xref: /openbmc/linux/fs/xfs/xfs_trans_buf.c (revision 293d5b43)
1 /*
2  * Copyright (c) 2000-2002,2005 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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_inode.h"
26 #include "xfs_trans.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_error.h"
30 #include "xfs_trace.h"
31 
32 /*
33  * Check to see if a buffer matching the given parameters is already
34  * a part of the given transaction.
35  */
36 STATIC struct xfs_buf *
37 xfs_trans_buf_item_match(
38 	struct xfs_trans	*tp,
39 	struct xfs_buftarg	*target,
40 	struct xfs_buf_map	*map,
41 	int			nmaps)
42 {
43 	struct xfs_log_item_desc *lidp;
44 	struct xfs_buf_log_item	*blip;
45 	int			len = 0;
46 	int			i;
47 
48 	for (i = 0; i < nmaps; i++)
49 		len += map[i].bm_len;
50 
51 	list_for_each_entry(lidp, &tp->t_items, lid_trans) {
52 		blip = (struct xfs_buf_log_item *)lidp->lid_item;
53 		if (blip->bli_item.li_type == XFS_LI_BUF &&
54 		    blip->bli_buf->b_target == target &&
55 		    XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
56 		    blip->bli_buf->b_length == len) {
57 			ASSERT(blip->bli_buf->b_map_count == nmaps);
58 			return blip->bli_buf;
59 		}
60 	}
61 
62 	return NULL;
63 }
64 
65 /*
66  * Add the locked buffer to the transaction.
67  *
68  * The buffer must be locked, and it cannot be associated with any
69  * transaction.
70  *
71  * If the buffer does not yet have a buf log item associated with it,
72  * then allocate one for it.  Then add the buf item to the transaction.
73  */
74 STATIC void
75 _xfs_trans_bjoin(
76 	struct xfs_trans	*tp,
77 	struct xfs_buf		*bp,
78 	int			reset_recur)
79 {
80 	struct xfs_buf_log_item	*bip;
81 
82 	ASSERT(bp->b_transp == NULL);
83 
84 	/*
85 	 * The xfs_buf_log_item pointer is stored in b_fsprivate.  If
86 	 * it doesn't have one yet, then allocate one and initialize it.
87 	 * The checks to see if one is there are in xfs_buf_item_init().
88 	 */
89 	xfs_buf_item_init(bp, tp->t_mountp);
90 	bip = bp->b_fspriv;
91 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
92 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
93 	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
94 	if (reset_recur)
95 		bip->bli_recur = 0;
96 
97 	/*
98 	 * Take a reference for this transaction on the buf item.
99 	 */
100 	atomic_inc(&bip->bli_refcount);
101 
102 	/*
103 	 * Get a log_item_desc to point at the new item.
104 	 */
105 	xfs_trans_add_item(tp, &bip->bli_item);
106 
107 	/*
108 	 * Initialize b_fsprivate2 so we can find it with incore_match()
109 	 * in xfs_trans_get_buf() and friends above.
110 	 */
111 	bp->b_transp = tp;
112 
113 }
114 
115 void
116 xfs_trans_bjoin(
117 	struct xfs_trans	*tp,
118 	struct xfs_buf		*bp)
119 {
120 	_xfs_trans_bjoin(tp, bp, 0);
121 	trace_xfs_trans_bjoin(bp->b_fspriv);
122 }
123 
124 /*
125  * Get and lock the buffer for the caller if it is not already
126  * locked within the given transaction.  If it is already locked
127  * within the transaction, just increment its lock recursion count
128  * and return a pointer to it.
129  *
130  * If the transaction pointer is NULL, make this just a normal
131  * get_buf() call.
132  */
133 struct xfs_buf *
134 xfs_trans_get_buf_map(
135 	struct xfs_trans	*tp,
136 	struct xfs_buftarg	*target,
137 	struct xfs_buf_map	*map,
138 	int			nmaps,
139 	xfs_buf_flags_t		flags)
140 {
141 	xfs_buf_t		*bp;
142 	xfs_buf_log_item_t	*bip;
143 
144 	if (!tp)
145 		return xfs_buf_get_map(target, map, nmaps, flags);
146 
147 	/*
148 	 * If we find the buffer in the cache with this transaction
149 	 * pointer in its b_fsprivate2 field, then we know we already
150 	 * have it locked.  In this case we just increment the lock
151 	 * recursion count and return the buffer to the caller.
152 	 */
153 	bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
154 	if (bp != NULL) {
155 		ASSERT(xfs_buf_islocked(bp));
156 		if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
157 			xfs_buf_stale(bp);
158 			bp->b_flags |= XBF_DONE;
159 		}
160 
161 		ASSERT(bp->b_transp == tp);
162 		bip = bp->b_fspriv;
163 		ASSERT(bip != NULL);
164 		ASSERT(atomic_read(&bip->bli_refcount) > 0);
165 		bip->bli_recur++;
166 		trace_xfs_trans_get_buf_recur(bip);
167 		return bp;
168 	}
169 
170 	bp = xfs_buf_get_map(target, map, nmaps, flags);
171 	if (bp == NULL) {
172 		return NULL;
173 	}
174 
175 	ASSERT(!bp->b_error);
176 
177 	_xfs_trans_bjoin(tp, bp, 1);
178 	trace_xfs_trans_get_buf(bp->b_fspriv);
179 	return bp;
180 }
181 
182 /*
183  * Get and lock the superblock buffer of this file system for the
184  * given transaction.
185  *
186  * We don't need to use incore_match() here, because the superblock
187  * buffer is a private buffer which we keep a pointer to in the
188  * mount structure.
189  */
190 xfs_buf_t *
191 xfs_trans_getsb(xfs_trans_t	*tp,
192 		struct xfs_mount *mp,
193 		int		flags)
194 {
195 	xfs_buf_t		*bp;
196 	xfs_buf_log_item_t	*bip;
197 
198 	/*
199 	 * Default to just trying to lock the superblock buffer
200 	 * if tp is NULL.
201 	 */
202 	if (tp == NULL)
203 		return xfs_getsb(mp, flags);
204 
205 	/*
206 	 * If the superblock buffer already has this transaction
207 	 * pointer in its b_fsprivate2 field, then we know we already
208 	 * have it locked.  In this case we just increment the lock
209 	 * recursion count and return the buffer to the caller.
210 	 */
211 	bp = mp->m_sb_bp;
212 	if (bp->b_transp == tp) {
213 		bip = bp->b_fspriv;
214 		ASSERT(bip != NULL);
215 		ASSERT(atomic_read(&bip->bli_refcount) > 0);
216 		bip->bli_recur++;
217 		trace_xfs_trans_getsb_recur(bip);
218 		return bp;
219 	}
220 
221 	bp = xfs_getsb(mp, flags);
222 	if (bp == NULL)
223 		return NULL;
224 
225 	_xfs_trans_bjoin(tp, bp, 1);
226 	trace_xfs_trans_getsb(bp->b_fspriv);
227 	return bp;
228 }
229 
230 /*
231  * Get and lock the buffer for the caller if it is not already
232  * locked within the given transaction.  If it has not yet been
233  * read in, read it from disk. If it is already locked
234  * within the transaction and already read in, just increment its
235  * lock recursion count and return a pointer to it.
236  *
237  * If the transaction pointer is NULL, make this just a normal
238  * read_buf() call.
239  */
240 int
241 xfs_trans_read_buf_map(
242 	struct xfs_mount	*mp,
243 	struct xfs_trans	*tp,
244 	struct xfs_buftarg	*target,
245 	struct xfs_buf_map	*map,
246 	int			nmaps,
247 	xfs_buf_flags_t		flags,
248 	struct xfs_buf		**bpp,
249 	const struct xfs_buf_ops *ops)
250 {
251 	struct xfs_buf		*bp = NULL;
252 	struct xfs_buf_log_item	*bip;
253 	int			error;
254 
255 	*bpp = NULL;
256 	/*
257 	 * If we find the buffer in the cache with this transaction
258 	 * pointer in its b_fsprivate2 field, then we know we already
259 	 * have it locked.  If it is already read in we just increment
260 	 * the lock recursion count and return the buffer to the caller.
261 	 * If the buffer is not yet read in, then we read it in, increment
262 	 * the lock recursion count, and return it to the caller.
263 	 */
264 	if (tp)
265 		bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
266 	if (bp) {
267 		ASSERT(xfs_buf_islocked(bp));
268 		ASSERT(bp->b_transp == tp);
269 		ASSERT(bp->b_fspriv != NULL);
270 		ASSERT(!bp->b_error);
271 		ASSERT(bp->b_flags & XBF_DONE);
272 
273 		/*
274 		 * We never locked this buf ourselves, so we shouldn't
275 		 * brelse it either. Just get out.
276 		 */
277 		if (XFS_FORCED_SHUTDOWN(mp)) {
278 			trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
279 			return -EIO;
280 		}
281 
282 		bip = bp->b_fspriv;
283 		bip->bli_recur++;
284 
285 		ASSERT(atomic_read(&bip->bli_refcount) > 0);
286 		trace_xfs_trans_read_buf_recur(bip);
287 		*bpp = bp;
288 		return 0;
289 	}
290 
291 	bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
292 	if (!bp) {
293 		if (!(flags & XBF_TRYLOCK))
294 			return -ENOMEM;
295 		return tp ? 0 : -EAGAIN;
296 	}
297 
298 	/*
299 	 * If we've had a read error, then the contents of the buffer are
300 	 * invalid and should not be used. To ensure that a followup read tries
301 	 * to pull the buffer from disk again, we clear the XBF_DONE flag and
302 	 * mark the buffer stale. This ensures that anyone who has a current
303 	 * reference to the buffer will interpret it's contents correctly and
304 	 * future cache lookups will also treat it as an empty, uninitialised
305 	 * buffer.
306 	 */
307 	if (bp->b_error) {
308 		error = bp->b_error;
309 		if (!XFS_FORCED_SHUTDOWN(mp))
310 			xfs_buf_ioerror_alert(bp, __func__);
311 		bp->b_flags &= ~XBF_DONE;
312 		xfs_buf_stale(bp);
313 
314 		if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
315 			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
316 		xfs_buf_relse(bp);
317 
318 		/* bad CRC means corrupted metadata */
319 		if (error == -EFSBADCRC)
320 			error = -EFSCORRUPTED;
321 		return error;
322 	}
323 
324 	if (XFS_FORCED_SHUTDOWN(mp)) {
325 		xfs_buf_relse(bp);
326 		trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
327 		return -EIO;
328 	}
329 
330 	if (tp) {
331 		_xfs_trans_bjoin(tp, bp, 1);
332 		trace_xfs_trans_read_buf(bp->b_fspriv);
333 	}
334 	*bpp = bp;
335 	return 0;
336 
337 }
338 
339 /*
340  * Release the buffer bp which was previously acquired with one of the
341  * xfs_trans_... buffer allocation routines if the buffer has not
342  * been modified within this transaction.  If the buffer is modified
343  * within this transaction, do decrement the recursion count but do
344  * not release the buffer even if the count goes to 0.  If the buffer is not
345  * modified within the transaction, decrement the recursion count and
346  * release the buffer if the recursion count goes to 0.
347  *
348  * If the buffer is to be released and it was not modified before
349  * this transaction began, then free the buf_log_item associated with it.
350  *
351  * If the transaction pointer is NULL, make this just a normal
352  * brelse() call.
353  */
354 void
355 xfs_trans_brelse(xfs_trans_t	*tp,
356 		 xfs_buf_t	*bp)
357 {
358 	xfs_buf_log_item_t	*bip;
359 
360 	/*
361 	 * Default to a normal brelse() call if the tp is NULL.
362 	 */
363 	if (tp == NULL) {
364 		ASSERT(bp->b_transp == NULL);
365 		xfs_buf_relse(bp);
366 		return;
367 	}
368 
369 	ASSERT(bp->b_transp == tp);
370 	bip = bp->b_fspriv;
371 	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
372 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
373 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
374 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
375 
376 	trace_xfs_trans_brelse(bip);
377 
378 	/*
379 	 * If the release is just for a recursive lock,
380 	 * then decrement the count and return.
381 	 */
382 	if (bip->bli_recur > 0) {
383 		bip->bli_recur--;
384 		return;
385 	}
386 
387 	/*
388 	 * If the buffer is dirty within this transaction, we can't
389 	 * release it until we commit.
390 	 */
391 	if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
392 		return;
393 
394 	/*
395 	 * If the buffer has been invalidated, then we can't release
396 	 * it until the transaction commits to disk unless it is re-dirtied
397 	 * as part of this transaction.  This prevents us from pulling
398 	 * the item from the AIL before we should.
399 	 */
400 	if (bip->bli_flags & XFS_BLI_STALE)
401 		return;
402 
403 	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
404 
405 	/*
406 	 * Free up the log item descriptor tracking the released item.
407 	 */
408 	xfs_trans_del_item(&bip->bli_item);
409 
410 	/*
411 	 * Clear the hold flag in the buf log item if it is set.
412 	 * We wouldn't want the next user of the buffer to
413 	 * get confused.
414 	 */
415 	if (bip->bli_flags & XFS_BLI_HOLD) {
416 		bip->bli_flags &= ~XFS_BLI_HOLD;
417 	}
418 
419 	/*
420 	 * Drop our reference to the buf log item.
421 	 */
422 	atomic_dec(&bip->bli_refcount);
423 
424 	/*
425 	 * If the buf item is not tracking data in the log, then
426 	 * we must free it before releasing the buffer back to the
427 	 * free pool.  Before releasing the buffer to the free pool,
428 	 * clear the transaction pointer in b_fsprivate2 to dissolve
429 	 * its relation to this transaction.
430 	 */
431 	if (!xfs_buf_item_dirty(bip)) {
432 /***
433 		ASSERT(bp->b_pincount == 0);
434 ***/
435 		ASSERT(atomic_read(&bip->bli_refcount) == 0);
436 		ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
437 		ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
438 		xfs_buf_item_relse(bp);
439 	}
440 
441 	bp->b_transp = NULL;
442 	xfs_buf_relse(bp);
443 }
444 
445 /*
446  * Mark the buffer as not needing to be unlocked when the buf item's
447  * iop_unlock() routine is called.  The buffer must already be locked
448  * and associated with the given transaction.
449  */
450 /* ARGSUSED */
451 void
452 xfs_trans_bhold(xfs_trans_t	*tp,
453 		xfs_buf_t	*bp)
454 {
455 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
456 
457 	ASSERT(bp->b_transp == tp);
458 	ASSERT(bip != NULL);
459 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
460 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
461 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
462 
463 	bip->bli_flags |= XFS_BLI_HOLD;
464 	trace_xfs_trans_bhold(bip);
465 }
466 
467 /*
468  * Cancel the previous buffer hold request made on this buffer
469  * for this transaction.
470  */
471 void
472 xfs_trans_bhold_release(xfs_trans_t	*tp,
473 			xfs_buf_t	*bp)
474 {
475 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
476 
477 	ASSERT(bp->b_transp == tp);
478 	ASSERT(bip != NULL);
479 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
480 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
481 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
482 	ASSERT(bip->bli_flags & XFS_BLI_HOLD);
483 
484 	bip->bli_flags &= ~XFS_BLI_HOLD;
485 	trace_xfs_trans_bhold_release(bip);
486 }
487 
488 /*
489  * This is called to mark bytes first through last inclusive of the given
490  * buffer as needing to be logged when the transaction is committed.
491  * The buffer must already be associated with the given transaction.
492  *
493  * First and last are numbers relative to the beginning of this buffer,
494  * so the first byte in the buffer is numbered 0 regardless of the
495  * value of b_blkno.
496  */
497 void
498 xfs_trans_log_buf(xfs_trans_t	*tp,
499 		  xfs_buf_t	*bp,
500 		  uint		first,
501 		  uint		last)
502 {
503 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
504 
505 	ASSERT(bp->b_transp == tp);
506 	ASSERT(bip != NULL);
507 	ASSERT(first <= last && last < BBTOB(bp->b_length));
508 	ASSERT(bp->b_iodone == NULL ||
509 	       bp->b_iodone == xfs_buf_iodone_callbacks);
510 
511 	/*
512 	 * Mark the buffer as needing to be written out eventually,
513 	 * and set its iodone function to remove the buffer's buf log
514 	 * item from the AIL and free it when the buffer is flushed
515 	 * to disk.  See xfs_buf_attach_iodone() for more details
516 	 * on li_cb and xfs_buf_iodone_callbacks().
517 	 * If we end up aborting this transaction, we trap this buffer
518 	 * inside the b_bdstrat callback so that this won't get written to
519 	 * disk.
520 	 */
521 	bp->b_flags |= XBF_DONE;
522 
523 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
524 	bp->b_iodone = xfs_buf_iodone_callbacks;
525 	bip->bli_item.li_cb = xfs_buf_iodone;
526 
527 	trace_xfs_trans_log_buf(bip);
528 
529 	/*
530 	 * If we invalidated the buffer within this transaction, then
531 	 * cancel the invalidation now that we're dirtying the buffer
532 	 * again.  There are no races with the code in xfs_buf_item_unpin(),
533 	 * because we have a reference to the buffer this entire time.
534 	 */
535 	if (bip->bli_flags & XFS_BLI_STALE) {
536 		bip->bli_flags &= ~XFS_BLI_STALE;
537 		ASSERT(bp->b_flags & XBF_STALE);
538 		bp->b_flags &= ~XBF_STALE;
539 		bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
540 	}
541 
542 	tp->t_flags |= XFS_TRANS_DIRTY;
543 	bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
544 
545 	/*
546 	 * If we have an ordered buffer we are not logging any dirty range but
547 	 * it still needs to be marked dirty and that it has been logged.
548 	 */
549 	bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
550 	if (!(bip->bli_flags & XFS_BLI_ORDERED))
551 		xfs_buf_item_log(bip, first, last);
552 }
553 
554 
555 /*
556  * Invalidate a buffer that is being used within a transaction.
557  *
558  * Typically this is because the blocks in the buffer are being freed, so we
559  * need to prevent it from being written out when we're done.  Allowing it
560  * to be written again might overwrite data in the free blocks if they are
561  * reallocated to a file.
562  *
563  * We prevent the buffer from being written out by marking it stale.  We can't
564  * get rid of the buf log item at this point because the buffer may still be
565  * pinned by another transaction.  If that is the case, then we'll wait until
566  * the buffer is committed to disk for the last time (we can tell by the ref
567  * count) and free it in xfs_buf_item_unpin().  Until that happens we will
568  * keep the buffer locked so that the buffer and buf log item are not reused.
569  *
570  * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
571  * the buf item.  This will be used at recovery time to determine that copies
572  * of the buffer in the log before this should not be replayed.
573  *
574  * We mark the item descriptor and the transaction dirty so that we'll hold
575  * the buffer until after the commit.
576  *
577  * Since we're invalidating the buffer, we also clear the state about which
578  * parts of the buffer have been logged.  We also clear the flag indicating
579  * that this is an inode buffer since the data in the buffer will no longer
580  * be valid.
581  *
582  * We set the stale bit in the buffer as well since we're getting rid of it.
583  */
584 void
585 xfs_trans_binval(
586 	xfs_trans_t	*tp,
587 	xfs_buf_t	*bp)
588 {
589 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
590 	int			i;
591 
592 	ASSERT(bp->b_transp == tp);
593 	ASSERT(bip != NULL);
594 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
595 
596 	trace_xfs_trans_binval(bip);
597 
598 	if (bip->bli_flags & XFS_BLI_STALE) {
599 		/*
600 		 * If the buffer is already invalidated, then
601 		 * just return.
602 		 */
603 		ASSERT(bp->b_flags & XBF_STALE);
604 		ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
605 		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
606 		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
607 		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
608 		ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
609 		ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
610 		return;
611 	}
612 
613 	xfs_buf_stale(bp);
614 
615 	bip->bli_flags |= XFS_BLI_STALE;
616 	bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
617 	bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
618 	bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
619 	bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
620 	for (i = 0; i < bip->bli_format_count; i++) {
621 		memset(bip->bli_formats[i].blf_data_map, 0,
622 		       (bip->bli_formats[i].blf_map_size * sizeof(uint)));
623 	}
624 	bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
625 	tp->t_flags |= XFS_TRANS_DIRTY;
626 }
627 
628 /*
629  * This call is used to indicate that the buffer contains on-disk inodes which
630  * must be handled specially during recovery.  They require special handling
631  * because only the di_next_unlinked from the inodes in the buffer should be
632  * recovered.  The rest of the data in the buffer is logged via the inodes
633  * themselves.
634  *
635  * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
636  * transferred to the buffer's log format structure so that we'll know what to
637  * do at recovery time.
638  */
639 void
640 xfs_trans_inode_buf(
641 	xfs_trans_t	*tp,
642 	xfs_buf_t	*bp)
643 {
644 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
645 
646 	ASSERT(bp->b_transp == tp);
647 	ASSERT(bip != NULL);
648 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
649 
650 	bip->bli_flags |= XFS_BLI_INODE_BUF;
651 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
652 }
653 
654 /*
655  * This call is used to indicate that the buffer is going to
656  * be staled and was an inode buffer. This means it gets
657  * special processing during unpin - where any inodes
658  * associated with the buffer should be removed from ail.
659  * There is also special processing during recovery,
660  * any replay of the inodes in the buffer needs to be
661  * prevented as the buffer may have been reused.
662  */
663 void
664 xfs_trans_stale_inode_buf(
665 	xfs_trans_t	*tp,
666 	xfs_buf_t	*bp)
667 {
668 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
669 
670 	ASSERT(bp->b_transp == tp);
671 	ASSERT(bip != NULL);
672 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
673 
674 	bip->bli_flags |= XFS_BLI_STALE_INODE;
675 	bip->bli_item.li_cb = xfs_buf_iodone;
676 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
677 }
678 
679 /*
680  * Mark the buffer as being one which contains newly allocated
681  * inodes.  We need to make sure that even if this buffer is
682  * relogged as an 'inode buf' we still recover all of the inode
683  * images in the face of a crash.  This works in coordination with
684  * xfs_buf_item_committed() to ensure that the buffer remains in the
685  * AIL at its original location even after it has been relogged.
686  */
687 /* ARGSUSED */
688 void
689 xfs_trans_inode_alloc_buf(
690 	xfs_trans_t	*tp,
691 	xfs_buf_t	*bp)
692 {
693 	xfs_buf_log_item_t	*bip = bp->b_fspriv;
694 
695 	ASSERT(bp->b_transp == tp);
696 	ASSERT(bip != NULL);
697 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
698 
699 	bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
700 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
701 }
702 
703 /*
704  * Mark the buffer as ordered for this transaction. This means
705  * that the contents of the buffer are not recorded in the transaction
706  * but it is tracked in the AIL as though it was. This allows us
707  * to record logical changes in transactions rather than the physical
708  * changes we make to the buffer without changing writeback ordering
709  * constraints of metadata buffers.
710  */
711 void
712 xfs_trans_ordered_buf(
713 	struct xfs_trans	*tp,
714 	struct xfs_buf		*bp)
715 {
716 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
717 
718 	ASSERT(bp->b_transp == tp);
719 	ASSERT(bip != NULL);
720 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
721 
722 	bip->bli_flags |= XFS_BLI_ORDERED;
723 	trace_xfs_buf_item_ordered(bip);
724 }
725 
726 /*
727  * Set the type of the buffer for log recovery so that it can correctly identify
728  * and hence attach the correct buffer ops to the buffer after replay.
729  */
730 void
731 xfs_trans_buf_set_type(
732 	struct xfs_trans	*tp,
733 	struct xfs_buf		*bp,
734 	enum xfs_blft		type)
735 {
736 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
737 
738 	if (!tp)
739 		return;
740 
741 	ASSERT(bp->b_transp == tp);
742 	ASSERT(bip != NULL);
743 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
744 
745 	xfs_blft_to_flags(&bip->__bli_format, type);
746 }
747 
748 void
749 xfs_trans_buf_copy_type(
750 	struct xfs_buf		*dst_bp,
751 	struct xfs_buf		*src_bp)
752 {
753 	struct xfs_buf_log_item	*sbip = src_bp->b_fspriv;
754 	struct xfs_buf_log_item	*dbip = dst_bp->b_fspriv;
755 	enum xfs_blft		type;
756 
757 	type = xfs_blft_from_flags(&sbip->__bli_format);
758 	xfs_blft_to_flags(&dbip->__bli_format, type);
759 }
760 
761 /*
762  * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
763  * dquots. However, unlike in inode buffer recovery, dquot buffers get
764  * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
765  * The only thing that makes dquot buffers different from regular
766  * buffers is that we must not replay dquot bufs when recovering
767  * if a _corresponding_ quotaoff has happened. We also have to distinguish
768  * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
769  * can be turned off independently.
770  */
771 /* ARGSUSED */
772 void
773 xfs_trans_dquot_buf(
774 	xfs_trans_t	*tp,
775 	xfs_buf_t	*bp,
776 	uint		type)
777 {
778 	struct xfs_buf_log_item	*bip = bp->b_fspriv;
779 
780 	ASSERT(type == XFS_BLF_UDQUOT_BUF ||
781 	       type == XFS_BLF_PDQUOT_BUF ||
782 	       type == XFS_BLF_GDQUOT_BUF);
783 
784 	bip->__bli_format.blf_flags |= type;
785 
786 	switch (type) {
787 	case XFS_BLF_UDQUOT_BUF:
788 		type = XFS_BLFT_UDQUOT_BUF;
789 		break;
790 	case XFS_BLF_PDQUOT_BUF:
791 		type = XFS_BLFT_PDQUOT_BUF;
792 		break;
793 	case XFS_BLF_GDQUOT_BUF:
794 		type = XFS_BLFT_GDQUOT_BUF;
795 		break;
796 	default:
797 		type = XFS_BLFT_UNKNOWN_BUF;
798 		break;
799 	}
800 
801 	xfs_trans_buf_set_type(tp, bp, type);
802 }
803