xref: /openbmc/linux/fs/xfs/xfs_trans_buf.c (revision f220d3eb)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2002,2005 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_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_buf_item.h"
16 #include "xfs_trans_priv.h"
17 #include "xfs_error.h"
18 #include "xfs_trace.h"
19 
20 /*
21  * Check to see if a buffer matching the given parameters is already
22  * a part of the given transaction.
23  */
24 STATIC struct xfs_buf *
25 xfs_trans_buf_item_match(
26 	struct xfs_trans	*tp,
27 	struct xfs_buftarg	*target,
28 	struct xfs_buf_map	*map,
29 	int			nmaps)
30 {
31 	struct xfs_log_item	*lip;
32 	struct xfs_buf_log_item	*blip;
33 	int			len = 0;
34 	int			i;
35 
36 	for (i = 0; i < nmaps; i++)
37 		len += map[i].bm_len;
38 
39 	list_for_each_entry(lip, &tp->t_items, li_trans) {
40 		blip = (struct xfs_buf_log_item *)lip;
41 		if (blip->bli_item.li_type == XFS_LI_BUF &&
42 		    blip->bli_buf->b_target == target &&
43 		    XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
44 		    blip->bli_buf->b_length == len) {
45 			ASSERT(blip->bli_buf->b_map_count == nmaps);
46 			return blip->bli_buf;
47 		}
48 	}
49 
50 	return NULL;
51 }
52 
53 /*
54  * Add the locked buffer to the transaction.
55  *
56  * The buffer must be locked, and it cannot be associated with any
57  * transaction.
58  *
59  * If the buffer does not yet have a buf log item associated with it,
60  * then allocate one for it.  Then add the buf item to the transaction.
61  */
62 STATIC void
63 _xfs_trans_bjoin(
64 	struct xfs_trans	*tp,
65 	struct xfs_buf		*bp,
66 	int			reset_recur)
67 {
68 	struct xfs_buf_log_item	*bip;
69 
70 	ASSERT(bp->b_transp == NULL);
71 
72 	/*
73 	 * The xfs_buf_log_item pointer is stored in b_log_item.  If
74 	 * it doesn't have one yet, then allocate one and initialize it.
75 	 * The checks to see if one is there are in xfs_buf_item_init().
76 	 */
77 	xfs_buf_item_init(bp, tp->t_mountp);
78 	bip = bp->b_log_item;
79 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
80 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
81 	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
82 	if (reset_recur)
83 		bip->bli_recur = 0;
84 
85 	/*
86 	 * Take a reference for this transaction on the buf item.
87 	 */
88 	atomic_inc(&bip->bli_refcount);
89 
90 	/*
91 	 * Attach the item to the transaction so we can find it in
92 	 * xfs_trans_get_buf() and friends.
93 	 */
94 	xfs_trans_add_item(tp, &bip->bli_item);
95 	bp->b_transp = tp;
96 
97 }
98 
99 void
100 xfs_trans_bjoin(
101 	struct xfs_trans	*tp,
102 	struct xfs_buf		*bp)
103 {
104 	_xfs_trans_bjoin(tp, bp, 0);
105 	trace_xfs_trans_bjoin(bp->b_log_item);
106 }
107 
108 /*
109  * Get and lock the buffer for the caller if it is not already
110  * locked within the given transaction.  If it is already locked
111  * within the transaction, just increment its lock recursion count
112  * and return a pointer to it.
113  *
114  * If the transaction pointer is NULL, make this just a normal
115  * get_buf() call.
116  */
117 struct xfs_buf *
118 xfs_trans_get_buf_map(
119 	struct xfs_trans	*tp,
120 	struct xfs_buftarg	*target,
121 	struct xfs_buf_map	*map,
122 	int			nmaps,
123 	xfs_buf_flags_t		flags)
124 {
125 	xfs_buf_t		*bp;
126 	struct xfs_buf_log_item	*bip;
127 
128 	if (!tp)
129 		return xfs_buf_get_map(target, map, nmaps, flags);
130 
131 	/*
132 	 * If we find the buffer in the cache with this transaction
133 	 * pointer in its b_fsprivate2 field, then we know we already
134 	 * have it locked.  In this case we just increment the lock
135 	 * recursion count and return the buffer to the caller.
136 	 */
137 	bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
138 	if (bp != NULL) {
139 		ASSERT(xfs_buf_islocked(bp));
140 		if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
141 			xfs_buf_stale(bp);
142 			bp->b_flags |= XBF_DONE;
143 		}
144 
145 		ASSERT(bp->b_transp == tp);
146 		bip = bp->b_log_item;
147 		ASSERT(bip != NULL);
148 		ASSERT(atomic_read(&bip->bli_refcount) > 0);
149 		bip->bli_recur++;
150 		trace_xfs_trans_get_buf_recur(bip);
151 		return bp;
152 	}
153 
154 	bp = xfs_buf_get_map(target, map, nmaps, flags);
155 	if (bp == NULL) {
156 		return NULL;
157 	}
158 
159 	ASSERT(!bp->b_error);
160 
161 	_xfs_trans_bjoin(tp, bp, 1);
162 	trace_xfs_trans_get_buf(bp->b_log_item);
163 	return bp;
164 }
165 
166 /*
167  * Get and lock the superblock buffer of this file system for the
168  * given transaction.
169  *
170  * We don't need to use incore_match() here, because the superblock
171  * buffer is a private buffer which we keep a pointer to in the
172  * mount structure.
173  */
174 xfs_buf_t *
175 xfs_trans_getsb(
176 	xfs_trans_t		*tp,
177 	struct xfs_mount	*mp,
178 	int			flags)
179 {
180 	xfs_buf_t		*bp;
181 	struct xfs_buf_log_item	*bip;
182 
183 	/*
184 	 * Default to just trying to lock the superblock buffer
185 	 * if tp is NULL.
186 	 */
187 	if (tp == NULL)
188 		return xfs_getsb(mp, flags);
189 
190 	/*
191 	 * If the superblock buffer already has this transaction
192 	 * pointer in its b_fsprivate2 field, then we know we already
193 	 * have it locked.  In this case we just increment the lock
194 	 * recursion count and return the buffer to the caller.
195 	 */
196 	bp = mp->m_sb_bp;
197 	if (bp->b_transp == tp) {
198 		bip = bp->b_log_item;
199 		ASSERT(bip != NULL);
200 		ASSERT(atomic_read(&bip->bli_refcount) > 0);
201 		bip->bli_recur++;
202 		trace_xfs_trans_getsb_recur(bip);
203 		return bp;
204 	}
205 
206 	bp = xfs_getsb(mp, flags);
207 	if (bp == NULL)
208 		return NULL;
209 
210 	_xfs_trans_bjoin(tp, bp, 1);
211 	trace_xfs_trans_getsb(bp->b_log_item);
212 	return bp;
213 }
214 
215 /*
216  * Get and lock the buffer for the caller if it is not already
217  * locked within the given transaction.  If it has not yet been
218  * read in, read it from disk. If it is already locked
219  * within the transaction and already read in, just increment its
220  * lock recursion count and return a pointer to it.
221  *
222  * If the transaction pointer is NULL, make this just a normal
223  * read_buf() call.
224  */
225 int
226 xfs_trans_read_buf_map(
227 	struct xfs_mount	*mp,
228 	struct xfs_trans	*tp,
229 	struct xfs_buftarg	*target,
230 	struct xfs_buf_map	*map,
231 	int			nmaps,
232 	xfs_buf_flags_t		flags,
233 	struct xfs_buf		**bpp,
234 	const struct xfs_buf_ops *ops)
235 {
236 	struct xfs_buf		*bp = NULL;
237 	struct xfs_buf_log_item	*bip;
238 	int			error;
239 
240 	*bpp = NULL;
241 	/*
242 	 * If we find the buffer in the cache with this transaction
243 	 * pointer in its b_fsprivate2 field, then we know we already
244 	 * have it locked.  If it is already read in we just increment
245 	 * the lock recursion count and return the buffer to the caller.
246 	 * If the buffer is not yet read in, then we read it in, increment
247 	 * the lock recursion count, and return it to the caller.
248 	 */
249 	if (tp)
250 		bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
251 	if (bp) {
252 		ASSERT(xfs_buf_islocked(bp));
253 		ASSERT(bp->b_transp == tp);
254 		ASSERT(bp->b_log_item != NULL);
255 		ASSERT(!bp->b_error);
256 		ASSERT(bp->b_flags & XBF_DONE);
257 
258 		/*
259 		 * We never locked this buf ourselves, so we shouldn't
260 		 * brelse it either. Just get out.
261 		 */
262 		if (XFS_FORCED_SHUTDOWN(mp)) {
263 			trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
264 			return -EIO;
265 		}
266 
267 		bip = bp->b_log_item;
268 		bip->bli_recur++;
269 
270 		ASSERT(atomic_read(&bip->bli_refcount) > 0);
271 		trace_xfs_trans_read_buf_recur(bip);
272 		*bpp = bp;
273 		return 0;
274 	}
275 
276 	bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
277 	if (!bp) {
278 		if (!(flags & XBF_TRYLOCK))
279 			return -ENOMEM;
280 		return tp ? 0 : -EAGAIN;
281 	}
282 
283 	/*
284 	 * If we've had a read error, then the contents of the buffer are
285 	 * invalid and should not be used. To ensure that a followup read tries
286 	 * to pull the buffer from disk again, we clear the XBF_DONE flag and
287 	 * mark the buffer stale. This ensures that anyone who has a current
288 	 * reference to the buffer will interpret it's contents correctly and
289 	 * future cache lookups will also treat it as an empty, uninitialised
290 	 * buffer.
291 	 */
292 	if (bp->b_error) {
293 		error = bp->b_error;
294 		if (!XFS_FORCED_SHUTDOWN(mp))
295 			xfs_buf_ioerror_alert(bp, __func__);
296 		bp->b_flags &= ~XBF_DONE;
297 		xfs_buf_stale(bp);
298 
299 		if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
300 			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
301 		xfs_buf_relse(bp);
302 
303 		/* bad CRC means corrupted metadata */
304 		if (error == -EFSBADCRC)
305 			error = -EFSCORRUPTED;
306 		return error;
307 	}
308 
309 	if (XFS_FORCED_SHUTDOWN(mp)) {
310 		xfs_buf_relse(bp);
311 		trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
312 		return -EIO;
313 	}
314 
315 	if (tp) {
316 		_xfs_trans_bjoin(tp, bp, 1);
317 		trace_xfs_trans_read_buf(bp->b_log_item);
318 	}
319 	*bpp = bp;
320 	return 0;
321 
322 }
323 
324 /*
325  * Release the buffer bp which was previously acquired with one of the
326  * xfs_trans_... buffer allocation routines if the buffer has not
327  * been modified within this transaction.  If the buffer is modified
328  * within this transaction, do decrement the recursion count but do
329  * not release the buffer even if the count goes to 0.  If the buffer is not
330  * modified within the transaction, decrement the recursion count and
331  * release the buffer if the recursion count goes to 0.
332  *
333  * If the buffer is to be released and it was not modified before
334  * this transaction began, then free the buf_log_item associated with it.
335  *
336  * If the transaction pointer is NULL, make this just a normal
337  * brelse() call.
338  */
339 void
340 xfs_trans_brelse(
341 	xfs_trans_t		*tp,
342 	xfs_buf_t		*bp)
343 {
344 	struct xfs_buf_log_item	*bip;
345 	int			freed;
346 
347 	/*
348 	 * Default to a normal brelse() call if the tp is NULL.
349 	 */
350 	if (tp == NULL) {
351 		ASSERT(bp->b_transp == NULL);
352 		xfs_buf_relse(bp);
353 		return;
354 	}
355 
356 	ASSERT(bp->b_transp == tp);
357 	bip = bp->b_log_item;
358 	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
359 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
360 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
361 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
362 
363 	trace_xfs_trans_brelse(bip);
364 
365 	/*
366 	 * If the release is just for a recursive lock,
367 	 * then decrement the count and return.
368 	 */
369 	if (bip->bli_recur > 0) {
370 		bip->bli_recur--;
371 		return;
372 	}
373 
374 	/*
375 	 * If the buffer is dirty within this transaction, we can't
376 	 * release it until we commit.
377 	 */
378 	if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
379 		return;
380 
381 	/*
382 	 * If the buffer has been invalidated, then we can't release
383 	 * it until the transaction commits to disk unless it is re-dirtied
384 	 * as part of this transaction.  This prevents us from pulling
385 	 * the item from the AIL before we should.
386 	 */
387 	if (bip->bli_flags & XFS_BLI_STALE)
388 		return;
389 
390 	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
391 
392 	/*
393 	 * Free up the log item descriptor tracking the released item.
394 	 */
395 	xfs_trans_del_item(&bip->bli_item);
396 
397 	/*
398 	 * Clear the hold flag in the buf log item if it is set.
399 	 * We wouldn't want the next user of the buffer to
400 	 * get confused.
401 	 */
402 	if (bip->bli_flags & XFS_BLI_HOLD) {
403 		bip->bli_flags &= ~XFS_BLI_HOLD;
404 	}
405 
406 	/*
407 	 * Drop our reference to the buf log item.
408 	 */
409 	freed = atomic_dec_and_test(&bip->bli_refcount);
410 
411 	/*
412 	 * If the buf item is not tracking data in the log, then we must free it
413 	 * before releasing the buffer back to the free pool.
414 	 *
415 	 * If the fs has shutdown and we dropped the last reference, it may fall
416 	 * on us to release a (possibly dirty) bli if it never made it to the
417 	 * AIL (e.g., the aborted unpin already happened and didn't release it
418 	 * due to our reference). Since we're already shutdown and need
419 	 * ail_lock, just force remove from the AIL and release the bli here.
420 	 */
421 	if (XFS_FORCED_SHUTDOWN(tp->t_mountp) && freed) {
422 		xfs_trans_ail_remove(&bip->bli_item, SHUTDOWN_LOG_IO_ERROR);
423 		xfs_buf_item_relse(bp);
424 	} else if (!(bip->bli_flags & XFS_BLI_DIRTY)) {
425 /***
426 		ASSERT(bp->b_pincount == 0);
427 ***/
428 		ASSERT(atomic_read(&bip->bli_refcount) == 0);
429 		ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags));
430 		ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
431 		xfs_buf_item_relse(bp);
432 	}
433 
434 	bp->b_transp = NULL;
435 	xfs_buf_relse(bp);
436 }
437 
438 /*
439  * Mark the buffer as not needing to be unlocked when the buf item's
440  * iop_unlock() routine is called.  The buffer must already be locked
441  * and associated with the given transaction.
442  */
443 /* ARGSUSED */
444 void
445 xfs_trans_bhold(
446 	xfs_trans_t		*tp,
447 	xfs_buf_t		*bp)
448 {
449 	struct xfs_buf_log_item	*bip = bp->b_log_item;
450 
451 	ASSERT(bp->b_transp == tp);
452 	ASSERT(bip != NULL);
453 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
454 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
455 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
456 
457 	bip->bli_flags |= XFS_BLI_HOLD;
458 	trace_xfs_trans_bhold(bip);
459 }
460 
461 /*
462  * Cancel the previous buffer hold request made on this buffer
463  * for this transaction.
464  */
465 void
466 xfs_trans_bhold_release(
467 	xfs_trans_t		*tp,
468 	xfs_buf_t		*bp)
469 {
470 	struct xfs_buf_log_item	*bip = bp->b_log_item;
471 
472 	ASSERT(bp->b_transp == tp);
473 	ASSERT(bip != NULL);
474 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
475 	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
476 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
477 	ASSERT(bip->bli_flags & XFS_BLI_HOLD);
478 
479 	bip->bli_flags &= ~XFS_BLI_HOLD;
480 	trace_xfs_trans_bhold_release(bip);
481 }
482 
483 /*
484  * Mark a buffer dirty in the transaction.
485  */
486 void
487 xfs_trans_dirty_buf(
488 	struct xfs_trans	*tp,
489 	struct xfs_buf		*bp)
490 {
491 	struct xfs_buf_log_item	*bip = bp->b_log_item;
492 
493 	ASSERT(bp->b_transp == tp);
494 	ASSERT(bip != NULL);
495 	ASSERT(bp->b_iodone == NULL ||
496 	       bp->b_iodone == xfs_buf_iodone_callbacks);
497 
498 	/*
499 	 * Mark the buffer as needing to be written out eventually,
500 	 * and set its iodone function to remove the buffer's buf log
501 	 * item from the AIL and free it when the buffer is flushed
502 	 * to disk.  See xfs_buf_attach_iodone() for more details
503 	 * on li_cb and xfs_buf_iodone_callbacks().
504 	 * If we end up aborting this transaction, we trap this buffer
505 	 * inside the b_bdstrat callback so that this won't get written to
506 	 * disk.
507 	 */
508 	bp->b_flags |= XBF_DONE;
509 
510 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
511 	bp->b_iodone = xfs_buf_iodone_callbacks;
512 	bip->bli_item.li_cb = xfs_buf_iodone;
513 
514 	/*
515 	 * If we invalidated the buffer within this transaction, then
516 	 * cancel the invalidation now that we're dirtying the buffer
517 	 * again.  There are no races with the code in xfs_buf_item_unpin(),
518 	 * because we have a reference to the buffer this entire time.
519 	 */
520 	if (bip->bli_flags & XFS_BLI_STALE) {
521 		bip->bli_flags &= ~XFS_BLI_STALE;
522 		ASSERT(bp->b_flags & XBF_STALE);
523 		bp->b_flags &= ~XBF_STALE;
524 		bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
525 	}
526 	bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
527 
528 	tp->t_flags |= XFS_TRANS_DIRTY;
529 	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
530 }
531 
532 /*
533  * This is called to mark bytes first through last inclusive of the given
534  * buffer as needing to be logged when the transaction is committed.
535  * The buffer must already be associated with the given transaction.
536  *
537  * First and last are numbers relative to the beginning of this buffer,
538  * so the first byte in the buffer is numbered 0 regardless of the
539  * value of b_blkno.
540  */
541 void
542 xfs_trans_log_buf(
543 	struct xfs_trans	*tp,
544 	struct xfs_buf		*bp,
545 	uint			first,
546 	uint			last)
547 {
548 	struct xfs_buf_log_item	*bip = bp->b_log_item;
549 
550 	ASSERT(first <= last && last < BBTOB(bp->b_length));
551 	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
552 
553 	xfs_trans_dirty_buf(tp, bp);
554 
555 	trace_xfs_trans_log_buf(bip);
556 	xfs_buf_item_log(bip, first, last);
557 }
558 
559 
560 /*
561  * Invalidate a buffer that is being used within a transaction.
562  *
563  * Typically this is because the blocks in the buffer are being freed, so we
564  * need to prevent it from being written out when we're done.  Allowing it
565  * to be written again might overwrite data in the free blocks if they are
566  * reallocated to a file.
567  *
568  * We prevent the buffer from being written out by marking it stale.  We can't
569  * get rid of the buf log item at this point because the buffer may still be
570  * pinned by another transaction.  If that is the case, then we'll wait until
571  * the buffer is committed to disk for the last time (we can tell by the ref
572  * count) and free it in xfs_buf_item_unpin().  Until that happens we will
573  * keep the buffer locked so that the buffer and buf log item are not reused.
574  *
575  * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
576  * the buf item.  This will be used at recovery time to determine that copies
577  * of the buffer in the log before this should not be replayed.
578  *
579  * We mark the item descriptor and the transaction dirty so that we'll hold
580  * the buffer until after the commit.
581  *
582  * Since we're invalidating the buffer, we also clear the state about which
583  * parts of the buffer have been logged.  We also clear the flag indicating
584  * that this is an inode buffer since the data in the buffer will no longer
585  * be valid.
586  *
587  * We set the stale bit in the buffer as well since we're getting rid of it.
588  */
589 void
590 xfs_trans_binval(
591 	xfs_trans_t		*tp,
592 	xfs_buf_t		*bp)
593 {
594 	struct xfs_buf_log_item	*bip = bp->b_log_item;
595 	int			i;
596 
597 	ASSERT(bp->b_transp == tp);
598 	ASSERT(bip != NULL);
599 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
600 
601 	trace_xfs_trans_binval(bip);
602 
603 	if (bip->bli_flags & XFS_BLI_STALE) {
604 		/*
605 		 * If the buffer is already invalidated, then
606 		 * just return.
607 		 */
608 		ASSERT(bp->b_flags & XBF_STALE);
609 		ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
610 		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
611 		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
612 		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
613 		ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
614 		ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
615 		return;
616 	}
617 
618 	xfs_buf_stale(bp);
619 
620 	bip->bli_flags |= XFS_BLI_STALE;
621 	bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
622 	bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
623 	bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
624 	bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
625 	for (i = 0; i < bip->bli_format_count; i++) {
626 		memset(bip->bli_formats[i].blf_data_map, 0,
627 		       (bip->bli_formats[i].blf_map_size * sizeof(uint)));
628 	}
629 	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
630 	tp->t_flags |= XFS_TRANS_DIRTY;
631 }
632 
633 /*
634  * This call is used to indicate that the buffer contains on-disk inodes which
635  * must be handled specially during recovery.  They require special handling
636  * because only the di_next_unlinked from the inodes in the buffer should be
637  * recovered.  The rest of the data in the buffer is logged via the inodes
638  * themselves.
639  *
640  * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
641  * transferred to the buffer's log format structure so that we'll know what to
642  * do at recovery time.
643  */
644 void
645 xfs_trans_inode_buf(
646 	xfs_trans_t		*tp,
647 	xfs_buf_t		*bp)
648 {
649 	struct xfs_buf_log_item	*bip = bp->b_log_item;
650 
651 	ASSERT(bp->b_transp == tp);
652 	ASSERT(bip != NULL);
653 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
654 
655 	bip->bli_flags |= XFS_BLI_INODE_BUF;
656 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
657 }
658 
659 /*
660  * This call is used to indicate that the buffer is going to
661  * be staled and was an inode buffer. This means it gets
662  * special processing during unpin - where any inodes
663  * associated with the buffer should be removed from ail.
664  * There is also special processing during recovery,
665  * any replay of the inodes in the buffer needs to be
666  * prevented as the buffer may have been reused.
667  */
668 void
669 xfs_trans_stale_inode_buf(
670 	xfs_trans_t		*tp,
671 	xfs_buf_t		*bp)
672 {
673 	struct xfs_buf_log_item	*bip = bp->b_log_item;
674 
675 	ASSERT(bp->b_transp == tp);
676 	ASSERT(bip != NULL);
677 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
678 
679 	bip->bli_flags |= XFS_BLI_STALE_INODE;
680 	bip->bli_item.li_cb = xfs_buf_iodone;
681 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
682 }
683 
684 /*
685  * Mark the buffer as being one which contains newly allocated
686  * inodes.  We need to make sure that even if this buffer is
687  * relogged as an 'inode buf' we still recover all of the inode
688  * images in the face of a crash.  This works in coordination with
689  * xfs_buf_item_committed() to ensure that the buffer remains in the
690  * AIL at its original location even after it has been relogged.
691  */
692 /* ARGSUSED */
693 void
694 xfs_trans_inode_alloc_buf(
695 	xfs_trans_t		*tp,
696 	xfs_buf_t		*bp)
697 {
698 	struct xfs_buf_log_item	*bip = bp->b_log_item;
699 
700 	ASSERT(bp->b_transp == tp);
701 	ASSERT(bip != NULL);
702 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
703 
704 	bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
705 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
706 }
707 
708 /*
709  * Mark the buffer as ordered for this transaction. This means that the contents
710  * of the buffer are not recorded in the transaction but it is tracked in the
711  * AIL as though it was. This allows us to record logical changes in
712  * transactions rather than the physical changes we make to the buffer without
713  * changing writeback ordering constraints of metadata buffers.
714  */
715 bool
716 xfs_trans_ordered_buf(
717 	struct xfs_trans	*tp,
718 	struct xfs_buf		*bp)
719 {
720 	struct xfs_buf_log_item	*bip = bp->b_log_item;
721 
722 	ASSERT(bp->b_transp == tp);
723 	ASSERT(bip != NULL);
724 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
725 
726 	if (xfs_buf_item_dirty_format(bip))
727 		return false;
728 
729 	bip->bli_flags |= XFS_BLI_ORDERED;
730 	trace_xfs_buf_item_ordered(bip);
731 
732 	/*
733 	 * We don't log a dirty range of an ordered buffer but it still needs
734 	 * to be marked dirty and that it has been logged.
735 	 */
736 	xfs_trans_dirty_buf(tp, bp);
737 	return true;
738 }
739 
740 /*
741  * Set the type of the buffer for log recovery so that it can correctly identify
742  * and hence attach the correct buffer ops to the buffer after replay.
743  */
744 void
745 xfs_trans_buf_set_type(
746 	struct xfs_trans	*tp,
747 	struct xfs_buf		*bp,
748 	enum xfs_blft		type)
749 {
750 	struct xfs_buf_log_item	*bip = bp->b_log_item;
751 
752 	if (!tp)
753 		return;
754 
755 	ASSERT(bp->b_transp == tp);
756 	ASSERT(bip != NULL);
757 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
758 
759 	xfs_blft_to_flags(&bip->__bli_format, type);
760 }
761 
762 void
763 xfs_trans_buf_copy_type(
764 	struct xfs_buf		*dst_bp,
765 	struct xfs_buf		*src_bp)
766 {
767 	struct xfs_buf_log_item	*sbip = src_bp->b_log_item;
768 	struct xfs_buf_log_item	*dbip = dst_bp->b_log_item;
769 	enum xfs_blft		type;
770 
771 	type = xfs_blft_from_flags(&sbip->__bli_format);
772 	xfs_blft_to_flags(&dbip->__bli_format, type);
773 }
774 
775 /*
776  * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
777  * dquots. However, unlike in inode buffer recovery, dquot buffers get
778  * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
779  * The only thing that makes dquot buffers different from regular
780  * buffers is that we must not replay dquot bufs when recovering
781  * if a _corresponding_ quotaoff has happened. We also have to distinguish
782  * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
783  * can be turned off independently.
784  */
785 /* ARGSUSED */
786 void
787 xfs_trans_dquot_buf(
788 	xfs_trans_t		*tp,
789 	xfs_buf_t		*bp,
790 	uint			type)
791 {
792 	struct xfs_buf_log_item	*bip = bp->b_log_item;
793 
794 	ASSERT(type == XFS_BLF_UDQUOT_BUF ||
795 	       type == XFS_BLF_PDQUOT_BUF ||
796 	       type == XFS_BLF_GDQUOT_BUF);
797 
798 	bip->__bli_format.blf_flags |= type;
799 
800 	switch (type) {
801 	case XFS_BLF_UDQUOT_BUF:
802 		type = XFS_BLFT_UDQUOT_BUF;
803 		break;
804 	case XFS_BLF_PDQUOT_BUF:
805 		type = XFS_BLFT_PDQUOT_BUF;
806 		break;
807 	case XFS_BLF_GDQUOT_BUF:
808 		type = XFS_BLFT_GDQUOT_BUF;
809 		break;
810 	default:
811 		type = XFS_BLFT_UNKNOWN_BUF;
812 		break;
813 	}
814 
815 	xfs_trans_buf_set_type(tp, bp, type);
816 }
817