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"
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"
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 kmem_free(efip
->efi_item
.li_lv_shadow
);
44 if (efip
->efi_format
.efi_nextents
> XFS_EFI_MAX_FAST_EXTENTS
)
47 kmem_zone_free(xfs_efi_zone
, efip
);
51 * This returns the number of iovecs needed to log the given efi item.
52 * We only need 1 iovec for an efi item. It just logs the efi_log_format
57 struct xfs_efi_log_item
*efip
)
59 return sizeof(struct xfs_efi_log_format
) +
60 (efip
->efi_format
.efi_nextents
- 1) * sizeof(xfs_extent_t
);
65 struct xfs_log_item
*lip
,
70 *nbytes
+= xfs_efi_item_sizeof(EFI_ITEM(lip
));
74 * This is called to fill in the vector of log iovecs for the
75 * given efi log item. We use only 1 iovec, and we point that
76 * at the efi_log_format structure embedded in the efi item.
77 * It is at this point that we assert that all of the extent
78 * slots in the efi item have been filled.
82 struct xfs_log_item
*lip
,
83 struct xfs_log_vec
*lv
)
85 struct xfs_efi_log_item
*efip
= EFI_ITEM(lip
);
86 struct xfs_log_iovec
*vecp
= NULL
;
88 ASSERT(atomic_read(&efip
->efi_next_extent
) ==
89 efip
->efi_format
.efi_nextents
);
91 efip
->efi_format
.efi_type
= XFS_LI_EFI
;
92 efip
->efi_format
.efi_size
= 1;
94 xlog_copy_iovec(lv
, &vecp
, XLOG_REG_TYPE_EFI_FORMAT
,
96 xfs_efi_item_sizeof(efip
));
101 * Pinning has no meaning for an efi item, so just return.
105 struct xfs_log_item
*lip
)
110 * The unpin operation is the last place an EFI is manipulated in the log. It is
111 * either inserted in the AIL or aborted in the event of a log I/O error. In
112 * either case, the EFI transaction has been successfully committed to make it
113 * this far. Therefore, we expect whoever committed the EFI to either construct
114 * and commit the EFD or drop the EFD's reference in the event of error. Simply
115 * drop the log's EFI reference now that the log is done with it.
119 struct xfs_log_item
*lip
,
122 struct xfs_efi_log_item
*efip
= EFI_ITEM(lip
);
123 xfs_efi_release(efip
);
127 * Efi items have no locking or pushing. However, since EFIs are pulled from
128 * the AIL when their corresponding EFDs are committed to disk, their situation
129 * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller
130 * will eventually flush the log. This should help in getting the EFI out of
135 struct xfs_log_item
*lip
,
136 struct list_head
*buffer_list
)
138 return XFS_ITEM_PINNED
;
142 * The EFI has been either committed or aborted if the transaction has been
143 * cancelled. If the transaction was cancelled, an EFD isn't going to be
144 * constructed and thus we free the EFI here directly.
148 struct xfs_log_item
*lip
)
150 if (lip
->li_flags
& XFS_LI_ABORTED
)
151 xfs_efi_item_free(EFI_ITEM(lip
));
155 * The EFI is logged only once and cannot be moved in the log, so simply return
156 * the lsn at which it's been logged.
159 xfs_efi_item_committed(
160 struct xfs_log_item
*lip
,
167 * The EFI dependency tracking op doesn't do squat. It can't because
168 * it doesn't know where the free extent is coming from. The dependency
169 * tracking has to be handled by the "enclosing" metadata object. For
170 * example, for inodes, the inode is locked throughout the extent freeing
171 * so the dependency should be recorded there.
174 xfs_efi_item_committing(
175 struct xfs_log_item
*lip
,
181 * This is the ops vector shared by all efi log items.
183 static const struct xfs_item_ops xfs_efi_item_ops
= {
184 .iop_size
= xfs_efi_item_size
,
185 .iop_format
= xfs_efi_item_format
,
186 .iop_pin
= xfs_efi_item_pin
,
187 .iop_unpin
= xfs_efi_item_unpin
,
188 .iop_unlock
= xfs_efi_item_unlock
,
189 .iop_committed
= xfs_efi_item_committed
,
190 .iop_push
= xfs_efi_item_push
,
191 .iop_committing
= xfs_efi_item_committing
196 * Allocate and initialize an efi item with the given number of extents.
198 struct xfs_efi_log_item
*
200 struct xfs_mount
*mp
,
204 struct xfs_efi_log_item
*efip
;
207 ASSERT(nextents
> 0);
208 if (nextents
> XFS_EFI_MAX_FAST_EXTENTS
) {
209 size
= (uint
)(sizeof(xfs_efi_log_item_t
) +
210 ((nextents
- 1) * sizeof(xfs_extent_t
)));
211 efip
= kmem_zalloc(size
, KM_SLEEP
);
213 efip
= kmem_zone_zalloc(xfs_efi_zone
, KM_SLEEP
);
216 xfs_log_item_init(mp
, &efip
->efi_item
, XFS_LI_EFI
, &xfs_efi_item_ops
);
217 efip
->efi_format
.efi_nextents
= nextents
;
218 efip
->efi_format
.efi_id
= (uintptr_t)(void *)efip
;
219 atomic_set(&efip
->efi_next_extent
, 0);
220 atomic_set(&efip
->efi_refcount
, 2);
226 * Copy an EFI format buffer from the given buf, and into the destination
227 * EFI format structure.
228 * The given buffer can be in 32 bit or 64 bit form (which has different padding),
229 * one of which will be the native format for this kernel.
230 * It will handle the conversion of formats if necessary.
233 xfs_efi_copy_format(xfs_log_iovec_t
*buf
, xfs_efi_log_format_t
*dst_efi_fmt
)
235 xfs_efi_log_format_t
*src_efi_fmt
= buf
->i_addr
;
237 uint len
= sizeof(xfs_efi_log_format_t
) +
238 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_t
);
239 uint len32
= sizeof(xfs_efi_log_format_32_t
) +
240 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_32_t
);
241 uint len64
= sizeof(xfs_efi_log_format_64_t
) +
242 (src_efi_fmt
->efi_nextents
- 1) * sizeof(xfs_extent_64_t
);
244 if (buf
->i_len
== len
) {
245 memcpy((char *)dst_efi_fmt
, (char*)src_efi_fmt
, len
);
247 } else if (buf
->i_len
== len32
) {
248 xfs_efi_log_format_32_t
*src_efi_fmt_32
= buf
->i_addr
;
250 dst_efi_fmt
->efi_type
= src_efi_fmt_32
->efi_type
;
251 dst_efi_fmt
->efi_size
= src_efi_fmt_32
->efi_size
;
252 dst_efi_fmt
->efi_nextents
= src_efi_fmt_32
->efi_nextents
;
253 dst_efi_fmt
->efi_id
= src_efi_fmt_32
->efi_id
;
254 for (i
= 0; i
< dst_efi_fmt
->efi_nextents
; i
++) {
255 dst_efi_fmt
->efi_extents
[i
].ext_start
=
256 src_efi_fmt_32
->efi_extents
[i
].ext_start
;
257 dst_efi_fmt
->efi_extents
[i
].ext_len
=
258 src_efi_fmt_32
->efi_extents
[i
].ext_len
;
261 } else if (buf
->i_len
== len64
) {
262 xfs_efi_log_format_64_t
*src_efi_fmt_64
= buf
->i_addr
;
264 dst_efi_fmt
->efi_type
= src_efi_fmt_64
->efi_type
;
265 dst_efi_fmt
->efi_size
= src_efi_fmt_64
->efi_size
;
266 dst_efi_fmt
->efi_nextents
= src_efi_fmt_64
->efi_nextents
;
267 dst_efi_fmt
->efi_id
= src_efi_fmt_64
->efi_id
;
268 for (i
= 0; i
< dst_efi_fmt
->efi_nextents
; i
++) {
269 dst_efi_fmt
->efi_extents
[i
].ext_start
=
270 src_efi_fmt_64
->efi_extents
[i
].ext_start
;
271 dst_efi_fmt
->efi_extents
[i
].ext_len
=
272 src_efi_fmt_64
->efi_extents
[i
].ext_len
;
276 return -EFSCORRUPTED
;
280 * Freeing the efi requires that we remove it from the AIL if it has already
281 * been placed there. However, the EFI may not yet have been placed in the AIL
282 * when called by xfs_efi_release() from EFD processing due to the ordering of
283 * committed vs unpin operations in bulk insert operations. Hence the reference
284 * count to ensure only the last caller frees the EFI.
288 struct xfs_efi_log_item
*efip
)
290 if (atomic_dec_and_test(&efip
->efi_refcount
)) {
291 xfs_trans_ail_remove(&efip
->efi_item
, SHUTDOWN_LOG_IO_ERROR
);
292 xfs_efi_item_free(efip
);
296 static inline struct xfs_efd_log_item
*EFD_ITEM(struct xfs_log_item
*lip
)
298 return container_of(lip
, struct xfs_efd_log_item
, efd_item
);
302 xfs_efd_item_free(struct xfs_efd_log_item
*efdp
)
304 kmem_free(efdp
->efd_item
.li_lv_shadow
);
305 if (efdp
->efd_format
.efd_nextents
> XFS_EFD_MAX_FAST_EXTENTS
)
308 kmem_zone_free(xfs_efd_zone
, efdp
);
312 * This returns the number of iovecs needed to log the given efd item.
313 * We only need 1 iovec for an efd item. It just logs the efd_log_format
318 struct xfs_efd_log_item
*efdp
)
320 return sizeof(xfs_efd_log_format_t
) +
321 (efdp
->efd_format
.efd_nextents
- 1) * sizeof(xfs_extent_t
);
326 struct xfs_log_item
*lip
,
331 *nbytes
+= xfs_efd_item_sizeof(EFD_ITEM(lip
));
335 * This is called to fill in the vector of log iovecs for the
336 * given efd log item. We use only 1 iovec, and we point that
337 * at the efd_log_format structure embedded in the efd item.
338 * It is at this point that we assert that all of the extent
339 * slots in the efd item have been filled.
343 struct xfs_log_item
*lip
,
344 struct xfs_log_vec
*lv
)
346 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
347 struct xfs_log_iovec
*vecp
= NULL
;
349 ASSERT(efdp
->efd_next_extent
== efdp
->efd_format
.efd_nextents
);
351 efdp
->efd_format
.efd_type
= XFS_LI_EFD
;
352 efdp
->efd_format
.efd_size
= 1;
354 xlog_copy_iovec(lv
, &vecp
, XLOG_REG_TYPE_EFD_FORMAT
,
356 xfs_efd_item_sizeof(efdp
));
360 * Pinning has no meaning for an efd item, so just return.
364 struct xfs_log_item
*lip
)
369 * Since pinning has no meaning for an efd item, unpinning does
374 struct xfs_log_item
*lip
,
380 * There isn't much you can do to push on an efd item. It is simply stuck
381 * waiting for the log to be flushed to disk.
385 struct xfs_log_item
*lip
,
386 struct list_head
*buffer_list
)
388 return XFS_ITEM_PINNED
;
392 * The EFD is either committed or aborted if the transaction is cancelled. If
393 * the transaction is cancelled, drop our reference to the EFI and free the EFD.
397 struct xfs_log_item
*lip
)
399 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
401 if (lip
->li_flags
& XFS_LI_ABORTED
) {
402 xfs_efi_release(efdp
->efd_efip
);
403 xfs_efd_item_free(efdp
);
408 * When the efd item is committed to disk, all we need to do is delete our
409 * reference to our partner efi item and then free ourselves. Since we're
410 * freeing ourselves we must return -1 to keep the transaction code from further
411 * referencing this item.
414 xfs_efd_item_committed(
415 struct xfs_log_item
*lip
,
418 struct xfs_efd_log_item
*efdp
= EFD_ITEM(lip
);
421 * Drop the EFI reference regardless of whether the EFD has been
422 * aborted. Once the EFD transaction is constructed, it is the sole
423 * responsibility of the EFD to release the EFI (even if the EFI is
424 * aborted due to log I/O error).
426 xfs_efi_release(efdp
->efd_efip
);
427 xfs_efd_item_free(efdp
);
429 return (xfs_lsn_t
)-1;
433 * The EFD dependency tracking op doesn't do squat. It can't because
434 * it doesn't know where the free extent is coming from. The dependency
435 * tracking has to be handled by the "enclosing" metadata object. For
436 * example, for inodes, the inode is locked throughout the extent freeing
437 * so the dependency should be recorded there.
440 xfs_efd_item_committing(
441 struct xfs_log_item
*lip
,
447 * This is the ops vector shared by all efd log items.
449 static const struct xfs_item_ops xfs_efd_item_ops
= {
450 .iop_size
= xfs_efd_item_size
,
451 .iop_format
= xfs_efd_item_format
,
452 .iop_pin
= xfs_efd_item_pin
,
453 .iop_unpin
= xfs_efd_item_unpin
,
454 .iop_unlock
= xfs_efd_item_unlock
,
455 .iop_committed
= xfs_efd_item_committed
,
456 .iop_push
= xfs_efd_item_push
,
457 .iop_committing
= xfs_efd_item_committing
461 * Allocate and initialize an efd item with the given number of extents.
463 struct xfs_efd_log_item
*
465 struct xfs_mount
*mp
,
466 struct xfs_efi_log_item
*efip
,
470 struct xfs_efd_log_item
*efdp
;
473 ASSERT(nextents
> 0);
474 if (nextents
> XFS_EFD_MAX_FAST_EXTENTS
) {
475 size
= (uint
)(sizeof(xfs_efd_log_item_t
) +
476 ((nextents
- 1) * sizeof(xfs_extent_t
)));
477 efdp
= kmem_zalloc(size
, KM_SLEEP
);
479 efdp
= kmem_zone_zalloc(xfs_efd_zone
, KM_SLEEP
);
482 xfs_log_item_init(mp
, &efdp
->efd_item
, XFS_LI_EFD
, &xfs_efd_item_ops
);
483 efdp
->efd_efip
= efip
;
484 efdp
->efd_format
.efd_nextents
= nextents
;
485 efdp
->efd_format
.efd_efi_id
= efip
->efi_format
.efi_id
;