1/*
2 * Copyright (c) 2000-2001,2005 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_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
23#include "xfs_mount.h"
24#include "xfs_trans.h"
25#include "xfs_trans_priv.h"
26#include "xfs_buf_item.h"
27#include "xfs_extfree_item.h"
28#include "xfs_log.h"
29
30
31kmem_zone_t	*xfs_efi_zone;
32kmem_zone_t	*xfs_efd_zone;
33
34static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
35{
36	return container_of(lip, struct xfs_efi_log_item, efi_item);
37}
38
39void
40xfs_efi_item_free(
41	struct xfs_efi_log_item	*efip)
42{
43	if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
44		kmem_free(efip);
45	else
46		kmem_zone_free(xfs_efi_zone, efip);
47}
48
49/*
50 * This returns the number of iovecs needed to log the given efi item.
51 * We only need 1 iovec for an efi item.  It just logs the efi_log_format
52 * structure.
53 */
54static inline int
55xfs_efi_item_sizeof(
56	struct xfs_efi_log_item *efip)
57{
58	return sizeof(struct xfs_efi_log_format) +
59	       (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
60}
61
62STATIC void
63xfs_efi_item_size(
64	struct xfs_log_item	*lip,
65	int			*nvecs,
66	int			*nbytes)
67{
68	*nvecs += 1;
69	*nbytes += xfs_efi_item_sizeof(EFI_ITEM(lip));
70}
71
72/*
73 * This is called to fill in the vector of log iovecs for the
74 * given efi log item. We use only 1 iovec, and we point that
75 * at the efi_log_format structure embedded in the efi item.
76 * It is at this point that we assert that all of the extent
77 * slots in the efi item have been filled.
78 */
79STATIC void
80xfs_efi_item_format(
81	struct xfs_log_item	*lip,
82	struct xfs_log_vec	*lv)
83{
84	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
85	struct xfs_log_iovec	*vecp = NULL;
86
87	ASSERT(atomic_read(&efip->efi_next_extent) ==
88				efip->efi_format.efi_nextents);
89
90	efip->efi_format.efi_type = XFS_LI_EFI;
91	efip->efi_format.efi_size = 1;
92
93	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
94			&efip->efi_format,
95			xfs_efi_item_sizeof(efip));
96}
97
98
99/*
100 * Pinning has no meaning for an efi item, so just return.
101 */
102STATIC void
103xfs_efi_item_pin(
104	struct xfs_log_item	*lip)
105{
106}
107
108/*
109 * The unpin operation is the last place an EFI is manipulated in the log. It is
110 * either inserted in the AIL or aborted in the event of a log I/O error. In
111 * either case, the EFI transaction has been successfully committed to make it
112 * this far. Therefore, we expect whoever committed the EFI to either construct
113 * and commit the EFD or drop the EFD's reference in the event of error. Simply
114 * drop the log's EFI reference now that the log is done with it.
115 */
116STATIC void
117xfs_efi_item_unpin(
118	struct xfs_log_item	*lip,
119	int			remove)
120{
121	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
122	xfs_efi_release(efip);
123}
124
125/*
126 * Efi items have no locking or pushing.  However, since EFIs are pulled from
127 * the AIL when their corresponding EFDs are committed to disk, their situation
128 * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
129 * will eventually flush the log.  This should help in getting the EFI out of
130 * the AIL.
131 */
132STATIC uint
133xfs_efi_item_push(
134	struct xfs_log_item	*lip,
135	struct list_head	*buffer_list)
136{
137	return XFS_ITEM_PINNED;
138}
139
140/*
141 * The EFI has been either committed or aborted if the transaction has been
142 * cancelled. If the transaction was cancelled, an EFD isn't going to be
143 * constructed and thus we free the EFI here directly.
144 */
145STATIC void
146xfs_efi_item_unlock(
147	struct xfs_log_item	*lip)
148{
149	if (lip->li_flags & XFS_LI_ABORTED)
150		xfs_efi_item_free(EFI_ITEM(lip));
151}
152
153/*
154 * The EFI is logged only once and cannot be moved in the log, so simply return
155 * the lsn at which it's been logged.
156 */
157STATIC xfs_lsn_t
158xfs_efi_item_committed(
159	struct xfs_log_item	*lip,
160	xfs_lsn_t		lsn)
161{
162	return lsn;
163}
164
165/*
166 * The EFI dependency tracking op doesn't do squat.  It can't because
167 * it doesn't know where the free extent is coming from.  The dependency
168 * tracking has to be handled by the "enclosing" metadata object.  For
169 * example, for inodes, the inode is locked throughout the extent freeing
170 * so the dependency should be recorded there.
171 */
172STATIC void
173xfs_efi_item_committing(
174	struct xfs_log_item	*lip,
175	xfs_lsn_t		lsn)
176{
177}
178
179/*
180 * This is the ops vector shared by all efi log items.
181 */
182static const struct xfs_item_ops xfs_efi_item_ops = {
183	.iop_size	= xfs_efi_item_size,
184	.iop_format	= xfs_efi_item_format,
185	.iop_pin	= xfs_efi_item_pin,
186	.iop_unpin	= xfs_efi_item_unpin,
187	.iop_unlock	= xfs_efi_item_unlock,
188	.iop_committed	= xfs_efi_item_committed,
189	.iop_push	= xfs_efi_item_push,
190	.iop_committing = xfs_efi_item_committing
191};
192
193
194/*
195 * Allocate and initialize an efi item with the given number of extents.
196 */
197struct xfs_efi_log_item *
198xfs_efi_init(
199	struct xfs_mount	*mp,
200	uint			nextents)
201
202{
203	struct xfs_efi_log_item	*efip;
204	uint			size;
205
206	ASSERT(nextents > 0);
207	if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
208		size = (uint)(sizeof(xfs_efi_log_item_t) +
209			((nextents - 1) * sizeof(xfs_extent_t)));
210		efip = kmem_zalloc(size, KM_SLEEP);
211	} else {
212		efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
213	}
214
215	xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
216	efip->efi_format.efi_nextents = nextents;
217	efip->efi_format.efi_id = (uintptr_t)(void *)efip;
218	atomic_set(&efip->efi_next_extent, 0);
219	atomic_set(&efip->efi_refcount, 2);
220
221	return efip;
222}
223
224/*
225 * Copy an EFI format buffer from the given buf, and into the destination
226 * EFI format structure.
227 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
228 * one of which will be the native format for this kernel.
229 * It will handle the conversion of formats if necessary.
230 */
231int
232xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
233{
234	xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
235	uint i;
236	uint len = sizeof(xfs_efi_log_format_t) +
237		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
238	uint len32 = sizeof(xfs_efi_log_format_32_t) +
239		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
240	uint len64 = sizeof(xfs_efi_log_format_64_t) +
241		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
242
243	if (buf->i_len == len) {
244		memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
245		return 0;
246	} else if (buf->i_len == len32) {
247		xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
248
249		dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
250		dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
251		dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
252		dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
253		for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
254			dst_efi_fmt->efi_extents[i].ext_start =
255				src_efi_fmt_32->efi_extents[i].ext_start;
256			dst_efi_fmt->efi_extents[i].ext_len =
257				src_efi_fmt_32->efi_extents[i].ext_len;
258		}
259		return 0;
260	} else if (buf->i_len == len64) {
261		xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
262
263		dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
264		dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
265		dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
266		dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
267		for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
268			dst_efi_fmt->efi_extents[i].ext_start =
269				src_efi_fmt_64->efi_extents[i].ext_start;
270			dst_efi_fmt->efi_extents[i].ext_len =
271				src_efi_fmt_64->efi_extents[i].ext_len;
272		}
273		return 0;
274	}
275	return -EFSCORRUPTED;
276}
277
278/*
279 * Freeing the efi requires that we remove it from the AIL if it has already
280 * been placed there. However, the EFI may not yet have been placed in the AIL
281 * when called by xfs_efi_release() from EFD processing due to the ordering of
282 * committed vs unpin operations in bulk insert operations. Hence the reference
283 * count to ensure only the last caller frees the EFI.
284 */
285void
286xfs_efi_release(
287	struct xfs_efi_log_item	*efip)
288{
289	if (atomic_dec_and_test(&efip->efi_refcount)) {
290		xfs_trans_ail_remove(&efip->efi_item, SHUTDOWN_LOG_IO_ERROR);
291		xfs_efi_item_free(efip);
292	}
293}
294
295static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
296{
297	return container_of(lip, struct xfs_efd_log_item, efd_item);
298}
299
300STATIC void
301xfs_efd_item_free(struct xfs_efd_log_item *efdp)
302{
303	if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
304		kmem_free(efdp);
305	else
306		kmem_zone_free(xfs_efd_zone, efdp);
307}
308
309/*
310 * This returns the number of iovecs needed to log the given efd item.
311 * We only need 1 iovec for an efd item.  It just logs the efd_log_format
312 * structure.
313 */
314static inline int
315xfs_efd_item_sizeof(
316	struct xfs_efd_log_item *efdp)
317{
318	return sizeof(xfs_efd_log_format_t) +
319	       (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
320}
321
322STATIC void
323xfs_efd_item_size(
324	struct xfs_log_item	*lip,
325	int			*nvecs,
326	int			*nbytes)
327{
328	*nvecs += 1;
329	*nbytes += xfs_efd_item_sizeof(EFD_ITEM(lip));
330}
331
332/*
333 * This is called to fill in the vector of log iovecs for the
334 * given efd log item. We use only 1 iovec, and we point that
335 * at the efd_log_format structure embedded in the efd item.
336 * It is at this point that we assert that all of the extent
337 * slots in the efd item have been filled.
338 */
339STATIC void
340xfs_efd_item_format(
341	struct xfs_log_item	*lip,
342	struct xfs_log_vec	*lv)
343{
344	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
345	struct xfs_log_iovec	*vecp = NULL;
346
347	ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
348
349	efdp->efd_format.efd_type = XFS_LI_EFD;
350	efdp->efd_format.efd_size = 1;
351
352	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
353			&efdp->efd_format,
354			xfs_efd_item_sizeof(efdp));
355}
356
357/*
358 * Pinning has no meaning for an efd item, so just return.
359 */
360STATIC void
361xfs_efd_item_pin(
362	struct xfs_log_item	*lip)
363{
364}
365
366/*
367 * Since pinning has no meaning for an efd item, unpinning does
368 * not either.
369 */
370STATIC void
371xfs_efd_item_unpin(
372	struct xfs_log_item	*lip,
373	int			remove)
374{
375}
376
377/*
378 * There isn't much you can do to push on an efd item.  It is simply stuck
379 * waiting for the log to be flushed to disk.
380 */
381STATIC uint
382xfs_efd_item_push(
383	struct xfs_log_item	*lip,
384	struct list_head	*buffer_list)
385{
386	return XFS_ITEM_PINNED;
387}
388
389/*
390 * The EFD is either committed or aborted if the transaction is cancelled. If
391 * the transaction is cancelled, drop our reference to the EFI and free the EFD.
392 */
393STATIC void
394xfs_efd_item_unlock(
395	struct xfs_log_item	*lip)
396{
397	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
398
399	if (lip->li_flags & XFS_LI_ABORTED) {
400		xfs_efi_release(efdp->efd_efip);
401		xfs_efd_item_free(efdp);
402	}
403}
404
405/*
406 * When the efd item is committed to disk, all we need to do is delete our
407 * reference to our partner efi item and then free ourselves. Since we're
408 * freeing ourselves we must return -1 to keep the transaction code from further
409 * referencing this item.
410 */
411STATIC xfs_lsn_t
412xfs_efd_item_committed(
413	struct xfs_log_item	*lip,
414	xfs_lsn_t		lsn)
415{
416	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
417
418	/*
419	 * Drop the EFI reference regardless of whether the EFD has been
420	 * aborted. Once the EFD transaction is constructed, it is the sole
421	 * responsibility of the EFD to release the EFI (even if the EFI is
422	 * aborted due to log I/O error).
423	 */
424	xfs_efi_release(efdp->efd_efip);
425	xfs_efd_item_free(efdp);
426
427	return (xfs_lsn_t)-1;
428}
429
430/*
431 * The EFD dependency tracking op doesn't do squat.  It can't because
432 * it doesn't know where the free extent is coming from.  The dependency
433 * tracking has to be handled by the "enclosing" metadata object.  For
434 * example, for inodes, the inode is locked throughout the extent freeing
435 * so the dependency should be recorded there.
436 */
437STATIC void
438xfs_efd_item_committing(
439	struct xfs_log_item	*lip,
440	xfs_lsn_t		lsn)
441{
442}
443
444/*
445 * This is the ops vector shared by all efd log items.
446 */
447static const struct xfs_item_ops xfs_efd_item_ops = {
448	.iop_size	= xfs_efd_item_size,
449	.iop_format	= xfs_efd_item_format,
450	.iop_pin	= xfs_efd_item_pin,
451	.iop_unpin	= xfs_efd_item_unpin,
452	.iop_unlock	= xfs_efd_item_unlock,
453	.iop_committed	= xfs_efd_item_committed,
454	.iop_push	= xfs_efd_item_push,
455	.iop_committing = xfs_efd_item_committing
456};
457
458/*
459 * Allocate and initialize an efd item with the given number of extents.
460 */
461struct xfs_efd_log_item *
462xfs_efd_init(
463	struct xfs_mount	*mp,
464	struct xfs_efi_log_item	*efip,
465	uint			nextents)
466
467{
468	struct xfs_efd_log_item	*efdp;
469	uint			size;
470
471	ASSERT(nextents > 0);
472	if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
473		size = (uint)(sizeof(xfs_efd_log_item_t) +
474			((nextents - 1) * sizeof(xfs_extent_t)));
475		efdp = kmem_zalloc(size, KM_SLEEP);
476	} else {
477		efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
478	}
479
480	xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
481	efdp->efd_efip = efip;
482	efdp->efd_format.efd_nextents = nextents;
483	efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
484
485	return efdp;
486}
487