Linux Audio

Check our new training course

Loading...
v3.1
 
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_types.h"
 
 
 
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
  27#include "xfs_mount.h"
  28#include "xfs_buf_item.h"
  29#include "xfs_trans_priv.h"
  30#include "xfs_error.h"
 
 
 
 
 
  31#include "xfs_trace.h"
 
 
 
  32
  33
  34kmem_zone_t	*xfs_buf_item_zone;
  35
  36static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
  37{
  38	return container_of(lip, struct xfs_buf_log_item, bli_item);
  39}
  40
 
 
 
 
 
 
 
 
 
 
 
  41
  42#ifdef XFS_TRANS_DEBUG
  43/*
  44 * This function uses an alternate strategy for tracking the bytes
  45 * that the user requests to be logged.  This can then be used
  46 * in conjunction with the bli_orig array in the buf log item to
  47 * catch bugs in our callers' code.
  48 *
  49 * We also double check the bits set in xfs_buf_item_log using a
  50 * simple algorithm to check that every byte is accounted for.
  51 */
  52STATIC void
  53xfs_buf_item_log_debug(
  54	xfs_buf_log_item_t	*bip,
  55	uint			first,
  56	uint			last)
  57{
  58	uint	x;
  59	uint	byte;
  60	uint	nbytes;
  61	uint	chunk_num;
  62	uint	word_num;
  63	uint	bit_num;
  64	uint	bit_set;
  65	uint	*wordp;
  66
  67	ASSERT(bip->bli_logged != NULL);
  68	byte = first;
  69	nbytes = last - first + 1;
  70	bfset(bip->bli_logged, first, nbytes);
  71	for (x = 0; x < nbytes; x++) {
  72		chunk_num = byte >> XFS_BLF_SHIFT;
  73		word_num = chunk_num >> BIT_TO_WORD_SHIFT;
  74		bit_num = chunk_num & (NBWORD - 1);
  75		wordp = &(bip->bli_format.blf_data_map[word_num]);
  76		bit_set = *wordp & (1 << bit_num);
  77		ASSERT(bit_set);
  78		byte++;
  79	}
  80}
  81
  82/*
  83 * This function is called when we flush something into a buffer without
  84 * logging it.  This happens for things like inodes which are logged
  85 * separately from the buffer.
  86 */
  87void
  88xfs_buf_item_flush_log_debug(
  89	xfs_buf_t	*bp,
  90	uint		first,
  91	uint		last)
  92{
  93	xfs_buf_log_item_t	*bip = bp->b_fspriv;
  94	uint			nbytes;
  95
  96	if (bip == NULL || (bip->bli_item.li_type != XFS_LI_BUF))
  97		return;
 
  98
  99	ASSERT(bip->bli_logged != NULL);
 100	nbytes = last - first + 1;
 101	bfset(bip->bli_logged, first, nbytes);
 102}
 103
 104/*
 105 * This function is called to verify that our callers have logged
 106 * all the bytes that they changed.
 107 *
 108 * It does this by comparing the original copy of the buffer stored in
 109 * the buf log item's bli_orig array to the current copy of the buffer
 110 * and ensuring that all bytes which mismatch are set in the bli_logged
 111 * array of the buf log item.
 112 */
 113STATIC void
 114xfs_buf_item_log_check(
 115	xfs_buf_log_item_t	*bip)
 116{
 117	char		*orig;
 118	char		*buffer;
 119	int		x;
 120	xfs_buf_t	*bp;
 121
 122	ASSERT(bip->bli_orig != NULL);
 123	ASSERT(bip->bli_logged != NULL);
 124
 125	bp = bip->bli_buf;
 126	ASSERT(XFS_BUF_COUNT(bp) > 0);
 127	ASSERT(bp->b_addr != NULL);
 128	orig = bip->bli_orig;
 129	buffer = bp->b_addr;
 130	for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
 131		if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) {
 132			xfs_emerg(bp->b_mount,
 133				"%s: bip %x buffer %x orig %x index %d",
 134				__func__, bip, bp, orig, x);
 135			ASSERT(0);
 136		}
 137	}
 138}
 139#else
 140#define		xfs_buf_item_log_debug(x,y,z)
 141#define		xfs_buf_item_log_check(x)
 142#endif
 143
 144STATIC void	xfs_buf_do_callbacks(struct xfs_buf *bp);
 
 
 145
 146/*
 147 * This returns the number of log iovecs needed to log the
 148 * given buf log item.
 149 *
 150 * It calculates this as 1 iovec for the buf log format structure
 151 * and 1 for each stretch of non-contiguous chunks to be logged.
 152 * Contiguous chunks are logged in a single iovec.
 153 *
 154 * If the XFS_BLI_STALE flag has been set, then log nothing.
 155 */
 156STATIC uint
 157xfs_buf_item_size(
 158	struct xfs_log_item	*lip)
 159{
 160	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 161	struct xfs_buf		*bp = bip->bli_buf;
 162	uint			nvecs;
 163	int			next_bit;
 164	int			last_bit;
 165
 166	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 167	if (bip->bli_flags & XFS_BLI_STALE) {
 168		/*
 169		 * The buffer is stale, so all we need to log
 170		 * is the buf log format structure with the
 171		 * cancel flag in it.
 172		 */
 173		trace_xfs_buf_item_size_stale(bip);
 174		ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
 175		return 1;
 176	}
 177
 178	ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
 179	nvecs = 1;
 180	last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
 181					 bip->bli_format.blf_map_size, 0);
 182	ASSERT(last_bit != -1);
 183	nvecs++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 184	while (last_bit != -1) {
 185		/*
 186		 * This takes the bit number to start looking from and
 187		 * returns the next set bit from there.  It returns -1
 188		 * if there are no more bits set or the start bit is
 189		 * beyond the end of the bitmap.
 190		 */
 191		next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
 192						 bip->bli_format.blf_map_size,
 193						 last_bit + 1);
 194		/*
 195		 * If we run out of bits, leave the loop,
 196		 * else if we find a new set of bits bump the number of vecs,
 197		 * else keep scanning the current set of bits.
 198		 */
 199		if (next_bit == -1) {
 200			last_bit = -1;
 201		} else if (next_bit != last_bit + 1) {
 202			last_bit = next_bit;
 203			nvecs++;
 204		} else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
 205			   (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
 206			    XFS_BLF_CHUNK)) {
 207			last_bit = next_bit;
 208			nvecs++;
 
 
 209		} else {
 210			last_bit++;
 
 211		}
 
 212	}
 213
 214	trace_xfs_buf_item_size(bip);
 215	return nvecs;
 216}
 217
 218/*
 219 * This is called to fill in the vector of log iovecs for the
 220 * given log buf item.  It fills the first entry with a buf log
 221 * format structure, and the rest point to contiguous chunks
 222 * within the buffer.
 
 
 
 
 
 
 
 
 
 
 
 
 223 */
 224STATIC void
 225xfs_buf_item_format(
 226	struct xfs_log_item	*lip,
 227	struct xfs_log_iovec	*vecp)
 
 228{
 229	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 230	struct xfs_buf	*bp = bip->bli_buf;
 231	uint		base_size;
 232	uint		nvecs;
 233	int		first_bit;
 234	int		last_bit;
 235	int		next_bit;
 236	uint		nbits;
 237	uint		buffer_offset;
 238
 239	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 240	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
 241	       (bip->bli_flags & XFS_BLI_STALE));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 242
 243	/*
 244	 * The size of the base structure is the size of the
 245	 * declared structure plus the space for the extra words
 246	 * of the bitmap.  We subtract one from the map size, because
 247	 * the first element of the bitmap is accounted for in the
 248	 * size of the base structure.
 249	 */
 250	base_size =
 251		(uint)(sizeof(xfs_buf_log_format_t) +
 252		       ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
 253	vecp->i_addr = &bip->bli_format;
 254	vecp->i_len = base_size;
 255	vecp->i_type = XLOG_REG_TYPE_BFORMAT;
 256	vecp++;
 257	nvecs = 1;
 258
 259	/*
 260	 * If it is an inode buffer, transfer the in-memory state to the
 261	 * format flags and clear the in-memory state. We do not transfer
 262	 * this state if the inode buffer allocation has not yet been committed
 263	 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
 264	 * correct replay of the inode allocation.
 265	 */
 266	if (bip->bli_flags & XFS_BLI_INODE_BUF) {
 267		if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
 268		      xfs_log_item_in_current_chkpt(lip)))
 269			bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
 270		bip->bli_flags &= ~XFS_BLI_INODE_BUF;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 271	}
 272
 
 
 
 273	if (bip->bli_flags & XFS_BLI_STALE) {
 274		/*
 275		 * The buffer is stale, so all we need to log
 276		 * is the buf log format structure with the
 277		 * cancel flag in it.
 278		 */
 279		trace_xfs_buf_item_format_stale(bip);
 280		ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
 281		bip->bli_format.blf_size = nvecs;
 282		return;
 283	}
 284
 
 285	/*
 286	 * Fill in an iovec for each set of contiguous chunks.
 287	 */
 288	first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
 289					 bip->bli_format.blf_map_size, 0);
 290	ASSERT(first_bit != -1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 291	last_bit = first_bit;
 292	nbits = 1;
 293	for (;;) {
 294		/*
 295		 * This takes the bit number to start looking from and
 296		 * returns the next set bit from there.  It returns -1
 297		 * if there are no more bits set or the start bit is
 298		 * beyond the end of the bitmap.
 299		 */
 300		next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
 301						 bip->bli_format.blf_map_size,
 302						 (uint)last_bit + 1);
 303		/*
 304		 * If we run out of bits fill in the last iovec and get
 305		 * out of the loop.
 306		 * Else if we start a new set of bits then fill in the
 307		 * iovec for the series we were looking at and start
 308		 * counting the bits in the new one.
 309		 * Else we're still in the same set of bits so just
 310		 * keep counting and scanning.
 311		 */
 312		if (next_bit == -1) {
 313			buffer_offset = first_bit * XFS_BLF_CHUNK;
 314			vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
 315			vecp->i_len = nbits * XFS_BLF_CHUNK;
 316			vecp->i_type = XLOG_REG_TYPE_BCHUNK;
 317			nvecs++;
 318			break;
 319		} else if (next_bit != last_bit + 1) {
 320			buffer_offset = first_bit * XFS_BLF_CHUNK;
 321			vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
 322			vecp->i_len = nbits * XFS_BLF_CHUNK;
 323			vecp->i_type = XLOG_REG_TYPE_BCHUNK;
 324			nvecs++;
 325			vecp++;
 326			first_bit = next_bit;
 327			last_bit = next_bit;
 328			nbits = 1;
 329		} else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
 330			   (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
 331			    XFS_BLF_CHUNK)) {
 332			buffer_offset = first_bit * XFS_BLF_CHUNK;
 333			vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
 334			vecp->i_len = nbits * XFS_BLF_CHUNK;
 335			vecp->i_type = XLOG_REG_TYPE_BCHUNK;
 336/* You would think we need to bump the nvecs here too, but we do not
 337 * this number is used by recovery, and it gets confused by the boundary
 338 * split here
 339 *			nvecs++;
 340 */
 341			vecp++;
 342			first_bit = next_bit;
 343			last_bit = next_bit;
 344			nbits = 1;
 345		} else {
 346			last_bit++;
 347			nbits++;
 348		}
 349	}
 350	bip->bli_format.blf_size = nvecs;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 351
 352	/*
 353	 * Check to make sure everything is consistent.
 354	 */
 355	trace_xfs_buf_item_format(bip);
 356	xfs_buf_item_log_check(bip);
 357}
 358
 359/*
 360 * This is called to pin the buffer associated with the buf log item in memory
 361 * so it cannot be written out.
 362 *
 363 * We also always take a reference to the buffer log item here so that the bli
 364 * is held while the item is pinned in memory. This means that we can
 365 * unconditionally drop the reference count a transaction holds when the
 366 * transaction is completed.
 
 
 
 
 
 
 
 
 367 */
 368STATIC void
 369xfs_buf_item_pin(
 370	struct xfs_log_item	*lip)
 371{
 372	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 373
 374	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 375	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
 
 376	       (bip->bli_flags & XFS_BLI_STALE));
 377
 378	trace_xfs_buf_item_pin(bip);
 379
 
 380	atomic_inc(&bip->bli_refcount);
 381	atomic_inc(&bip->bli_buf->b_pin_count);
 382}
 383
 384/*
 385 * This is called to unpin the buffer associated with the buf log
 386 * item which was previously pinned with a call to xfs_buf_item_pin().
 
 387 *
 388 * Also drop the reference to the buf item for the current transaction.
 389 * If the XFS_BLI_STALE flag is set and we are the last reference,
 390 * then free up the buf log item and unlock the buffer.
 
 
 
 
 391 *
 392 * If the remove flag is set we are called from uncommit in the
 393 * forced-shutdown path.  If that is true and the reference count on
 394 * the log item is going to drop to zero we need to free the item's
 395 * descriptor in the transaction.
 
 
 396 */
 397STATIC void
 398xfs_buf_item_unpin(
 399	struct xfs_log_item	*lip,
 400	int			remove)
 401{
 402	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 403	xfs_buf_t	*bp = bip->bli_buf;
 404	struct xfs_ail	*ailp = lip->li_ailp;
 405	int		stale = bip->bli_flags & XFS_BLI_STALE;
 406	int		freed;
 407
 408	ASSERT(bp->b_fspriv == bip);
 409	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 410
 411	trace_xfs_buf_item_unpin(bip);
 412
 413	freed = atomic_dec_and_test(&bip->bli_refcount);
 414
 415	if (atomic_dec_and_test(&bp->b_pin_count))
 416		wake_up_all(&bp->b_waiters);
 417
 418	if (freed && stale) {
 
 
 
 
 
 
 
 
 
 419		ASSERT(bip->bli_flags & XFS_BLI_STALE);
 420		ASSERT(xfs_buf_islocked(bp));
 421		ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
 422		ASSERT(XFS_BUF_ISSTALE(bp));
 423		ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
 
 424
 425		trace_xfs_buf_item_unpin_stale(bip);
 426
 427		if (remove) {
 428			/*
 429			 * If we are in a transaction context, we have to
 430			 * remove the log item from the transaction as we are
 431			 * about to release our reference to the buffer.  If we
 432			 * don't, the unlock that occurs later in
 433			 * xfs_trans_uncommit() will try to reference the
 434			 * buffer which we no longer have a hold on.
 435			 */
 436			if (lip->li_desc)
 437				xfs_trans_del_item(lip);
 438
 439			/*
 440			 * Since the transaction no longer refers to the buffer,
 441			 * the buffer should no longer refer to the transaction.
 442			 */
 443			bp->b_transp = NULL;
 444		}
 445
 446		/*
 447		 * If we get called here because of an IO error, we may
 448		 * or may not have the item on the AIL. xfs_trans_ail_delete()
 449		 * will take care of that situation.
 450		 * xfs_trans_ail_delete() drops the AIL lock.
 451		 */
 452		if (bip->bli_flags & XFS_BLI_STALE_INODE) {
 453			xfs_buf_do_callbacks(bp);
 454			bp->b_fspriv = NULL;
 455			bp->b_iodone = NULL;
 456		} else {
 457			spin_lock(&ailp->xa_lock);
 458			xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
 459			xfs_buf_item_relse(bp);
 460			ASSERT(bp->b_fspriv == NULL);
 461		}
 462		xfs_buf_relse(bp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 463	}
 
 
 
 
 
 
 
 464}
 465
 466/*
 467 * This is called to attempt to lock the buffer associated with this
 468 * buf log item.  Don't sleep on the buffer lock.  If we can't get
 469 * the lock right away, return 0.  If we can get the lock, take a
 470 * reference to the buffer. If this is a delayed write buffer that
 471 * needs AIL help to be written back, invoke the pushbuf routine
 472 * rather than the normal success path.
 473 */
 474STATIC uint
 475xfs_buf_item_trylock(
 476	struct xfs_log_item	*lip)
 
 477{
 478	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 479	struct xfs_buf		*bp = bip->bli_buf;
 
 480
 481	if (xfs_buf_ispinned(bp))
 482		return XFS_ITEM_PINNED;
 483	if (!xfs_buf_trylock(bp))
 
 
 
 
 
 
 
 
 
 484		return XFS_ITEM_LOCKED;
 485
 486	/* take a reference to the buffer.  */
 487	xfs_buf_hold(bp);
 488
 489	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 490	trace_xfs_buf_item_trylock(bip);
 491	if (XFS_BUF_ISDELAYWRITE(bp))
 492		return XFS_ITEM_PUSHBUF;
 493	return XFS_ITEM_SUCCESS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 494}
 495
 496/*
 497 * Release the buffer associated with the buf log item.  If there is no dirty
 498 * logged data associated with the buffer recorded in the buf log item, then
 499 * free the buf log item and remove the reference to it in the buffer.
 500 *
 501 * This call ignores the recursion count.  It is only called when the buffer
 502 * should REALLY be unlocked, regardless of the recursion count.
 503 *
 504 * We unconditionally drop the transaction's reference to the log item. If the
 505 * item was logged, then another reference was taken when it was pinned, so we
 506 * can safely drop the transaction reference now.  This also allows us to avoid
 507 * potential races with the unpin code freeing the bli by not referencing the
 508 * bli after we've dropped the reference count.
 509 *
 510 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
 511 * if necessary but do not unlock the buffer.  This is for support of
 512 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
 513 * free the item.
 514 */
 515STATIC void
 516xfs_buf_item_unlock(
 517	struct xfs_log_item	*lip)
 518{
 519	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 520	struct xfs_buf		*bp = bip->bli_buf;
 521	int			aborted;
 522	uint			hold;
 
 
 
 
 
 
 
 523
 524	/* Clear the buffer's association with this transaction. */
 525	bp->b_transp = NULL;
 526
 527	/*
 528	 * If this is a transaction abort, don't return early.  Instead, allow
 529	 * the brelse to happen.  Normally it would be done for stale
 530	 * (cancelled) buffers at unpin time, but we'll never go through the
 531	 * pin/unpin cycle if we abort inside commit.
 532	 */
 533	aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
 
 
 534
 535	/*
 536	 * Before possibly freeing the buf item, determine if we should
 537	 * release the buffer at the end of this routine.
 538	 */
 539	hold = bip->bli_flags & XFS_BLI_HOLD;
 540
 541	/* Clear the per transaction state. */
 542	bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
 543
 544	/*
 545	 * If the buf item is marked stale, then don't do anything.  We'll
 546	 * unlock the buffer and free the buf item when the buffer is unpinned
 547	 * for the last time.
 548	 */
 549	if (bip->bli_flags & XFS_BLI_STALE) {
 550		trace_xfs_buf_item_unlock_stale(bip);
 551		ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
 552		if (!aborted) {
 553			atomic_dec(&bip->bli_refcount);
 554			return;
 555		}
 556	}
 557
 558	trace_xfs_buf_item_unlock(bip);
 559
 560	/*
 561	 * If the buf item isn't tracking any data, free it, otherwise drop the
 562	 * reference we hold to it.
 
 
 563	 */
 564	if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
 565			     bip->bli_format.blf_map_size))
 566		xfs_buf_item_relse(bp);
 567	else
 568		atomic_dec(&bip->bli_refcount);
 
 569
 570	if (!hold)
 571		xfs_buf_relse(bp);
 
 
 
 
 572}
 573
 574/*
 575 * This is called to find out where the oldest active copy of the
 576 * buf log item in the on disk log resides now that the last log
 577 * write of it completed at the given lsn.
 578 * We always re-log all the dirty data in a buffer, so usually the
 579 * latest copy in the on disk log is the only one that matters.  For
 580 * those cases we simply return the given lsn.
 581 *
 582 * The one exception to this is for buffers full of newly allocated
 583 * inodes.  These buffers are only relogged with the XFS_BLI_INODE_BUF
 584 * flag set, indicating that only the di_next_unlinked fields from the
 585 * inodes in the buffers will be replayed during recovery.  If the
 586 * original newly allocated inode images have not yet been flushed
 587 * when the buffer is so relogged, then we need to make sure that we
 588 * keep the old images in the 'active' portion of the log.  We do this
 589 * by returning the original lsn of that transaction here rather than
 590 * the current one.
 591 */
 592STATIC xfs_lsn_t
 593xfs_buf_item_committed(
 594	struct xfs_log_item	*lip,
 595	xfs_lsn_t		lsn)
 596{
 597	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 598
 599	trace_xfs_buf_item_committed(bip);
 600
 601	if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
 602		return lip->li_lsn;
 603	return lsn;
 604}
 605
 606/*
 607 * The buffer is locked, but is not a delayed write buffer. This happens
 608 * if we race with IO completion and hence we don't want to try to write it
 609 * again. Just release the buffer.
 610 */
 611STATIC void
 612xfs_buf_item_push(
 613	struct xfs_log_item	*lip)
 614{
 615	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 616	struct xfs_buf		*bp = bip->bli_buf;
 617
 618	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 619	ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
 620
 621	trace_xfs_buf_item_push(bip);
 622
 623	xfs_buf_relse(bp);
 624}
 625
 626/*
 627 * The buffer is locked and is a delayed write buffer. Promote the buffer
 628 * in the delayed write queue as the caller knows that they must invoke
 629 * the xfsbufd to get this buffer written. We have to unlock the buffer
 630 * to allow the xfsbufd to write it, too.
 631 */
 632STATIC bool
 633xfs_buf_item_pushbuf(
 634	struct xfs_log_item	*lip)
 635{
 636	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 637	struct xfs_buf		*bp = bip->bli_buf;
 
 
 638
 639	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 640	ASSERT(XFS_BUF_ISDELAYWRITE(bp));
 641
 642	trace_xfs_buf_item_pushbuf(bip);
 643
 644	xfs_buf_delwri_promote(bp);
 645	xfs_buf_relse(bp);
 646	return true;
 647}
 
 
 
 648
 649STATIC void
 650xfs_buf_item_committing(
 651	struct xfs_log_item	*lip,
 652	xfs_lsn_t		commit_lsn)
 653{
 654}
 
 
 
 655
 656/*
 657 * This is the ops vector shared by all buf log items.
 658 */
 659static struct xfs_item_ops xfs_buf_item_ops = {
 660	.iop_size	= xfs_buf_item_size,
 
 661	.iop_format	= xfs_buf_item_format,
 662	.iop_pin	= xfs_buf_item_pin,
 663	.iop_unpin	= xfs_buf_item_unpin,
 664	.iop_trylock	= xfs_buf_item_trylock,
 665	.iop_unlock	= xfs_buf_item_unlock,
 666	.iop_committed	= xfs_buf_item_committed,
 667	.iop_push	= xfs_buf_item_push,
 668	.iop_pushbuf	= xfs_buf_item_pushbuf,
 669	.iop_committing = xfs_buf_item_committing
 670};
 671
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 672
 673/*
 674 * Allocate a new buf log item to go with the given buffer.
 675 * Set the buffer's b_fsprivate field to point to the new
 676 * buf log item.  If there are other item's attached to the
 677 * buffer (see xfs_buf_attach_iodone() below), then put the
 678 * buf log item at the front.
 679 */
 680void
 681xfs_buf_item_init(
 682	xfs_buf_t	*bp,
 683	xfs_mount_t	*mp)
 684{
 685	xfs_log_item_t		*lip = bp->b_fspriv;
 686	xfs_buf_log_item_t	*bip;
 687	int			chunks;
 688	int			map_size;
 
 689
 690	/*
 691	 * Check to see if there is already a buf log item for
 692	 * this buffer.  If there is, it is guaranteed to be
 693	 * the first.  If we do already have one, there is
 694	 * nothing to do here so return.
 695	 */
 696	ASSERT(bp->b_target->bt_mount == mp);
 697	if (lip != NULL && lip->li_type == XFS_LI_BUF)
 698		return;
 699
 700	/*
 701	 * chunks is the number of XFS_BLF_CHUNK size pieces
 702	 * the buffer can be divided into. Make sure not to
 703	 * truncate any pieces.  map_size is the size of the
 704	 * bitmap needed to describe the chunks of the buffer.
 705	 */
 706	chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
 707	map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
 708
 709	bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
 710						    KM_SLEEP);
 711	xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
 712	bip->bli_buf = bp;
 713	xfs_buf_hold(bp);
 714	bip->bli_format.blf_type = XFS_LI_BUF;
 715	bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
 716	bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
 717	bip->bli_format.blf_map_size = map_size;
 718
 719#ifdef XFS_TRANS_DEBUG
 720	/*
 721	 * Allocate the arrays for tracking what needs to be logged
 722	 * and what our callers request to be logged.  bli_orig
 723	 * holds a copy of the original, clean buffer for comparison
 724	 * against, and bli_logged keeps a 1 bit flag per byte in
 725	 * the buffer to indicate which bytes the callers have asked
 726	 * to have logged.
 727	 */
 728	bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
 729	memcpy(bip->bli_orig, bp->b_addr, XFS_BUF_COUNT(bp));
 730	bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
 731#endif
 732
 733	/*
 734	 * Put the buf item into the list of items attached to the
 735	 * buffer at the front.
 
 
 
 
 
 736	 */
 737	if (bp->b_fspriv)
 738		bip->bli_item.li_bio_list = bp->b_fspriv;
 739	bp->b_fspriv = bip;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 740}
 741
 742
 743/*
 744 * Mark bytes first through last inclusive as dirty in the buf
 745 * item's bitmap.
 746 */
 747void
 748xfs_buf_item_log(
 749	xfs_buf_log_item_t	*bip,
 750	uint			first,
 751	uint			last)
 
 752{
 753	uint		first_bit;
 754	uint		last_bit;
 755	uint		bits_to_set;
 756	uint		bits_set;
 757	uint		word_num;
 758	uint		*wordp;
 759	uint		bit;
 760	uint		end_bit;
 761	uint		mask;
 762
 763	/*
 764	 * Mark the item as having some dirty data for
 765	 * quick reference in xfs_buf_item_dirty.
 766	 */
 767	bip->bli_flags |= XFS_BLI_DIRTY;
 768
 769	/*
 770	 * Convert byte offsets to bit numbers.
 771	 */
 772	first_bit = first >> XFS_BLF_SHIFT;
 773	last_bit = last >> XFS_BLF_SHIFT;
 774
 775	/*
 776	 * Calculate the total number of bits to be set.
 777	 */
 778	bits_to_set = last_bit - first_bit + 1;
 779
 780	/*
 781	 * Get a pointer to the first word in the bitmap
 782	 * to set a bit in.
 783	 */
 784	word_num = first_bit >> BIT_TO_WORD_SHIFT;
 785	wordp = &(bip->bli_format.blf_data_map[word_num]);
 786
 787	/*
 788	 * Calculate the starting bit in the first word.
 789	 */
 790	bit = first_bit & (uint)(NBWORD - 1);
 791
 792	/*
 793	 * First set any bits in the first word of our range.
 794	 * If it starts at bit 0 of the word, it will be
 795	 * set below rather than here.  That is what the variable
 796	 * bit tells us. The variable bits_set tracks the number
 797	 * of bits that have been set so far.  End_bit is the number
 798	 * of the last bit to be set in this word plus one.
 799	 */
 800	if (bit) {
 801		end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
 802		mask = ((1 << (end_bit - bit)) - 1) << bit;
 803		*wordp |= mask;
 804		wordp++;
 805		bits_set = end_bit - bit;
 806	} else {
 807		bits_set = 0;
 808	}
 809
 810	/*
 811	 * Now set bits a whole word at a time that are between
 812	 * first_bit and last_bit.
 813	 */
 814	while ((bits_to_set - bits_set) >= NBWORD) {
 815		*wordp |= 0xffffffff;
 816		bits_set += NBWORD;
 817		wordp++;
 818	}
 819
 820	/*
 821	 * Finally, set any bits left to be set in one last partial word.
 822	 */
 823	end_bit = bits_to_set - bits_set;
 824	if (end_bit) {
 825		mask = (1 << end_bit) - 1;
 826		*wordp |= mask;
 827	}
 828
 829	xfs_buf_item_log_debug(bip, first, last);
 830}
 831
 832
 833/*
 834 * Return 1 if the buffer has some data that has been logged (at any
 835 * point, not just the current transaction) and 0 if not.
 836 */
 837uint
 838xfs_buf_item_dirty(
 839	xfs_buf_log_item_t	*bip)
 840{
 841	return (bip->bli_flags & XFS_BLI_DIRTY);
 842}
 843
 844STATIC void
 845xfs_buf_item_free(
 846	xfs_buf_log_item_t	*bip)
 847{
 848#ifdef XFS_TRANS_DEBUG
 849	kmem_free(bip->bli_orig);
 850	kmem_free(bip->bli_logged);
 851#endif /* XFS_TRANS_DEBUG */
 852
 853	kmem_zone_free(xfs_buf_item_zone, bip);
 854}
 855
 856/*
 857 * This is called when the buf log item is no longer needed.  It should
 858 * free the buf log item associated with the given buffer and clear
 859 * the buffer's pointer to the buf log item.  If there are no more
 860 * items in the list, clear the b_iodone field of the buffer (see
 861 * xfs_buf_attach_iodone() below).
 862 */
 863void
 864xfs_buf_item_relse(
 865	xfs_buf_t	*bp)
 
 
 866{
 867	xfs_buf_log_item_t	*bip;
 
 
 
 868
 869	trace_xfs_buf_item_relse(bp, _RET_IP_);
 
 
 
 
 
 
 
 870
 871	bip = bp->b_fspriv;
 872	bp->b_fspriv = bip->bli_item.li_bio_list;
 873	if (bp->b_fspriv == NULL)
 874		bp->b_iodone = NULL;
 
 875
 876	xfs_buf_rele(bp);
 877	xfs_buf_item_free(bip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 878}
 879
 880
 881/*
 882 * Add the given log item with its callback to the list of callbacks
 883 * to be called when the buffer's I/O completes.  If it is not set
 884 * already, set the buffer's b_iodone() routine to be
 885 * xfs_buf_iodone_callbacks() and link the log item into the list of
 886 * items rooted at b_fsprivate.  Items are always added as the second
 887 * entry in the list if there is a first, because the buf item code
 888 * assumes that the buf log item is first.
 889 */
 890void
 891xfs_buf_attach_iodone(
 892	xfs_buf_t	*bp,
 893	void		(*cb)(xfs_buf_t *, xfs_log_item_t *),
 894	xfs_log_item_t	*lip)
 895{
 896	xfs_log_item_t	*head_lip;
 897
 898	ASSERT(xfs_buf_islocked(bp));
 899
 900	lip->li_cb = cb;
 901	head_lip = bp->b_fspriv;
 902	if (head_lip) {
 903		lip->li_bio_list = head_lip->li_bio_list;
 904		head_lip->li_bio_list = lip;
 905	} else {
 906		bp->b_fspriv = lip;
 907	}
 908
 909	ASSERT(bp->b_iodone == NULL ||
 910	       bp->b_iodone == xfs_buf_iodone_callbacks);
 911	bp->b_iodone = xfs_buf_iodone_callbacks;
 912}
 913
 914/*
 915 * We can have many callbacks on a buffer. Running the callbacks individually
 916 * can cause a lot of contention on the AIL lock, so we allow for a single
 917 * callback to be able to scan the remaining lip->li_bio_list for other items
 918 * of the same type and callback to be processed in the first call.
 919 *
 920 * As a result, the loop walking the callback list below will also modify the
 921 * list. it removes the first item from the list and then runs the callback.
 922 * The loop then restarts from the new head of the list. This allows the
 923 * callback to scan and modify the list attached to the buffer and we don't
 924 * have to care about maintaining a next item pointer.
 925 */
 926STATIC void
 927xfs_buf_do_callbacks(
 928	struct xfs_buf		*bp)
 929{
 930	struct xfs_log_item	*lip;
 931
 932	while ((lip = bp->b_fspriv) != NULL) {
 933		bp->b_fspriv = lip->li_bio_list;
 934		ASSERT(lip->li_cb != NULL);
 935		/*
 936		 * Clear the next pointer so we don't have any
 937		 * confusion if the item is added to another buf.
 938		 * Don't touch the log item after calling its
 939		 * callback, because it could have freed itself.
 940		 */
 941		lip->li_bio_list = NULL;
 942		lip->li_cb(bp, lip);
 943	}
 944}
 945
 946/*
 947 * This is the iodone() function for buffers which have had callbacks
 948 * attached to them by xfs_buf_attach_iodone().  It should remove each
 949 * log item from the buffer's list and call the callback of each in turn.
 950 * When done, the buffer's fsprivate field is set to NULL and the buffer
 951 * is unlocked with a call to iodone().
 952 */
 953void
 954xfs_buf_iodone_callbacks(
 955	struct xfs_buf		*bp)
 956{
 957	struct xfs_log_item	*lip = bp->b_fspriv;
 958	struct xfs_mount	*mp = lip->li_mountp;
 959	static ulong		lasttime;
 960	static xfs_buftarg_t	*lasttarg;
 961
 962	if (likely(!xfs_buf_geterror(bp)))
 963		goto do_callbacks;
 964
 965	/*
 966	 * If we've already decided to shutdown the filesystem because of
 967	 * I/O errors, there's no point in giving this a retry.
 968	 */
 969	if (XFS_FORCED_SHUTDOWN(mp)) {
 970		XFS_BUF_SUPER_STALE(bp);
 971		trace_xfs_buf_item_iodone(bp, _RET_IP_);
 972		goto do_callbacks;
 973	}
 974
 975	if (bp->b_target != lasttarg ||
 976	    time_after(jiffies, (lasttime + 5*HZ))) {
 977		lasttime = jiffies;
 978		xfs_alert(mp, "Device %s: metadata write error block 0x%llx",
 979			xfs_buf_target_name(bp->b_target),
 980		      (__uint64_t)XFS_BUF_ADDR(bp));
 981	}
 982	lasttarg = bp->b_target;
 983
 984	/*
 985	 * If the write was asynchronous then no one will be looking for the
 986	 * error.  Clear the error state and write the buffer out again.
 987	 *
 988	 * During sync or umount we'll write all pending buffers again
 989	 * synchronous, which will catch these errors if they keep hanging
 990	 * around.
 991	 */
 992	if (XFS_BUF_ISASYNC(bp)) {
 993		xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
 994
 995		if (!XFS_BUF_ISSTALE(bp)) {
 996			XFS_BUF_DELAYWRITE(bp);
 997			XFS_BUF_DONE(bp);
 998		}
 999		ASSERT(bp->b_iodone != NULL);
1000		trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1001		xfs_buf_relse(bp);
1002		return;
1003	}
1004
1005	/*
1006	 * If the write of the buffer was synchronous, we want to make
1007	 * sure to return the error to the caller of xfs_bwrite().
1008	 */
1009	XFS_BUF_STALE(bp);
1010	XFS_BUF_DONE(bp);
1011	XFS_BUF_UNDELAYWRITE(bp);
1012
1013	trace_xfs_buf_error_relse(bp, _RET_IP_);
1014
1015do_callbacks:
1016	xfs_buf_do_callbacks(bp);
1017	bp->b_fspriv = NULL;
1018	bp->b_iodone = NULL;
1019	xfs_buf_ioend(bp, 0);
1020}
1021
1022/*
1023 * This is the iodone() function for buffers which have been
1024 * logged.  It is called when they are eventually flushed out.
1025 * It should remove the buf item from the AIL, and free the buf item.
1026 * It is called by xfs_buf_iodone_callbacks() above which will take
1027 * care of cleaning up the buffer itself.
1028 */
1029void
1030xfs_buf_iodone(
1031	struct xfs_buf		*bp,
1032	struct xfs_log_item	*lip)
1033{
1034	struct xfs_ail		*ailp = lip->li_ailp;
1035
1036	ASSERT(BUF_ITEM(lip)->bli_buf == bp);
1037
1038	xfs_buf_rele(bp);
1039
1040	/*
1041	 * If we are forcibly shutting down, this may well be
1042	 * off the AIL already. That's because we simulate the
1043	 * log-committed callbacks to unpin these buffers. Or we may never
1044	 * have put this item on AIL because of the transaction was
1045	 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1046	 *
1047	 * Either way, AIL is useless if we're forcing a shutdown.
 
 
 
1048	 */
1049	spin_lock(&ailp->xa_lock);
1050	xfs_trans_ail_delete(ailp, lip);
1051	xfs_buf_item_free(BUF_ITEM(lip));
 
1052}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * All Rights Reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_bit.h"
 
 
 
 
 
  13#include "xfs_mount.h"
  14#include "xfs_trans.h"
  15#include "xfs_trans_priv.h"
  16#include "xfs_buf_item.h"
  17#include "xfs_inode.h"
  18#include "xfs_inode_item.h"
  19#include "xfs_quota.h"
  20#include "xfs_dquot_item.h"
  21#include "xfs_dquot.h"
  22#include "xfs_trace.h"
  23#include "xfs_log.h"
  24#include "xfs_log_priv.h"
  25#include "xfs_error.h"
  26
  27
  28struct kmem_cache	*xfs_buf_item_cache;
  29
  30static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
  31{
  32	return container_of(lip, struct xfs_buf_log_item, bli_item);
  33}
  34
  35/* Is this log iovec plausibly large enough to contain the buffer log format? */
  36bool
  37xfs_buf_log_check_iovec(
  38	struct xfs_log_iovec		*iovec)
  39{
  40	struct xfs_buf_log_format	*blfp = iovec->i_addr;
  41	char				*bmp_end;
  42	char				*item_end;
  43
  44	if (offsetof(struct xfs_buf_log_format, blf_data_map) > iovec->i_len)
  45		return false;
  46
  47	item_end = (char *)iovec->i_addr + iovec->i_len;
  48	bmp_end = (char *)&blfp->blf_data_map[blfp->blf_map_size];
  49	return bmp_end <= item_end;
  50}
  51
  52static inline int
  53xfs_buf_log_format_size(
  54	struct xfs_buf_log_format *blfp)
 
 
 
 
 
 
 
  55{
  56	return offsetof(struct xfs_buf_log_format, blf_data_map) +
  57			(blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  58}
  59
  60static inline bool
  61xfs_buf_item_straddle(
  62	struct xfs_buf		*bp,
  63	uint			offset,
  64	int			first_bit,
  65	int			nbits)
 
 
 
 
  66{
  67	void			*first, *last;
 
  68
  69	first = xfs_buf_offset(bp, offset + (first_bit << XFS_BLF_SHIFT));
  70	last = xfs_buf_offset(bp,
  71			offset + ((first_bit + nbits) << XFS_BLF_SHIFT));
  72
  73	if (last - first != nbits * XFS_BLF_CHUNK)
  74		return true;
  75	return false;
  76}
  77
  78/*
  79 * Return the number of log iovecs and space needed to log the given buf log
  80 * item segment.
  81 *
  82 * It calculates this as 1 iovec for the buf log format structure and 1 for each
  83 * stretch of non-contiguous chunks to be logged.  Contiguous chunks are logged
  84 * in a single iovec.
 
  85 */
  86STATIC void
  87xfs_buf_item_size_segment(
  88	struct xfs_buf_log_item		*bip,
  89	struct xfs_buf_log_format	*blfp,
  90	uint				offset,
  91	int				*nvecs,
  92	int				*nbytes)
  93{
  94	struct xfs_buf			*bp = bip->bli_buf;
  95	int				first_bit;
  96	int				nbits;
  97	int				next_bit;
  98	int				last_bit;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  99
 100	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
 101	if (first_bit == -1)
 102		return;
 103
 104	(*nvecs)++;
 105	*nbytes += xfs_buf_log_format_size(blfp);
 106
 107	do {
 108		nbits = xfs_contig_bits(blfp->blf_data_map,
 109					blfp->blf_map_size, first_bit);
 110		ASSERT(nbits > 0);
 
 
 
 
 
 
 
 
 
 
 
 
 111
 
 
 112		/*
 113		 * Straddling a page is rare because we don't log contiguous
 114		 * chunks of unmapped buffers anywhere.
 
 115		 */
 116		if (nbits > 1 &&
 117		    xfs_buf_item_straddle(bp, offset, first_bit, nbits))
 118			goto slow_scan;
 
 119
 120		(*nvecs)++;
 121		*nbytes += nbits * XFS_BLF_CHUNK;
 122
 123		/*
 124		 * This takes the bit number to start looking from and
 125		 * returns the next set bit from there.  It returns -1
 126		 * if there are no more bits set or the start bit is
 127		 * beyond the end of the bitmap.
 128		 */
 129		first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 130					(uint)first_bit + nbits + 1);
 131	} while (first_bit != -1);
 132
 133	return;
 134
 135slow_scan:
 136	/* Count the first bit we jumped out of the above loop from */
 137	(*nvecs)++;
 138	*nbytes += XFS_BLF_CHUNK;
 139	last_bit = first_bit;
 140	while (last_bit != -1) {
 141		/*
 142		 * This takes the bit number to start looking from and
 143		 * returns the next set bit from there.  It returns -1
 144		 * if there are no more bits set or the start bit is
 145		 * beyond the end of the bitmap.
 146		 */
 147		next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 148					last_bit + 1);
 
 149		/*
 150		 * If we run out of bits, leave the loop,
 151		 * else if we find a new set of bits bump the number of vecs,
 152		 * else keep scanning the current set of bits.
 153		 */
 154		if (next_bit == -1) {
 155			break;
 156		} else if (next_bit != last_bit + 1 ||
 157		           xfs_buf_item_straddle(bp, offset, first_bit, nbits)) {
 
 
 
 
 158			last_bit = next_bit;
 159			first_bit = next_bit;
 160			(*nvecs)++;
 161			nbits = 1;
 162		} else {
 163			last_bit++;
 164			nbits++;
 165		}
 166		*nbytes += XFS_BLF_CHUNK;
 167	}
 
 
 
 168}
 169
 170/*
 171 * Return the number of log iovecs and space needed to log the given buf log
 172 * item.
 173 *
 174 * Discontiguous buffers need a format structure per region that is being
 175 * logged. This makes the changes in the buffer appear to log recovery as though
 176 * they came from separate buffers, just like would occur if multiple buffers
 177 * were used instead of a single discontiguous buffer. This enables
 178 * discontiguous buffers to be in-memory constructs, completely transparent to
 179 * what ends up on disk.
 180 *
 181 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
 182 * format structures. If the item has previously been logged and has dirty
 183 * regions, we do not relog them in stale buffers. This has the effect of
 184 * reducing the size of the relogged item by the amount of dirty data tracked
 185 * by the log item. This can result in the committing transaction reducing the
 186 * amount of space being consumed by the CIL.
 187 */
 188STATIC void
 189xfs_buf_item_size(
 190	struct xfs_log_item	*lip,
 191	int			*nvecs,
 192	int			*nbytes)
 193{
 194	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 195	struct xfs_buf		*bp = bip->bli_buf;
 196	int			i;
 197	int			bytes;
 198	uint			offset = 0;
 
 
 
 
 199
 200	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 201	if (bip->bli_flags & XFS_BLI_STALE) {
 202		/*
 203		 * The buffer is stale, so all we need to log is the buf log
 204		 * format structure with the cancel flag in it as we are never
 205		 * going to replay the changes tracked in the log item.
 206		 */
 207		trace_xfs_buf_item_size_stale(bip);
 208		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
 209		*nvecs += bip->bli_format_count;
 210		for (i = 0; i < bip->bli_format_count; i++) {
 211			*nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
 212		}
 213		return;
 214	}
 215
 216	ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
 217
 218	if (bip->bli_flags & XFS_BLI_ORDERED) {
 219		/*
 220		 * The buffer has been logged just to order it. It is not being
 221		 * included in the transaction commit, so no vectors are used at
 222		 * all.
 223		 */
 224		trace_xfs_buf_item_size_ordered(bip);
 225		*nvecs = XFS_LOG_VEC_ORDERED;
 226		return;
 227	}
 228
 229	/*
 230	 * The vector count is based on the number of buffer vectors we have
 231	 * dirty bits in. This will only be greater than one when we have a
 232	 * compound buffer with more than one segment dirty. Hence for compound
 233	 * buffers we need to track which segment the dirty bits correspond to,
 234	 * and when we move from one segment to the next increment the vector
 235	 * count for the extra buf log format structure that will need to be
 236	 * written.
 237	 */
 238	bytes = 0;
 239	for (i = 0; i < bip->bli_format_count; i++) {
 240		xfs_buf_item_size_segment(bip, &bip->bli_formats[i], offset,
 241					  nvecs, &bytes);
 242		offset += BBTOB(bp->b_maps[i].bm_len);
 243	}
 244
 245	/*
 246	 * Round up the buffer size required to minimise the number of memory
 247	 * allocations that need to be done as this item grows when relogged by
 248	 * repeated modifications.
 
 
 249	 */
 250	*nbytes = round_up(bytes, 512);
 251	trace_xfs_buf_item_size(bip);
 252}
 253
 254static inline void
 255xfs_buf_item_copy_iovec(
 256	struct xfs_log_vec	*lv,
 257	struct xfs_log_iovec	**vecp,
 258	struct xfs_buf		*bp,
 259	uint			offset,
 260	int			first_bit,
 261	uint			nbits)
 262{
 263	offset += first_bit * XFS_BLF_CHUNK;
 264	xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
 265			xfs_buf_offset(bp, offset),
 266			nbits * XFS_BLF_CHUNK);
 267}
 268
 269static void
 270xfs_buf_item_format_segment(
 271	struct xfs_buf_log_item	*bip,
 272	struct xfs_log_vec	*lv,
 273	struct xfs_log_iovec	**vecp,
 274	uint			offset,
 275	struct xfs_buf_log_format *blfp)
 276{
 277	struct xfs_buf		*bp = bip->bli_buf;
 278	uint			base_size;
 279	int			first_bit;
 280	int			last_bit;
 281	int			next_bit;
 282	uint			nbits;
 283
 284	/* copy the flags across from the base format item */
 285	blfp->blf_flags = bip->__bli_format.blf_flags;
 286
 287	/*
 288	 * Base size is the actual size of the ondisk structure - it reflects
 289	 * the actual size of the dirty bitmap rather than the size of the in
 290	 * memory structure.
 291	 */
 292	base_size = xfs_buf_log_format_size(blfp);
 293
 294	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
 295	if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
 296		/*
 297		 * If the map is not be dirty in the transaction, mark
 298		 * the size as zero and do not advance the vector pointer.
 299		 */
 300		return;
 301	}
 302
 303	blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
 304	blfp->blf_size = 1;
 305
 306	if (bip->bli_flags & XFS_BLI_STALE) {
 307		/*
 308		 * The buffer is stale, so all we need to log
 309		 * is the buf log format structure with the
 310		 * cancel flag in it.
 311		 */
 312		trace_xfs_buf_item_format_stale(bip);
 313		ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
 
 314		return;
 315	}
 316
 317
 318	/*
 319	 * Fill in an iovec for each set of contiguous chunks.
 320	 */
 321	do {
 322		ASSERT(first_bit >= 0);
 323		nbits = xfs_contig_bits(blfp->blf_data_map,
 324					blfp->blf_map_size, first_bit);
 325		ASSERT(nbits > 0);
 326
 327		/*
 328		 * Straddling a page is rare because we don't log contiguous
 329		 * chunks of unmapped buffers anywhere.
 330		 */
 331		if (nbits > 1 &&
 332		    xfs_buf_item_straddle(bp, offset, first_bit, nbits))
 333			goto slow_scan;
 334
 335		xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 336					first_bit, nbits);
 337		blfp->blf_size++;
 338
 339		/*
 340		 * This takes the bit number to start looking from and
 341		 * returns the next set bit from there.  It returns -1
 342		 * if there are no more bits set or the start bit is
 343		 * beyond the end of the bitmap.
 344		 */
 345		first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 346					(uint)first_bit + nbits + 1);
 347	} while (first_bit != -1);
 348
 349	return;
 350
 351slow_scan:
 352	ASSERT(bp->b_addr == NULL);
 353	last_bit = first_bit;
 354	nbits = 1;
 355	for (;;) {
 356		/*
 357		 * This takes the bit number to start looking from and
 358		 * returns the next set bit from there.  It returns -1
 359		 * if there are no more bits set or the start bit is
 360		 * beyond the end of the bitmap.
 361		 */
 362		next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 363					(uint)last_bit + 1);
 
 364		/*
 365		 * If we run out of bits fill in the last iovec and get out of
 366		 * the loop.  Else if we start a new set of bits then fill in
 367		 * the iovec for the series we were looking at and start
 368		 * counting the bits in the new one.  Else we're still in the
 369		 * same set of bits so just keep counting and scanning.
 
 
 370		 */
 371		if (next_bit == -1) {
 372			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 373						first_bit, nbits);
 374			blfp->blf_size++;
 
 
 375			break;
 376		} else if (next_bit != last_bit + 1 ||
 377		           xfs_buf_item_straddle(bp, offset, first_bit, nbits)) {
 378			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 379						first_bit, nbits);
 380			blfp->blf_size++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 381			first_bit = next_bit;
 382			last_bit = next_bit;
 383			nbits = 1;
 384		} else {
 385			last_bit++;
 386			nbits++;
 387		}
 388	}
 389}
 390
 391/*
 392 * This is called to fill in the vector of log iovecs for the
 393 * given log buf item.  It fills the first entry with a buf log
 394 * format structure, and the rest point to contiguous chunks
 395 * within the buffer.
 396 */
 397STATIC void
 398xfs_buf_item_format(
 399	struct xfs_log_item	*lip,
 400	struct xfs_log_vec	*lv)
 401{
 402	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 403	struct xfs_buf		*bp = bip->bli_buf;
 404	struct xfs_log_iovec	*vecp = NULL;
 405	uint			offset = 0;
 406	int			i;
 407
 408	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 409	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
 410	       (bip->bli_flags & XFS_BLI_STALE));
 411	ASSERT((bip->bli_flags & XFS_BLI_STALE) ||
 412	       (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF
 413	        && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF));
 414	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) ||
 415	       (bip->bli_flags & XFS_BLI_STALE));
 416
 417
 418	/*
 419	 * If it is an inode buffer, transfer the in-memory state to the
 420	 * format flags and clear the in-memory state.
 421	 *
 422	 * For buffer based inode allocation, we do not transfer
 423	 * this state if the inode buffer allocation has not yet been committed
 424	 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
 425	 * correct replay of the inode allocation.
 426	 *
 427	 * For icreate item based inode allocation, the buffers aren't written
 428	 * to the journal during allocation, and hence we should always tag the
 429	 * buffer as an inode buffer so that the correct unlinked list replay
 430	 * occurs during recovery.
 431	 */
 432	if (bip->bli_flags & XFS_BLI_INODE_BUF) {
 433		if (xfs_has_v3inodes(lip->li_log->l_mp) ||
 434		    !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
 435		      xfs_log_item_in_current_chkpt(lip)))
 436			bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
 437		bip->bli_flags &= ~XFS_BLI_INODE_BUF;
 438	}
 439
 440	for (i = 0; i < bip->bli_format_count; i++) {
 441		xfs_buf_item_format_segment(bip, lv, &vecp, offset,
 442					    &bip->bli_formats[i]);
 443		offset += BBTOB(bp->b_maps[i].bm_len);
 444	}
 445
 446	/*
 447	 * Check to make sure everything is consistent.
 448	 */
 449	trace_xfs_buf_item_format(bip);
 
 450}
 451
 452/*
 453 * This is called to pin the buffer associated with the buf log item in memory
 454 * so it cannot be written out.
 455 *
 456 * We take a reference to the buffer log item here so that the BLI life cycle
 457 * extends at least until the buffer is unpinned via xfs_buf_item_unpin() and
 458 * inserted into the AIL.
 459 *
 460 * We also need to take a reference to the buffer itself as the BLI unpin
 461 * processing requires accessing the buffer after the BLI has dropped the final
 462 * BLI reference. See xfs_buf_item_unpin() for an explanation.
 463 * If unpins race to drop the final BLI reference and only the
 464 * BLI owns a reference to the buffer, then the loser of the race can have the
 465 * buffer fgreed from under it (e.g. on shutdown). Taking a buffer reference per
 466 * pin count ensures the life cycle of the buffer extends for as
 467 * long as we hold the buffer pin reference in xfs_buf_item_unpin().
 468 */
 469STATIC void
 470xfs_buf_item_pin(
 471	struct xfs_log_item	*lip)
 472{
 473	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 474
 475	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 476	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
 477	       (bip->bli_flags & XFS_BLI_ORDERED) ||
 478	       (bip->bli_flags & XFS_BLI_STALE));
 479
 480	trace_xfs_buf_item_pin(bip);
 481
 482	xfs_buf_hold(bip->bli_buf);
 483	atomic_inc(&bip->bli_refcount);
 484	atomic_inc(&bip->bli_buf->b_pin_count);
 485}
 486
 487/*
 488 * This is called to unpin the buffer associated with the buf log item which was
 489 * previously pinned with a call to xfs_buf_item_pin().  We enter this function
 490 * with a buffer pin count, a buffer reference and a BLI reference.
 491 *
 492 * We must drop the BLI reference before we unpin the buffer because the AIL
 493 * doesn't acquire a BLI reference whenever it accesses it. Therefore if the
 494 * refcount drops to zero, the bli could still be AIL resident and the buffer
 495 * submitted for I/O at any point before we return. This can result in IO
 496 * completion freeing the buffer while we are still trying to access it here.
 497 * This race condition can also occur in shutdown situations where we abort and
 498 * unpin buffers from contexts other that journal IO completion.
 499 *
 500 * Hence we have to hold a buffer reference per pin count to ensure that the
 501 * buffer cannot be freed until we have finished processing the unpin operation.
 502 * The reference is taken in xfs_buf_item_pin(), and we must hold it until we
 503 * are done processing the buffer state. In the case of an abort (remove =
 504 * true) then we re-use the current pin reference as the IO reference we hand
 505 * off to IO failure handling.
 506 */
 507STATIC void
 508xfs_buf_item_unpin(
 509	struct xfs_log_item	*lip,
 510	int			remove)
 511{
 512	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 513	struct xfs_buf		*bp = bip->bli_buf;
 514	int			stale = bip->bli_flags & XFS_BLI_STALE;
 515	int			freed;
 
 516
 517	ASSERT(bp->b_log_item == bip);
 518	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 519
 520	trace_xfs_buf_item_unpin(bip);
 521
 522	freed = atomic_dec_and_test(&bip->bli_refcount);
 
 523	if (atomic_dec_and_test(&bp->b_pin_count))
 524		wake_up_all(&bp->b_waiters);
 525
 526	/*
 527	 * Nothing to do but drop the buffer pin reference if the BLI is
 528	 * still active.
 529	 */
 530	if (!freed) {
 531		xfs_buf_rele(bp);
 532		return;
 533	}
 534
 535	if (stale) {
 536		ASSERT(bip->bli_flags & XFS_BLI_STALE);
 537		ASSERT(xfs_buf_islocked(bp));
 538		ASSERT(bp->b_flags & XBF_STALE);
 539		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
 540		ASSERT(list_empty(&lip->li_trans));
 541		ASSERT(!bp->b_transp);
 542
 543		trace_xfs_buf_item_unpin_stale(bip);
 544
 545		/*
 546		 * The buffer has been locked and referenced since it was marked
 547		 * stale so we own both lock and reference exclusively here. We
 548		 * do not need the pin reference any more, so drop it now so
 549		 * that we only have one reference to drop once item completion
 550		 * processing is complete.
 551		 */
 552		xfs_buf_rele(bp);
 
 
 
 
 
 
 
 
 
 
 553
 554		/*
 555		 * If we get called here because of an IO error, we may or may
 556		 * not have the item on the AIL. xfs_trans_ail_delete() will
 557		 * take care of that situation. xfs_trans_ail_delete() drops
 558		 * the AIL lock.
 559		 */
 560		if (bip->bli_flags & XFS_BLI_STALE_INODE) {
 561			xfs_buf_item_done(bp);
 562			xfs_buf_inode_iodone(bp);
 563			ASSERT(list_empty(&bp->b_li_list));
 564		} else {
 565			xfs_trans_ail_delete(lip, SHUTDOWN_LOG_IO_ERROR);
 
 566			xfs_buf_item_relse(bp);
 567			ASSERT(bp->b_log_item == NULL);
 568		}
 569		xfs_buf_relse(bp);
 570		return;
 571	}
 572
 573	if (remove) {
 574		/*
 575		 * We need to simulate an async IO failures here to ensure that
 576		 * the correct error completion is run on this buffer. This
 577		 * requires a reference to the buffer and for the buffer to be
 578		 * locked. We can safely pass ownership of the pin reference to
 579		 * the IO to ensure that nothing can free the buffer while we
 580		 * wait for the lock and then run the IO failure completion.
 581		 */
 582		xfs_buf_lock(bp);
 583		bp->b_flags |= XBF_ASYNC;
 584		xfs_buf_ioend_fail(bp);
 585		return;
 586	}
 587
 588	/*
 589	 * BLI has no more active references - it will be moved to the AIL to
 590	 * manage the remaining BLI/buffer life cycle. There is nothing left for
 591	 * us to do here so drop the pin reference to the buffer.
 592	 */
 593	xfs_buf_rele(bp);
 594}
 595
 
 
 
 
 
 
 
 
 596STATIC uint
 597xfs_buf_item_push(
 598	struct xfs_log_item	*lip,
 599	struct list_head	*buffer_list)
 600{
 601	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 602	struct xfs_buf		*bp = bip->bli_buf;
 603	uint			rval = XFS_ITEM_SUCCESS;
 604
 605	if (xfs_buf_ispinned(bp))
 606		return XFS_ITEM_PINNED;
 607	if (!xfs_buf_trylock(bp)) {
 608		/*
 609		 * If we have just raced with a buffer being pinned and it has
 610		 * been marked stale, we could end up stalling until someone else
 611		 * issues a log force to unpin the stale buffer. Check for the
 612		 * race condition here so xfsaild recognizes the buffer is pinned
 613		 * and queues a log force to move it along.
 614		 */
 615		if (xfs_buf_ispinned(bp))
 616			return XFS_ITEM_PINNED;
 617		return XFS_ITEM_LOCKED;
 618	}
 
 
 619
 620	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 621
 622	trace_xfs_buf_item_push(bip);
 623
 624	/* has a previous flush failed due to IO errors? */
 625	if (bp->b_flags & XBF_WRITE_FAIL) {
 626		xfs_buf_alert_ratelimited(bp, "XFS: Failing async write",
 627	    "Failing async write on buffer block 0x%llx. Retrying async write.",
 628					  (long long)xfs_buf_daddr(bp));
 629	}
 630
 631	if (!xfs_buf_delwri_queue(bp, buffer_list))
 632		rval = XFS_ITEM_FLUSHING;
 633	xfs_buf_unlock(bp);
 634	return rval;
 635}
 636
 637/*
 638 * Drop the buffer log item refcount and take appropriate action. This helper
 639 * determines whether the bli must be freed or not, since a decrement to zero
 640 * does not necessarily mean the bli is unused.
 641 *
 642 * Return true if the bli is freed, false otherwise.
 643 */
 644bool
 645xfs_buf_item_put(
 646	struct xfs_buf_log_item	*bip)
 647{
 648	struct xfs_log_item	*lip = &bip->bli_item;
 649	bool			aborted;
 650	bool			dirty;
 651
 652	/* drop the bli ref and return if it wasn't the last one */
 653	if (!atomic_dec_and_test(&bip->bli_refcount))
 654		return false;
 655
 656	/*
 657	 * We dropped the last ref and must free the item if clean or aborted.
 658	 * If the bli is dirty and non-aborted, the buffer was clean in the
 659	 * transaction but still awaiting writeback from previous changes. In
 660	 * that case, the bli is freed on buffer writeback completion.
 661	 */
 662	aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) ||
 663			xlog_is_shutdown(lip->li_log);
 664	dirty = bip->bli_flags & XFS_BLI_DIRTY;
 665	if (dirty && !aborted)
 666		return false;
 667
 668	/*
 669	 * The bli is aborted or clean. An aborted item may be in the AIL
 670	 * regardless of dirty state.  For example, consider an aborted
 671	 * transaction that invalidated a dirty bli and cleared the dirty
 672	 * state.
 673	 */
 674	if (aborted)
 675		xfs_trans_ail_delete(lip, 0);
 676	xfs_buf_item_relse(bip->bli_buf);
 677	return true;
 678}
 679
 680/*
 681 * Release the buffer associated with the buf log item.  If there is no dirty
 682 * logged data associated with the buffer recorded in the buf log item, then
 683 * free the buf log item and remove the reference to it in the buffer.
 684 *
 685 * This call ignores the recursion count.  It is only called when the buffer
 686 * should REALLY be unlocked, regardless of the recursion count.
 687 *
 688 * We unconditionally drop the transaction's reference to the log item. If the
 689 * item was logged, then another reference was taken when it was pinned, so we
 690 * can safely drop the transaction reference now.  This also allows us to avoid
 691 * potential races with the unpin code freeing the bli by not referencing the
 692 * bli after we've dropped the reference count.
 693 *
 694 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
 695 * if necessary but do not unlock the buffer.  This is for support of
 696 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
 697 * free the item.
 698 */
 699STATIC void
 700xfs_buf_item_release(
 701	struct xfs_log_item	*lip)
 702{
 703	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 704	struct xfs_buf		*bp = bip->bli_buf;
 705	bool			released;
 706	bool			hold = bip->bli_flags & XFS_BLI_HOLD;
 707	bool			stale = bip->bli_flags & XFS_BLI_STALE;
 708#if defined(DEBUG) || defined(XFS_WARN)
 709	bool			ordered = bip->bli_flags & XFS_BLI_ORDERED;
 710	bool			dirty = bip->bli_flags & XFS_BLI_DIRTY;
 711	bool			aborted = test_bit(XFS_LI_ABORTED,
 712						   &lip->li_flags);
 713#endif
 714
 715	trace_xfs_buf_item_release(bip);
 
 716
 717	/*
 718	 * The bli dirty state should match whether the blf has logged segments
 719	 * except for ordered buffers, where only the bli should be dirty.
 
 
 720	 */
 721	ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) ||
 722	       (ordered && dirty && !xfs_buf_item_dirty_format(bip)));
 723	ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
 724
 725	/*
 726	 * Clear the buffer's association with this transaction and
 727	 * per-transaction state from the bli, which has been copied above.
 728	 */
 729	bp->b_transp = NULL;
 730	bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 731
 732	/*
 733	 * Unref the item and unlock the buffer unless held or stale. Stale
 734	 * buffers remain locked until final unpin unless the bli is freed by
 735	 * the unref call. The latter implies shutdown because buffer
 736	 * invalidation dirties the bli and transaction.
 737	 */
 738	released = xfs_buf_item_put(bip);
 739	if (hold || (stale && !released))
 740		return;
 741	ASSERT(!stale || aborted);
 742	xfs_buf_relse(bp);
 743}
 744
 745STATIC void
 746xfs_buf_item_committing(
 747	struct xfs_log_item	*lip,
 748	xfs_csn_t		seq)
 749{
 750	return xfs_buf_item_release(lip);
 751}
 752
 753/*
 754 * This is called to find out where the oldest active copy of the
 755 * buf log item in the on disk log resides now that the last log
 756 * write of it completed at the given lsn.
 757 * We always re-log all the dirty data in a buffer, so usually the
 758 * latest copy in the on disk log is the only one that matters.  For
 759 * those cases we simply return the given lsn.
 760 *
 761 * The one exception to this is for buffers full of newly allocated
 762 * inodes.  These buffers are only relogged with the XFS_BLI_INODE_BUF
 763 * flag set, indicating that only the di_next_unlinked fields from the
 764 * inodes in the buffers will be replayed during recovery.  If the
 765 * original newly allocated inode images have not yet been flushed
 766 * when the buffer is so relogged, then we need to make sure that we
 767 * keep the old images in the 'active' portion of the log.  We do this
 768 * by returning the original lsn of that transaction here rather than
 769 * the current one.
 770 */
 771STATIC xfs_lsn_t
 772xfs_buf_item_committed(
 773	struct xfs_log_item	*lip,
 774	xfs_lsn_t		lsn)
 775{
 776	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 777
 778	trace_xfs_buf_item_committed(bip);
 779
 780	if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
 781		return lip->li_lsn;
 782	return lsn;
 783}
 784
 785#ifdef DEBUG_EXPENSIVE
 786static int
 787xfs_buf_item_precommit(
 788	struct xfs_trans	*tp,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 789	struct xfs_log_item	*lip)
 790{
 791	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 792	struct xfs_buf		*bp = bip->bli_buf;
 793	struct xfs_mount	*mp = bp->b_mount;
 794	xfs_failaddr_t		fa;
 795
 796	if (!bp->b_ops || !bp->b_ops->verify_struct)
 797		return 0;
 798	if (bip->bli_flags & XFS_BLI_STALE)
 799		return 0;
 800
 801	fa = bp->b_ops->verify_struct(bp);
 802	if (fa) {
 803		xfs_buf_verifier_error(bp, -EFSCORRUPTED, bp->b_ops->name,
 804				bp->b_addr, BBTOB(bp->b_length), fa);
 805		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 806		ASSERT(fa == NULL);
 807	}
 808
 809	return 0;
 
 
 
 
 810}
 811#else
 812# define xfs_buf_item_precommit	NULL
 813#endif
 814
 815static const struct xfs_item_ops xfs_buf_item_ops = {
 
 
 
 816	.iop_size	= xfs_buf_item_size,
 817	.iop_precommit	= xfs_buf_item_precommit,
 818	.iop_format	= xfs_buf_item_format,
 819	.iop_pin	= xfs_buf_item_pin,
 820	.iop_unpin	= xfs_buf_item_unpin,
 821	.iop_release	= xfs_buf_item_release,
 822	.iop_committing	= xfs_buf_item_committing,
 823	.iop_committed	= xfs_buf_item_committed,
 824	.iop_push	= xfs_buf_item_push,
 
 
 825};
 826
 827STATIC void
 828xfs_buf_item_get_format(
 829	struct xfs_buf_log_item	*bip,
 830	int			count)
 831{
 832	ASSERT(bip->bli_formats == NULL);
 833	bip->bli_format_count = count;
 834
 835	if (count == 1) {
 836		bip->bli_formats = &bip->__bli_format;
 837		return;
 838	}
 839
 840	bip->bli_formats = kzalloc(count * sizeof(struct xfs_buf_log_format),
 841				GFP_KERNEL | __GFP_NOFAIL);
 842}
 843
 844STATIC void
 845xfs_buf_item_free_format(
 846	struct xfs_buf_log_item	*bip)
 847{
 848	if (bip->bli_formats != &bip->__bli_format) {
 849		kfree(bip->bli_formats);
 850		bip->bli_formats = NULL;
 851	}
 852}
 853
 854/*
 855 * Allocate a new buf log item to go with the given buffer.
 856 * Set the buffer's b_log_item field to point to the new
 857 * buf log item.
 
 
 858 */
 859int
 860xfs_buf_item_init(
 861	struct xfs_buf	*bp,
 862	struct xfs_mount *mp)
 863{
 864	struct xfs_buf_log_item	*bip = bp->b_log_item;
 
 865	int			chunks;
 866	int			map_size;
 867	int			i;
 868
 869	/*
 870	 * Check to see if there is already a buf log item for
 871	 * this buffer. If we do already have one, there is
 
 872	 * nothing to do here so return.
 873	 */
 874	ASSERT(bp->b_mount == mp);
 875	if (bip) {
 876		ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
 877		ASSERT(!bp->b_transp);
 878		ASSERT(bip->bli_buf == bp);
 879		return 0;
 880	}
 
 
 
 
 
 881
 882	bip = kmem_cache_zalloc(xfs_buf_item_cache, GFP_KERNEL | __GFP_NOFAIL);
 
 883	xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
 884	bip->bli_buf = bp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 885
 886	/*
 887	 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
 888	 * can be divided into. Make sure not to truncate any pieces.
 889	 * map_size is the size of the bitmap needed to describe the
 890	 * chunks of the buffer.
 891	 *
 892	 * Discontiguous buffer support follows the layout of the underlying
 893	 * buffer. This makes the implementation as simple as possible.
 894	 */
 895	xfs_buf_item_get_format(bip, bp->b_map_count);
 896
 897	for (i = 0; i < bip->bli_format_count; i++) {
 898		chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
 899				      XFS_BLF_CHUNK);
 900		map_size = DIV_ROUND_UP(chunks, NBWORD);
 901
 902		if (map_size > XFS_BLF_DATAMAP_SIZE) {
 903			kmem_cache_free(xfs_buf_item_cache, bip);
 904			xfs_err(mp,
 905	"buffer item dirty bitmap (%u uints) too small to reflect %u bytes!",
 906					map_size,
 907					BBTOB(bp->b_maps[i].bm_len));
 908			return -EFSCORRUPTED;
 909		}
 910
 911		bip->bli_formats[i].blf_type = XFS_LI_BUF;
 912		bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
 913		bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
 914		bip->bli_formats[i].blf_map_size = map_size;
 915	}
 916
 917	bp->b_log_item = bip;
 918	xfs_buf_hold(bp);
 919	return 0;
 920}
 921
 922
 923/*
 924 * Mark bytes first through last inclusive as dirty in the buf
 925 * item's bitmap.
 926 */
 927static void
 928xfs_buf_item_log_segment(
 
 929	uint			first,
 930	uint			last,
 931	uint			*map)
 932{
 933	uint		first_bit;
 934	uint		last_bit;
 935	uint		bits_to_set;
 936	uint		bits_set;
 937	uint		word_num;
 938	uint		*wordp;
 939	uint		bit;
 940	uint		end_bit;
 941	uint		mask;
 942
 943	ASSERT(first < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD);
 944	ASSERT(last < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD);
 
 
 
 945
 946	/*
 947	 * Convert byte offsets to bit numbers.
 948	 */
 949	first_bit = first >> XFS_BLF_SHIFT;
 950	last_bit = last >> XFS_BLF_SHIFT;
 951
 952	/*
 953	 * Calculate the total number of bits to be set.
 954	 */
 955	bits_to_set = last_bit - first_bit + 1;
 956
 957	/*
 958	 * Get a pointer to the first word in the bitmap
 959	 * to set a bit in.
 960	 */
 961	word_num = first_bit >> BIT_TO_WORD_SHIFT;
 962	wordp = &map[word_num];
 963
 964	/*
 965	 * Calculate the starting bit in the first word.
 966	 */
 967	bit = first_bit & (uint)(NBWORD - 1);
 968
 969	/*
 970	 * First set any bits in the first word of our range.
 971	 * If it starts at bit 0 of the word, it will be
 972	 * set below rather than here.  That is what the variable
 973	 * bit tells us. The variable bits_set tracks the number
 974	 * of bits that have been set so far.  End_bit is the number
 975	 * of the last bit to be set in this word plus one.
 976	 */
 977	if (bit) {
 978		end_bit = min(bit + bits_to_set, (uint)NBWORD);
 979		mask = ((1U << (end_bit - bit)) - 1) << bit;
 980		*wordp |= mask;
 981		wordp++;
 982		bits_set = end_bit - bit;
 983	} else {
 984		bits_set = 0;
 985	}
 986
 987	/*
 988	 * Now set bits a whole word at a time that are between
 989	 * first_bit and last_bit.
 990	 */
 991	while ((bits_to_set - bits_set) >= NBWORD) {
 992		*wordp = 0xffffffff;
 993		bits_set += NBWORD;
 994		wordp++;
 995	}
 996
 997	/*
 998	 * Finally, set any bits left to be set in one last partial word.
 999	 */
1000	end_bit = bits_to_set - bits_set;
1001	if (end_bit) {
1002		mask = (1U << end_bit) - 1;
1003		*wordp |= mask;
1004	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1005}
1006
1007/*
1008 * Mark bytes first through last inclusive as dirty in the buf
1009 * item's bitmap.
 
 
 
1010 */
1011void
1012xfs_buf_item_log(
1013	struct xfs_buf_log_item	*bip,
1014	uint			first,
1015	uint			last)
1016{
1017	int			i;
1018	uint			start;
1019	uint			end;
1020	struct xfs_buf		*bp = bip->bli_buf;
1021
1022	/*
1023	 * walk each buffer segment and mark them dirty appropriately.
1024	 */
1025	start = 0;
1026	for (i = 0; i < bip->bli_format_count; i++) {
1027		if (start > last)
1028			break;
1029		end = start + BBTOB(bp->b_maps[i].bm_len) - 1;
1030
1031		/* skip to the map that includes the first byte to log */
1032		if (first > end) {
1033			start += BBTOB(bp->b_maps[i].bm_len);
1034			continue;
1035		}
1036
1037		/*
1038		 * Trim the range to this segment and mark it in the bitmap.
1039		 * Note that we must convert buffer offsets to segment relative
1040		 * offsets (e.g., the first byte of each segment is byte 0 of
1041		 * that segment).
1042		 */
1043		if (first < start)
1044			first = start;
1045		if (end > last)
1046			end = last;
1047		xfs_buf_item_log_segment(first - start, end - start,
1048					 &bip->bli_formats[i].blf_data_map[0]);
1049
1050		start += BBTOB(bp->b_maps[i].bm_len);
1051	}
1052}
1053
1054
1055/*
1056 * Return true if the buffer has any ranges logged/dirtied by a transaction,
1057 * false otherwise.
 
 
 
 
 
1058 */
1059bool
1060xfs_buf_item_dirty_format(
1061	struct xfs_buf_log_item	*bip)
1062{
1063	int			i;
1064
1065	for (i = 0; i < bip->bli_format_count; i++) {
1066		if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
1067			     bip->bli_formats[i].blf_map_size))
1068			return true;
 
 
 
 
 
 
 
1069	}
1070
1071	return false;
 
 
1072}
1073
 
 
 
 
 
 
 
 
 
 
 
 
1074STATIC void
1075xfs_buf_item_free(
1076	struct xfs_buf_log_item	*bip)
1077{
1078	xfs_buf_item_free_format(bip);
1079	kvfree(bip->bli_item.li_lv_shadow);
1080	kmem_cache_free(xfs_buf_item_cache, bip);
 
 
 
 
 
 
 
 
 
 
 
1081}
1082
1083/*
1084 * xfs_buf_item_relse() is called when the buf log item is no longer needed.
 
 
 
 
1085 */
1086void
1087xfs_buf_item_relse(
1088	struct xfs_buf	*bp)
1089{
1090	struct xfs_buf_log_item	*bip = bp->b_log_item;
 
 
 
 
 
 
1091
1092	trace_xfs_buf_item_relse(bp, _RET_IP_);
1093	ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1094
1095	if (atomic_read(&bip->bli_refcount))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1096		return;
1097	bp->b_log_item = NULL;
1098	xfs_buf_rele(bp);
1099	xfs_buf_item_free(bip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1100}
1101
 
 
 
 
 
 
 
1102void
1103xfs_buf_item_done(
1104	struct xfs_buf		*bp)
 
1105{
 
 
 
 
 
 
1106	/*
1107	 * If we are forcibly shutting down, this may well be off the AIL
1108	 * already. That's because we simulate the log-committed callbacks to
1109	 * unpin these buffers. Or we may never have put this item on AIL
1110	 * because of the transaction was aborted forcibly.
1111	 * xfs_trans_ail_delete() takes care of these.
1112	 *
1113	 * Either way, AIL is useless if we're forcing a shutdown.
1114	 *
1115	 * Note that log recovery writes might have buffer items that are not on
1116	 * the AIL even when the file system is not shut down.
1117	 */
1118	xfs_trans_ail_delete(&bp->b_log_item->bli_item,
1119			     (bp->b_flags & _XBF_LOGRECOVERY) ? 0 :
1120			     SHUTDOWN_CORRUPT_INCORE);
1121	xfs_buf_item_relse(bp);
1122}