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_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_extfree_item.h"
30 #include "xfs_btree.h"
34 kmem_zone_t
*xfs_efi_zone
;
35 kmem_zone_t
*xfs_efd_zone
;
37 static inline struct xfs_efi_log_item
*EFI_ITEM(struct xfs_log_item
*lip
)
39 return container_of(lip
, struct xfs_efi_log_item
, efi_item
);
44 struct xfs_efi_log_item
*efip
)
46 kmem_free(efip
->efi_item
.li_lv_shadow
);
47 if (efip
->efi_format
.efi_nextents
> XFS_EFI_MAX_FAST_EXTENTS
)
50 kmem_zone_free(xfs_efi_zone
, efip
);
54 * This returns the number of iovecs needed to log the given efi item.
55 * We only need 1 iovec for an efi item. It just logs the efi_log_format
60 struct xfs_efi_log_item
*efip
)
62 return sizeof(struct xfs_efi_log_format
) +
63 (efip
->efi_format
.efi_nextents
- 1) * sizeof(xfs_extent_t
);
68 struct xfs_log_item
*lip
,
73 *nbytes
+= xfs_efi_item_sizeof(EFI_ITEM(lip
));
77 * This is called to fill in the vector of log iovecs for the
78 * given efi log item. We use only 1 iovec, and we point that
79 * at the efi_log_format structure embedded in the efi item.
80 * It is at this point that we assert that all of the extent
81 * slots in the efi item have been filled.
85 struct xfs_log_item
*lip
,
86 struct xfs_log_vec
*lv
)
88 struct xfs_efi_log_item
*efip
= EFI_ITEM(lip
);
89 struct xfs_log_iovec
*vecp
= NULL
;
91 ASSERT(atomic_read(&efip
->efi_next_extent
) ==
92 efip
->efi_format
.efi_nextents
);
94 efip
->efi_format
.efi_type
= XFS_LI_EFI
;
95 efip
->efi_format
.efi_size
= 1;
97 xlog_copy_iovec(lv
, &vecp
, XLOG_REG_TYPE_EFI_FORMAT
,
99 xfs_efi_item_sizeof(efip
));
104 * Pinning has no meaning for an efi item, so just return.
108 struct xfs_log_item
*lip
)
113 * The unpin operation is the last place an EFI is manipulated in the log. It is
114 * either inserted in the AIL or aborted in the event of a log I/O error. In
115 * either case, the EFI transaction has been successfully committed to make it
116 * this far. Therefore, we expect whoever committed the EFI to either construct
117 * and commit the EFD or drop the EFD's reference in the event of error. Simply
118 * drop the log's EFI reference now that the log is done with it.
122 struct xfs_log_item
*lip
,
125 struct xfs_efi_log_item
*efip
= EFI_ITEM(lip
);
126 xfs_efi_release(efip
);
130 * Efi items have no locking or pushing. However, since EFIs are pulled from
131 * the AIL when their corresponding EFDs are committed to disk, their situation
132 * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller
133 * will eventually flush the log. This should help in getting the EFI out of
138 struct xfs_log_item
*lip
,
139 struct list_head
*buffer_list
)
141 return XFS_ITEM_PINNED
;
145 * The EFI has been either committed or aborted if the transaction has been
146 * cancelled. If the transaction was cancelled, an EFD isn't going to be
147 * constructed and thus we free the EFI here directly.
151 struct xfs_log_item
*lip
)
153 if (lip
->li_flags
& XFS_LI_ABORTED
)
154 xfs_efi_item_free(EFI_ITEM(lip
));
158 * The EFI is logged only once and cannot be moved in the log, so simply return
159 * the lsn at which it's been logged.
162 xfs_efi_item_committed(
163 struct xfs_log_item
*lip
,
170 * The EFI dependency tracking op doesn't do squat. It can't because
171 * it doesn't know where the free extent is coming from. The dependency
172 * tracking has to be handled by the "enclosing" metadata object. For
173 * example, for inodes, the inode is locked throughout the extent freeing
174 * so the dependency should be recorded there.
177 xfs_efi_item_committing(
178 struct xfs_log_item
*lip
,
184 * This is the ops vector shared by all efi log items.
186 static const struct xfs_item_ops xfs_efi_item_ops
= {
187 .iop_size
= xfs_efi_item_size
,
188 .iop_format
= xfs_efi_item_format
,
189 .iop_pin
= xfs_efi_item_pin
,
190 .iop_unpin
= xfs_efi_item_unpin
,
191 .iop_unlock
= xfs_efi_item_unlock
,
192 .iop_committed
= xfs_efi_item_committed
,
193 .iop_push
= xfs_efi_item_push
,
194 .iop_committing
= xfs_efi_item_committing
199 * Allocate and initialize an efi item with the given number of extents.
201 struct xfs_efi_log_item
*
203 struct xfs_mount
*mp
,
207 struct xfs_efi_log_item
*efip
;
210 ASSERT(nextents
> 0);
211 if (nextents
> XFS_EFI_MAX_FAST_EXTENTS
) {
212 size
= (uint
)(sizeof(xfs_efi_log_item_t
) +
213 ((nextents
- 1) * sizeof(xfs_extent_t
)));
214 efip
= kmem_zalloc(size
, KM_SLEEP
);
216 efip
= kmem_zone_zalloc(xfs_efi_zone
, KM_SLEEP
);
219 xfs_log_item_init(mp
, &efip
->efi_item
, XFS_LI_EFI
, &xfs_efi_item_ops
);
220 efip
->efi_format
.efi_nextents
= nextents
;
221 efip
->efi_format
.efi_id
= (uintptr_t)(void *)efip
;
222 atomic_set(&efip
->efi_next_extent
, 0);
223 atomic_set(&efip
->efi_refcount
, 2);
229 * Copy an EFI format buffer from the given buf, and into the destination
230 * EFI format structure.
231 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
232 * one of which will be the native format for this kernel.
233 * It will handle the conversion of formats if necessary.
236 xfs_efi_copy_format(xfs_log_iovec_t
*buf
, xfs_efi_log_format_t
*dst_efi_fmt
)
238 xfs_efi_log_format_t
*src_efi_fmt
= buf
->i_addr
;
240 uint len
= sizeof(xfs_efi_log_format_t
) +
241 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_t
);
242 uint len32
= sizeof(xfs_efi_log_format_32_t
) +
243 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_32_t
);
244 uint len64
= sizeof(xfs_efi_log_format_64_t
) +
245 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_64_t
);
247 if (buf
->i_len
== len
) {
248 memcpy((char *)dst_efi_fmt
, (char*)src_efi_fmt
, len
);
250 } else if (buf
->i_len
== len32
) {
251 xfs_efi_log_format_32_t
*src_efi_fmt_32
= buf
->i_addr
;
253 dst_efi_fmt
->efi_type
= src_efi_fmt_32
->efi_type
;
254 dst_efi_fmt
->efi_size
= src_efi_fmt_32
->efi_size
;
255 dst_efi_fmt
->efi_nextents
= src_efi_fmt_32
->efi_nextents
;
256 dst_efi_fmt
->efi_id
= src_efi_fmt_32
->efi_id
;
257 for (i
= 0; i
< dst_efi_fmt
->efi_nextents
; i
++) {
258 dst_efi_fmt
->efi_extents
[i
].ext_start
=
259 src_efi_fmt_32
->efi_extents
[i
].ext_start
;
260 dst_efi_fmt
->efi_extents
[i
].ext_len
=
261 src_efi_fmt_32
->efi_extents
[i
].ext_len
;
264 } else if (buf
->i_len
== len64
) {
265 xfs_efi_log_format_64_t
*src_efi_fmt_64
= buf
->i_addr
;
267 dst_efi_fmt
->efi_type
= src_efi_fmt_64
->efi_type
;
268 dst_efi_fmt
->efi_size
= src_efi_fmt_64
->efi_size
;
269 dst_efi_fmt
->efi_nextents
= src_efi_fmt_64
->efi_nextents
;
270 dst_efi_fmt
->efi_id
= src_efi_fmt_64
->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_64
->efi_extents
[i
].ext_start
;
274 dst_efi_fmt
->efi_extents
[i
].ext_len
=
275 src_efi_fmt_64
->efi_extents
[i
].ext_len
;
279 return -EFSCORRUPTED
;
283 * Freeing the efi requires that we remove it from the AIL if it has already
284 * been placed there. However, the EFI may not yet have been placed in the AIL
285 * when called by xfs_efi_release() from EFD processing due to the ordering of
286 * committed vs unpin operations in bulk insert operations. Hence the reference
287 * count to ensure only the last caller frees the EFI.
291 struct xfs_efi_log_item
*efip
)
293 if (atomic_dec_and_test(&efip
->efi_refcount
)) {
294 xfs_trans_ail_remove(&efip
->efi_item
, SHUTDOWN_LOG_IO_ERROR
);
295 xfs_efi_item_free(efip
);
299 static inline struct xfs_efd_log_item
*EFD_ITEM(struct xfs_log_item
*lip
)
301 return container_of(lip
, struct xfs_efd_log_item
, efd_item
);
305 xfs_efd_item_free(struct xfs_efd_log_item
*efdp
)
307 kmem_free(efdp
->efd_item
.li_lv_shadow
);
308 if (efdp
->efd_format
.efd_nextents
> XFS_EFD_MAX_FAST_EXTENTS
)
311 kmem_zone_free(xfs_efd_zone
, efdp
);
315 * This returns the number of iovecs needed to log the given efd item.
316 * We only need 1 iovec for an efd item. It just logs the efd_log_format
321 struct xfs_efd_log_item
*efdp
)
323 return sizeof(xfs_efd_log_format_t
) +
324 (efdp
->efd_format
.efd_nextents
- 1) * sizeof(xfs_extent_t
);
329 struct xfs_log_item
*lip
,
334 *nbytes
+= xfs_efd_item_sizeof(EFD_ITEM(lip
));
338 * This is called to fill in the vector of log iovecs for the
339 * given efd log item. We use only 1 iovec, and we point that
340 * at the efd_log_format structure embedded in the efd item.
341 * It is at this point that we assert that all of the extent
342 * slots in the efd item have been filled.
346 struct xfs_log_item
*lip
,
347 struct xfs_log_vec
*lv
)
349 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
350 struct xfs_log_iovec
*vecp
= NULL
;
352 ASSERT(efdp
->efd_next_extent
== efdp
->efd_format
.efd_nextents
);
354 efdp
->efd_format
.efd_type
= XFS_LI_EFD
;
355 efdp
->efd_format
.efd_size
= 1;
357 xlog_copy_iovec(lv
, &vecp
, XLOG_REG_TYPE_EFD_FORMAT
,
359 xfs_efd_item_sizeof(efdp
));
363 * Pinning has no meaning for an efd item, so just return.
367 struct xfs_log_item
*lip
)
372 * Since pinning has no meaning for an efd item, unpinning does
377 struct xfs_log_item
*lip
,
383 * There isn't much you can do to push on an efd item. It is simply stuck
384 * waiting for the log to be flushed to disk.
388 struct xfs_log_item
*lip
,
389 struct list_head
*buffer_list
)
391 return XFS_ITEM_PINNED
;
395 * The EFD is either committed or aborted if the transaction is cancelled. If
396 * the transaction is cancelled, drop our reference to the EFI and free the EFD.
400 struct xfs_log_item
*lip
)
402 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
404 if (lip
->li_flags
& XFS_LI_ABORTED
) {
405 xfs_efi_release(efdp
->efd_efip
);
406 xfs_efd_item_free(efdp
);
411 * When the efd item is committed to disk, all we need to do is delete our
412 * reference to our partner efi item and then free ourselves. Since we're
413 * freeing ourselves we must return -1 to keep the transaction code from further
414 * referencing this item.
417 xfs_efd_item_committed(
418 struct xfs_log_item
*lip
,
421 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
424 * Drop the EFI reference regardless of whether the EFD has been
425 * aborted. Once the EFD transaction is constructed, it is the sole
426 * responsibility of the EFD to release the EFI (even if the EFI is
427 * aborted due to log I/O error).
429 xfs_efi_release(efdp
->efd_efip
);
430 xfs_efd_item_free(efdp
);
432 return (xfs_lsn_t
)-1;
436 * The EFD dependency tracking op doesn't do squat. It can't because
437 * it doesn't know where the free extent is coming from. The dependency
438 * tracking has to be handled by the "enclosing" metadata object. For
439 * example, for inodes, the inode is locked throughout the extent freeing
440 * so the dependency should be recorded there.
443 xfs_efd_item_committing(
444 struct xfs_log_item
*lip
,
450 * This is the ops vector shared by all efd log items.
452 static const struct xfs_item_ops xfs_efd_item_ops
= {
453 .iop_size
= xfs_efd_item_size
,
454 .iop_format
= xfs_efd_item_format
,
455 .iop_pin
= xfs_efd_item_pin
,
456 .iop_unpin
= xfs_efd_item_unpin
,
457 .iop_unlock
= xfs_efd_item_unlock
,
458 .iop_committed
= xfs_efd_item_committed
,
459 .iop_push
= xfs_efd_item_push
,
460 .iop_committing
= xfs_efd_item_committing
464 * Allocate and initialize an efd item with the given number of extents.
466 struct xfs_efd_log_item
*
468 struct xfs_mount
*mp
,
469 struct xfs_efi_log_item
*efip
,
473 struct xfs_efd_log_item
*efdp
;
476 ASSERT(nextents
> 0);
477 if (nextents
> XFS_EFD_MAX_FAST_EXTENTS
) {
478 size
= (uint
)(sizeof(xfs_efd_log_item_t
) +
479 ((nextents
- 1) * sizeof(xfs_extent_t
)));
480 efdp
= kmem_zalloc(size
, KM_SLEEP
);
482 efdp
= kmem_zone_zalloc(xfs_efd_zone
, KM_SLEEP
);
485 xfs_log_item_init(mp
, &efdp
->efd_item
, XFS_LI_EFD
, &xfs_efd_item_ops
);
486 efdp
->efd_efip
= efip
;
487 efdp
->efd_format
.efd_nextents
= nextents
;
488 efdp
->efd_format
.efd_efi_id
= efip
->efi_format
.efi_id
;
494 * Process an extent free intent item that was recovered from
495 * the log. We need to free the extents that it describes.
499 struct xfs_mount
*mp
,
500 struct xfs_efi_log_item
*efip
)
502 struct xfs_efd_log_item
*efdp
;
503 struct xfs_trans
*tp
;
507 xfs_fsblock_t startblock_fsb
;
508 struct xfs_owner_info oinfo
;
510 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
513 * First check the validity of the extents described by the
514 * EFI. If any are bad, then assume that all are bad and
517 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
518 extp
= &efip
->efi_format
.efi_extents
[i
];
519 startblock_fsb
= XFS_BB_TO_FSB(mp
,
520 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
521 if (startblock_fsb
== 0 ||
522 extp
->ext_len
== 0 ||
523 startblock_fsb
>= mp
->m_sb
.sb_dblocks
||
524 extp
->ext_len
>= mp
->m_sb
.sb_agblocks
) {
526 * This will pull the EFI from the AIL and
527 * free the memory associated with it.
529 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
530 xfs_efi_release(efip
);
535 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, 0, 0, 0, &tp
);
538 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
540 xfs_rmap_skip_owner_update(&oinfo
);
541 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
542 extp
= &efip
->efi_format
.efi_extents
[i
];
543 error
= xfs_trans_free_extent(tp
, efdp
, extp
->ext_start
,
544 extp
->ext_len
, &oinfo
);
550 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
551 error
= xfs_trans_commit(tp
);
555 xfs_trans_cancel(tp
);