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}
v6.9.4
   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
  26
  27struct kmem_cache	*xfs_buf_item_cache;
  28
  29static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
  30{
  31	return container_of(lip, struct xfs_buf_log_item, bli_item);
  32}
  33
  34/* Is this log iovec plausibly large enough to contain the buffer log format? */
  35bool
  36xfs_buf_log_check_iovec(
  37	struct xfs_log_iovec		*iovec)
  38{
  39	struct xfs_buf_log_format	*blfp = iovec->i_addr;
  40	char				*bmp_end;
  41	char				*item_end;
  42
  43	if (offsetof(struct xfs_buf_log_format, blf_data_map) > iovec->i_len)
  44		return false;
  45
  46	item_end = (char *)iovec->i_addr + iovec->i_len;
  47	bmp_end = (char *)&blfp->blf_data_map[blfp->blf_map_size];
  48	return bmp_end <= item_end;
  49}
  50
  51static inline int
  52xfs_buf_log_format_size(
  53	struct xfs_buf_log_format *blfp)
 
 
 
 
  54{
  55	return offsetof(struct xfs_buf_log_format, blf_data_map) +
  56			(blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  57}
  58
  59static inline bool
  60xfs_buf_item_straddle(
  61	struct xfs_buf		*bp,
  62	uint			offset,
  63	int			first_bit,
  64	int			nbits)
 
 
 
 
  65{
  66	void			*first, *last;
 
  67
  68	first = xfs_buf_offset(bp, offset + (first_bit << XFS_BLF_SHIFT));
  69	last = xfs_buf_offset(bp,
  70			offset + ((first_bit + nbits) << XFS_BLF_SHIFT));
  71
  72	if (last - first != nbits * XFS_BLF_CHUNK)
  73		return true;
  74	return false;
  75}
  76
  77/*
  78 * Return the number of log iovecs and space needed to log the given buf log
  79 * item segment.
  80 *
  81 * It calculates this as 1 iovec for the buf log format structure and 1 for each
  82 * stretch of non-contiguous chunks to be logged.  Contiguous chunks are logged
  83 * in a single iovec.
 
  84 */
  85STATIC void
  86xfs_buf_item_size_segment(
  87	struct xfs_buf_log_item		*bip,
  88	struct xfs_buf_log_format	*blfp,
  89	uint				offset,
  90	int				*nvecs,
  91	int				*nbytes)
  92{
  93	struct xfs_buf			*bp = bip->bli_buf;
  94	int				first_bit;
  95	int				nbits;
  96	int				next_bit;
  97	int				last_bit;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  98
  99	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
 100	if (first_bit == -1)
 101		return;
 102
 103	(*nvecs)++;
 104	*nbytes += xfs_buf_log_format_size(blfp);
 105
 106	do {
 107		nbits = xfs_contig_bits(blfp->blf_data_map,
 108					blfp->blf_map_size, first_bit);
 109		ASSERT(nbits > 0);
 
 
 
 
 
 
 
 
 
 
 
 
 110
 
 
 111		/*
 112		 * Straddling a page is rare because we don't log contiguous
 113		 * chunks of unmapped buffers anywhere.
 
 114		 */
 115		if (nbits > 1 &&
 116		    xfs_buf_item_straddle(bp, offset, first_bit, nbits))
 117			goto slow_scan;
 
 118
 119		(*nvecs)++;
 120		*nbytes += nbits * XFS_BLF_CHUNK;
 121
 122		/*
 123		 * This takes the bit number to start looking from and
 124		 * returns the next set bit from there.  It returns -1
 125		 * if there are no more bits set or the start bit is
 126		 * beyond the end of the bitmap.
 127		 */
 128		first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 129					(uint)first_bit + nbits + 1);
 130	} while (first_bit != -1);
 131
 132	return;
 133
 134slow_scan:
 135	/* Count the first bit we jumped out of the above loop from */
 136	(*nvecs)++;
 137	*nbytes += XFS_BLF_CHUNK;
 138	last_bit = first_bit;
 139	while (last_bit != -1) {
 140		/*
 141		 * This takes the bit number to start looking from and
 142		 * returns the next set bit from there.  It returns -1
 143		 * if there are no more bits set or the start bit is
 144		 * beyond the end of the bitmap.
 145		 */
 146		next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 147					last_bit + 1);
 
 148		/*
 149		 * If we run out of bits, leave the loop,
 150		 * else if we find a new set of bits bump the number of vecs,
 151		 * else keep scanning the current set of bits.
 152		 */
 153		if (next_bit == -1) {
 154			break;
 155		} else if (next_bit != last_bit + 1 ||
 156		           xfs_buf_item_straddle(bp, offset, first_bit, nbits)) {
 
 
 
 
 157			last_bit = next_bit;
 158			first_bit = next_bit;
 159			(*nvecs)++;
 160			nbits = 1;
 161		} else {
 162			last_bit++;
 163			nbits++;
 164		}
 165		*nbytes += XFS_BLF_CHUNK;
 166	}
 
 
 
 167}
 168
 169/*
 170 * Return the number of log iovecs and space needed to log the given buf log
 171 * item.
 172 *
 173 * Discontiguous buffers need a format structure per region that is being
 174 * logged. This makes the changes in the buffer appear to log recovery as though
 175 * they came from separate buffers, just like would occur if multiple buffers
 176 * were used instead of a single discontiguous buffer. This enables
 177 * discontiguous buffers to be in-memory constructs, completely transparent to
 178 * what ends up on disk.
 179 *
 180 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
 181 * format structures. If the item has previously been logged and has dirty
 182 * regions, we do not relog them in stale buffers. This has the effect of
 183 * reducing the size of the relogged item by the amount of dirty data tracked
 184 * by the log item. This can result in the committing transaction reducing the
 185 * amount of space being consumed by the CIL.
 186 */
 187STATIC void
 188xfs_buf_item_size(
 189	struct xfs_log_item	*lip,
 190	int			*nvecs,
 191	int			*nbytes)
 192{
 193	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 194	struct xfs_buf		*bp = bip->bli_buf;
 195	int			i;
 196	int			bytes;
 197	uint			offset = 0;
 
 
 
 
 198
 199	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 200	if (bip->bli_flags & XFS_BLI_STALE) {
 201		/*
 202		 * The buffer is stale, so all we need to log is the buf log
 203		 * format structure with the cancel flag in it as we are never
 204		 * going to replay the changes tracked in the log item.
 205		 */
 206		trace_xfs_buf_item_size_stale(bip);
 207		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
 208		*nvecs += bip->bli_format_count;
 209		for (i = 0; i < bip->bli_format_count; i++) {
 210			*nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
 211		}
 212		return;
 213	}
 214
 215	ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
 216
 217	if (bip->bli_flags & XFS_BLI_ORDERED) {
 218		/*
 219		 * The buffer has been logged just to order it. It is not being
 220		 * included in the transaction commit, so no vectors are used at
 221		 * all.
 222		 */
 223		trace_xfs_buf_item_size_ordered(bip);
 224		*nvecs = XFS_LOG_VEC_ORDERED;
 225		return;
 226	}
 227
 228	/*
 229	 * The vector count is based on the number of buffer vectors we have
 230	 * dirty bits in. This will only be greater than one when we have a
 231	 * compound buffer with more than one segment dirty. Hence for compound
 232	 * buffers we need to track which segment the dirty bits correspond to,
 233	 * and when we move from one segment to the next increment the vector
 234	 * count for the extra buf log format structure that will need to be
 235	 * written.
 236	 */
 237	bytes = 0;
 238	for (i = 0; i < bip->bli_format_count; i++) {
 239		xfs_buf_item_size_segment(bip, &bip->bli_formats[i], offset,
 240					  nvecs, &bytes);
 241		offset += BBTOB(bp->b_maps[i].bm_len);
 242	}
 243
 244	/*
 245	 * Round up the buffer size required to minimise the number of memory
 246	 * allocations that need to be done as this item grows when relogged by
 247	 * repeated modifications.
 
 
 248	 */
 249	*nbytes = round_up(bytes, 512);
 250	trace_xfs_buf_item_size(bip);
 251}
 252
 253static inline void
 254xfs_buf_item_copy_iovec(
 255	struct xfs_log_vec	*lv,
 256	struct xfs_log_iovec	**vecp,
 257	struct xfs_buf		*bp,
 258	uint			offset,
 259	int			first_bit,
 260	uint			nbits)
 261{
 262	offset += first_bit * XFS_BLF_CHUNK;
 263	xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
 264			xfs_buf_offset(bp, offset),
 265			nbits * XFS_BLF_CHUNK);
 266}
 267
 268static void
 269xfs_buf_item_format_segment(
 270	struct xfs_buf_log_item	*bip,
 271	struct xfs_log_vec	*lv,
 272	struct xfs_log_iovec	**vecp,
 273	uint			offset,
 274	struct xfs_buf_log_format *blfp)
 275{
 276	struct xfs_buf		*bp = bip->bli_buf;
 277	uint			base_size;
 278	int			first_bit;
 279	int			last_bit;
 280	int			next_bit;
 281	uint			nbits;
 282
 283	/* copy the flags across from the base format item */
 284	blfp->blf_flags = bip->__bli_format.blf_flags;
 285
 286	/*
 287	 * Base size is the actual size of the ondisk structure - it reflects
 288	 * the actual size of the dirty bitmap rather than the size of the in
 289	 * memory structure.
 290	 */
 291	base_size = xfs_buf_log_format_size(blfp);
 292
 293	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
 294	if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
 295		/*
 296		 * If the map is not be dirty in the transaction, mark
 297		 * the size as zero and do not advance the vector pointer.
 298		 */
 299		return;
 300	}
 301
 302	blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
 303	blfp->blf_size = 1;
 304
 305	if (bip->bli_flags & XFS_BLI_STALE) {
 306		/*
 307		 * The buffer is stale, so all we need to log
 308		 * is the buf log format structure with the
 309		 * cancel flag in it.
 310		 */
 311		trace_xfs_buf_item_format_stale(bip);
 312		ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
 
 313		return;
 314	}
 315
 316
 317	/*
 318	 * Fill in an iovec for each set of contiguous chunks.
 319	 */
 320	do {
 321		ASSERT(first_bit >= 0);
 322		nbits = xfs_contig_bits(blfp->blf_data_map,
 323					blfp->blf_map_size, first_bit);
 324		ASSERT(nbits > 0);
 325
 326		/*
 327		 * Straddling a page is rare because we don't log contiguous
 328		 * chunks of unmapped buffers anywhere.
 329		 */
 330		if (nbits > 1 &&
 331		    xfs_buf_item_straddle(bp, offset, first_bit, nbits))
 332			goto slow_scan;
 333
 334		xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 335					first_bit, nbits);
 336		blfp->blf_size++;
 337
 338		/*
 339		 * This takes the bit number to start looking from and
 340		 * returns the next set bit from there.  It returns -1
 341		 * if there are no more bits set or the start bit is
 342		 * beyond the end of the bitmap.
 343		 */
 344		first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 345					(uint)first_bit + nbits + 1);
 346	} while (first_bit != -1);
 347
 348	return;
 349
 350slow_scan:
 351	ASSERT(bp->b_addr == NULL);
 352	last_bit = first_bit;
 353	nbits = 1;
 354	for (;;) {
 355		/*
 356		 * This takes the bit number to start looking from and
 357		 * returns the next set bit from there.  It returns -1
 358		 * if there are no more bits set or the start bit is
 359		 * beyond the end of the bitmap.
 360		 */
 361		next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
 362					(uint)last_bit + 1);
 
 363		/*
 364		 * If we run out of bits fill in the last iovec and get out of
 365		 * the loop.  Else if we start a new set of bits then fill in
 366		 * the iovec for the series we were looking at and start
 367		 * counting the bits in the new one.  Else we're still in the
 368		 * same set of bits so just keep counting and scanning.
 
 
 369		 */
 370		if (next_bit == -1) {
 371			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 372						first_bit, nbits);
 373			blfp->blf_size++;
 
 
 374			break;
 375		} else if (next_bit != last_bit + 1 ||
 376		           xfs_buf_item_straddle(bp, offset, first_bit, nbits)) {
 377			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 378						first_bit, nbits);
 379			blfp->blf_size++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 380			first_bit = next_bit;
 381			last_bit = next_bit;
 382			nbits = 1;
 383		} else {
 384			last_bit++;
 385			nbits++;
 386		}
 387	}
 388}
 389
 390/*
 391 * This is called to fill in the vector of log iovecs for the
 392 * given log buf item.  It fills the first entry with a buf log
 393 * format structure, and the rest point to contiguous chunks
 394 * within the buffer.
 395 */
 396STATIC void
 397xfs_buf_item_format(
 398	struct xfs_log_item	*lip,
 399	struct xfs_log_vec	*lv)
 400{
 401	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 402	struct xfs_buf		*bp = bip->bli_buf;
 403	struct xfs_log_iovec	*vecp = NULL;
 404	uint			offset = 0;
 405	int			i;
 406
 407	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 408	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
 409	       (bip->bli_flags & XFS_BLI_STALE));
 410	ASSERT((bip->bli_flags & XFS_BLI_STALE) ||
 411	       (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF
 412	        && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF));
 413	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) ||
 414	       (bip->bli_flags & XFS_BLI_STALE));
 415
 416
 417	/*
 418	 * If it is an inode buffer, transfer the in-memory state to the
 419	 * format flags and clear the in-memory state.
 420	 *
 421	 * For buffer based inode allocation, we do not transfer
 422	 * this state if the inode buffer allocation has not yet been committed
 423	 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
 424	 * correct replay of the inode allocation.
 425	 *
 426	 * For icreate item based inode allocation, the buffers aren't written
 427	 * to the journal during allocation, and hence we should always tag the
 428	 * buffer as an inode buffer so that the correct unlinked list replay
 429	 * occurs during recovery.
 430	 */
 431	if (bip->bli_flags & XFS_BLI_INODE_BUF) {
 432		if (xfs_has_v3inodes(lip->li_log->l_mp) ||
 433		    !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
 434		      xfs_log_item_in_current_chkpt(lip)))
 435			bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
 436		bip->bli_flags &= ~XFS_BLI_INODE_BUF;
 437	}
 438
 439	for (i = 0; i < bip->bli_format_count; i++) {
 440		xfs_buf_item_format_segment(bip, lv, &vecp, offset,
 441					    &bip->bli_formats[i]);
 442		offset += BBTOB(bp->b_maps[i].bm_len);
 443	}
 444
 445	/*
 446	 * Check to make sure everything is consistent.
 447	 */
 448	trace_xfs_buf_item_format(bip);
 
 449}
 450
 451/*
 452 * This is called to pin the buffer associated with the buf log item in memory
 453 * so it cannot be written out.
 454 *
 455 * We take a reference to the buffer log item here so that the BLI life cycle
 456 * extends at least until the buffer is unpinned via xfs_buf_item_unpin() and
 457 * inserted into the AIL.
 458 *
 459 * We also need to take a reference to the buffer itself as the BLI unpin
 460 * processing requires accessing the buffer after the BLI has dropped the final
 461 * BLI reference. See xfs_buf_item_unpin() for an explanation.
 462 * If unpins race to drop the final BLI reference and only the
 463 * BLI owns a reference to the buffer, then the loser of the race can have the
 464 * buffer fgreed from under it (e.g. on shutdown). Taking a buffer reference per
 465 * pin count ensures the life cycle of the buffer extends for as
 466 * long as we hold the buffer pin reference in xfs_buf_item_unpin().
 467 */
 468STATIC void
 469xfs_buf_item_pin(
 470	struct xfs_log_item	*lip)
 471{
 472	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 473
 474	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 475	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
 476	       (bip->bli_flags & XFS_BLI_ORDERED) ||
 477	       (bip->bli_flags & XFS_BLI_STALE));
 478
 479	trace_xfs_buf_item_pin(bip);
 480
 481	xfs_buf_hold(bip->bli_buf);
 482	atomic_inc(&bip->bli_refcount);
 483	atomic_inc(&bip->bli_buf->b_pin_count);
 484}
 485
 486/*
 487 * This is called to unpin the buffer associated with the buf log item which was
 488 * previously pinned with a call to xfs_buf_item_pin().  We enter this function
 489 * with a buffer pin count, a buffer reference and a BLI reference.
 490 *
 491 * We must drop the BLI reference before we unpin the buffer because the AIL
 492 * doesn't acquire a BLI reference whenever it accesses it. Therefore if the
 493 * refcount drops to zero, the bli could still be AIL resident and the buffer
 494 * submitted for I/O at any point before we return. This can result in IO
 495 * completion freeing the buffer while we are still trying to access it here.
 496 * This race condition can also occur in shutdown situations where we abort and
 497 * unpin buffers from contexts other that journal IO completion.
 498 *
 499 * Hence we have to hold a buffer reference per pin count to ensure that the
 500 * buffer cannot be freed until we have finished processing the unpin operation.
 501 * The reference is taken in xfs_buf_item_pin(), and we must hold it until we
 502 * are done processing the buffer state. In the case of an abort (remove =
 503 * true) then we re-use the current pin reference as the IO reference we hand
 504 * off to IO failure handling.
 505 */
 506STATIC void
 507xfs_buf_item_unpin(
 508	struct xfs_log_item	*lip,
 509	int			remove)
 510{
 511	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 512	struct xfs_buf		*bp = bip->bli_buf;
 513	int			stale = bip->bli_flags & XFS_BLI_STALE;
 514	int			freed;
 
 515
 516	ASSERT(bp->b_log_item == bip);
 517	ASSERT(atomic_read(&bip->bli_refcount) > 0);
 518
 519	trace_xfs_buf_item_unpin(bip);
 520
 521	freed = atomic_dec_and_test(&bip->bli_refcount);
 
 522	if (atomic_dec_and_test(&bp->b_pin_count))
 523		wake_up_all(&bp->b_waiters);
 524
 525	/*
 526	 * Nothing to do but drop the buffer pin reference if the BLI is
 527	 * still active.
 528	 */
 529	if (!freed) {
 530		xfs_buf_rele(bp);
 531		return;
 532	}
 533
 534	if (stale) {
 535		ASSERT(bip->bli_flags & XFS_BLI_STALE);
 536		ASSERT(xfs_buf_islocked(bp));
 537		ASSERT(bp->b_flags & XBF_STALE);
 538		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
 539		ASSERT(list_empty(&lip->li_trans));
 540		ASSERT(!bp->b_transp);
 541
 542		trace_xfs_buf_item_unpin_stale(bip);
 543
 544		/*
 545		 * The buffer has been locked and referenced since it was marked
 546		 * stale so we own both lock and reference exclusively here. We
 547		 * do not need the pin reference any more, so drop it now so
 548		 * that we only have one reference to drop once item completion
 549		 * processing is complete.
 550		 */
 551		xfs_buf_rele(bp);
 
 
 
 
 
 
 
 
 
 
 552
 553		/*
 554		 * If we get called here because of an IO error, we may or may
 555		 * not have the item on the AIL. xfs_trans_ail_delete() will
 556		 * take care of that situation. xfs_trans_ail_delete() drops
 557		 * the AIL lock.
 558		 */
 559		if (bip->bli_flags & XFS_BLI_STALE_INODE) {
 560			xfs_buf_item_done(bp);
 561			xfs_buf_inode_iodone(bp);
 562			ASSERT(list_empty(&bp->b_li_list));
 563		} else {
 564			xfs_trans_ail_delete(lip, SHUTDOWN_LOG_IO_ERROR);
 
 565			xfs_buf_item_relse(bp);
 566			ASSERT(bp->b_log_item == NULL);
 567		}
 568		xfs_buf_relse(bp);
 569		return;
 570	}
 571
 572	if (remove) {
 573		/*
 574		 * We need to simulate an async IO failures here to ensure that
 575		 * the correct error completion is run on this buffer. This
 576		 * requires a reference to the buffer and for the buffer to be
 577		 * locked. We can safely pass ownership of the pin reference to
 578		 * the IO to ensure that nothing can free the buffer while we
 579		 * wait for the lock and then run the IO failure completion.
 580		 */
 581		xfs_buf_lock(bp);
 582		bp->b_flags |= XBF_ASYNC;
 583		xfs_buf_ioend_fail(bp);
 584		return;
 585	}
 586
 587	/*
 588	 * BLI has no more active references - it will be moved to the AIL to
 589	 * manage the remaining BLI/buffer life cycle. There is nothing left for
 590	 * us to do here so drop the pin reference to the buffer.
 591	 */
 592	xfs_buf_rele(bp);
 593}
 594
 
 
 
 
 
 
 
 
 595STATIC uint
 596xfs_buf_item_push(
 597	struct xfs_log_item	*lip,
 598	struct list_head	*buffer_list)
 599{
 600	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 601	struct xfs_buf		*bp = bip->bli_buf;
 602	uint			rval = XFS_ITEM_SUCCESS;
 603
 604	if (xfs_buf_ispinned(bp))
 605		return XFS_ITEM_PINNED;
 606	if (!xfs_buf_trylock(bp)) {
 607		/*
 608		 * If we have just raced with a buffer being pinned and it has
 609		 * been marked stale, we could end up stalling until someone else
 610		 * issues a log force to unpin the stale buffer. Check for the
 611		 * race condition here so xfsaild recognizes the buffer is pinned
 612		 * and queues a log force to move it along.
 613		 */
 614		if (xfs_buf_ispinned(bp))
 615			return XFS_ITEM_PINNED;
 616		return XFS_ITEM_LOCKED;
 617	}
 
 
 618
 619	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 620
 621	trace_xfs_buf_item_push(bip);
 622
 623	/* has a previous flush failed due to IO errors? */
 624	if (bp->b_flags & XBF_WRITE_FAIL) {
 625		xfs_buf_alert_ratelimited(bp, "XFS: Failing async write",
 626	    "Failing async write on buffer block 0x%llx. Retrying async write.",
 627					  (long long)xfs_buf_daddr(bp));
 628	}
 629
 630	if (!xfs_buf_delwri_queue(bp, buffer_list))
 631		rval = XFS_ITEM_FLUSHING;
 632	xfs_buf_unlock(bp);
 633	return rval;
 634}
 635
 636/*
 637 * Drop the buffer log item refcount and take appropriate action. This helper
 638 * determines whether the bli must be freed or not, since a decrement to zero
 639 * does not necessarily mean the bli is unused.
 640 *
 641 * Return true if the bli is freed, false otherwise.
 642 */
 643bool
 644xfs_buf_item_put(
 645	struct xfs_buf_log_item	*bip)
 646{
 647	struct xfs_log_item	*lip = &bip->bli_item;
 648	bool			aborted;
 649	bool			dirty;
 650
 651	/* drop the bli ref and return if it wasn't the last one */
 652	if (!atomic_dec_and_test(&bip->bli_refcount))
 653		return false;
 654
 655	/*
 656	 * We dropped the last ref and must free the item if clean or aborted.
 657	 * If the bli is dirty and non-aborted, the buffer was clean in the
 658	 * transaction but still awaiting writeback from previous changes. In
 659	 * that case, the bli is freed on buffer writeback completion.
 660	 */
 661	aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) ||
 662			xlog_is_shutdown(lip->li_log);
 663	dirty = bip->bli_flags & XFS_BLI_DIRTY;
 664	if (dirty && !aborted)
 665		return false;
 666
 667	/*
 668	 * The bli is aborted or clean. An aborted item may be in the AIL
 669	 * regardless of dirty state.  For example, consider an aborted
 670	 * transaction that invalidated a dirty bli and cleared the dirty
 671	 * state.
 672	 */
 673	if (aborted)
 674		xfs_trans_ail_delete(lip, 0);
 675	xfs_buf_item_relse(bip->bli_buf);
 676	return true;
 677}
 678
 679/*
 680 * Release the buffer associated with the buf log item.  If there is no dirty
 681 * logged data associated with the buffer recorded in the buf log item, then
 682 * free the buf log item and remove the reference to it in the buffer.
 683 *
 684 * This call ignores the recursion count.  It is only called when the buffer
 685 * should REALLY be unlocked, regardless of the recursion count.
 686 *
 687 * We unconditionally drop the transaction's reference to the log item. If the
 688 * item was logged, then another reference was taken when it was pinned, so we
 689 * can safely drop the transaction reference now.  This also allows us to avoid
 690 * potential races with the unpin code freeing the bli by not referencing the
 691 * bli after we've dropped the reference count.
 692 *
 693 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
 694 * if necessary but do not unlock the buffer.  This is for support of
 695 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
 696 * free the item.
 697 */
 698STATIC void
 699xfs_buf_item_release(
 700	struct xfs_log_item	*lip)
 701{
 702	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 703	struct xfs_buf		*bp = bip->bli_buf;
 704	bool			released;
 705	bool			hold = bip->bli_flags & XFS_BLI_HOLD;
 706	bool			stale = bip->bli_flags & XFS_BLI_STALE;
 707#if defined(DEBUG) || defined(XFS_WARN)
 708	bool			ordered = bip->bli_flags & XFS_BLI_ORDERED;
 709	bool			dirty = bip->bli_flags & XFS_BLI_DIRTY;
 710	bool			aborted = test_bit(XFS_LI_ABORTED,
 711						   &lip->li_flags);
 712#endif
 713
 714	trace_xfs_buf_item_release(bip);
 
 
 
 
 
 
 715
 716	/*
 717	 * The bli dirty state should match whether the blf has logged segments
 718	 * except for ordered buffers, where only the bli should be dirty.
 719	 */
 720	ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) ||
 721	       (ordered && dirty && !xfs_buf_item_dirty_format(bip)));
 722	ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
 
 723
 724	/*
 725	 * Clear the buffer's association with this transaction and
 726	 * per-transaction state from the bli, which has been copied above.
 
 727	 */
 728	bp->b_transp = NULL;
 729	bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
 
 
 
 
 
 
 
 
 730
 731	/*
 732	 * Unref the item and unlock the buffer unless held or stale. Stale
 733	 * buffers remain locked until final unpin unless the bli is freed by
 734	 * the unref call. The latter implies shutdown because buffer
 735	 * invalidation dirties the bli and transaction.
 736	 */
 737	released = xfs_buf_item_put(bip);
 738	if (hold || (stale && !released))
 739		return;
 740	ASSERT(!stale || aborted);
 741	xfs_buf_relse(bp);
 742}
 743
 744STATIC void
 745xfs_buf_item_committing(
 746	struct xfs_log_item	*lip,
 747	xfs_csn_t		seq)
 748{
 749	return xfs_buf_item_release(lip);
 750}
 751
 752/*
 753 * This is called to find out where the oldest active copy of the
 754 * buf log item in the on disk log resides now that the last log
 755 * write of it completed at the given lsn.
 756 * We always re-log all the dirty data in a buffer, so usually the
 757 * latest copy in the on disk log is the only one that matters.  For
 758 * those cases we simply return the given lsn.
 759 *
 760 * The one exception to this is for buffers full of newly allocated
 761 * inodes.  These buffers are only relogged with the XFS_BLI_INODE_BUF
 762 * flag set, indicating that only the di_next_unlinked fields from the
 763 * inodes in the buffers will be replayed during recovery.  If the
 764 * original newly allocated inode images have not yet been flushed
 765 * when the buffer is so relogged, then we need to make sure that we
 766 * keep the old images in the 'active' portion of the log.  We do this
 767 * by returning the original lsn of that transaction here rather than
 768 * the current one.
 769 */
 770STATIC xfs_lsn_t
 771xfs_buf_item_committed(
 772	struct xfs_log_item	*lip,
 773	xfs_lsn_t		lsn)
 774{
 775	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 776
 777	trace_xfs_buf_item_committed(bip);
 778
 779	if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
 780		return lip->li_lsn;
 781	return lsn;
 782}
 783
 784static const struct xfs_item_ops xfs_buf_item_ops = {
 785	.iop_size	= xfs_buf_item_size,
 786	.iop_format	= xfs_buf_item_format,
 787	.iop_pin	= xfs_buf_item_pin,
 788	.iop_unpin	= xfs_buf_item_unpin,
 789	.iop_release	= xfs_buf_item_release,
 790	.iop_committing	= xfs_buf_item_committing,
 791	.iop_committed	= xfs_buf_item_committed,
 792	.iop_push	= xfs_buf_item_push,
 793};
 
 
 
 
 
 
 
 
 
 794
 795STATIC void
 796xfs_buf_item_get_format(
 797	struct xfs_buf_log_item	*bip,
 798	int			count)
 
 
 
 
 
 799{
 800	ASSERT(bip->bli_formats == NULL);
 801	bip->bli_format_count = count;
 802
 803	if (count == 1) {
 804		bip->bli_formats = &bip->__bli_format;
 805		return;
 806	}
 807
 808	bip->bli_formats = kzalloc(count * sizeof(struct xfs_buf_log_format),
 809				GFP_KERNEL | __GFP_NOFAIL);
 
 810}
 811
 812STATIC void
 813xfs_buf_item_free_format(
 814	struct xfs_buf_log_item	*bip)
 
 815{
 816	if (bip->bli_formats != &bip->__bli_format) {
 817		kfree(bip->bli_formats);
 818		bip->bli_formats = NULL;
 819	}
 820}
 821
 822/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 823 * Allocate a new buf log item to go with the given buffer.
 824 * Set the buffer's b_log_item field to point to the new
 825 * buf log item.
 
 
 826 */
 827int
 828xfs_buf_item_init(
 829	struct xfs_buf	*bp,
 830	struct xfs_mount *mp)
 831{
 832	struct xfs_buf_log_item	*bip = bp->b_log_item;
 
 833	int			chunks;
 834	int			map_size;
 835	int			i;
 836
 837	/*
 838	 * Check to see if there is already a buf log item for
 839	 * this buffer. If we do already have one, there is
 
 840	 * nothing to do here so return.
 841	 */
 842	ASSERT(bp->b_mount == mp);
 843	if (bip) {
 844		ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
 845		ASSERT(!bp->b_transp);
 846		ASSERT(bip->bli_buf == bp);
 847		return 0;
 848	}
 
 
 
 
 
 849
 850	bip = kmem_cache_zalloc(xfs_buf_item_cache, GFP_KERNEL | __GFP_NOFAIL);
 
 851	xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
 852	bip->bli_buf = bp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 853
 854	/*
 855	 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
 856	 * can be divided into. Make sure not to truncate any pieces.
 857	 * map_size is the size of the bitmap needed to describe the
 858	 * chunks of the buffer.
 859	 *
 860	 * Discontiguous buffer support follows the layout of the underlying
 861	 * buffer. This makes the implementation as simple as possible.
 862	 */
 863	xfs_buf_item_get_format(bip, bp->b_map_count);
 864
 865	for (i = 0; i < bip->bli_format_count; i++) {
 866		chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
 867				      XFS_BLF_CHUNK);
 868		map_size = DIV_ROUND_UP(chunks, NBWORD);
 869
 870		if (map_size > XFS_BLF_DATAMAP_SIZE) {
 871			kmem_cache_free(xfs_buf_item_cache, bip);
 872			xfs_err(mp,
 873	"buffer item dirty bitmap (%u uints) too small to reflect %u bytes!",
 874					map_size,
 875					BBTOB(bp->b_maps[i].bm_len));
 876			return -EFSCORRUPTED;
 877		}
 878
 879		bip->bli_formats[i].blf_type = XFS_LI_BUF;
 880		bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
 881		bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
 882		bip->bli_formats[i].blf_map_size = map_size;
 883	}
 884
 885	bp->b_log_item = bip;
 886	xfs_buf_hold(bp);
 887	return 0;
 888}
 889
 890
 891/*
 892 * Mark bytes first through last inclusive as dirty in the buf
 893 * item's bitmap.
 894 */
 895static void
 896xfs_buf_item_log_segment(
 
 897	uint			first,
 898	uint			last,
 899	uint			*map)
 900{
 901	uint		first_bit;
 902	uint		last_bit;
 903	uint		bits_to_set;
 904	uint		bits_set;
 905	uint		word_num;
 906	uint		*wordp;
 907	uint		bit;
 908	uint		end_bit;
 909	uint		mask;
 910
 911	ASSERT(first < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD);
 912	ASSERT(last < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD);
 
 
 
 913
 914	/*
 915	 * Convert byte offsets to bit numbers.
 916	 */
 917	first_bit = first >> XFS_BLF_SHIFT;
 918	last_bit = last >> XFS_BLF_SHIFT;
 919
 920	/*
 921	 * Calculate the total number of bits to be set.
 922	 */
 923	bits_to_set = last_bit - first_bit + 1;
 924
 925	/*
 926	 * Get a pointer to the first word in the bitmap
 927	 * to set a bit in.
 928	 */
 929	word_num = first_bit >> BIT_TO_WORD_SHIFT;
 930	wordp = &map[word_num];
 931
 932	/*
 933	 * Calculate the starting bit in the first word.
 934	 */
 935	bit = first_bit & (uint)(NBWORD - 1);
 936
 937	/*
 938	 * First set any bits in the first word of our range.
 939	 * If it starts at bit 0 of the word, it will be
 940	 * set below rather than here.  That is what the variable
 941	 * bit tells us. The variable bits_set tracks the number
 942	 * of bits that have been set so far.  End_bit is the number
 943	 * of the last bit to be set in this word plus one.
 944	 */
 945	if (bit) {
 946		end_bit = min(bit + bits_to_set, (uint)NBWORD);
 947		mask = ((1U << (end_bit - bit)) - 1) << bit;
 948		*wordp |= mask;
 949		wordp++;
 950		bits_set = end_bit - bit;
 951	} else {
 952		bits_set = 0;
 953	}
 954
 955	/*
 956	 * Now set bits a whole word at a time that are between
 957	 * first_bit and last_bit.
 958	 */
 959	while ((bits_to_set - bits_set) >= NBWORD) {
 960		*wordp = 0xffffffff;
 961		bits_set += NBWORD;
 962		wordp++;
 963	}
 964
 965	/*
 966	 * Finally, set any bits left to be set in one last partial word.
 967	 */
 968	end_bit = bits_to_set - bits_set;
 969	if (end_bit) {
 970		mask = (1U << end_bit) - 1;
 971		*wordp |= mask;
 972	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 973}
 974
 975/*
 976 * Mark bytes first through last inclusive as dirty in the buf
 977 * item's bitmap.
 
 
 
 978 */
 979void
 980xfs_buf_item_log(
 981	struct xfs_buf_log_item	*bip,
 982	uint			first,
 983	uint			last)
 984{
 985	int			i;
 986	uint			start;
 987	uint			end;
 988	struct xfs_buf		*bp = bip->bli_buf;
 989
 990	/*
 991	 * walk each buffer segment and mark them dirty appropriately.
 992	 */
 993	start = 0;
 994	for (i = 0; i < bip->bli_format_count; i++) {
 995		if (start > last)
 996			break;
 997		end = start + BBTOB(bp->b_maps[i].bm_len) - 1;
 998
 999		/* skip to the map that includes the first byte to log */
1000		if (first > end) {
1001			start += BBTOB(bp->b_maps[i].bm_len);
1002			continue;
1003		}
1004
1005		/*
1006		 * Trim the range to this segment and mark it in the bitmap.
1007		 * Note that we must convert buffer offsets to segment relative
1008		 * offsets (e.g., the first byte of each segment is byte 0 of
1009		 * that segment).
1010		 */
1011		if (first < start)
1012			first = start;
1013		if (end > last)
1014			end = last;
1015		xfs_buf_item_log_segment(first - start, end - start,
1016					 &bip->bli_formats[i].blf_data_map[0]);
1017
1018		start += BBTOB(bp->b_maps[i].bm_len);
1019	}
1020}
1021
1022
1023/*
1024 * Return true if the buffer has any ranges logged/dirtied by a transaction,
1025 * false otherwise.
 
 
 
 
 
1026 */
1027bool
1028xfs_buf_item_dirty_format(
1029	struct xfs_buf_log_item	*bip)
1030{
1031	int			i;
1032
1033	for (i = 0; i < bip->bli_format_count; i++) {
1034		if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
1035			     bip->bli_formats[i].blf_map_size))
1036			return true;
 
 
 
 
 
 
 
1037	}
1038
1039	return false;
 
 
1040}
1041
 
 
 
 
 
 
 
 
 
 
 
 
1042STATIC void
1043xfs_buf_item_free(
1044	struct xfs_buf_log_item	*bip)
1045{
1046	xfs_buf_item_free_format(bip);
1047	kvfree(bip->bli_item.li_lv_shadow);
1048	kmem_cache_free(xfs_buf_item_cache, bip);
 
 
 
 
 
 
 
 
 
 
 
1049}
1050
1051/*
1052 * xfs_buf_item_relse() is called when the buf log item is no longer needed.
 
 
 
 
1053 */
1054void
1055xfs_buf_item_relse(
1056	struct xfs_buf	*bp)
1057{
1058	struct xfs_buf_log_item	*bip = bp->b_log_item;
 
 
 
 
 
 
1059
1060	trace_xfs_buf_item_relse(bp, _RET_IP_);
1061	ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1062
1063	if (atomic_read(&bip->bli_refcount))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1064		return;
1065	bp->b_log_item = NULL;
1066	xfs_buf_rele(bp);
1067	xfs_buf_item_free(bip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1068}
1069
 
 
 
 
 
 
 
1070void
1071xfs_buf_item_done(
1072	struct xfs_buf		*bp)
 
1073{
 
 
 
 
 
 
1074	/*
1075	 * If we are forcibly shutting down, this may well be off the AIL
1076	 * already. That's because we simulate the log-committed callbacks to
1077	 * unpin these buffers. Or we may never have put this item on AIL
1078	 * because of the transaction was aborted forcibly.
1079	 * xfs_trans_ail_delete() takes care of these.
1080	 *
1081	 * Either way, AIL is useless if we're forcing a shutdown.
1082	 *
1083	 * Note that log recovery writes might have buffer items that are not on
1084	 * the AIL even when the file system is not shut down.
1085	 */
1086	xfs_trans_ail_delete(&bp->b_log_item->bli_item,
1087			     (bp->b_flags & _XBF_LOGRECOVERY) ? 0 :
1088			     SHUTDOWN_CORRUPT_INCORE);
1089	xfs_buf_item_relse(bp);
1090}