Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
5880f2d7 DW |
2 | /* |
3 | * Copyright (C) 2016 Oracle. All Rights Reserved. | |
5880f2d7 | 4 | * Author: Darrick J. Wong <darrick.wong@oracle.com> |
5880f2d7 DW |
5 | */ |
6 | #include "xfs.h" | |
7 | #include "xfs_fs.h" | |
8 | #include "xfs_format.h" | |
9 | #include "xfs_log_format.h" | |
10 | #include "xfs_trans_resv.h" | |
9e88b5d8 | 11 | #include "xfs_bit.h" |
b31c2bdc | 12 | #include "xfs_shared.h" |
5880f2d7 | 13 | #include "xfs_mount.h" |
9c194644 | 14 | #include "xfs_defer.h" |
5880f2d7 DW |
15 | #include "xfs_trans.h" |
16 | #include "xfs_trans_priv.h" | |
17 | #include "xfs_buf_item.h" | |
18 | #include "xfs_rmap_item.h" | |
19 | #include "xfs_log.h" | |
9c194644 | 20 | #include "xfs_rmap.h" |
5880f2d7 DW |
21 | |
22 | ||
23 | kmem_zone_t *xfs_rui_zone; | |
24 | kmem_zone_t *xfs_rud_zone; | |
25 | ||
26 | static inline struct xfs_rui_log_item *RUI_ITEM(struct xfs_log_item *lip) | |
27 | { | |
28 | return container_of(lip, struct xfs_rui_log_item, rui_item); | |
29 | } | |
30 | ||
31 | void | |
32 | xfs_rui_item_free( | |
33 | struct xfs_rui_log_item *ruip) | |
34 | { | |
35 | if (ruip->rui_format.rui_nextents > XFS_RUI_MAX_FAST_EXTENTS) | |
36 | kmem_free(ruip); | |
37 | else | |
38 | kmem_zone_free(xfs_rui_zone, ruip); | |
39 | } | |
40 | ||
0612d116 DC |
41 | /* |
42 | * Freeing the RUI requires that we remove it from the AIL if it has already | |
43 | * been placed there. However, the RUI may not yet have been placed in the AIL | |
44 | * when called by xfs_rui_release() from RUD processing due to the ordering of | |
45 | * committed vs unpin operations in bulk insert operations. Hence the reference | |
46 | * count to ensure only the last caller frees the RUI. | |
47 | */ | |
48 | void | |
49 | xfs_rui_release( | |
50 | struct xfs_rui_log_item *ruip) | |
51 | { | |
52 | ASSERT(atomic_read(&ruip->rui_refcount) > 0); | |
53 | if (atomic_dec_and_test(&ruip->rui_refcount)) { | |
54 | xfs_trans_ail_remove(&ruip->rui_item, SHUTDOWN_LOG_IO_ERROR); | |
55 | xfs_rui_item_free(ruip); | |
56 | } | |
57 | } | |
58 | ||
5880f2d7 DW |
59 | STATIC void |
60 | xfs_rui_item_size( | |
61 | struct xfs_log_item *lip, | |
62 | int *nvecs, | |
63 | int *nbytes) | |
64 | { | |
cd00158c DW |
65 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); |
66 | ||
5880f2d7 | 67 | *nvecs += 1; |
cd00158c | 68 | *nbytes += xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents); |
5880f2d7 DW |
69 | } |
70 | ||
71 | /* | |
72 | * This is called to fill in the vector of log iovecs for the | |
73 | * given rui log item. We use only 1 iovec, and we point that | |
74 | * at the rui_log_format structure embedded in the rui item. | |
75 | * It is at this point that we assert that all of the extent | |
76 | * slots in the rui item have been filled. | |
77 | */ | |
78 | STATIC void | |
79 | xfs_rui_item_format( | |
80 | struct xfs_log_item *lip, | |
81 | struct xfs_log_vec *lv) | |
82 | { | |
83 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); | |
84 | struct xfs_log_iovec *vecp = NULL; | |
85 | ||
86 | ASSERT(atomic_read(&ruip->rui_next_extent) == | |
87 | ruip->rui_format.rui_nextents); | |
88 | ||
89 | ruip->rui_format.rui_type = XFS_LI_RUI; | |
90 | ruip->rui_format.rui_size = 1; | |
91 | ||
92 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUI_FORMAT, &ruip->rui_format, | |
cd00158c | 93 | xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents)); |
5880f2d7 DW |
94 | } |
95 | ||
5880f2d7 DW |
96 | /* |
97 | * The unpin operation is the last place an RUI is manipulated in the log. It is | |
98 | * either inserted in the AIL or aborted in the event of a log I/O error. In | |
99 | * either case, the RUI transaction has been successfully committed to make it | |
100 | * this far. Therefore, we expect whoever committed the RUI to either construct | |
101 | * and commit the RUD or drop the RUD's reference in the event of error. Simply | |
102 | * drop the log's RUI reference now that the log is done with it. | |
103 | */ | |
104 | STATIC void | |
105 | xfs_rui_item_unpin( | |
106 | struct xfs_log_item *lip, | |
107 | int remove) | |
108 | { | |
109 | struct xfs_rui_log_item *ruip = RUI_ITEM(lip); | |
110 | ||
111 | xfs_rui_release(ruip); | |
112 | } | |
113 | ||
5880f2d7 DW |
114 | /* |
115 | * The RUI has been either committed or aborted if the transaction has been | |
116 | * cancelled. If the transaction was cancelled, an RUD isn't going to be | |
117 | * constructed and thus we free the RUI here directly. | |
118 | */ | |
119 | STATIC void | |
ddf92053 | 120 | xfs_rui_item_release( |
5880f2d7 DW |
121 | struct xfs_log_item *lip) |
122 | { | |
ddf92053 | 123 | xfs_rui_release(RUI_ITEM(lip)); |
5880f2d7 DW |
124 | } |
125 | ||
5880f2d7 DW |
126 | /* |
127 | * This is the ops vector shared by all rui log items. | |
128 | */ | |
129 | static const struct xfs_item_ops xfs_rui_item_ops = { | |
130 | .iop_size = xfs_rui_item_size, | |
131 | .iop_format = xfs_rui_item_format, | |
5880f2d7 | 132 | .iop_unpin = xfs_rui_item_unpin, |
ddf92053 | 133 | .iop_release = xfs_rui_item_release, |
5880f2d7 DW |
134 | }; |
135 | ||
136 | /* | |
137 | * Allocate and initialize an rui item with the given number of extents. | |
138 | */ | |
139 | struct xfs_rui_log_item * | |
140 | xfs_rui_init( | |
141 | struct xfs_mount *mp, | |
142 | uint nextents) | |
143 | ||
144 | { | |
145 | struct xfs_rui_log_item *ruip; | |
5880f2d7 DW |
146 | |
147 | ASSERT(nextents > 0); | |
cd00158c DW |
148 | if (nextents > XFS_RUI_MAX_FAST_EXTENTS) |
149 | ruip = kmem_zalloc(xfs_rui_log_item_sizeof(nextents), KM_SLEEP); | |
150 | else | |
5880f2d7 | 151 | ruip = kmem_zone_zalloc(xfs_rui_zone, KM_SLEEP); |
5880f2d7 DW |
152 | |
153 | xfs_log_item_init(mp, &ruip->rui_item, XFS_LI_RUI, &xfs_rui_item_ops); | |
154 | ruip->rui_format.rui_nextents = nextents; | |
155 | ruip->rui_format.rui_id = (uintptr_t)(void *)ruip; | |
156 | atomic_set(&ruip->rui_next_extent, 0); | |
157 | atomic_set(&ruip->rui_refcount, 2); | |
158 | ||
159 | return ruip; | |
160 | } | |
161 | ||
162 | /* | |
163 | * Copy an RUI format buffer from the given buf, and into the destination | |
164 | * RUI format structure. The RUI/RUD items were designed not to need any | |
165 | * special alignment handling. | |
166 | */ | |
167 | int | |
168 | xfs_rui_copy_format( | |
169 | struct xfs_log_iovec *buf, | |
170 | struct xfs_rui_log_format *dst_rui_fmt) | |
171 | { | |
172 | struct xfs_rui_log_format *src_rui_fmt; | |
173 | uint len; | |
174 | ||
175 | src_rui_fmt = buf->i_addr; | |
cd00158c | 176 | len = xfs_rui_log_format_sizeof(src_rui_fmt->rui_nextents); |
5880f2d7 DW |
177 | |
178 | if (buf->i_len != len) | |
179 | return -EFSCORRUPTED; | |
180 | ||
cd00158c | 181 | memcpy(dst_rui_fmt, src_rui_fmt, len); |
5880f2d7 DW |
182 | return 0; |
183 | } | |
184 | ||
5880f2d7 DW |
185 | static inline struct xfs_rud_log_item *RUD_ITEM(struct xfs_log_item *lip) |
186 | { | |
187 | return container_of(lip, struct xfs_rud_log_item, rud_item); | |
188 | } | |
189 | ||
5880f2d7 DW |
190 | STATIC void |
191 | xfs_rud_item_size( | |
192 | struct xfs_log_item *lip, | |
193 | int *nvecs, | |
194 | int *nbytes) | |
195 | { | |
196 | *nvecs += 1; | |
722e2517 | 197 | *nbytes += sizeof(struct xfs_rud_log_format); |
5880f2d7 DW |
198 | } |
199 | ||
200 | /* | |
201 | * This is called to fill in the vector of log iovecs for the | |
202 | * given rud log item. We use only 1 iovec, and we point that | |
203 | * at the rud_log_format structure embedded in the rud item. | |
204 | * It is at this point that we assert that all of the extent | |
205 | * slots in the rud item have been filled. | |
206 | */ | |
207 | STATIC void | |
208 | xfs_rud_item_format( | |
209 | struct xfs_log_item *lip, | |
210 | struct xfs_log_vec *lv) | |
211 | { | |
212 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
213 | struct xfs_log_iovec *vecp = NULL; | |
214 | ||
5880f2d7 DW |
215 | rudp->rud_format.rud_type = XFS_LI_RUD; |
216 | rudp->rud_format.rud_size = 1; | |
217 | ||
218 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUD_FORMAT, &rudp->rud_format, | |
722e2517 | 219 | sizeof(struct xfs_rud_log_format)); |
5880f2d7 DW |
220 | } |
221 | ||
5880f2d7 DW |
222 | /* |
223 | * The RUD is either committed or aborted if the transaction is cancelled. If | |
224 | * the transaction is cancelled, drop our reference to the RUI and free the | |
225 | * RUD. | |
226 | */ | |
227 | STATIC void | |
ddf92053 | 228 | xfs_rud_item_release( |
5880f2d7 DW |
229 | struct xfs_log_item *lip) |
230 | { | |
231 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
232 | ||
ddf92053 CH |
233 | xfs_rui_release(rudp->rud_ruip); |
234 | kmem_zone_free(xfs_rud_zone, rudp); | |
5880f2d7 DW |
235 | } |
236 | ||
237 | /* | |
238 | * When the rud item is committed to disk, all we need to do is delete our | |
239 | * reference to our partner rui item and then free ourselves. Since we're | |
240 | * freeing ourselves we must return -1 to keep the transaction code from | |
241 | * further referencing this item. | |
242 | */ | |
243 | STATIC xfs_lsn_t | |
244 | xfs_rud_item_committed( | |
245 | struct xfs_log_item *lip, | |
246 | xfs_lsn_t lsn) | |
247 | { | |
248 | struct xfs_rud_log_item *rudp = RUD_ITEM(lip); | |
249 | ||
250 | /* | |
251 | * Drop the RUI reference regardless of whether the RUD has been | |
252 | * aborted. Once the RUD transaction is constructed, it is the sole | |
253 | * responsibility of the RUD to release the RUI (even if the RUI is | |
254 | * aborted due to log I/O error). | |
255 | */ | |
256 | xfs_rui_release(rudp->rud_ruip); | |
722e2517 | 257 | kmem_zone_free(xfs_rud_zone, rudp); |
5880f2d7 DW |
258 | |
259 | return (xfs_lsn_t)-1; | |
260 | } | |
261 | ||
5880f2d7 DW |
262 | /* |
263 | * This is the ops vector shared by all rud log items. | |
264 | */ | |
265 | static const struct xfs_item_ops xfs_rud_item_ops = { | |
266 | .iop_size = xfs_rud_item_size, | |
267 | .iop_format = xfs_rud_item_format, | |
ddf92053 | 268 | .iop_release = xfs_rud_item_release, |
5880f2d7 | 269 | .iop_committed = xfs_rud_item_committed, |
5880f2d7 DW |
270 | }; |
271 | ||
272 | /* | |
273 | * Allocate and initialize an rud item with the given number of extents. | |
274 | */ | |
275 | struct xfs_rud_log_item * | |
276 | xfs_rud_init( | |
277 | struct xfs_mount *mp, | |
722e2517 | 278 | struct xfs_rui_log_item *ruip) |
5880f2d7 DW |
279 | |
280 | { | |
281 | struct xfs_rud_log_item *rudp; | |
5880f2d7 | 282 | |
722e2517 | 283 | rudp = kmem_zone_zalloc(xfs_rud_zone, KM_SLEEP); |
5880f2d7 DW |
284 | xfs_log_item_init(mp, &rudp->rud_item, XFS_LI_RUD, &xfs_rud_item_ops); |
285 | rudp->rud_ruip = ruip; | |
5880f2d7 DW |
286 | rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id; |
287 | ||
288 | return rudp; | |
289 | } | |
9e88b5d8 DW |
290 | |
291 | /* | |
292 | * Process an rmap update intent item that was recovered from the log. | |
293 | * We need to update the rmapbt. | |
294 | */ | |
295 | int | |
296 | xfs_rui_recover( | |
297 | struct xfs_mount *mp, | |
298 | struct xfs_rui_log_item *ruip) | |
299 | { | |
300 | int i; | |
301 | int error = 0; | |
302 | struct xfs_map_extent *rmap; | |
303 | xfs_fsblock_t startblock_fsb; | |
304 | bool op_ok; | |
9c194644 DW |
305 | struct xfs_rud_log_item *rudp; |
306 | enum xfs_rmap_intent_type type; | |
307 | int whichfork; | |
308 | xfs_exntst_t state; | |
309 | struct xfs_trans *tp; | |
310 | struct xfs_btree_cur *rcur = NULL; | |
9e88b5d8 DW |
311 | |
312 | ASSERT(!test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags)); | |
313 | ||
314 | /* | |
315 | * First check the validity of the extents described by the | |
316 | * RUI. If any are bad, then assume that all are bad and | |
317 | * just toss the RUI. | |
318 | */ | |
319 | for (i = 0; i < ruip->rui_format.rui_nextents; i++) { | |
e127fafd | 320 | rmap = &ruip->rui_format.rui_extents[i]; |
9e88b5d8 DW |
321 | startblock_fsb = XFS_BB_TO_FSB(mp, |
322 | XFS_FSB_TO_DADDR(mp, rmap->me_startblock)); | |
323 | switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) { | |
324 | case XFS_RMAP_EXTENT_MAP: | |
0e07c039 | 325 | case XFS_RMAP_EXTENT_MAP_SHARED: |
9e88b5d8 | 326 | case XFS_RMAP_EXTENT_UNMAP: |
0e07c039 | 327 | case XFS_RMAP_EXTENT_UNMAP_SHARED: |
9e88b5d8 | 328 | case XFS_RMAP_EXTENT_CONVERT: |
0e07c039 | 329 | case XFS_RMAP_EXTENT_CONVERT_SHARED: |
9e88b5d8 DW |
330 | case XFS_RMAP_EXTENT_ALLOC: |
331 | case XFS_RMAP_EXTENT_FREE: | |
332 | op_ok = true; | |
333 | break; | |
334 | default: | |
335 | op_ok = false; | |
336 | break; | |
337 | } | |
e127fafd DW |
338 | if (!op_ok || startblock_fsb == 0 || |
339 | rmap->me_len == 0 || | |
340 | startblock_fsb >= mp->m_sb.sb_dblocks || | |
341 | rmap->me_len >= mp->m_sb.sb_agblocks || | |
9e88b5d8 DW |
342 | (rmap->me_flags & ~XFS_RMAP_EXTENT_FLAGS)) { |
343 | /* | |
344 | * This will pull the RUI from the AIL and | |
345 | * free the memory associated with it. | |
346 | */ | |
347 | set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags); | |
348 | xfs_rui_release(ruip); | |
349 | return -EIO; | |
350 | } | |
351 | } | |
352 | ||
b31c2bdc DW |
353 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, |
354 | mp->m_rmap_maxlevels, 0, XFS_TRANS_RESERVE, &tp); | |
9c194644 DW |
355 | if (error) |
356 | return error; | |
722e2517 | 357 | rudp = xfs_trans_get_rud(tp, ruip); |
9c194644 DW |
358 | |
359 | for (i = 0; i < ruip->rui_format.rui_nextents; i++) { | |
e127fafd | 360 | rmap = &ruip->rui_format.rui_extents[i]; |
9c194644 DW |
361 | state = (rmap->me_flags & XFS_RMAP_EXTENT_UNWRITTEN) ? |
362 | XFS_EXT_UNWRITTEN : XFS_EXT_NORM; | |
363 | whichfork = (rmap->me_flags & XFS_RMAP_EXTENT_ATTR_FORK) ? | |
364 | XFS_ATTR_FORK : XFS_DATA_FORK; | |
365 | switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) { | |
366 | case XFS_RMAP_EXTENT_MAP: | |
367 | type = XFS_RMAP_MAP; | |
368 | break; | |
ceeb9c83 DW |
369 | case XFS_RMAP_EXTENT_MAP_SHARED: |
370 | type = XFS_RMAP_MAP_SHARED; | |
371 | break; | |
9c194644 DW |
372 | case XFS_RMAP_EXTENT_UNMAP: |
373 | type = XFS_RMAP_UNMAP; | |
374 | break; | |
ceeb9c83 DW |
375 | case XFS_RMAP_EXTENT_UNMAP_SHARED: |
376 | type = XFS_RMAP_UNMAP_SHARED; | |
377 | break; | |
9c194644 DW |
378 | case XFS_RMAP_EXTENT_CONVERT: |
379 | type = XFS_RMAP_CONVERT; | |
380 | break; | |
3f165b33 DW |
381 | case XFS_RMAP_EXTENT_CONVERT_SHARED: |
382 | type = XFS_RMAP_CONVERT_SHARED; | |
383 | break; | |
9c194644 DW |
384 | case XFS_RMAP_EXTENT_ALLOC: |
385 | type = XFS_RMAP_ALLOC; | |
386 | break; | |
387 | case XFS_RMAP_EXTENT_FREE: | |
388 | type = XFS_RMAP_FREE; | |
389 | break; | |
390 | default: | |
391 | error = -EFSCORRUPTED; | |
392 | goto abort_error; | |
393 | } | |
394 | error = xfs_trans_log_finish_rmap_update(tp, rudp, type, | |
395 | rmap->me_owner, whichfork, | |
396 | rmap->me_startoff, rmap->me_startblock, | |
397 | rmap->me_len, state, &rcur); | |
398 | if (error) | |
399 | goto abort_error; | |
400 | ||
401 | } | |
402 | ||
403 | xfs_rmap_finish_one_cleanup(tp, rcur, error); | |
9e88b5d8 | 404 | set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags); |
9c194644 DW |
405 | error = xfs_trans_commit(tp); |
406 | return error; | |
407 | ||
408 | abort_error: | |
409 | xfs_rmap_finish_one_cleanup(tp, rcur, error); | |
410 | xfs_trans_cancel(tp); | |
9e88b5d8 DW |
411 | return error; |
412 | } |