Linux Audio

Check our new training course

Loading...
v6.13.7
  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_trans.h"
 14#include "xfs_buf_item.h"
 15#include "xfs_trans_priv.h"
 16#include "xfs_trace.h"
 17
 18/*
 19 * Check to see if a buffer matching the given parameters is already
 20 * a part of the given transaction.
 21 */
 22STATIC struct xfs_buf *
 23xfs_trans_buf_item_match(
 24	struct xfs_trans	*tp,
 25	struct xfs_buftarg	*target,
 26	struct xfs_buf_map	*map,
 27	int			nmaps)
 28{
 29	struct xfs_log_item	*lip;
 30	struct xfs_buf_log_item	*blip;
 31	int			len = 0;
 32	int			i;
 33
 34	for (i = 0; i < nmaps; i++)
 35		len += map[i].bm_len;
 36
 37	list_for_each_entry(lip, &tp->t_items, li_trans) {
 38		blip = (struct xfs_buf_log_item *)lip;
 39		if (blip->bli_item.li_type == XFS_LI_BUF &&
 40		    blip->bli_buf->b_target == target &&
 41		    xfs_buf_daddr(blip->bli_buf) == map[0].bm_bn &&
 42		    blip->bli_buf->b_length == len) {
 43			ASSERT(blip->bli_buf->b_map_count == nmaps);
 44			return blip->bli_buf;
 45		}
 46	}
 47
 48	return NULL;
 49}
 50
 51/*
 52 * Add the locked buffer to the transaction.
 53 *
 54 * The buffer must be locked, and it cannot be associated with any
 55 * transaction.
 56 *
 57 * If the buffer does not yet have a buf log item associated with it,
 58 * then allocate one for it.  Then add the buf item to the transaction.
 59 */
 60STATIC void
 61_xfs_trans_bjoin(
 62	struct xfs_trans	*tp,
 63	struct xfs_buf		*bp,
 64	int			reset_recur)
 65{
 66	struct xfs_buf_log_item	*bip;
 67
 68	ASSERT(bp->b_transp == NULL);
 69
 70	/*
 71	 * The xfs_buf_log_item pointer is stored in b_log_item.  If
 72	 * it doesn't have one yet, then allocate one and initialize it.
 73	 * The checks to see if one is there are in xfs_buf_item_init().
 74	 */
 75	xfs_buf_item_init(bp, tp->t_mountp);
 76	bip = bp->b_log_item;
 77	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 78	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
 79	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
 80	if (reset_recur)
 81		bip->bli_recur = 0;
 82
 83	/*
 84	 * Take a reference for this transaction on the buf item.
 85	 */
 86	atomic_inc(&bip->bli_refcount);
 87
 88	/*
 89	 * Attach the item to the transaction so we can find it in
 90	 * xfs_trans_get_buf() and friends.
 91	 */
 92	xfs_trans_add_item(tp, &bip->bli_item);
 93	bp->b_transp = tp;
 94
 95}
 96
 97void
 98xfs_trans_bjoin(
 99	struct xfs_trans	*tp,
100	struct xfs_buf		*bp)
101{
102	_xfs_trans_bjoin(tp, bp, 0);
103	trace_xfs_trans_bjoin(bp->b_log_item);
104}
105
106/*
107 * Get and lock the buffer for the caller if it is not already
108 * locked within the given transaction.  If it is already locked
109 * within the transaction, just increment its lock recursion count
110 * and return a pointer to it.
111 *
112 * If the transaction pointer is NULL, make this just a normal
113 * get_buf() call.
114 */
115int
116xfs_trans_get_buf_map(
117	struct xfs_trans	*tp,
118	struct xfs_buftarg	*target,
119	struct xfs_buf_map	*map,
120	int			nmaps,
121	xfs_buf_flags_t		flags,
122	struct xfs_buf		**bpp)
123{
124	struct xfs_buf		*bp;
125	struct xfs_buf_log_item	*bip;
126	int			error;
127
128	*bpp = NULL;
129	if (!tp)
130		return xfs_buf_get_map(target, map, nmaps, flags, bpp);
131
132	/*
133	 * If we find the buffer in the cache with this transaction
134	 * pointer in its b_fsprivate2 field, then we know we already
135	 * have it locked.  In this case we just increment the lock
136	 * recursion count and return the buffer to the caller.
137	 */
138	bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
139	if (bp != NULL) {
140		ASSERT(xfs_buf_islocked(bp));
141		if (xfs_is_shutdown(tp->t_mountp)) {
142			xfs_buf_stale(bp);
143			bp->b_flags |= XBF_DONE;
144		}
145
146		ASSERT(bp->b_transp == tp);
147		bip = bp->b_log_item;
148		ASSERT(bip != NULL);
149		ASSERT(atomic_read(&bip->bli_refcount) > 0);
150		bip->bli_recur++;
151		trace_xfs_trans_get_buf_recur(bip);
152		*bpp = bp;
153		return 0;
154	}
155
156	error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
157	if (error)
158		return error;
159
160	ASSERT(!bp->b_error);
161
162	_xfs_trans_bjoin(tp, bp, 1);
163	trace_xfs_trans_get_buf(bp->b_log_item);
164	*bpp = bp;
165	return 0;
166}
167
168/*
169 * Get and lock the superblock buffer for the given transaction.
 
 
 
 
 
170 */
171static struct xfs_buf *
172__xfs_trans_getsb(
173	struct xfs_trans	*tp,
174	struct xfs_buf		*bp)
175{
 
 
 
176	/*
177	 * Just increment the lock recursion count if the buffer is already
178	 * attached to this transaction.
179	 */
180	if (bp->b_transp == tp) {
181		struct xfs_buf_log_item	*bip = bp->b_log_item;
182
 
 
 
 
 
 
 
 
 
183		ASSERT(bip != NULL);
184		ASSERT(atomic_read(&bip->bli_refcount) > 0);
185		bip->bli_recur++;
186
187		trace_xfs_trans_getsb_recur(bip);
188	} else {
189		xfs_buf_lock(bp);
190		xfs_buf_hold(bp);
191		_xfs_trans_bjoin(tp, bp, 1);
192
193		trace_xfs_trans_getsb(bp->b_log_item);
194	}
195
196	return bp;
197}
198
199struct xfs_buf *
200xfs_trans_getsb(
201	struct xfs_trans	*tp)
202{
203	return __xfs_trans_getsb(tp, tp->t_mountp->m_sb_bp);
204}
205
206struct xfs_buf *
207xfs_trans_getrtsb(
208	struct xfs_trans	*tp)
209{
210	if (!tp->t_mountp->m_rtsb_bp)
211		return NULL;
212	return __xfs_trans_getsb(tp, tp->t_mountp->m_rtsb_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 */
225int
226xfs_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_is_shutdown(mp)) {
263			trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
264			return -EIO;
265		}
266
267		/*
268		 * Check if the caller is trying to read a buffer that is
269		 * already attached to the transaction yet has no buffer ops
270		 * assigned.  Ops are usually attached when the buffer is
271		 * attached to the transaction, or by the read caller if
272		 * special circumstances.  That didn't happen, which is not
273		 * how this is supposed to go.
274		 *
275		 * If the buffer passes verification we'll let this go, but if
276		 * not we have to shut down.  Let the transaction cleanup code
277		 * release this buffer when it kills the tranaction.
278		 */
279		ASSERT(bp->b_ops != NULL);
280		error = xfs_buf_reverify(bp, ops);
281		if (error) {
282			xfs_buf_ioerror_alert(bp, __return_address);
283
284			if (tp->t_flags & XFS_TRANS_DIRTY)
285				xfs_force_shutdown(tp->t_mountp,
286						SHUTDOWN_META_IO_ERROR);
287
288			/* bad CRC means corrupted metadata */
289			if (error == -EFSBADCRC)
290				error = -EFSCORRUPTED;
291			return error;
292		}
293
294		bip = bp->b_log_item;
295		bip->bli_recur++;
296
297		ASSERT(atomic_read(&bip->bli_refcount) > 0);
298		trace_xfs_trans_read_buf_recur(bip);
299		ASSERT(bp->b_ops != NULL || ops == NULL);
300		*bpp = bp;
301		return 0;
302	}
303
304	error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
305			__return_address);
306	switch (error) {
307	case 0:
308		break;
309	default:
310		if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
311			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
312		fallthrough;
313	case -ENOMEM:
314	case -EAGAIN:
315		return error;
316	}
317
318	if (xfs_is_shutdown(mp)) {
319		xfs_buf_relse(bp);
320		trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
321		return -EIO;
322	}
323
324	if (tp) {
325		_xfs_trans_bjoin(tp, bp, 1);
326		trace_xfs_trans_read_buf(bp->b_log_item);
327	}
328	ASSERT(bp->b_ops != NULL || ops == NULL);
329	*bpp = bp;
330	return 0;
331
332}
333
334/* Has this buffer been dirtied by anyone? */
335bool
336xfs_trans_buf_is_dirty(
337	struct xfs_buf		*bp)
338{
339	struct xfs_buf_log_item	*bip = bp->b_log_item;
340
341	if (!bip)
342		return false;
343	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
344	return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
345}
346
347/*
348 * Release a buffer previously joined to the transaction. If the buffer is
349 * modified within this transaction, decrement the recursion count but do not
350 * release the buffer even if the count goes to 0. If the buffer is not modified
351 * within the transaction, decrement the recursion count and release the buffer
352 * if the recursion count goes to 0.
353 *
354 * If the buffer is to be released and it was not already dirty before this
355 * transaction began, then also free the buf_log_item associated with it.
356 *
357 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
358 */
359void
360xfs_trans_brelse(
361	struct xfs_trans	*tp,
362	struct xfs_buf		*bp)
363{
364	struct xfs_buf_log_item	*bip = bp->b_log_item;
365
366	ASSERT(bp->b_transp == tp);
367
368	if (!tp) {
369		xfs_buf_relse(bp);
370		return;
371	}
372
373	trace_xfs_trans_brelse(bip);
374	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
375	ASSERT(atomic_read(&bip->bli_refcount) > 0);
376
377	/*
378	 * If the release is for a recursive lookup, then decrement the count
379	 * and return.
380	 */
381	if (bip->bli_recur > 0) {
382		bip->bli_recur--;
383		return;
384	}
385
386	/*
387	 * If the buffer is invalidated or dirty in this transaction, we can't
388	 * release it until we commit.
389	 */
390	if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
391		return;
392	if (bip->bli_flags & XFS_BLI_STALE)
393		return;
394
395	/*
396	 * Unlink the log item from the transaction and clear the hold flag, if
397	 * set. We wouldn't want the next user of the buffer to get confused.
398	 */
399	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
400	xfs_trans_del_item(&bip->bli_item);
401	bip->bli_flags &= ~XFS_BLI_HOLD;
402
403	/* drop the reference to the bli */
404	xfs_buf_item_put(bip);
405
406	bp->b_transp = NULL;
407	xfs_buf_relse(bp);
408}
409
410/*
411 * Forcibly detach a buffer previously joined to the transaction.  The caller
412 * will retain its locked reference to the buffer after this function returns.
413 * The buffer must be completely clean and must not be held to the transaction.
414 */
415void
416xfs_trans_bdetach(
417	struct xfs_trans	*tp,
418	struct xfs_buf		*bp)
419{
420	struct xfs_buf_log_item	*bip = bp->b_log_item;
421
422	ASSERT(tp != NULL);
423	ASSERT(bp->b_transp == tp);
424	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
425	ASSERT(atomic_read(&bip->bli_refcount) > 0);
426
427	trace_xfs_trans_bdetach(bip);
428
429	/*
430	 * Erase all recursion count, since we're removing this buffer from the
431	 * transaction.
432	 */
433	bip->bli_recur = 0;
434
435	/*
436	 * The buffer must be completely clean.  Specifically, it had better
437	 * not be dirty, stale, logged, ordered, or held to the transaction.
438	 */
439	ASSERT(!test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
440	ASSERT(!(bip->bli_flags & XFS_BLI_DIRTY));
441	ASSERT(!(bip->bli_flags & XFS_BLI_HOLD));
442	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
443	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
444	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
445
446	/* Unlink the log item from the transaction and drop the log item. */
447	xfs_trans_del_item(&bip->bli_item);
448	xfs_buf_item_put(bip);
449	bp->b_transp = NULL;
450}
451
452/*
453 * Mark the buffer as not needing to be unlocked when the buf item's
454 * iop_committing() routine is called.  The buffer must already be locked
455 * and associated with the given transaction.
456 */
457/* ARGSUSED */
458void
459xfs_trans_bhold(
460	xfs_trans_t		*tp,
461	struct xfs_buf		*bp)
462{
463	struct xfs_buf_log_item	*bip = bp->b_log_item;
464
465	ASSERT(bp->b_transp == tp);
466	ASSERT(bip != NULL);
467	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
468	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
469	ASSERT(atomic_read(&bip->bli_refcount) > 0);
470
471	bip->bli_flags |= XFS_BLI_HOLD;
472	trace_xfs_trans_bhold(bip);
473}
474
475/*
476 * Cancel the previous buffer hold request made on this buffer
477 * for this transaction.
478 */
479void
480xfs_trans_bhold_release(
481	xfs_trans_t		*tp,
482	struct xfs_buf		*bp)
483{
484	struct xfs_buf_log_item	*bip = bp->b_log_item;
485
486	ASSERT(bp->b_transp == tp);
487	ASSERT(bip != NULL);
488	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
489	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
490	ASSERT(atomic_read(&bip->bli_refcount) > 0);
491	ASSERT(bip->bli_flags & XFS_BLI_HOLD);
492
493	bip->bli_flags &= ~XFS_BLI_HOLD;
494	trace_xfs_trans_bhold_release(bip);
495}
496
497/*
498 * Mark a buffer dirty in the transaction.
499 */
500void
501xfs_trans_dirty_buf(
502	struct xfs_trans	*tp,
503	struct xfs_buf		*bp)
504{
505	struct xfs_buf_log_item	*bip = bp->b_log_item;
506
507	ASSERT(bp->b_transp == tp);
508	ASSERT(bip != NULL);
509
510	/*
511	 * Mark the buffer as needing to be written out eventually,
512	 * and set its iodone function to remove the buffer's buf log
513	 * item from the AIL and free it when the buffer is flushed
514	 * to disk.
515	 */
516	bp->b_flags |= XBF_DONE;
517
518	ASSERT(atomic_read(&bip->bli_refcount) > 0);
519
520	/*
521	 * If we invalidated the buffer within this transaction, then
522	 * cancel the invalidation now that we're dirtying the buffer
523	 * again.  There are no races with the code in xfs_buf_item_unpin(),
524	 * because we have a reference to the buffer this entire time.
525	 */
526	if (bip->bli_flags & XFS_BLI_STALE) {
527		bip->bli_flags &= ~XFS_BLI_STALE;
528		ASSERT(bp->b_flags & XBF_STALE);
529		bp->b_flags &= ~XBF_STALE;
530		bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
531	}
532	bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
533
534	tp->t_flags |= XFS_TRANS_DIRTY;
535	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
536}
537
538/*
539 * This is called to mark bytes first through last inclusive of the given
540 * buffer as needing to be logged when the transaction is committed.
541 * The buffer must already be associated with the given transaction.
542 *
543 * First and last are numbers relative to the beginning of this buffer,
544 * so the first byte in the buffer is numbered 0 regardless of the
545 * value of b_blkno.
546 */
547void
548xfs_trans_log_buf(
549	struct xfs_trans	*tp,
550	struct xfs_buf		*bp,
551	uint			first,
552	uint			last)
553{
554	struct xfs_buf_log_item	*bip = bp->b_log_item;
555
556	ASSERT(first <= last && last < BBTOB(bp->b_length));
557	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
558
559	xfs_trans_dirty_buf(tp, bp);
560
561	trace_xfs_trans_log_buf(bip);
562	xfs_buf_item_log(bip, first, last);
563}
564
565
566/*
567 * Invalidate a buffer that is being used within a transaction.
568 *
569 * Typically this is because the blocks in the buffer are being freed, so we
570 * need to prevent it from being written out when we're done.  Allowing it
571 * to be written again might overwrite data in the free blocks if they are
572 * reallocated to a file.
573 *
574 * We prevent the buffer from being written out by marking it stale.  We can't
575 * get rid of the buf log item at this point because the buffer may still be
576 * pinned by another transaction.  If that is the case, then we'll wait until
577 * the buffer is committed to disk for the last time (we can tell by the ref
578 * count) and free it in xfs_buf_item_unpin().  Until that happens we will
579 * keep the buffer locked so that the buffer and buf log item are not reused.
580 *
581 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
582 * the buf item.  This will be used at recovery time to determine that copies
583 * of the buffer in the log before this should not be replayed.
584 *
585 * We mark the item descriptor and the transaction dirty so that we'll hold
586 * the buffer until after the commit.
587 *
588 * Since we're invalidating the buffer, we also clear the state about which
589 * parts of the buffer have been logged.  We also clear the flag indicating
590 * that this is an inode buffer since the data in the buffer will no longer
591 * be valid.
592 *
593 * We set the stale bit in the buffer as well since we're getting rid of it.
594 */
595void
596xfs_trans_binval(
597	xfs_trans_t		*tp,
598	struct xfs_buf		*bp)
599{
600	struct xfs_buf_log_item	*bip = bp->b_log_item;
601	int			i;
602
603	ASSERT(bp->b_transp == tp);
604	ASSERT(bip != NULL);
605	ASSERT(atomic_read(&bip->bli_refcount) > 0);
606
607	trace_xfs_trans_binval(bip);
608
609	if (bip->bli_flags & XFS_BLI_STALE) {
610		/*
611		 * If the buffer is already invalidated, then
612		 * just return.
613		 */
614		ASSERT(bp->b_flags & XBF_STALE);
615		ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
616		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
617		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
618		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
619		ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
620		ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
621		return;
622	}
623
624	xfs_buf_stale(bp);
625
626	bip->bli_flags |= XFS_BLI_STALE;
627	bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
628	bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
629	bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
630	bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
631	for (i = 0; i < bip->bli_format_count; i++) {
632		memset(bip->bli_formats[i].blf_data_map, 0,
633		       (bip->bli_formats[i].blf_map_size * sizeof(uint)));
634	}
635	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
636	tp->t_flags |= XFS_TRANS_DIRTY;
637}
638
639/*
640 * This call is used to indicate that the buffer contains on-disk inodes which
641 * must be handled specially during recovery.  They require special handling
642 * because only the di_next_unlinked from the inodes in the buffer should be
643 * recovered.  The rest of the data in the buffer is logged via the inodes
644 * themselves.
645 *
646 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
647 * transferred to the buffer's log format structure so that we'll know what to
648 * do at recovery time.
649 */
650void
651xfs_trans_inode_buf(
652	xfs_trans_t		*tp,
653	struct xfs_buf		*bp)
654{
655	struct xfs_buf_log_item	*bip = bp->b_log_item;
656
657	ASSERT(bp->b_transp == tp);
658	ASSERT(bip != NULL);
659	ASSERT(atomic_read(&bip->bli_refcount) > 0);
660
661	bip->bli_flags |= XFS_BLI_INODE_BUF;
662	bp->b_flags |= _XBF_INODES;
663	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
664}
665
666/*
667 * This call is used to indicate that the buffer is going to
668 * be staled and was an inode buffer. This means it gets
669 * special processing during unpin - where any inodes
670 * associated with the buffer should be removed from ail.
671 * There is also special processing during recovery,
672 * any replay of the inodes in the buffer needs to be
673 * prevented as the buffer may have been reused.
674 */
675void
676xfs_trans_stale_inode_buf(
677	xfs_trans_t		*tp,
678	struct xfs_buf		*bp)
679{
680	struct xfs_buf_log_item	*bip = bp->b_log_item;
681
682	ASSERT(bp->b_transp == tp);
683	ASSERT(bip != NULL);
684	ASSERT(atomic_read(&bip->bli_refcount) > 0);
685
686	bip->bli_flags |= XFS_BLI_STALE_INODE;
687	bp->b_flags |= _XBF_INODES;
688	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
689}
690
691/*
692 * Mark the buffer as being one which contains newly allocated
693 * inodes.  We need to make sure that even if this buffer is
694 * relogged as an 'inode buf' we still recover all of the inode
695 * images in the face of a crash.  This works in coordination with
696 * xfs_buf_item_committed() to ensure that the buffer remains in the
697 * AIL at its original location even after it has been relogged.
698 */
699/* ARGSUSED */
700void
701xfs_trans_inode_alloc_buf(
702	xfs_trans_t		*tp,
703	struct xfs_buf		*bp)
704{
705	struct xfs_buf_log_item	*bip = bp->b_log_item;
706
707	ASSERT(bp->b_transp == tp);
708	ASSERT(bip != NULL);
709	ASSERT(atomic_read(&bip->bli_refcount) > 0);
710
711	bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
712	bp->b_flags |= _XBF_INODES;
713	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
714}
715
716/*
717 * Mark the buffer as ordered for this transaction. This means that the contents
718 * of the buffer are not recorded in the transaction but it is tracked in the
719 * AIL as though it was. This allows us to record logical changes in
720 * transactions rather than the physical changes we make to the buffer without
721 * changing writeback ordering constraints of metadata buffers.
722 */
723bool
724xfs_trans_ordered_buf(
725	struct xfs_trans	*tp,
726	struct xfs_buf		*bp)
727{
728	struct xfs_buf_log_item	*bip = bp->b_log_item;
729
730	ASSERT(bp->b_transp == tp);
731	ASSERT(bip != NULL);
732	ASSERT(atomic_read(&bip->bli_refcount) > 0);
733
734	if (xfs_buf_item_dirty_format(bip))
735		return false;
736
737	bip->bli_flags |= XFS_BLI_ORDERED;
738	trace_xfs_buf_item_ordered(bip);
739
740	/*
741	 * We don't log a dirty range of an ordered buffer but it still needs
742	 * to be marked dirty and that it has been logged.
743	 */
744	xfs_trans_dirty_buf(tp, bp);
745	return true;
746}
747
748/*
749 * Set the type of the buffer for log recovery so that it can correctly identify
750 * and hence attach the correct buffer ops to the buffer after replay.
751 */
752void
753xfs_trans_buf_set_type(
754	struct xfs_trans	*tp,
755	struct xfs_buf		*bp,
756	enum xfs_blft		type)
757{
758	struct xfs_buf_log_item	*bip = bp->b_log_item;
759
760	if (!tp)
761		return;
762
763	ASSERT(bp->b_transp == tp);
764	ASSERT(bip != NULL);
765	ASSERT(atomic_read(&bip->bli_refcount) > 0);
766
767	xfs_blft_to_flags(&bip->__bli_format, type);
768}
769
770void
771xfs_trans_buf_copy_type(
772	struct xfs_buf		*dst_bp,
773	struct xfs_buf		*src_bp)
774{
775	struct xfs_buf_log_item	*sbip = src_bp->b_log_item;
776	struct xfs_buf_log_item	*dbip = dst_bp->b_log_item;
777	enum xfs_blft		type;
778
779	type = xfs_blft_from_flags(&sbip->__bli_format);
780	xfs_blft_to_flags(&dbip->__bli_format, type);
781}
782
783/*
784 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
785 * dquots. However, unlike in inode buffer recovery, dquot buffers get
786 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
787 * The only thing that makes dquot buffers different from regular
788 * buffers is that we must not replay dquot bufs when recovering
789 * if a _corresponding_ quotaoff has happened. We also have to distinguish
790 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
791 * can be turned off independently.
792 */
793/* ARGSUSED */
794void
795xfs_trans_dquot_buf(
796	xfs_trans_t		*tp,
797	struct xfs_buf		*bp,
798	uint			type)
799{
800	struct xfs_buf_log_item	*bip = bp->b_log_item;
801
802	ASSERT(type == XFS_BLF_UDQUOT_BUF ||
803	       type == XFS_BLF_PDQUOT_BUF ||
804	       type == XFS_BLF_GDQUOT_BUF);
805
806	bip->__bli_format.blf_flags |= type;
807
808	switch (type) {
809	case XFS_BLF_UDQUOT_BUF:
810		type = XFS_BLFT_UDQUOT_BUF;
811		break;
812	case XFS_BLF_PDQUOT_BUF:
813		type = XFS_BLFT_PDQUOT_BUF;
814		break;
815	case XFS_BLF_GDQUOT_BUF:
816		type = XFS_BLFT_GDQUOT_BUF;
817		break;
818	default:
819		type = XFS_BLFT_UNKNOWN_BUF;
820		break;
821	}
822
823	bp->b_flags |= _XBF_DQUOTS;
824	xfs_trans_buf_set_type(tp, bp, type);
825}
v5.9
  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_trans.h"
 14#include "xfs_buf_item.h"
 15#include "xfs_trans_priv.h"
 16#include "xfs_trace.h"
 17
 18/*
 19 * Check to see if a buffer matching the given parameters is already
 20 * a part of the given transaction.
 21 */
 22STATIC struct xfs_buf *
 23xfs_trans_buf_item_match(
 24	struct xfs_trans	*tp,
 25	struct xfs_buftarg	*target,
 26	struct xfs_buf_map	*map,
 27	int			nmaps)
 28{
 29	struct xfs_log_item	*lip;
 30	struct xfs_buf_log_item	*blip;
 31	int			len = 0;
 32	int			i;
 33
 34	for (i = 0; i < nmaps; i++)
 35		len += map[i].bm_len;
 36
 37	list_for_each_entry(lip, &tp->t_items, li_trans) {
 38		blip = (struct xfs_buf_log_item *)lip;
 39		if (blip->bli_item.li_type == XFS_LI_BUF &&
 40		    blip->bli_buf->b_target == target &&
 41		    XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
 42		    blip->bli_buf->b_length == len) {
 43			ASSERT(blip->bli_buf->b_map_count == nmaps);
 44			return blip->bli_buf;
 45		}
 46	}
 47
 48	return NULL;
 49}
 50
 51/*
 52 * Add the locked buffer to the transaction.
 53 *
 54 * The buffer must be locked, and it cannot be associated with any
 55 * transaction.
 56 *
 57 * If the buffer does not yet have a buf log item associated with it,
 58 * then allocate one for it.  Then add the buf item to the transaction.
 59 */
 60STATIC void
 61_xfs_trans_bjoin(
 62	struct xfs_trans	*tp,
 63	struct xfs_buf		*bp,
 64	int			reset_recur)
 65{
 66	struct xfs_buf_log_item	*bip;
 67
 68	ASSERT(bp->b_transp == NULL);
 69
 70	/*
 71	 * The xfs_buf_log_item pointer is stored in b_log_item.  If
 72	 * it doesn't have one yet, then allocate one and initialize it.
 73	 * The checks to see if one is there are in xfs_buf_item_init().
 74	 */
 75	xfs_buf_item_init(bp, tp->t_mountp);
 76	bip = bp->b_log_item;
 77	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 78	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
 79	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
 80	if (reset_recur)
 81		bip->bli_recur = 0;
 82
 83	/*
 84	 * Take a reference for this transaction on the buf item.
 85	 */
 86	atomic_inc(&bip->bli_refcount);
 87
 88	/*
 89	 * Attach the item to the transaction so we can find it in
 90	 * xfs_trans_get_buf() and friends.
 91	 */
 92	xfs_trans_add_item(tp, &bip->bli_item);
 93	bp->b_transp = tp;
 94
 95}
 96
 97void
 98xfs_trans_bjoin(
 99	struct xfs_trans	*tp,
100	struct xfs_buf		*bp)
101{
102	_xfs_trans_bjoin(tp, bp, 0);
103	trace_xfs_trans_bjoin(bp->b_log_item);
104}
105
106/*
107 * Get and lock the buffer for the caller if it is not already
108 * locked within the given transaction.  If it is already locked
109 * within the transaction, just increment its lock recursion count
110 * and return a pointer to it.
111 *
112 * If the transaction pointer is NULL, make this just a normal
113 * get_buf() call.
114 */
115int
116xfs_trans_get_buf_map(
117	struct xfs_trans	*tp,
118	struct xfs_buftarg	*target,
119	struct xfs_buf_map	*map,
120	int			nmaps,
121	xfs_buf_flags_t		flags,
122	struct xfs_buf		**bpp)
123{
124	xfs_buf_t		*bp;
125	struct xfs_buf_log_item	*bip;
126	int			error;
127
128	*bpp = NULL;
129	if (!tp)
130		return xfs_buf_get_map(target, map, nmaps, flags, bpp);
131
132	/*
133	 * If we find the buffer in the cache with this transaction
134	 * pointer in its b_fsprivate2 field, then we know we already
135	 * have it locked.  In this case we just increment the lock
136	 * recursion count and return the buffer to the caller.
137	 */
138	bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
139	if (bp != NULL) {
140		ASSERT(xfs_buf_islocked(bp));
141		if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
142			xfs_buf_stale(bp);
143			bp->b_flags |= XBF_DONE;
144		}
145
146		ASSERT(bp->b_transp == tp);
147		bip = bp->b_log_item;
148		ASSERT(bip != NULL);
149		ASSERT(atomic_read(&bip->bli_refcount) > 0);
150		bip->bli_recur++;
151		trace_xfs_trans_get_buf_recur(bip);
152		*bpp = bp;
153		return 0;
154	}
155
156	error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
157	if (error)
158		return error;
159
160	ASSERT(!bp->b_error);
161
162	_xfs_trans_bjoin(tp, bp, 1);
163	trace_xfs_trans_get_buf(bp->b_log_item);
164	*bpp = bp;
165	return 0;
166}
167
168/*
169 * Get and lock the superblock buffer of this file system for the
170 * given transaction.
171 *
172 * We don't need to use incore_match() here, because the superblock
173 * buffer is a private buffer which we keep a pointer to in the
174 * mount structure.
175 */
176xfs_buf_t *
177xfs_trans_getsb(
178	xfs_trans_t		*tp,
179	struct xfs_mount	*mp)
180{
181	xfs_buf_t		*bp;
182	struct xfs_buf_log_item	*bip;
183
184	/*
185	 * Default to just trying to lock the superblock buffer
186	 * if tp is NULL.
187	 */
188	if (tp == NULL)
189		return xfs_getsb(mp);
190
191	/*
192	 * If the superblock buffer already has this transaction
193	 * pointer in its b_fsprivate2 field, then we know we already
194	 * have it locked.  In this case we just increment the lock
195	 * recursion count and return the buffer to the caller.
196	 */
197	bp = mp->m_sb_bp;
198	if (bp->b_transp == tp) {
199		bip = bp->b_log_item;
200		ASSERT(bip != NULL);
201		ASSERT(atomic_read(&bip->bli_refcount) > 0);
202		bip->bli_recur++;
 
203		trace_xfs_trans_getsb_recur(bip);
204		return bp;
 
 
 
 
 
205	}
206
207	bp = xfs_getsb(mp);
208	if (bp == NULL)
 
 
 
 
 
 
 
 
 
 
 
 
 
209		return NULL;
210
211	_xfs_trans_bjoin(tp, bp, 1);
212	trace_xfs_trans_getsb(bp->b_log_item);
213	return bp;
214}
215
216/*
217 * Get and lock the buffer for the caller if it is not already
218 * locked within the given transaction.  If it has not yet been
219 * read in, read it from disk. If it is already locked
220 * within the transaction and already read in, just increment its
221 * lock recursion count and return a pointer to it.
222 *
223 * If the transaction pointer is NULL, make this just a normal
224 * read_buf() call.
225 */
226int
227xfs_trans_read_buf_map(
228	struct xfs_mount	*mp,
229	struct xfs_trans	*tp,
230	struct xfs_buftarg	*target,
231	struct xfs_buf_map	*map,
232	int			nmaps,
233	xfs_buf_flags_t		flags,
234	struct xfs_buf		**bpp,
235	const struct xfs_buf_ops *ops)
236{
237	struct xfs_buf		*bp = NULL;
238	struct xfs_buf_log_item	*bip;
239	int			error;
240
241	*bpp = NULL;
242	/*
243	 * If we find the buffer in the cache with this transaction
244	 * pointer in its b_fsprivate2 field, then we know we already
245	 * have it locked.  If it is already read in we just increment
246	 * the lock recursion count and return the buffer to the caller.
247	 * If the buffer is not yet read in, then we read it in, increment
248	 * the lock recursion count, and return it to the caller.
249	 */
250	if (tp)
251		bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
252	if (bp) {
253		ASSERT(xfs_buf_islocked(bp));
254		ASSERT(bp->b_transp == tp);
255		ASSERT(bp->b_log_item != NULL);
256		ASSERT(!bp->b_error);
257		ASSERT(bp->b_flags & XBF_DONE);
258
259		/*
260		 * We never locked this buf ourselves, so we shouldn't
261		 * brelse it either. Just get out.
262		 */
263		if (XFS_FORCED_SHUTDOWN(mp)) {
264			trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
265			return -EIO;
266		}
267
268		/*
269		 * Check if the caller is trying to read a buffer that is
270		 * already attached to the transaction yet has no buffer ops
271		 * assigned.  Ops are usually attached when the buffer is
272		 * attached to the transaction, or by the read caller if
273		 * special circumstances.  That didn't happen, which is not
274		 * how this is supposed to go.
275		 *
276		 * If the buffer passes verification we'll let this go, but if
277		 * not we have to shut down.  Let the transaction cleanup code
278		 * release this buffer when it kills the tranaction.
279		 */
280		ASSERT(bp->b_ops != NULL);
281		error = xfs_buf_reverify(bp, ops);
282		if (error) {
283			xfs_buf_ioerror_alert(bp, __return_address);
284
285			if (tp->t_flags & XFS_TRANS_DIRTY)
286				xfs_force_shutdown(tp->t_mountp,
287						SHUTDOWN_META_IO_ERROR);
288
289			/* bad CRC means corrupted metadata */
290			if (error == -EFSBADCRC)
291				error = -EFSCORRUPTED;
292			return error;
293		}
294
295		bip = bp->b_log_item;
296		bip->bli_recur++;
297
298		ASSERT(atomic_read(&bip->bli_refcount) > 0);
299		trace_xfs_trans_read_buf_recur(bip);
300		ASSERT(bp->b_ops != NULL || ops == NULL);
301		*bpp = bp;
302		return 0;
303	}
304
305	error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
306			__return_address);
307	switch (error) {
308	case 0:
309		break;
310	default:
311		if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
312			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
313		/* fall through */
314	case -ENOMEM:
315	case -EAGAIN:
316		return error;
317	}
318
319	if (XFS_FORCED_SHUTDOWN(mp)) {
320		xfs_buf_relse(bp);
321		trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
322		return -EIO;
323	}
324
325	if (tp) {
326		_xfs_trans_bjoin(tp, bp, 1);
327		trace_xfs_trans_read_buf(bp->b_log_item);
328	}
329	ASSERT(bp->b_ops != NULL || ops == NULL);
330	*bpp = bp;
331	return 0;
332
333}
334
335/* Has this buffer been dirtied by anyone? */
336bool
337xfs_trans_buf_is_dirty(
338	struct xfs_buf		*bp)
339{
340	struct xfs_buf_log_item	*bip = bp->b_log_item;
341
342	if (!bip)
343		return false;
344	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
345	return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
346}
347
348/*
349 * Release a buffer previously joined to the transaction. If the buffer is
350 * modified within this transaction, decrement the recursion count but do not
351 * release the buffer even if the count goes to 0. If the buffer is not modified
352 * within the transaction, decrement the recursion count and release the buffer
353 * if the recursion count goes to 0.
354 *
355 * If the buffer is to be released and it was not already dirty before this
356 * transaction began, then also free the buf_log_item associated with it.
357 *
358 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
359 */
360void
361xfs_trans_brelse(
362	struct xfs_trans	*tp,
363	struct xfs_buf		*bp)
364{
365	struct xfs_buf_log_item	*bip = bp->b_log_item;
366
367	ASSERT(bp->b_transp == tp);
368
369	if (!tp) {
370		xfs_buf_relse(bp);
371		return;
372	}
373
374	trace_xfs_trans_brelse(bip);
375	ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
376	ASSERT(atomic_read(&bip->bli_refcount) > 0);
377
378	/*
379	 * If the release is for a recursive lookup, then decrement the count
380	 * and return.
381	 */
382	if (bip->bli_recur > 0) {
383		bip->bli_recur--;
384		return;
385	}
386
387	/*
388	 * If the buffer is invalidated or dirty in this transaction, we can't
389	 * release it until we commit.
390	 */
391	if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
392		return;
393	if (bip->bli_flags & XFS_BLI_STALE)
394		return;
395
396	/*
397	 * Unlink the log item from the transaction and clear the hold flag, if
398	 * set. We wouldn't want the next user of the buffer to get confused.
399	 */
400	ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
401	xfs_trans_del_item(&bip->bli_item);
402	bip->bli_flags &= ~XFS_BLI_HOLD;
403
404	/* drop the reference to the bli */
405	xfs_buf_item_put(bip);
406
407	bp->b_transp = NULL;
408	xfs_buf_relse(bp);
409}
410
411/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
412 * Mark the buffer as not needing to be unlocked when the buf item's
413 * iop_committing() routine is called.  The buffer must already be locked
414 * and associated with the given transaction.
415 */
416/* ARGSUSED */
417void
418xfs_trans_bhold(
419	xfs_trans_t		*tp,
420	xfs_buf_t		*bp)
421{
422	struct xfs_buf_log_item	*bip = bp->b_log_item;
423
424	ASSERT(bp->b_transp == tp);
425	ASSERT(bip != NULL);
426	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
427	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
428	ASSERT(atomic_read(&bip->bli_refcount) > 0);
429
430	bip->bli_flags |= XFS_BLI_HOLD;
431	trace_xfs_trans_bhold(bip);
432}
433
434/*
435 * Cancel the previous buffer hold request made on this buffer
436 * for this transaction.
437 */
438void
439xfs_trans_bhold_release(
440	xfs_trans_t		*tp,
441	xfs_buf_t		*bp)
442{
443	struct xfs_buf_log_item	*bip = bp->b_log_item;
444
445	ASSERT(bp->b_transp == tp);
446	ASSERT(bip != NULL);
447	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
448	ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
449	ASSERT(atomic_read(&bip->bli_refcount) > 0);
450	ASSERT(bip->bli_flags & XFS_BLI_HOLD);
451
452	bip->bli_flags &= ~XFS_BLI_HOLD;
453	trace_xfs_trans_bhold_release(bip);
454}
455
456/*
457 * Mark a buffer dirty in the transaction.
458 */
459void
460xfs_trans_dirty_buf(
461	struct xfs_trans	*tp,
462	struct xfs_buf		*bp)
463{
464	struct xfs_buf_log_item	*bip = bp->b_log_item;
465
466	ASSERT(bp->b_transp == tp);
467	ASSERT(bip != NULL);
468
469	/*
470	 * Mark the buffer as needing to be written out eventually,
471	 * and set its iodone function to remove the buffer's buf log
472	 * item from the AIL and free it when the buffer is flushed
473	 * to disk.
474	 */
475	bp->b_flags |= XBF_DONE;
476
477	ASSERT(atomic_read(&bip->bli_refcount) > 0);
478
479	/*
480	 * If we invalidated the buffer within this transaction, then
481	 * cancel the invalidation now that we're dirtying the buffer
482	 * again.  There are no races with the code in xfs_buf_item_unpin(),
483	 * because we have a reference to the buffer this entire time.
484	 */
485	if (bip->bli_flags & XFS_BLI_STALE) {
486		bip->bli_flags &= ~XFS_BLI_STALE;
487		ASSERT(bp->b_flags & XBF_STALE);
488		bp->b_flags &= ~XBF_STALE;
489		bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
490	}
491	bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
492
493	tp->t_flags |= XFS_TRANS_DIRTY;
494	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
495}
496
497/*
498 * This is called to mark bytes first through last inclusive of the given
499 * buffer as needing to be logged when the transaction is committed.
500 * The buffer must already be associated with the given transaction.
501 *
502 * First and last are numbers relative to the beginning of this buffer,
503 * so the first byte in the buffer is numbered 0 regardless of the
504 * value of b_blkno.
505 */
506void
507xfs_trans_log_buf(
508	struct xfs_trans	*tp,
509	struct xfs_buf		*bp,
510	uint			first,
511	uint			last)
512{
513	struct xfs_buf_log_item	*bip = bp->b_log_item;
514
515	ASSERT(first <= last && last < BBTOB(bp->b_length));
516	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
517
518	xfs_trans_dirty_buf(tp, bp);
519
520	trace_xfs_trans_log_buf(bip);
521	xfs_buf_item_log(bip, first, last);
522}
523
524
525/*
526 * Invalidate a buffer that is being used within a transaction.
527 *
528 * Typically this is because the blocks in the buffer are being freed, so we
529 * need to prevent it from being written out when we're done.  Allowing it
530 * to be written again might overwrite data in the free blocks if they are
531 * reallocated to a file.
532 *
533 * We prevent the buffer from being written out by marking it stale.  We can't
534 * get rid of the buf log item at this point because the buffer may still be
535 * pinned by another transaction.  If that is the case, then we'll wait until
536 * the buffer is committed to disk for the last time (we can tell by the ref
537 * count) and free it in xfs_buf_item_unpin().  Until that happens we will
538 * keep the buffer locked so that the buffer and buf log item are not reused.
539 *
540 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
541 * the buf item.  This will be used at recovery time to determine that copies
542 * of the buffer in the log before this should not be replayed.
543 *
544 * We mark the item descriptor and the transaction dirty so that we'll hold
545 * the buffer until after the commit.
546 *
547 * Since we're invalidating the buffer, we also clear the state about which
548 * parts of the buffer have been logged.  We also clear the flag indicating
549 * that this is an inode buffer since the data in the buffer will no longer
550 * be valid.
551 *
552 * We set the stale bit in the buffer as well since we're getting rid of it.
553 */
554void
555xfs_trans_binval(
556	xfs_trans_t		*tp,
557	xfs_buf_t		*bp)
558{
559	struct xfs_buf_log_item	*bip = bp->b_log_item;
560	int			i;
561
562	ASSERT(bp->b_transp == tp);
563	ASSERT(bip != NULL);
564	ASSERT(atomic_read(&bip->bli_refcount) > 0);
565
566	trace_xfs_trans_binval(bip);
567
568	if (bip->bli_flags & XFS_BLI_STALE) {
569		/*
570		 * If the buffer is already invalidated, then
571		 * just return.
572		 */
573		ASSERT(bp->b_flags & XBF_STALE);
574		ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
575		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
576		ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
577		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
578		ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
579		ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
580		return;
581	}
582
583	xfs_buf_stale(bp);
584
585	bip->bli_flags |= XFS_BLI_STALE;
586	bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
587	bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
588	bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
589	bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
590	for (i = 0; i < bip->bli_format_count; i++) {
591		memset(bip->bli_formats[i].blf_data_map, 0,
592		       (bip->bli_formats[i].blf_map_size * sizeof(uint)));
593	}
594	set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
595	tp->t_flags |= XFS_TRANS_DIRTY;
596}
597
598/*
599 * This call is used to indicate that the buffer contains on-disk inodes which
600 * must be handled specially during recovery.  They require special handling
601 * because only the di_next_unlinked from the inodes in the buffer should be
602 * recovered.  The rest of the data in the buffer is logged via the inodes
603 * themselves.
604 *
605 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
606 * transferred to the buffer's log format structure so that we'll know what to
607 * do at recovery time.
608 */
609void
610xfs_trans_inode_buf(
611	xfs_trans_t		*tp,
612	xfs_buf_t		*bp)
613{
614	struct xfs_buf_log_item	*bip = bp->b_log_item;
615
616	ASSERT(bp->b_transp == tp);
617	ASSERT(bip != NULL);
618	ASSERT(atomic_read(&bip->bli_refcount) > 0);
619
620	bip->bli_flags |= XFS_BLI_INODE_BUF;
621	bp->b_flags |= _XBF_INODES;
622	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
623}
624
625/*
626 * This call is used to indicate that the buffer is going to
627 * be staled and was an inode buffer. This means it gets
628 * special processing during unpin - where any inodes
629 * associated with the buffer should be removed from ail.
630 * There is also special processing during recovery,
631 * any replay of the inodes in the buffer needs to be
632 * prevented as the buffer may have been reused.
633 */
634void
635xfs_trans_stale_inode_buf(
636	xfs_trans_t		*tp,
637	xfs_buf_t		*bp)
638{
639	struct xfs_buf_log_item	*bip = bp->b_log_item;
640
641	ASSERT(bp->b_transp == tp);
642	ASSERT(bip != NULL);
643	ASSERT(atomic_read(&bip->bli_refcount) > 0);
644
645	bip->bli_flags |= XFS_BLI_STALE_INODE;
646	bp->b_flags |= _XBF_INODES;
647	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
648}
649
650/*
651 * Mark the buffer as being one which contains newly allocated
652 * inodes.  We need to make sure that even if this buffer is
653 * relogged as an 'inode buf' we still recover all of the inode
654 * images in the face of a crash.  This works in coordination with
655 * xfs_buf_item_committed() to ensure that the buffer remains in the
656 * AIL at its original location even after it has been relogged.
657 */
658/* ARGSUSED */
659void
660xfs_trans_inode_alloc_buf(
661	xfs_trans_t		*tp,
662	xfs_buf_t		*bp)
663{
664	struct xfs_buf_log_item	*bip = bp->b_log_item;
665
666	ASSERT(bp->b_transp == tp);
667	ASSERT(bip != NULL);
668	ASSERT(atomic_read(&bip->bli_refcount) > 0);
669
670	bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
671	bp->b_flags |= _XBF_INODES;
672	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
673}
674
675/*
676 * Mark the buffer as ordered for this transaction. This means that the contents
677 * of the buffer are not recorded in the transaction but it is tracked in the
678 * AIL as though it was. This allows us to record logical changes in
679 * transactions rather than the physical changes we make to the buffer without
680 * changing writeback ordering constraints of metadata buffers.
681 */
682bool
683xfs_trans_ordered_buf(
684	struct xfs_trans	*tp,
685	struct xfs_buf		*bp)
686{
687	struct xfs_buf_log_item	*bip = bp->b_log_item;
688
689	ASSERT(bp->b_transp == tp);
690	ASSERT(bip != NULL);
691	ASSERT(atomic_read(&bip->bli_refcount) > 0);
692
693	if (xfs_buf_item_dirty_format(bip))
694		return false;
695
696	bip->bli_flags |= XFS_BLI_ORDERED;
697	trace_xfs_buf_item_ordered(bip);
698
699	/*
700	 * We don't log a dirty range of an ordered buffer but it still needs
701	 * to be marked dirty and that it has been logged.
702	 */
703	xfs_trans_dirty_buf(tp, bp);
704	return true;
705}
706
707/*
708 * Set the type of the buffer for log recovery so that it can correctly identify
709 * and hence attach the correct buffer ops to the buffer after replay.
710 */
711void
712xfs_trans_buf_set_type(
713	struct xfs_trans	*tp,
714	struct xfs_buf		*bp,
715	enum xfs_blft		type)
716{
717	struct xfs_buf_log_item	*bip = bp->b_log_item;
718
719	if (!tp)
720		return;
721
722	ASSERT(bp->b_transp == tp);
723	ASSERT(bip != NULL);
724	ASSERT(atomic_read(&bip->bli_refcount) > 0);
725
726	xfs_blft_to_flags(&bip->__bli_format, type);
727}
728
729void
730xfs_trans_buf_copy_type(
731	struct xfs_buf		*dst_bp,
732	struct xfs_buf		*src_bp)
733{
734	struct xfs_buf_log_item	*sbip = src_bp->b_log_item;
735	struct xfs_buf_log_item	*dbip = dst_bp->b_log_item;
736	enum xfs_blft		type;
737
738	type = xfs_blft_from_flags(&sbip->__bli_format);
739	xfs_blft_to_flags(&dbip->__bli_format, type);
740}
741
742/*
743 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
744 * dquots. However, unlike in inode buffer recovery, dquot buffers get
745 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
746 * The only thing that makes dquot buffers different from regular
747 * buffers is that we must not replay dquot bufs when recovering
748 * if a _corresponding_ quotaoff has happened. We also have to distinguish
749 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
750 * can be turned off independently.
751 */
752/* ARGSUSED */
753void
754xfs_trans_dquot_buf(
755	xfs_trans_t		*tp,
756	xfs_buf_t		*bp,
757	uint			type)
758{
759	struct xfs_buf_log_item	*bip = bp->b_log_item;
760
761	ASSERT(type == XFS_BLF_UDQUOT_BUF ||
762	       type == XFS_BLF_PDQUOT_BUF ||
763	       type == XFS_BLF_GDQUOT_BUF);
764
765	bip->__bli_format.blf_flags |= type;
766
767	switch (type) {
768	case XFS_BLF_UDQUOT_BUF:
769		type = XFS_BLFT_UDQUOT_BUF;
770		break;
771	case XFS_BLF_PDQUOT_BUF:
772		type = XFS_BLFT_PDQUOT_BUF;
773		break;
774	case XFS_BLF_GDQUOT_BUF:
775		type = XFS_BLFT_GDQUOT_BUF;
776		break;
777	default:
778		type = XFS_BLFT_UNKNOWN_BUF;
779		break;
780	}
781
782	bp->b_flags |= _XBF_DQUOTS;
783	xfs_trans_buf_set_type(tp, bp, type);
784}