		 */
5132		if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5133			sync_dirty_buffer(iloc.bh);
5134		if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5135			EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5136					 "IO error syncing inode");
5137			err = -EIO;
5138		}
5139		brelse(iloc.bh);
5140	}
5141	return err;
5142}
5143
5144/*
5145 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5146 * buffers that are attached to a page stradding i_size and are undergoing
5147 * commit. In that case we have to wait for commit to finish and try again.
5148 */
5149static void ext4_wait_for_tail_page_commit(struct inode *inode)
5150{
5151	struct page *page;
5152	unsigned offset;
5153	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5154	tid_t commit_tid = 0;
5155	int ret;
5156
5157	offset = inode->i_size & (PAGE_SIZE - 1);
5158	/*
5159	 * All buffers in the last page remain valid? Then there's nothing to
5160	 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5161	 * blocksize case
5162	 */
5163	if (offset > PAGE_SIZE - (1 << inode->i_blkbits))
5164		return;
5165	while (1) {
5166		page = find_lock_page(inode->i_mapping,
5167				      inode->i_size >> PAGE_SHIFT);
5168		if (!page)
5169			return;
5170		ret = __ext4_journalled_invalidatepage(page, offset,
5171						PAGE_SIZE - offset);
5172		unlock_page(page);
5173		put_page(page);
5174		if (ret != -EBUSY)
5175			return;
5176		commit_tid = 0;
5177		read_lock(&journal->j_state_lock);
5178		if (journal->j_committing_transaction)
5179			commit_tid = journal->j_committing_transaction->t_tid;
5180		read_unlock(&journal->j_state_lock);
5181		if (commit_tid)
5182			jbd2_log_wait_commit(journal, commit_tid);
5183	}
5184}
5185
5186/*
5187 * ext4_setattr()
5188 *
5189 * Called from notify_change.
5190 *
5191 * We want to trap VFS attempts to truncate the file as soon as
5192 * possible.  In particular, we want to make sure that when the VFS
5193 * shrinks i_size, we put the inode on the orphan list and modify
5194 * i_disksize immediately, so that during the subsequent flushing of
5195 * dirty pages and freeing of disk blocks, we can guarantee that any
5196 * commit will leave the blocks being flushed in an unused state on
5197 * disk.  (On recovery, the inode will get truncated and the blocks will
5198 * be freed, so we have a strong guarantee that no future commit will
5199 * leave these blocks visible to the user.)
5200 *
5201 * Another thing we have to assure is that if we are in ordered mode
5202 * and inode is still attached to the committing transaction, we must
5203 * we start writeout of all the dirty pages which are being truncated.
5204 * This way we are sure that all the data written in the previous
5205 * transaction are already on disk (truncate waits for pages under
5206 * writeback).
5207 *
5208 * Called with inode->i_mutex down.
5209 */
5210int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5211{
5212	struct inode *inode = d_inode(dentry);
5213	int error, rc = 0;
5214	int orphan = 0;
5215	const unsigned int ia_valid = attr->ia_valid;
5216
5217	error = setattr_prepare(dentry, attr);
5218	if (error)
5219		return error;
5220
5221	if (is_quota_modification(inode, attr)) {
5222		error = dquot_initialize(inode);
5223		if (error)
5224			return error;
5225	}
5226	if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5227	    (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5228		handle_t *handle;
5229
5230		/* (user+group)*(old+new) structure, inode write (sb,
5231		 * inode block, ? - but truncate inode update has it) */
5232		handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5233			(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5234			 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5235		if (IS_ERR(handle)) {
5236			error = PTR_ERR(handle);
5237			goto err_out;
5238		}
5239		error = dquot_transfer(inode, attr);
5240		if (error) {
5241			ext4_journal_stop(handle);
5242			return error;
5243		}
5244		/* Update corresponding info in inode so that everything is in
5245		 * one transaction */
5246		if (attr->ia_valid & ATTR_UID)
5247			inode->i_uid = attr->ia_uid;
5248		if (attr->ia_valid & ATTR_GID)
5249			inode->i_gid = attr->ia_gid;
5250		error = ext4_mark_inode_dirty(handle, inode);
5251		ext4_journal_stop(handle);
5252	}
5253
5254	if (attr->ia_valid & ATTR_SIZE) {
5255		handle_t *handle;
5256		loff_t oldsize = inode->i_size;
5257		int shrink = (attr->ia_size <= inode->i_size);
5258
5259		if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5260			struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5261
5262			if (attr->ia_size > sbi->s_bitmap_maxbytes)
5263				return -EFBIG;
5264		}
5265		if (!S_ISREG(inode->i_mode))
5266			return -EINVAL;
 
 
 
 
5267
5268		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5269			inode_inc_iversion(inode);
 
 
 
 
 
 
 
 
 
 
 
 
5270
5271		if (ext4_should_order_data(inode) &&
5272		    (attr->ia_size < inode->i_size)) {
5273			error = ext4_begin_ordered_truncate(inode,
5274							    attr->ia_size);
5275			if (error)
5276				goto err_out;
5277		}
5278		if (attr->ia_size != inode->i_size) {
5279			handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5280			if (IS_ERR(handle)) {
5281				error = PTR_ERR(handle);
5282				goto err_out;
5283			}
5284			if (ext4_handle_valid(handle) && shrink) {
5285				error = ext4_orphan_add(handle, inode);
5286				orphan = 1;
5287			}
5288			/*
5289			 * Update c/mtime on truncate up, ext4_truncate() will
5290			 * update c/mtime in shrink case below
5291			 */
5292			if (!shrink) {
5293				inode->i_mtime = current_time(inode);
5294				inode->i_ctime = inode->i_mtime;
5295			}
5296			down_write(&EXT4_I(inode)->i_data_sem);
5297			EXT4_I(inode)->i_disksize = attr->ia_size;
5298			rc = ext4_mark_inode_dirty(handle, inode);
5299			if (!error)
5300				error = rc;
5301			/*
5302			 * We have to update i_size under i_data_sem together
5303			 * with i_disksize to avoid races with writeback code
5304			 * running ext4_wb_update_i_disksize().
5305			 */
5306			if (!error)
5307				i_size_write(inode, attr->ia_size);
5308			up_write(&EXT4_I(inode)->i_data_sem);
5309			ext4_journal_stop(handle);
5310			if (error) {
5311				if (orphan)
 
 
5312					ext4_orphan_del(NULL, inode);
 
 
 
 
 
5313				goto err_out;
5314			}
5315		}
5316		if (!shrink)
5317			pagecache_isize_extended(inode, oldsize, inode->i_size);
5318
5319		/*
5320		 * Blocks are going to be removed from the inode. Wait
5321		 * for dio in flight.  Temporarily disable
5322		 * dioread_nolock to prevent livelock.
5323		 */
5324		if (orphan) {
5325			if (!ext4_should_journal_data(inode)) {
5326				ext4_inode_block_unlocked_dio(inode);
5327				inode_dio_wait(inode);
5328				ext4_inode_resume_unlocked_dio(inode);
5329			} else
5330				ext4_wait_for_tail_page_commit(inode);
5331		}
5332		down_write(&EXT4_I(inode)->i_mmap_sem);
5333		/*
5334		 * Truncate pagecache after we've waited for commit
5335		 * in data=journal mode to make pages freeable.
5336		 */
5337		truncate_pagecache(inode, inode->i_size);
5338		if (shrink) {
5339			rc = ext4_truncate(inode);
5340			if (rc)
5341				error = rc;
5342		}
5343		up_write(&EXT4_I(inode)->i_mmap_sem);
5344	}
5345
5346	if (!error) {
5347		setattr_copy(inode, attr);
5348		mark_inode_dirty(inode);
5349	}
5350
5351	/*
5352	 * If the call to ext4_truncate failed to get a transaction handle at
5353	 * all, we need to clean up the in-core orphan list manually.
5354	 */
5355	if (orphan && inode->i_nlink)
5356		ext4_orphan_del(NULL, inode);
5357
5358	if (!error && (ia_valid & ATTR_MODE))
5359		rc = posix_acl_chmod(inode, inode->i_mode);
5360
5361err_out:
5362	ext4_std_error(inode->i_sb, error);
5363	if (!error)
5364		error = rc;
5365	return error;
5366}
5367
5368int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5369		 struct kstat *stat)
5370{
5371	struct inode *inode;
5372	unsigned long long delalloc_blocks;
5373
5374	inode = d_inode(dentry);
5375	generic_fillattr(inode, stat);
5376
5377	/*
5378	 * If there is inline data in the inode, the inode will normally not
5379	 * have data blocks allocated (it may have an external xattr block).
5380	 * Report at least one sector for such files, so tools like tar, rsync,
5381	 * others doen't incorrectly think the file is completely sparse.
5382	 */
5383	if (unlikely(ext4_has_inline_data(inode)))
5384		stat->blocks += (stat->size + 511) >> 9;
5385
5386	/*
5387	 * We can't update i_blocks if the block allocation is delayed
5388	 * otherwise in the case of system crash before the real block
5389	 * allocation is done, we will have i_blocks inconsistent with
5390	 * on-disk file blocks.
5391	 * We always keep i_blocks updated together with real
5392	 * allocation. But to not confuse with user, stat
5393	 * will return the blocks that include the delayed allocation
5394	 * blocks for this file.
5395	 */
5396	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5397				   EXT4_I(inode)->i_reserved_data_blocks);
5398	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5399	return 0;
5400}
5401
5402static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5403				   int pextents)
5404{
5405	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5406		return ext4_ind_trans_blocks(inode, lblocks);
5407	return ext4_ext_index_trans_blocks(inode, pextents);
5408}
5409
5410/*
5411 * Account for index blocks, block groups bitmaps and block group
5412 * descriptor blocks if modify datablocks and index blocks
5413 * worse case, the indexs blocks spread over different block groups
5414 *
5415 * If datablocks are discontiguous, they are possible to spread over
5416 * different block groups too. If they are contiguous, with flexbg,
5417 * they could still across block group boundary.
5418 *
5419 * Also account for superblock, inode, quota and xattr blocks
5420 */
5421static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5422				  int pextents)
5423{
5424	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5425	int gdpblocks;
5426	int idxblocks;
5427	int ret = 0;
5428
5429	/*
5430	 * How many index blocks need to touch to map @lblocks logical blocks
5431	 * to @pextents physical extents?
 
 
 
 
5432	 */
5433	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5434
5435	ret = idxblocks;
5436
5437	/*
5438	 * Now let's see how many group bitmaps and group descriptors need
5439	 * to account
5440	 */
5441	groups = idxblocks + pextents;
 
 
 
 
 
5442	gdpblocks = groups;
5443	if (groups > ngroups)
5444		groups = ngroups;
5445	if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5446		gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5447
5448	/* bitmaps and block group descriptor blocks */
5449	ret += groups + gdpblocks;
5450
5451	/* Blocks for super block, inode, quota and xattr blocks */
5452	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5453
5454	return ret;
5455}
5456
5457/*
5458 * Calculate the total number of credits to reserve to fit
5459 * the modification of a single pages into a single transaction,
5460 * which may include multiple chunks of block allocations.
5461 *
5462 * This could be called via ext4_write_begin()
5463 *
5464 * We need to consider the worse case, when
5465 * one new block per extent.
5466 */
5467int ext4_writepage_trans_blocks(struct inode *inode)
5468{
5469	int bpp = ext4_journal_blocks_per_page(inode);
5470	int ret;
5471
5472	ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5473
5474	/* Account for data blocks for journalled mode */
5475	if (ext4_should_journal_data(inode))
5476		ret += bpp;
5477	return ret;
5478}
5479
5480/*
5481 * Calculate the journal credits for a chunk of data modification.
5482 *
5483 * This is called from DIO, fallocate or whoever calling
5484 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5485 *
5486 * journal buffers for data blocks are not included here, as DIO
5487 * and fallocate do no need to journal data buffers.
5488 */
5489int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5490{
5491	return ext4_meta_trans_blocks(inode, nrblocks, 1);
5492}
5493
5494/*
5495 * The caller must have previously called ext4_reserve_inode_write().
5496 * Give this, we know that the caller already has write access to iloc->bh.
5497 */
5498int ext4_mark_iloc_dirty(handle_t *handle,
5499			 struct inode *inode, struct ext4_iloc *iloc)
5500{
5501	int err = 0;
5502
5503	if (IS_I_VERSION(inode))
5504		inode_inc_iversion(inode);
5505
5506	/* the do_update_inode consumes one bh->b_count */
5507	get_bh(iloc->bh);
5508
5509	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5510	err = ext4_do_update_inode(handle, inode, iloc);
5511	put_bh(iloc->bh);
5512	return err;
5513}
5514
5515/*
5516 * On success, We end up with an outstanding reference count against
5517 * iloc->bh.  This _must_ be cleaned up later.
5518 */
5519
5520int
5521ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5522			 struct ext4_iloc *iloc)
5523{
5524	int err;
5525
5526	err = ext4_get_inode_loc(inode, iloc);
5527	if (!err) {
5528		BUFFER_TRACE(iloc->bh, "get_write_access");
5529		err = ext4_journal_get_write_access(handle, iloc->bh);
5530		if (err) {
5531			brelse(iloc->bh);
5532			iloc->bh = NULL;
5533		}
5534	}
5535	ext4_std_error(inode->i_sb, err);
5536	return err;
5537}
5538
5539/*
5540 * Expand an inode by new_extra_isize bytes.
5541 * Returns 0 on success or negative error number on failure.
5542 */
5543static int ext4_expand_extra_isize(struct inode *inode,
5544				   unsigned int new_extra_isize,
5545				   struct ext4_iloc iloc,
5546				   handle_t *handle)
5547{
5548	struct ext4_inode *raw_inode;
5549	struct ext4_xattr_ibody_header *header;
5550
5551	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5552		return 0;
5553
5554	raw_inode = ext4_raw_inode(&iloc);
5555
5556	header = IHDR(inode, raw_inode);
5557
5558	/* No extended attributes present */
5559	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5560	    header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5561		memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5562			new_extra_isize);
5563		EXT4_I(inode)->i_extra_isize = new_extra_isize;
5564		return 0;
5565	}
5566
5567	/* try to expand with EAs present */
5568	return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5569					  raw_inode, handle);
5570}
5571
5572/*
5573 * What we do here is to mark the in-core inode as clean with respect to inode
5574 * dirtiness (it may still be data-dirty).
5575 * This means that the in-core inode may be reaped by prune_icache
5576 * without having to perform any I/O.  This is a very good thing,
5577 * because *any* task may call prune_icache - even ones which
5578 * have a transaction open against a different journal.
5579 *
5580 * Is this cheating?  Not really.  Sure, we haven't written the
5581 * inode out, but prune_icache isn't a user-visible syncing function.
5582 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5583 * we start and wait on commits.
 
 
 
 
 
 
 
 
5584 */
5585int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5586{
5587	struct ext4_iloc iloc;
5588	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5589	static unsigned int mnt_count;
5590	int err, ret;
5591
5592	might_sleep();
5593	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5594	err = ext4_reserve_inode_write(handle, inode, &iloc);
5595	if (err)
5596		return err;
5597	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5598	    !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5599		/*
5600		 * In nojournal mode, we can immediately attempt to expand
5601		 * the inode.  When journaled, we first need to obtain extra
5602		 * buffer credits since we may write into the EA block
5603		 * with this same handle. If journal_extend fails, then it will
5604		 * only result in a minor loss of functionality for that inode.
5605		 * If this is felt to be critical, then e2fsck should be run to
5606		 * force a large enough s_min_extra_isize.
5607		 */
5608		if (!ext4_handle_valid(handle) ||
5609		    jbd2_journal_extend(handle,
5610			     EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) == 0) {
5611			ret = ext4_expand_extra_isize(inode,
5612						      sbi->s_want_extra_isize,
5613						      iloc, handle);
5614			if (ret) {
 
 
5615				if (mnt_count !=
5616					le16_to_cpu(sbi->s_es->s_mnt_count)) {
5617					ext4_warning(inode->i_sb,
5618					"Unable to expand inode %lu. Delete"
5619					" some EAs or run e2fsck.",
5620					inode->i_ino);
5621					mnt_count =
5622					  le16_to_cpu(sbi->s_es->s_mnt_count);
5623				}
5624			}
5625		}
5626	}
5627	return ext4_mark_iloc_dirty(handle, inode, &iloc);
 
 
5628}
5629
5630/*
5631 * ext4_dirty_inode() is called from __mark_inode_dirty()
5632 *
5633 * We're really interested in the case where a file is being extended.
5634 * i_size has been changed by generic_commit_write() and we thus need
5635 * to include the updated inode in the current transaction.
5636 *
5637 * Also, dquot_alloc_block() will always dirty the inode when blocks
5638 * are allocated to the file.
5639 *
5640 * If the inode is marked synchronous, we don't honour that here - doing
5641 * so would cause a commit on atime updates, which we don't bother doing.
5642 * We handle synchronous inodes at the highest possible level.
5643 *
5644 * If only the I_DIRTY_TIME flag is set, we can skip everything.  If
5645 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5646 * to copy into the on-disk inode structure are the timestamp files.
5647 */
5648void ext4_dirty_inode(struct inode *inode, int flags)
5649{
5650	handle_t *handle;
5651
5652	if (flags == I_DIRTY_TIME)
5653		return;
5654	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5655	if (IS_ERR(handle))
5656		goto out;
5657
5658	ext4_mark_inode_dirty(handle, inode);
5659
5660	ext4_journal_stop(handle);
5661out:
5662	return;
5663}
5664
5665#if 0
5666/*
5667 * Bind an inode's backing buffer_head into this transaction, to prevent
5668 * it from being flushed to disk early.  Unlike
5669 * ext4_reserve_inode_write, this leaves behind no bh reference and
5670 * returns no iloc structure, so the caller needs to repeat the iloc
5671 * lookup to mark the inode dirty later.
5672 */
5673static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5674{
5675	struct ext4_iloc iloc;
5676
5677	int err = 0;
5678	if (handle) {
5679		err = ext4_get_inode_loc(inode, &iloc);
5680		if (!err) {
5681			BUFFER_TRACE(iloc.bh, "get_write_access");
5682			err = jbd2_journal_get_write_access(handle, iloc.bh);
5683			if (!err)
5684				err = ext4_handle_dirty_metadata(handle,
5685								 NULL,
5686								 iloc.bh);
5687			brelse(iloc.bh);
5688		}
5689	}
5690	ext4_std_error(inode->i_sb, err);
5691	return err;
5692}
5693#endif
5694
5695int ext4_change_inode_journal_flag(struct inode *inode, int val)
5696{
5697	journal_t *journal;
5698	handle_t *handle;
5699	int err;
5700	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5701
5702	/*
5703	 * We have to be very careful here: changing a data block's
5704	 * journaling status dynamically is dangerous.  If we write a
5705	 * data block to the journal, change the status and then delete
5706	 * that block, we risk forgetting to revoke the old log record
5707	 * from the journal and so a subsequent replay can corrupt data.
5708	 * So, first we make sure that the journal is empty and that
5709	 * nobody is changing anything.
5710	 */
5711
5712	journal = EXT4_JOURNAL(inode);
5713	if (!journal)
5714		return 0;
5715	if (is_journal_aborted(journal))
5716		return -EROFS;
5717
5718	/* Wait for all existing dio workers */
5719	ext4_inode_block_unlocked_dio(inode);
5720	inode_dio_wait(inode);
5721
5722	/*
5723	 * Before flushing the journal and switching inode's aops, we have
5724	 * to flush all dirty data the inode has. There can be outstanding
5725	 * delayed allocations, there can be unwritten extents created by
5726	 * fallocate or buffered writes in dioread_nolock mode covered by
5727	 * dirty data which can be converted only after flushing the dirty
5728	 * data (and journalled aops don't know how to handle these cases).
5729	 */
5730	if (val) {
5731		down_write(&EXT4_I(inode)->i_mmap_sem);
5732		err = filemap_write_and_wait(inode->i_mapping);
5733		if (err < 0) {
5734			up_write(&EXT4_I(inode)->i_mmap_sem);
5735			ext4_inode_resume_unlocked_dio(inode);
5736			return err;
5737		}
5738	}
5739
5740	percpu_down_write(&sbi->s_journal_flag_rwsem);
5741	jbd2_journal_lock_updates(journal);
 
5742
5743	/*
5744	 * OK, there are no updates running now, and all cached data is
5745	 * synced to disk.  We are now in a completely consistent state
5746	 * which doesn't have anything in the journal, and we know that
5747	 * no filesystem updates are running, so it is safe to modify
5748	 * the inode's in-core data-journaling state flag now.
5749	 */
5750
5751	if (val)
5752		ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5753	else {
5754		err = jbd2_journal_flush(journal);
5755		if (err < 0) {
5756			jbd2_journal_unlock_updates(journal);
5757			percpu_up_write(&sbi->s_journal_flag_rwsem);
5758			ext4_inode_resume_unlocked_dio(inode);
5759			return err;
5760		}
5761		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5762	}
5763	ext4_set_aops(inode);
5764	/*
5765	 * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated.
5766	 * E.g. S_DAX may get cleared / set.
5767	 */
5768	ext4_set_inode_flags(inode);
5769
5770	jbd2_journal_unlock_updates(journal);
5771	percpu_up_write(&sbi->s_journal_flag_rwsem);
5772
5773	if (val)
5774		up_write(&EXT4_I(inode)->i_mmap_sem);
5775	ext4_inode_resume_unlocked_dio(inode);
5776
5777	/* Finally we can mark the inode as dirty. */
5778
5779	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5780	if (IS_ERR(handle))
5781		return PTR_ERR(handle);
5782
5783	err = ext4_mark_inode_dirty(handle, inode);
5784	ext4_handle_sync(handle);
5785	ext4_journal_stop(handle);
5786	ext4_std_error(inode->i_sb, err);
5787
5788	return err;
5789}
5790
5791static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5792{
5793	return !buffer_mapped(bh);
5794}
5795
5796int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5797{
5798	struct page *page = vmf->page;
5799	loff_t size;
5800	unsigned long len;
5801	int ret;
5802	struct file *file = vma->vm_file;
5803	struct inode *inode = file_inode(file);
5804	struct address_space *mapping = inode->i_mapping;
5805	handle_t *handle;
5806	get_block_t *get_block;
5807	int retries = 0;
5808
5809	sb_start_pagefault(inode->i_sb);
5810	file_update_time(vma->vm_file);
5811
5812	down_read(&EXT4_I(inode)->i_mmap_sem);
 
5813	/* Delalloc case is easy... */
5814	if (test_opt(inode->i_sb, DELALLOC) &&
5815	    !ext4_should_journal_data(inode) &&
5816	    !ext4_nonda_switch(inode->i_sb)) {
5817		do {
5818			ret = block_page_mkwrite(vma, vmf,
5819						   ext4_da_get_block_prep);
5820		} while (ret == -ENOSPC &&
5821		       ext4_should_retry_alloc(inode->i_sb, &retries));
5822		goto out_ret;
5823	}
5824
5825	lock_page(page);
5826	size = i_size_read(inode);
5827	/* Page got truncated from under us? */
5828	if (page->mapping != mapping || page_offset(page) > size) {
5829		unlock_page(page);
5830		ret = VM_FAULT_NOPAGE;
5831		goto out;
5832	}
5833
5834	if (page->index == size >> PAGE_SHIFT)
5835		len = size & ~PAGE_MASK;
5836	else
5837		len = PAGE_SIZE;
5838	/*
5839	 * Return if we have all the buffers mapped. This avoids the need to do
5840	 * journal_start/journal_stop which can block and take a long time
5841	 */
5842	if (page_has_buffers(page)) {
5843		if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5844					    0, len, NULL,
5845					    ext4_bh_unmapped)) {
5846			/* Wait so that we don't change page under IO */
5847			wait_for_stable_page(page);
5848			ret = VM_FAULT_LOCKED;
5849			goto out;
5850		}
5851	}
5852	unlock_page(page);
5853	/* OK, we need to fill the hole... */
5854	if (ext4_should_dioread_nolock(inode))
5855		get_block = ext4_get_block_unwritten;
5856	else
5857		get_block = ext4_get_block;
5858retry_alloc:
5859	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5860				    ext4_writepage_trans_blocks(inode));
5861	if (IS_ERR(handle)) {
5862		ret = VM_FAULT_SIGBUS;
5863		goto out;
5864	}
5865	ret = block_page_mkwrite(vma, vmf, get_block);
5866	if (!ret && ext4_should_journal_data(inode)) {
5867		if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5868			  PAGE_SIZE, NULL, do_journal_get_write_access)) {
5869			unlock_page(page);
5870			ret = VM_FAULT_SIGBUS;
5871			ext4_journal_stop(handle);
5872			goto out;
5873		}
5874		ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5875	}
5876	ext4_journal_stop(handle);
5877	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5878		goto retry_alloc;
5879out_ret:
5880	ret = block_page_mkwrite_return(ret);
5881out:
5882	up_read(&EXT4_I(inode)->i_mmap_sem);
5883	sb_end_pagefault(inode->i_sb);
5884	return ret;
5885}
5886
5887int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
5888{
5889	struct inode *inode = file_inode(vma->vm_file);
5890	int err;
5891
5892	down_read(&EXT4_I(inode)->i_mmap_sem);
5893	err = filemap_fault(vma, vmf);
5894	up_read(&EXT4_I(inode)->i_mmap_sem);
5895
5896	return err;
5897}
5898
5899/*
5900 * Find the first extent at or after @lblk in an inode that is not a hole.
5901 * Search for @map_len blocks at most. The extent is returned in @result.
5902 *
5903 * The function returns 1 if we found an extent. The function returns 0 in
5904 * case there is no extent at or after @lblk and in that case also sets
5905 * @result->es_len to 0. In case of error, the error code is returned.
5906 */
5907int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
5908			 unsigned int map_len, struct extent_status *result)
5909{
5910	struct ext4_map_blocks map;
5911	struct extent_status es = {};
5912	int ret;
5913
5914	map.m_lblk = lblk;
5915	map.m_len = map_len;
5916
5917	/*
5918	 * For non-extent based files this loop may iterate several times since
5919	 * we do not determine full hole size.
5920	 */
5921	while (map.m_len > 0) {
5922		ret = ext4_map_blocks(NULL, inode, &map, 0);
5923		if (ret < 0)
5924			return ret;
5925		/* There's extent covering m_lblk? Just return it. */
5926		if (ret > 0) {
5927			int status;
5928
5929			ext4_es_store_pblock(result, map.m_pblk);
5930			result->es_lblk = map.m_lblk;
5931			result->es_len = map.m_len;
5932			if (map.m_flags & EXT4_MAP_UNWRITTEN)
5933				status = EXTENT_STATUS_UNWRITTEN;
5934			else
5935				status = EXTENT_STATUS_WRITTEN;
5936			ext4_es_store_status(result, status);
5937			return 1;
5938		}
5939		ext4_es_find_delayed_extent_range(inode, map.m_lblk,
5940						  map.m_lblk + map.m_len - 1,
5941						  &es);
5942		/* Is delalloc data before next block in extent tree? */
5943		if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
5944			ext4_lblk_t offset = 0;
5945
5946			if (es.es_lblk < lblk)
5947				offset = lblk - es.es_lblk;
5948			result->es_lblk = es.es_lblk + offset;
5949			ext4_es_store_pblock(result,
5950					     ext4_es_pblock(&es) + offset);
5951			result->es_len = es.es_len - offset;
5952			ext4_es_store_status(result, ext4_es_status(&es));
5953
5954			return 1;
5955		}
5956		/* There's a hole at m_lblk, advance us after it */
5957		map.m_lblk += map.m_len;
5958		map_len -= map.m_len;
5959		map.m_len = map_len;
5960		cond_resched();
5961	}
5962	result->es_len = 0;
5963	return 0;
5964}