2 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
22 #include "xfs_trans.h"
23 #include "xfs_buf_item.h"
26 #include "xfs_mount.h"
27 #include "xfs_trans_priv.h"
28 #include "xfs_extfree_item.h"
31 kmem_zone_t
*xfs_efi_zone
;
32 kmem_zone_t
*xfs_efd_zone
;
34 static inline struct xfs_efi_log_item
*EFI_ITEM(struct xfs_log_item
*lip
)
36 return container_of(lip
, struct xfs_efi_log_item
, efi_item
);
41 struct xfs_efi_log_item
*efip
)
43 if (efip
->efi_format
.efi_nextents
> XFS_EFI_MAX_FAST_EXTENTS
)
46 kmem_zone_free(xfs_efi_zone
, efip
);
50 * Freeing the efi requires that we remove it from the AIL if it has already
51 * been placed there. However, the EFI may not yet have been placed in the AIL
52 * when called by xfs_efi_release() from EFD processing due to the ordering of
53 * committed vs unpin operations in bulk insert operations. Hence the reference
54 * count to ensure only the last caller frees the EFI.
58 struct xfs_efi_log_item
*efip
)
60 struct xfs_ail
*ailp
= efip
->efi_item
.li_ailp
;
62 if (atomic_dec_and_test(&efip
->efi_refcount
)) {
63 spin_lock(&ailp
->xa_lock
);
64 /* xfs_trans_ail_delete() drops the AIL lock. */
65 xfs_trans_ail_delete(ailp
, &efip
->efi_item
,
66 SHUTDOWN_LOG_IO_ERROR
);
67 xfs_efi_item_free(efip
);
72 * This returns the number of iovecs needed to log the given efi item.
73 * We only need 1 iovec for an efi item. It just logs the efi_log_format
78 struct xfs_log_item
*lip
)
84 * This is called to fill in the vector of log iovecs for the
85 * given efi log item. We use only 1 iovec, and we point that
86 * at the efi_log_format structure embedded in the efi item.
87 * It is at this point that we assert that all of the extent
88 * slots in the efi item have been filled.
92 struct xfs_log_item
*lip
,
93 struct xfs_log_iovec
*log_vector
)
95 struct xfs_efi_log_item
*efip
= EFI_ITEM(lip
);
98 ASSERT(atomic_read(&efip
->efi_next_extent
) ==
99 efip
->efi_format
.efi_nextents
);
101 efip
->efi_format
.efi_type
= XFS_LI_EFI
;
103 size
= sizeof(xfs_efi_log_format_t
);
104 size
+= (efip
->efi_format
.efi_nextents
- 1) * sizeof(xfs_extent_t
);
105 efip
->efi_format
.efi_size
= 1;
107 log_vector
->i_addr
= &efip
->efi_format
;
108 log_vector
->i_len
= size
;
109 log_vector
->i_type
= XLOG_REG_TYPE_EFI_FORMAT
;
110 ASSERT(size
>= sizeof(xfs_efi_log_format_t
));
115 * Pinning has no meaning for an efi item, so just return.
119 struct xfs_log_item
*lip
)
124 * While EFIs cannot really be pinned, the unpin operation is the last place at
125 * which the EFI is manipulated during a transaction. If we are being asked to
126 * remove the EFI it's because the transaction has been cancelled and by
127 * definition that means the EFI cannot be in the AIL so remove it from the
128 * transaction and free it. Otherwise coordinate with xfs_efi_release()
129 * to determine who gets to free the EFI.
133 struct xfs_log_item
*lip
,
136 struct xfs_efi_log_item
*efip
= EFI_ITEM(lip
);
139 ASSERT(!(lip
->li_flags
& XFS_LI_IN_AIL
));
141 xfs_trans_del_item(lip
);
142 xfs_efi_item_free(efip
);
145 __xfs_efi_release(efip
);
149 * Efi items have no locking or pushing. However, since EFIs are pulled from
150 * the AIL when their corresponding EFDs are committed to disk, their situation
151 * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller
152 * will eventually flush the log. This should help in getting the EFI out of
157 struct xfs_log_item
*lip
,
158 struct list_head
*buffer_list
)
160 return XFS_ITEM_PINNED
;
165 struct xfs_log_item
*lip
)
167 if (lip
->li_flags
& XFS_LI_ABORTED
)
168 xfs_efi_item_free(EFI_ITEM(lip
));
172 * The EFI is logged only once and cannot be moved in the log, so simply return
173 * the lsn at which it's been logged.
176 xfs_efi_item_committed(
177 struct xfs_log_item
*lip
,
184 * The EFI dependency tracking op doesn't do squat. It can't because
185 * it doesn't know where the free extent is coming from. The dependency
186 * tracking has to be handled by the "enclosing" metadata object. For
187 * example, for inodes, the inode is locked throughout the extent freeing
188 * so the dependency should be recorded there.
191 xfs_efi_item_committing(
192 struct xfs_log_item
*lip
,
198 * This is the ops vector shared by all efi log items.
200 static const struct xfs_item_ops xfs_efi_item_ops
= {
201 .iop_size
= xfs_efi_item_size
,
202 .iop_format
= xfs_efi_item_format
,
203 .iop_pin
= xfs_efi_item_pin
,
204 .iop_unpin
= xfs_efi_item_unpin
,
205 .iop_unlock
= xfs_efi_item_unlock
,
206 .iop_committed
= xfs_efi_item_committed
,
207 .iop_push
= xfs_efi_item_push
,
208 .iop_committing
= xfs_efi_item_committing
213 * Allocate and initialize an efi item with the given number of extents.
215 struct xfs_efi_log_item
*
217 struct xfs_mount
*mp
,
221 struct xfs_efi_log_item
*efip
;
224 ASSERT(nextents
> 0);
225 if (nextents
> XFS_EFI_MAX_FAST_EXTENTS
) {
226 size
= (uint
)(sizeof(xfs_efi_log_item_t
) +
227 ((nextents
- 1) * sizeof(xfs_extent_t
)));
228 efip
= kmem_zalloc(size
, KM_SLEEP
);
230 efip
= kmem_zone_zalloc(xfs_efi_zone
, KM_SLEEP
);
233 xfs_log_item_init(mp
, &efip
->efi_item
, XFS_LI_EFI
, &xfs_efi_item_ops
);
234 efip
->efi_format
.efi_nextents
= nextents
;
235 efip
->efi_format
.efi_id
= (__psint_t
)(void*)efip
;
236 atomic_set(&efip
->efi_next_extent
, 0);
237 atomic_set(&efip
->efi_refcount
, 2);
243 * Copy an EFI format buffer from the given buf, and into the destination
244 * EFI format structure.
245 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
246 * one of which will be the native format for this kernel.
247 * It will handle the conversion of formats if necessary.
250 xfs_efi_copy_format(xfs_log_iovec_t
*buf
, xfs_efi_log_format_t
*dst_efi_fmt
)
252 xfs_efi_log_format_t
*src_efi_fmt
= buf
->i_addr
;
254 uint len
= sizeof(xfs_efi_log_format_t
) +
255 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_t
);
256 uint len32
= sizeof(xfs_efi_log_format_32_t
) +
257 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_32_t
);
258 uint len64
= sizeof(xfs_efi_log_format_64_t
) +
259 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_64_t
);
261 if (buf
->i_len
== len
) {
262 memcpy((char *)dst_efi_fmt
, (char*)src_efi_fmt
, len
);
264 } else if (buf
->i_len
== len32
) {
265 xfs_efi_log_format_32_t
*src_efi_fmt_32
= buf
->i_addr
;
267 dst_efi_fmt
->efi_type
= src_efi_fmt_32
->efi_type
;
268 dst_efi_fmt
->efi_size
= src_efi_fmt_32
->efi_size
;
269 dst_efi_fmt
->efi_nextents
= src_efi_fmt_32
->efi_nextents
;
270 dst_efi_fmt
->efi_id
= src_efi_fmt_32
->efi_id
;
271 for (i
= 0; i
< dst_efi_fmt
->efi_nextents
; i
++) {
272 dst_efi_fmt
->efi_extents
[i
].ext_start
=
273 src_efi_fmt_32
->efi_extents
[i
].ext_start
;
274 dst_efi_fmt
->efi_extents
[i
].ext_len
=
275 src_efi_fmt_32
->efi_extents
[i
].ext_len
;
278 } else if (buf
->i_len
== len64
) {
279 xfs_efi_log_format_64_t
*src_efi_fmt_64
= buf
->i_addr
;
281 dst_efi_fmt
->efi_type
= src_efi_fmt_64
->efi_type
;
282 dst_efi_fmt
->efi_size
= src_efi_fmt_64
->efi_size
;
283 dst_efi_fmt
->efi_nextents
= src_efi_fmt_64
->efi_nextents
;
284 dst_efi_fmt
->efi_id
= src_efi_fmt_64
->efi_id
;
285 for (i
= 0; i
< dst_efi_fmt
->efi_nextents
; i
++) {
286 dst_efi_fmt
->efi_extents
[i
].ext_start
=
287 src_efi_fmt_64
->efi_extents
[i
].ext_start
;
288 dst_efi_fmt
->efi_extents
[i
].ext_len
=
289 src_efi_fmt_64
->efi_extents
[i
].ext_len
;
297 * This is called by the efd item code below to release references to the given
298 * efi item. Each efd calls this with the number of extents that it has
299 * logged, and when the sum of these reaches the total number of extents logged
300 * by this efi item we can free the efi item.
303 xfs_efi_release(xfs_efi_log_item_t
*efip
,
306 ASSERT(atomic_read(&efip
->efi_next_extent
) >= nextents
);
307 if (atomic_sub_and_test(nextents
, &efip
->efi_next_extent
)) {
308 /* recovery needs us to drop the EFI reference, too */
309 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
))
310 __xfs_efi_release(efip
);
312 __xfs_efi_release(efip
);
313 /* efip may now have been freed, do not reference it again. */
317 static inline struct xfs_efd_log_item
*EFD_ITEM(struct xfs_log_item
*lip
)
319 return container_of(lip
, struct xfs_efd_log_item
, efd_item
);
323 xfs_efd_item_free(struct xfs_efd_log_item
*efdp
)
325 if (efdp
->efd_format
.efd_nextents
> XFS_EFD_MAX_FAST_EXTENTS
)
328 kmem_zone_free(xfs_efd_zone
, efdp
);
332 * This returns the number of iovecs needed to log the given efd item.
333 * We only need 1 iovec for an efd item. It just logs the efd_log_format
338 struct xfs_log_item
*lip
)
344 * This is called to fill in the vector of log iovecs for the
345 * given efd log item. We use only 1 iovec, and we point that
346 * at the efd_log_format structure embedded in the efd item.
347 * It is at this point that we assert that all of the extent
348 * slots in the efd item have been filled.
352 struct xfs_log_item
*lip
,
353 struct xfs_log_iovec
*log_vector
)
355 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
358 ASSERT(efdp
->efd_next_extent
== efdp
->efd_format
.efd_nextents
);
360 efdp
->efd_format
.efd_type
= XFS_LI_EFD
;
362 size
= sizeof(xfs_efd_log_format_t
);
363 size
+= (efdp
->efd_format
.efd_nextents
- 1) * sizeof(xfs_extent_t
);
364 efdp
->efd_format
.efd_size
= 1;
366 log_vector
->i_addr
= &efdp
->efd_format
;
367 log_vector
->i_len
= size
;
368 log_vector
->i_type
= XLOG_REG_TYPE_EFD_FORMAT
;
369 ASSERT(size
>= sizeof(xfs_efd_log_format_t
));
373 * Pinning has no meaning for an efd item, so just return.
377 struct xfs_log_item
*lip
)
382 * Since pinning has no meaning for an efd item, unpinning does
387 struct xfs_log_item
*lip
,
393 * There isn't much you can do to push on an efd item. It is simply stuck
394 * waiting for the log to be flushed to disk.
398 struct xfs_log_item
*lip
,
399 struct list_head
*buffer_list
)
401 return XFS_ITEM_PINNED
;
406 struct xfs_log_item
*lip
)
408 if (lip
->li_flags
& XFS_LI_ABORTED
)
409 xfs_efd_item_free(EFD_ITEM(lip
));
413 * When the efd item is committed to disk, all we need to do
414 * is delete our reference to our partner efi item and then
415 * free ourselves. Since we're freeing ourselves we must
416 * return -1 to keep the transaction code from further referencing
420 xfs_efd_item_committed(
421 struct xfs_log_item
*lip
,
424 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
427 * If we got a log I/O error, it's always the case that the LR with the
428 * EFI got unpinned and freed before the EFD got aborted.
430 if (!(lip
->li_flags
& XFS_LI_ABORTED
))
431 xfs_efi_release(efdp
->efd_efip
, efdp
->efd_format
.efd_nextents
);
433 xfs_efd_item_free(efdp
);
434 return (xfs_lsn_t
)-1;
438 * The EFD dependency tracking op doesn't do squat. It can't because
439 * it doesn't know where the free extent is coming from. The dependency
440 * tracking has to be handled by the "enclosing" metadata object. For
441 * example, for inodes, the inode is locked throughout the extent freeing
442 * so the dependency should be recorded there.
445 xfs_efd_item_committing(
446 struct xfs_log_item
*lip
,
452 * This is the ops vector shared by all efd log items.
454 static const struct xfs_item_ops xfs_efd_item_ops
= {
455 .iop_size
= xfs_efd_item_size
,
456 .iop_format
= xfs_efd_item_format
,
457 .iop_pin
= xfs_efd_item_pin
,
458 .iop_unpin
= xfs_efd_item_unpin
,
459 .iop_unlock
= xfs_efd_item_unlock
,
460 .iop_committed
= xfs_efd_item_committed
,
461 .iop_push
= xfs_efd_item_push
,
462 .iop_committing
= xfs_efd_item_committing
466 * Allocate and initialize an efd item with the given number of extents.
468 struct xfs_efd_log_item
*
470 struct xfs_mount
*mp
,
471 struct xfs_efi_log_item
*efip
,
475 struct xfs_efd_log_item
*efdp
;
478 ASSERT(nextents
> 0);
479 if (nextents
> XFS_EFD_MAX_FAST_EXTENTS
) {
480 size
= (uint
)(sizeof(xfs_efd_log_item_t
) +
481 ((nextents
- 1) * sizeof(xfs_extent_t
)));
482 efdp
= kmem_zalloc(size
, KM_SLEEP
);
484 efdp
= kmem_zone_zalloc(xfs_efd_zone
, KM_SLEEP
);
487 xfs_log_item_init(mp
, &efdp
->efd_item
, XFS_LI_EFD
, &xfs_efd_item_ops
);
488 efdp
->efd_efip
= efip
;
489 efdp
->efd_format
.efd_nextents
= nextents
;
490 efdp
->efd_format
.efd_efi_id
= efip
->efi_format
.efi_id
;