Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
baf4bcac DW |
2 | /* |
3 | * Copyright (C) 2016 Oracle. All Rights Reserved. | |
baf4bcac | 4 | * Author: Darrick J. Wong <darrick.wong@oracle.com> |
baf4bcac 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" | |
f997ee21 | 11 | #include "xfs_bit.h" |
b31c2bdc | 12 | #include "xfs_shared.h" |
baf4bcac | 13 | #include "xfs_mount.h" |
f997ee21 | 14 | #include "xfs_defer.h" |
baf4bcac DW |
15 | #include "xfs_trans.h" |
16 | #include "xfs_trans_priv.h" | |
17 | #include "xfs_buf_item.h" | |
18 | #include "xfs_refcount_item.h" | |
19 | #include "xfs_log.h" | |
f997ee21 | 20 | #include "xfs_refcount.h" |
baf4bcac DW |
21 | |
22 | ||
23 | kmem_zone_t *xfs_cui_zone; | |
24 | kmem_zone_t *xfs_cud_zone; | |
25 | ||
26 | static inline struct xfs_cui_log_item *CUI_ITEM(struct xfs_log_item *lip) | |
27 | { | |
28 | return container_of(lip, struct xfs_cui_log_item, cui_item); | |
29 | } | |
30 | ||
31 | void | |
32 | xfs_cui_item_free( | |
33 | struct xfs_cui_log_item *cuip) | |
34 | { | |
35 | if (cuip->cui_format.cui_nextents > XFS_CUI_MAX_FAST_EXTENTS) | |
36 | kmem_free(cuip); | |
37 | else | |
38 | kmem_zone_free(xfs_cui_zone, cuip); | |
39 | } | |
40 | ||
0612d116 DC |
41 | /* |
42 | * Freeing the CUI requires that we remove it from the AIL if it has already | |
43 | * been placed there. However, the CUI may not yet have been placed in the AIL | |
44 | * when called by xfs_cui_release() from CUD 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 CUI. | |
47 | */ | |
48 | void | |
49 | xfs_cui_release( | |
50 | struct xfs_cui_log_item *cuip) | |
51 | { | |
52 | ASSERT(atomic_read(&cuip->cui_refcount) > 0); | |
53 | if (atomic_dec_and_test(&cuip->cui_refcount)) { | |
54 | xfs_trans_ail_remove(&cuip->cui_item, SHUTDOWN_LOG_IO_ERROR); | |
55 | xfs_cui_item_free(cuip); | |
56 | } | |
57 | } | |
58 | ||
59 | ||
baf4bcac DW |
60 | STATIC void |
61 | xfs_cui_item_size( | |
62 | struct xfs_log_item *lip, | |
63 | int *nvecs, | |
64 | int *nbytes) | |
65 | { | |
66 | struct xfs_cui_log_item *cuip = CUI_ITEM(lip); | |
67 | ||
68 | *nvecs += 1; | |
69 | *nbytes += xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents); | |
70 | } | |
71 | ||
72 | /* | |
73 | * This is called to fill in the vector of log iovecs for the | |
74 | * given cui log item. We use only 1 iovec, and we point that | |
75 | * at the cui_log_format structure embedded in the cui item. | |
76 | * It is at this point that we assert that all of the extent | |
77 | * slots in the cui item have been filled. | |
78 | */ | |
79 | STATIC void | |
80 | xfs_cui_item_format( | |
81 | struct xfs_log_item *lip, | |
82 | struct xfs_log_vec *lv) | |
83 | { | |
84 | struct xfs_cui_log_item *cuip = CUI_ITEM(lip); | |
85 | struct xfs_log_iovec *vecp = NULL; | |
86 | ||
87 | ASSERT(atomic_read(&cuip->cui_next_extent) == | |
88 | cuip->cui_format.cui_nextents); | |
89 | ||
90 | cuip->cui_format.cui_type = XFS_LI_CUI; | |
91 | cuip->cui_format.cui_size = 1; | |
92 | ||
93 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUI_FORMAT, &cuip->cui_format, | |
94 | xfs_cui_log_format_sizeof(cuip->cui_format.cui_nextents)); | |
95 | } | |
96 | ||
baf4bcac DW |
97 | /* |
98 | * The unpin operation is the last place an CUI is manipulated in the log. It is | |
99 | * either inserted in the AIL or aborted in the event of a log I/O error. In | |
100 | * either case, the CUI transaction has been successfully committed to make it | |
101 | * this far. Therefore, we expect whoever committed the CUI to either construct | |
102 | * and commit the CUD or drop the CUD's reference in the event of error. Simply | |
103 | * drop the log's CUI reference now that the log is done with it. | |
104 | */ | |
105 | STATIC void | |
106 | xfs_cui_item_unpin( | |
107 | struct xfs_log_item *lip, | |
108 | int remove) | |
109 | { | |
110 | struct xfs_cui_log_item *cuip = CUI_ITEM(lip); | |
111 | ||
112 | xfs_cui_release(cuip); | |
113 | } | |
114 | ||
baf4bcac DW |
115 | /* |
116 | * The CUI has been either committed or aborted if the transaction has been | |
117 | * cancelled. If the transaction was cancelled, an CUD isn't going to be | |
118 | * constructed and thus we free the CUI here directly. | |
119 | */ | |
120 | STATIC void | |
ddf92053 | 121 | xfs_cui_item_release( |
baf4bcac DW |
122 | struct xfs_log_item *lip) |
123 | { | |
ddf92053 | 124 | xfs_cui_release(CUI_ITEM(lip)); |
baf4bcac DW |
125 | } |
126 | ||
baf4bcac DW |
127 | static const struct xfs_item_ops xfs_cui_item_ops = { |
128 | .iop_size = xfs_cui_item_size, | |
129 | .iop_format = xfs_cui_item_format, | |
baf4bcac | 130 | .iop_unpin = xfs_cui_item_unpin, |
ddf92053 | 131 | .iop_release = xfs_cui_item_release, |
baf4bcac DW |
132 | }; |
133 | ||
134 | /* | |
135 | * Allocate and initialize an cui item with the given number of extents. | |
136 | */ | |
137 | struct xfs_cui_log_item * | |
138 | xfs_cui_init( | |
139 | struct xfs_mount *mp, | |
140 | uint nextents) | |
141 | ||
142 | { | |
143 | struct xfs_cui_log_item *cuip; | |
144 | ||
145 | ASSERT(nextents > 0); | |
146 | if (nextents > XFS_CUI_MAX_FAST_EXTENTS) | |
147 | cuip = kmem_zalloc(xfs_cui_log_item_sizeof(nextents), | |
148 | KM_SLEEP); | |
149 | else | |
150 | cuip = kmem_zone_zalloc(xfs_cui_zone, KM_SLEEP); | |
151 | ||
152 | xfs_log_item_init(mp, &cuip->cui_item, XFS_LI_CUI, &xfs_cui_item_ops); | |
153 | cuip->cui_format.cui_nextents = nextents; | |
154 | cuip->cui_format.cui_id = (uintptr_t)(void *)cuip; | |
155 | atomic_set(&cuip->cui_next_extent, 0); | |
156 | atomic_set(&cuip->cui_refcount, 2); | |
157 | ||
158 | return cuip; | |
159 | } | |
160 | ||
baf4bcac DW |
161 | static inline struct xfs_cud_log_item *CUD_ITEM(struct xfs_log_item *lip) |
162 | { | |
163 | return container_of(lip, struct xfs_cud_log_item, cud_item); | |
164 | } | |
165 | ||
166 | STATIC void | |
167 | xfs_cud_item_size( | |
168 | struct xfs_log_item *lip, | |
169 | int *nvecs, | |
170 | int *nbytes) | |
171 | { | |
172 | *nvecs += 1; | |
173 | *nbytes += sizeof(struct xfs_cud_log_format); | |
174 | } | |
175 | ||
176 | /* | |
177 | * This is called to fill in the vector of log iovecs for the | |
178 | * given cud log item. We use only 1 iovec, and we point that | |
179 | * at the cud_log_format structure embedded in the cud item. | |
180 | * It is at this point that we assert that all of the extent | |
181 | * slots in the cud item have been filled. | |
182 | */ | |
183 | STATIC void | |
184 | xfs_cud_item_format( | |
185 | struct xfs_log_item *lip, | |
186 | struct xfs_log_vec *lv) | |
187 | { | |
188 | struct xfs_cud_log_item *cudp = CUD_ITEM(lip); | |
189 | struct xfs_log_iovec *vecp = NULL; | |
190 | ||
191 | cudp->cud_format.cud_type = XFS_LI_CUD; | |
192 | cudp->cud_format.cud_size = 1; | |
193 | ||
194 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_CUD_FORMAT, &cudp->cud_format, | |
195 | sizeof(struct xfs_cud_log_format)); | |
196 | } | |
197 | ||
baf4bcac DW |
198 | /* |
199 | * The CUD is either committed or aborted if the transaction is cancelled. If | |
200 | * the transaction is cancelled, drop our reference to the CUI and free the | |
201 | * CUD. | |
202 | */ | |
203 | STATIC void | |
ddf92053 | 204 | xfs_cud_item_release( |
baf4bcac DW |
205 | struct xfs_log_item *lip) |
206 | { | |
207 | struct xfs_cud_log_item *cudp = CUD_ITEM(lip); | |
208 | ||
ddf92053 CH |
209 | xfs_cui_release(cudp->cud_cuip); |
210 | kmem_zone_free(xfs_cud_zone, cudp); | |
baf4bcac DW |
211 | } |
212 | ||
baf4bcac | 213 | static const struct xfs_item_ops xfs_cud_item_ops = { |
9ce632a2 | 214 | .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED, |
baf4bcac DW |
215 | .iop_size = xfs_cud_item_size, |
216 | .iop_format = xfs_cud_item_format, | |
ddf92053 | 217 | .iop_release = xfs_cud_item_release, |
baf4bcac DW |
218 | }; |
219 | ||
220 | /* | |
221 | * Allocate and initialize an cud item with the given number of extents. | |
222 | */ | |
223 | struct xfs_cud_log_item * | |
224 | xfs_cud_init( | |
225 | struct xfs_mount *mp, | |
226 | struct xfs_cui_log_item *cuip) | |
227 | ||
228 | { | |
229 | struct xfs_cud_log_item *cudp; | |
230 | ||
231 | cudp = kmem_zone_zalloc(xfs_cud_zone, KM_SLEEP); | |
232 | xfs_log_item_init(mp, &cudp->cud_item, XFS_LI_CUD, &xfs_cud_item_ops); | |
233 | cudp->cud_cuip = cuip; | |
234 | cudp->cud_format.cud_cui_id = cuip->cui_format.cui_id; | |
235 | ||
236 | return cudp; | |
237 | } | |
f997ee21 DW |
238 | |
239 | /* | |
240 | * Process a refcount update intent item that was recovered from the log. | |
241 | * We need to update the refcountbt. | |
242 | */ | |
243 | int | |
244 | xfs_cui_recover( | |
fbfa977d BF |
245 | struct xfs_trans *parent_tp, |
246 | struct xfs_cui_log_item *cuip) | |
f997ee21 DW |
247 | { |
248 | int i; | |
249 | int error = 0; | |
33ba6129 | 250 | unsigned int refc_type; |
f997ee21 DW |
251 | struct xfs_phys_extent *refc; |
252 | xfs_fsblock_t startblock_fsb; | |
253 | bool op_ok; | |
33ba6129 DW |
254 | struct xfs_cud_log_item *cudp; |
255 | struct xfs_trans *tp; | |
256 | struct xfs_btree_cur *rcur = NULL; | |
257 | enum xfs_refcount_intent_type type; | |
33ba6129 DW |
258 | xfs_fsblock_t new_fsb; |
259 | xfs_extlen_t new_len; | |
260 | struct xfs_bmbt_irec irec; | |
33ba6129 | 261 | bool requeue_only = false; |
fbfa977d | 262 | struct xfs_mount *mp = parent_tp->t_mountp; |
f997ee21 DW |
263 | |
264 | ASSERT(!test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags)); | |
265 | ||
266 | /* | |
267 | * First check the validity of the extents described by the | |
268 | * CUI. If any are bad, then assume that all are bad and | |
269 | * just toss the CUI. | |
270 | */ | |
271 | for (i = 0; i < cuip->cui_format.cui_nextents; i++) { | |
272 | refc = &cuip->cui_format.cui_extents[i]; | |
273 | startblock_fsb = XFS_BB_TO_FSB(mp, | |
274 | XFS_FSB_TO_DADDR(mp, refc->pe_startblock)); | |
275 | switch (refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK) { | |
276 | case XFS_REFCOUNT_INCREASE: | |
277 | case XFS_REFCOUNT_DECREASE: | |
278 | case XFS_REFCOUNT_ALLOC_COW: | |
279 | case XFS_REFCOUNT_FREE_COW: | |
280 | op_ok = true; | |
281 | break; | |
282 | default: | |
283 | op_ok = false; | |
284 | break; | |
285 | } | |
286 | if (!op_ok || startblock_fsb == 0 || | |
287 | refc->pe_len == 0 || | |
288 | startblock_fsb >= mp->m_sb.sb_dblocks || | |
289 | refc->pe_len >= mp->m_sb.sb_agblocks || | |
290 | (refc->pe_flags & ~XFS_REFCOUNT_EXTENT_FLAGS)) { | |
291 | /* | |
292 | * This will pull the CUI from the AIL and | |
293 | * free the memory associated with it. | |
294 | */ | |
295 | set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags); | |
296 | xfs_cui_release(cuip); | |
297 | return -EIO; | |
298 | } | |
299 | } | |
300 | ||
33ba6129 DW |
301 | /* |
302 | * Under normal operation, refcount updates are deferred, so we | |
303 | * wouldn't be adding them directly to a transaction. All | |
304 | * refcount updates manage reservation usage internally and | |
305 | * dynamically by deferring work that won't fit in the | |
306 | * transaction. Normally, any work that needs to be deferred | |
307 | * gets attached to the same defer_ops that scheduled the | |
308 | * refcount update. However, we're in log recovery here, so we | |
b31c2bdc DW |
309 | * we use the passed in defer_ops and to finish up any work that |
310 | * doesn't fit. We need to reserve enough blocks to handle a | |
311 | * full btree split on either end of the refcount range. | |
33ba6129 | 312 | */ |
b31c2bdc DW |
313 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, |
314 | mp->m_refc_maxlevels * 2, 0, XFS_TRANS_RESERVE, &tp); | |
33ba6129 DW |
315 | if (error) |
316 | return error; | |
91ef75b6 BF |
317 | /* |
318 | * Recovery stashes all deferred ops during intent processing and | |
319 | * finishes them on completion. Transfer current dfops state to this | |
320 | * transaction and transfer the result back before we return. | |
321 | */ | |
ce356d64 | 322 | xfs_defer_move(tp, parent_tp); |
33ba6129 DW |
323 | cudp = xfs_trans_get_cud(tp, cuip); |
324 | ||
33ba6129 DW |
325 | for (i = 0; i < cuip->cui_format.cui_nextents; i++) { |
326 | refc = &cuip->cui_format.cui_extents[i]; | |
327 | refc_type = refc->pe_flags & XFS_REFCOUNT_EXTENT_TYPE_MASK; | |
328 | switch (refc_type) { | |
329 | case XFS_REFCOUNT_INCREASE: | |
330 | case XFS_REFCOUNT_DECREASE: | |
331 | case XFS_REFCOUNT_ALLOC_COW: | |
332 | case XFS_REFCOUNT_FREE_COW: | |
333 | type = refc_type; | |
334 | break; | |
335 | default: | |
336 | error = -EFSCORRUPTED; | |
337 | goto abort_error; | |
338 | } | |
339 | if (requeue_only) { | |
340 | new_fsb = refc->pe_startblock; | |
341 | new_len = refc->pe_len; | |
342 | } else | |
343 | error = xfs_trans_log_finish_refcount_update(tp, cudp, | |
7dbddbac BF |
344 | type, refc->pe_startblock, refc->pe_len, |
345 | &new_fsb, &new_len, &rcur); | |
33ba6129 DW |
346 | if (error) |
347 | goto abort_error; | |
348 | ||
349 | /* Requeue what we didn't finish. */ | |
350 | if (new_len > 0) { | |
351 | irec.br_startblock = new_fsb; | |
352 | irec.br_blockcount = new_len; | |
353 | switch (type) { | |
354 | case XFS_REFCOUNT_INCREASE: | |
0f37d178 | 355 | error = xfs_refcount_increase_extent(tp, &irec); |
33ba6129 DW |
356 | break; |
357 | case XFS_REFCOUNT_DECREASE: | |
0f37d178 | 358 | error = xfs_refcount_decrease_extent(tp, &irec); |
33ba6129 | 359 | break; |
174edb0e | 360 | case XFS_REFCOUNT_ALLOC_COW: |
0f37d178 | 361 | error = xfs_refcount_alloc_cow_extent(tp, |
174edb0e DW |
362 | irec.br_startblock, |
363 | irec.br_blockcount); | |
364 | break; | |
365 | case XFS_REFCOUNT_FREE_COW: | |
0f37d178 | 366 | error = xfs_refcount_free_cow_extent(tp, |
174edb0e DW |
367 | irec.br_startblock, |
368 | irec.br_blockcount); | |
369 | break; | |
33ba6129 DW |
370 | default: |
371 | ASSERT(0); | |
372 | } | |
373 | if (error) | |
374 | goto abort_error; | |
375 | requeue_only = true; | |
376 | } | |
377 | } | |
378 | ||
379 | xfs_refcount_finish_one_cleanup(tp, rcur, error); | |
f997ee21 | 380 | set_bit(XFS_CUI_RECOVERED, &cuip->cui_flags); |
ce356d64 | 381 | xfs_defer_move(parent_tp, tp); |
33ba6129 DW |
382 | error = xfs_trans_commit(tp); |
383 | return error; | |
384 | ||
385 | abort_error: | |
386 | xfs_refcount_finish_one_cleanup(tp, rcur, error); | |
ce356d64 | 387 | xfs_defer_move(parent_tp, tp); |
33ba6129 | 388 | xfs_trans_cancel(tp); |
f997ee21 DW |
389 | return error; |
390 | } |