]>
Commit | Line | Data |
---|---|---|
5880f2d7 DW |
1 | /* |
2 | * Copyright (C) 2016 Oracle. All Rights Reserved. | |
3 | * | |
4 | * Author: Darrick J. Wong <darrick.wong@oracle.com> | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version 2 | |
9 | * of the License, or (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it would be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write the Free Software Foundation, | |
18 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. | |
19 | */ | |
20 | #include "xfs.h" | |
21 | #include "xfs_fs.h" | |
22 | #include "xfs_format.h" | |
23 | #include "xfs_log_format.h" | |
24 | #include "xfs_trans_resv.h" | |
25 | #include "xfs_mount.h" | |
26 | #include "xfs_trans.h" | |
27 | #include "xfs_trans_priv.h" | |
28 | #include "xfs_buf_item.h" | |
29 | #include "xfs_rmap_item.h" | |
30 | #include "xfs_log.h" | |
31 | ||
32 | ||
33 | kmem_zone_t *xfs_rui_zone; | |
34 | kmem_zone_t *xfs_rud_zone; | |
35 | ||
36 | static inline struct xfs_rui_log_item *RUI_ITEM(struct xfs_log_item *lip) | |
37 | { | |
38 | return container_of(lip, struct xfs_rui_log_item, rui_item); | |
39 | } | |
40 | ||
41 | void | |
42 | xfs_rui_item_free( | |
43 | struct xfs_rui_log_item *ruip) | |
44 | { | |
45 | if (ruip->rui_format.rui_nextents > XFS_RUI_MAX_FAST_EXTENTS) | |
46 | kmem_free(ruip); | |
47 | else | |
48 | kmem_zone_free(xfs_rui_zone, ruip); | |
49 | } | |
50 | ||
51 | /* | |
52 | * This returns the number of iovecs needed to log the given rui item. | |
53 | * We only need 1 iovec for an rui item. It just logs the rui_log_format | |
54 | * structure. | |
55 | */ | |
56 | static inline int | |
57 | xfs_rui_item_sizeof( | |
58 | struct xfs_rui_log_item *ruip) | |
59 | { | |
60 | return sizeof(struct xfs_rui_log_format) + | |
61 | (ruip->rui_format.rui_nextents - 1) * | |
62 | sizeof(struct xfs_map_extent); | |
63 | } | |
64 | ||
65 | STATIC void | |
66 | xfs_rui_item_size( | |
67 | struct xfs_log_item *lip, | |
68 | int *nvecs, | |
69 | int *nbytes) | |
70 | { | |
71 | *nvecs += 1; | |
72 | *nbytes += xfs_rui_item_sizeof(RUI_ITEM(lip)); | |
73 | } | |
74 | ||
75 | /* | |
76 | * This is called to fill in the vector of log iovecs for the | |
77 | * given rui log item. We use only 1 iovec, and we point that | |
78 | * at the rui_log_format structure embedded in the rui item. | |
79 | * It is at this point that we assert that all of the extent | |
80 | * slots in the rui item have been filled. | |
81 | */ | |
82 | STATIC void | |
83 | xfs_rui_item_format( | |
84 | struct xfs_log_item *lip, | |
85 | struct xfs_log_vec *lv) | |
86 | { | |
87 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); | |
88 | struct xfs_log_iovec *vecp = NULL; | |
89 | ||
90 | ASSERT(atomic_read(&ruip->rui_next_extent) == | |
91 | ruip->rui_format.rui_nextents); | |
92 | ||
93 | ruip->rui_format.rui_type = XFS_LI_RUI; | |
94 | ruip->rui_format.rui_size = 1; | |
95 | ||
96 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUI_FORMAT, &ruip->rui_format, | |
97 | xfs_rui_item_sizeof(ruip)); | |
98 | } | |
99 | ||
100 | /* | |
101 | * Pinning has no meaning for an rui item, so just return. | |
102 | */ | |
103 | STATIC void | |
104 | xfs_rui_item_pin( | |
105 | struct xfs_log_item *lip) | |
106 | { | |
107 | } | |
108 | ||
109 | /* | |
110 | * The unpin operation is the last place an RUI 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 RUI transaction has been successfully committed to make it | |
113 | * this far. Therefore, we expect whoever committed the RUI to either construct | |
114 | * and commit the RUD or drop the RUD's reference in the event of error. Simply | |
115 | * drop the log's RUI reference now that the log is done with it. | |
116 | */ | |
117 | STATIC void | |
118 | xfs_rui_item_unpin( | |
119 | struct xfs_log_item *lip, | |
120 | int remove) | |
121 | { | |
122 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); | |
123 | ||
124 | xfs_rui_release(ruip); | |
125 | } | |
126 | ||
127 | /* | |
128 | * RUI items have no locking or pushing. However, since RUIs are pulled from | |
129 | * the AIL when their corresponding RUDs are committed to disk, their situation | |
130 | * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller | |
131 | * will eventually flush the log. This should help in getting the RUI out of | |
132 | * the AIL. | |
133 | */ | |
134 | STATIC uint | |
135 | xfs_rui_item_push( | |
136 | struct xfs_log_item *lip, | |
137 | struct list_head *buffer_list) | |
138 | { | |
139 | return XFS_ITEM_PINNED; | |
140 | } | |
141 | ||
142 | /* | |
143 | * The RUI has been either committed or aborted if the transaction has been | |
144 | * cancelled. If the transaction was cancelled, an RUD isn't going to be | |
145 | * constructed and thus we free the RUI here directly. | |
146 | */ | |
147 | STATIC void | |
148 | xfs_rui_item_unlock( | |
149 | struct xfs_log_item *lip) | |
150 | { | |
151 | if (lip->li_flags & XFS_LI_ABORTED) | |
152 | xfs_rui_item_free(RUI_ITEM(lip)); | |
153 | } | |
154 | ||
155 | /* | |
156 | * The RUI is logged only once and cannot be moved in the log, so simply return | |
157 | * the lsn at which it's been logged. | |
158 | */ | |
159 | STATIC xfs_lsn_t | |
160 | xfs_rui_item_committed( | |
161 | struct xfs_log_item *lip, | |
162 | xfs_lsn_t lsn) | |
163 | { | |
164 | return lsn; | |
165 | } | |
166 | ||
167 | /* | |
168 | * The RUI dependency tracking op doesn't do squat. It can't because | |
169 | * it doesn't know where the free extent is coming from. The dependency | |
170 | * tracking has to be handled by the "enclosing" metadata object. For | |
171 | * example, for inodes, the inode is locked throughout the extent freeing | |
172 | * so the dependency should be recorded there. | |
173 | */ | |
174 | STATIC void | |
175 | xfs_rui_item_committing( | |
176 | struct xfs_log_item *lip, | |
177 | xfs_lsn_t lsn) | |
178 | { | |
179 | } | |
180 | ||
181 | /* | |
182 | * This is the ops vector shared by all rui log items. | |
183 | */ | |
184 | static const struct xfs_item_ops xfs_rui_item_ops = { | |
185 | .iop_size = xfs_rui_item_size, | |
186 | .iop_format = xfs_rui_item_format, | |
187 | .iop_pin = xfs_rui_item_pin, | |
188 | .iop_unpin = xfs_rui_item_unpin, | |
189 | .iop_unlock = xfs_rui_item_unlock, | |
190 | .iop_committed = xfs_rui_item_committed, | |
191 | .iop_push = xfs_rui_item_push, | |
192 | .iop_committing = xfs_rui_item_committing, | |
193 | }; | |
194 | ||
195 | /* | |
196 | * Allocate and initialize an rui item with the given number of extents. | |
197 | */ | |
198 | struct xfs_rui_log_item * | |
199 | xfs_rui_init( | |
200 | struct xfs_mount *mp, | |
201 | uint nextents) | |
202 | ||
203 | { | |
204 | struct xfs_rui_log_item *ruip; | |
205 | uint size; | |
206 | ||
207 | ASSERT(nextents > 0); | |
208 | if (nextents > XFS_RUI_MAX_FAST_EXTENTS) { | |
209 | size = (uint)(sizeof(struct xfs_rui_log_item) + | |
210 | ((nextents - 1) * sizeof(struct xfs_map_extent))); | |
211 | ruip = kmem_zalloc(size, KM_SLEEP); | |
212 | } else { | |
213 | ruip = kmem_zone_zalloc(xfs_rui_zone, KM_SLEEP); | |
214 | } | |
215 | ||
216 | xfs_log_item_init(mp, &ruip->rui_item, XFS_LI_RUI, &xfs_rui_item_ops); | |
217 | ruip->rui_format.rui_nextents = nextents; | |
218 | ruip->rui_format.rui_id = (uintptr_t)(void *)ruip; | |
219 | atomic_set(&ruip->rui_next_extent, 0); | |
220 | atomic_set(&ruip->rui_refcount, 2); | |
221 | ||
222 | return ruip; | |
223 | } | |
224 | ||
225 | /* | |
226 | * Copy an RUI format buffer from the given buf, and into the destination | |
227 | * RUI format structure. The RUI/RUD items were designed not to need any | |
228 | * special alignment handling. | |
229 | */ | |
230 | int | |
231 | xfs_rui_copy_format( | |
232 | struct xfs_log_iovec *buf, | |
233 | struct xfs_rui_log_format *dst_rui_fmt) | |
234 | { | |
235 | struct xfs_rui_log_format *src_rui_fmt; | |
236 | uint len; | |
237 | ||
238 | src_rui_fmt = buf->i_addr; | |
239 | len = sizeof(struct xfs_rui_log_format) + | |
240 | (src_rui_fmt->rui_nextents - 1) * | |
241 | sizeof(struct xfs_map_extent); | |
242 | ||
243 | if (buf->i_len != len) | |
244 | return -EFSCORRUPTED; | |
245 | ||
246 | memcpy((char *)dst_rui_fmt, (char *)src_rui_fmt, len); | |
247 | return 0; | |
248 | } | |
249 | ||
250 | /* | |
251 | * Freeing the RUI requires that we remove it from the AIL if it has already | |
252 | * been placed there. However, the RUI may not yet have been placed in the AIL | |
253 | * when called by xfs_rui_release() from RUD processing due to the ordering of | |
254 | * committed vs unpin operations in bulk insert operations. Hence the reference | |
255 | * count to ensure only the last caller frees the RUI. | |
256 | */ | |
257 | void | |
258 | xfs_rui_release( | |
259 | struct xfs_rui_log_item *ruip) | |
260 | { | |
261 | if (atomic_dec_and_test(&ruip->rui_refcount)) { | |
262 | xfs_trans_ail_remove(&ruip->rui_item, SHUTDOWN_LOG_IO_ERROR); | |
263 | xfs_rui_item_free(ruip); | |
264 | } | |
265 | } | |
266 | ||
267 | static inline struct xfs_rud_log_item *RUD_ITEM(struct xfs_log_item *lip) | |
268 | { | |
269 | return container_of(lip, struct xfs_rud_log_item, rud_item); | |
270 | } | |
271 | ||
272 | STATIC void | |
273 | xfs_rud_item_free(struct xfs_rud_log_item *rudp) | |
274 | { | |
275 | if (rudp->rud_format.rud_nextents > XFS_RUD_MAX_FAST_EXTENTS) | |
276 | kmem_free(rudp); | |
277 | else | |
278 | kmem_zone_free(xfs_rud_zone, rudp); | |
279 | } | |
280 | ||
281 | /* | |
282 | * This returns the number of iovecs needed to log the given rud item. | |
283 | * We only need 1 iovec for an rud item. It just logs the rud_log_format | |
284 | * structure. | |
285 | */ | |
286 | static inline int | |
287 | xfs_rud_item_sizeof( | |
288 | struct xfs_rud_log_item *rudp) | |
289 | { | |
290 | return sizeof(struct xfs_rud_log_format) + | |
291 | (rudp->rud_format.rud_nextents - 1) * | |
292 | sizeof(struct xfs_map_extent); | |
293 | } | |
294 | ||
295 | STATIC void | |
296 | xfs_rud_item_size( | |
297 | struct xfs_log_item *lip, | |
298 | int *nvecs, | |
299 | int *nbytes) | |
300 | { | |
301 | *nvecs += 1; | |
302 | *nbytes += xfs_rud_item_sizeof(RUD_ITEM(lip)); | |
303 | } | |
304 | ||
305 | /* | |
306 | * This is called to fill in the vector of log iovecs for the | |
307 | * given rud log item. We use only 1 iovec, and we point that | |
308 | * at the rud_log_format structure embedded in the rud item. | |
309 | * It is at this point that we assert that all of the extent | |
310 | * slots in the rud item have been filled. | |
311 | */ | |
312 | STATIC void | |
313 | xfs_rud_item_format( | |
314 | struct xfs_log_item *lip, | |
315 | struct xfs_log_vec *lv) | |
316 | { | |
317 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
318 | struct xfs_log_iovec *vecp = NULL; | |
319 | ||
320 | ASSERT(rudp->rud_next_extent == rudp->rud_format.rud_nextents); | |
321 | ||
322 | rudp->rud_format.rud_type = XFS_LI_RUD; | |
323 | rudp->rud_format.rud_size = 1; | |
324 | ||
325 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUD_FORMAT, &rudp->rud_format, | |
326 | xfs_rud_item_sizeof(rudp)); | |
327 | } | |
328 | ||
329 | /* | |
330 | * Pinning has no meaning for an rud item, so just return. | |
331 | */ | |
332 | STATIC void | |
333 | xfs_rud_item_pin( | |
334 | struct xfs_log_item *lip) | |
335 | { | |
336 | } | |
337 | ||
338 | /* | |
339 | * Since pinning has no meaning for an rud item, unpinning does | |
340 | * not either. | |
341 | */ | |
342 | STATIC void | |
343 | xfs_rud_item_unpin( | |
344 | struct xfs_log_item *lip, | |
345 | int remove) | |
346 | { | |
347 | } | |
348 | ||
349 | /* | |
350 | * There isn't much you can do to push on an rud item. It is simply stuck | |
351 | * waiting for the log to be flushed to disk. | |
352 | */ | |
353 | STATIC uint | |
354 | xfs_rud_item_push( | |
355 | struct xfs_log_item *lip, | |
356 | struct list_head *buffer_list) | |
357 | { | |
358 | return XFS_ITEM_PINNED; | |
359 | } | |
360 | ||
361 | /* | |
362 | * The RUD is either committed or aborted if the transaction is cancelled. If | |
363 | * the transaction is cancelled, drop our reference to the RUI and free the | |
364 | * RUD. | |
365 | */ | |
366 | STATIC void | |
367 | xfs_rud_item_unlock( | |
368 | struct xfs_log_item *lip) | |
369 | { | |
370 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
371 | ||
372 | if (lip->li_flags & XFS_LI_ABORTED) { | |
373 | xfs_rui_release(rudp->rud_ruip); | |
374 | xfs_rud_item_free(rudp); | |
375 | } | |
376 | } | |
377 | ||
378 | /* | |
379 | * When the rud item is committed to disk, all we need to do is delete our | |
380 | * reference to our partner rui item and then free ourselves. Since we're | |
381 | * freeing ourselves we must return -1 to keep the transaction code from | |
382 | * further referencing this item. | |
383 | */ | |
384 | STATIC xfs_lsn_t | |
385 | xfs_rud_item_committed( | |
386 | struct xfs_log_item *lip, | |
387 | xfs_lsn_t lsn) | |
388 | { | |
389 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
390 | ||
391 | /* | |
392 | * Drop the RUI reference regardless of whether the RUD has been | |
393 | * aborted. Once the RUD transaction is constructed, it is the sole | |
394 | * responsibility of the RUD to release the RUI (even if the RUI is | |
395 | * aborted due to log I/O error). | |
396 | */ | |
397 | xfs_rui_release(rudp->rud_ruip); | |
398 | xfs_rud_item_free(rudp); | |
399 | ||
400 | return (xfs_lsn_t)-1; | |
401 | } | |
402 | ||
403 | /* | |
404 | * The RUD dependency tracking op doesn't do squat. It can't because | |
405 | * it doesn't know where the free extent is coming from. The dependency | |
406 | * tracking has to be handled by the "enclosing" metadata object. For | |
407 | * example, for inodes, the inode is locked throughout the extent freeing | |
408 | * so the dependency should be recorded there. | |
409 | */ | |
410 | STATIC void | |
411 | xfs_rud_item_committing( | |
412 | struct xfs_log_item *lip, | |
413 | xfs_lsn_t lsn) | |
414 | { | |
415 | } | |
416 | ||
417 | /* | |
418 | * This is the ops vector shared by all rud log items. | |
419 | */ | |
420 | static const struct xfs_item_ops xfs_rud_item_ops = { | |
421 | .iop_size = xfs_rud_item_size, | |
422 | .iop_format = xfs_rud_item_format, | |
423 | .iop_pin = xfs_rud_item_pin, | |
424 | .iop_unpin = xfs_rud_item_unpin, | |
425 | .iop_unlock = xfs_rud_item_unlock, | |
426 | .iop_committed = xfs_rud_item_committed, | |
427 | .iop_push = xfs_rud_item_push, | |
428 | .iop_committing = xfs_rud_item_committing, | |
429 | }; | |
430 | ||
431 | /* | |
432 | * Allocate and initialize an rud item with the given number of extents. | |
433 | */ | |
434 | struct xfs_rud_log_item * | |
435 | xfs_rud_init( | |
436 | struct xfs_mount *mp, | |
437 | struct xfs_rui_log_item *ruip, | |
438 | uint nextents) | |
439 | ||
440 | { | |
441 | struct xfs_rud_log_item *rudp; | |
442 | uint size; | |
443 | ||
444 | ASSERT(nextents > 0); | |
445 | if (nextents > XFS_RUD_MAX_FAST_EXTENTS) { | |
446 | size = (uint)(sizeof(struct xfs_rud_log_item) + | |
447 | ((nextents - 1) * sizeof(struct xfs_map_extent))); | |
448 | rudp = kmem_zalloc(size, KM_SLEEP); | |
449 | } else { | |
450 | rudp = kmem_zone_zalloc(xfs_rud_zone, KM_SLEEP); | |
451 | } | |
452 | ||
453 | xfs_log_item_init(mp, &rudp->rud_item, XFS_LI_RUD, &xfs_rud_item_ops); | |
454 | rudp->rud_ruip = ruip; | |
455 | rudp->rud_format.rud_nextents = nextents; | |
456 | rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id; | |
457 | ||
458 | return rudp; | |
459 | } |