1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. |
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" |
11 | #include "xfs_bit.h" |
12 | #include "xfs_shared.h" |
13 | #include "xfs_mount.h" |
14 | #include "xfs_ag.h" |
15 | #include "xfs_defer.h" |
16 | #include "xfs_trans.h" |
17 | #include "xfs_trans_priv.h" |
18 | #include "xfs_extfree_item.h" |
19 | #include "xfs_log.h" |
20 | #include "xfs_btree.h" |
21 | #include "xfs_rmap.h" |
22 | #include "xfs_alloc.h" |
23 | #include "xfs_bmap.h" |
24 | #include "xfs_trace.h" |
25 | #include "xfs_error.h" |
26 | #include "xfs_log_priv.h" |
27 | #include "xfs_log_recover.h" |
28 | |
29 | struct kmem_cache *xfs_efi_cache; |
30 | struct kmem_cache *xfs_efd_cache; |
31 | |
32 | static const struct xfs_item_ops xfs_efi_item_ops; |
33 | |
34 | static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip) |
35 | { |
36 | return container_of(lip, struct xfs_efi_log_item, efi_item); |
37 | } |
38 | |
39 | STATIC void |
40 | xfs_efi_item_free( |
41 | struct xfs_efi_log_item *efip) |
42 | { |
43 | kmem_free(ptr: efip->efi_item.li_lv_shadow); |
44 | if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS) |
45 | kmem_free(ptr: efip); |
46 | else |
47 | kmem_cache_free(s: xfs_efi_cache, objp: efip); |
48 | } |
49 | |
50 | /* |
51 | * Freeing the efi requires that we remove it from the AIL if it has already |
52 | * been placed there. However, the EFI may not yet have been placed in the AIL |
53 | * when called by xfs_efi_release() from EFD processing due to the ordering of |
54 | * committed vs unpin operations in bulk insert operations. Hence the reference |
55 | * count to ensure only the last caller frees the EFI. |
56 | */ |
57 | STATIC void |
58 | xfs_efi_release( |
59 | struct xfs_efi_log_item *efip) |
60 | { |
61 | ASSERT(atomic_read(&efip->efi_refcount) > 0); |
62 | if (!atomic_dec_and_test(v: &efip->efi_refcount)) |
63 | return; |
64 | |
65 | xfs_trans_ail_delete(lip: &efip->efi_item, shutdown_type: 0); |
66 | xfs_efi_item_free(efip); |
67 | } |
68 | |
69 | STATIC void |
70 | xfs_efi_item_size( |
71 | struct xfs_log_item *lip, |
72 | int *nvecs, |
73 | int *nbytes) |
74 | { |
75 | struct xfs_efi_log_item *efip = EFI_ITEM(lip); |
76 | |
77 | *nvecs += 1; |
78 | *nbytes += xfs_efi_log_format_sizeof(efip->efi_format.efi_nextents); |
79 | } |
80 | |
81 | /* |
82 | * This is called to fill in the vector of log iovecs for the |
83 | * given efi log item. We use only 1 iovec, and we point that |
84 | * at the efi_log_format structure embedded in the efi item. |
85 | * It is at this point that we assert that all of the extent |
86 | * slots in the efi item have been filled. |
87 | */ |
88 | STATIC void |
89 | xfs_efi_item_format( |
90 | struct xfs_log_item *lip, |
91 | struct xfs_log_vec *lv) |
92 | { |
93 | struct xfs_efi_log_item *efip = EFI_ITEM(lip); |
94 | struct xfs_log_iovec *vecp = NULL; |
95 | |
96 | ASSERT(atomic_read(&efip->efi_next_extent) == |
97 | efip->efi_format.efi_nextents); |
98 | |
99 | efip->efi_format.efi_type = XFS_LI_EFI; |
100 | efip->efi_format.efi_size = 1; |
101 | |
102 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT, |
103 | &efip->efi_format, |
104 | xfs_efi_log_format_sizeof(efip->efi_format.efi_nextents)); |
105 | } |
106 | |
107 | |
108 | /* |
109 | * The unpin operation is the last place an EFI is manipulated in the log. It is |
110 | * either inserted in the AIL or aborted in the event of a log I/O error. In |
111 | * either case, the EFI transaction has been successfully committed to make it |
112 | * this far. Therefore, we expect whoever committed the EFI to either construct |
113 | * and commit the EFD or drop the EFD's reference in the event of error. Simply |
114 | * drop the log's EFI reference now that the log is done with it. |
115 | */ |
116 | STATIC void |
117 | xfs_efi_item_unpin( |
118 | struct xfs_log_item *lip, |
119 | int remove) |
120 | { |
121 | struct xfs_efi_log_item *efip = EFI_ITEM(lip); |
122 | xfs_efi_release(efip); |
123 | } |
124 | |
125 | /* |
126 | * The EFI has been either committed or aborted if the transaction has been |
127 | * cancelled. If the transaction was cancelled, an EFD isn't going to be |
128 | * constructed and thus we free the EFI here directly. |
129 | */ |
130 | STATIC void |
131 | xfs_efi_item_release( |
132 | struct xfs_log_item *lip) |
133 | { |
134 | xfs_efi_release(efip: EFI_ITEM(lip)); |
135 | } |
136 | |
137 | /* |
138 | * Allocate and initialize an efi item with the given number of extents. |
139 | */ |
140 | STATIC struct xfs_efi_log_item * |
141 | xfs_efi_init( |
142 | struct xfs_mount *mp, |
143 | uint nextents) |
144 | |
145 | { |
146 | struct xfs_efi_log_item *efip; |
147 | |
148 | ASSERT(nextents > 0); |
149 | if (nextents > XFS_EFI_MAX_FAST_EXTENTS) { |
150 | efip = kzalloc(size: xfs_efi_log_item_sizeof(nr: nextents), |
151 | GFP_KERNEL | __GFP_NOFAIL); |
152 | } else { |
153 | efip = kmem_cache_zalloc(k: xfs_efi_cache, |
154 | GFP_KERNEL | __GFP_NOFAIL); |
155 | } |
156 | |
157 | xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops); |
158 | efip->efi_format.efi_nextents = nextents; |
159 | efip->efi_format.efi_id = (uintptr_t)(void *)efip; |
160 | atomic_set(v: &efip->efi_next_extent, i: 0); |
161 | atomic_set(v: &efip->efi_refcount, i: 2); |
162 | |
163 | return efip; |
164 | } |
165 | |
166 | /* |
167 | * Copy an EFI format buffer from the given buf, and into the destination |
168 | * EFI format structure. |
169 | * The given buffer can be in 32 bit or 64 bit form (which has different padding), |
170 | * one of which will be the native format for this kernel. |
171 | * It will handle the conversion of formats if necessary. |
172 | */ |
173 | STATIC int |
174 | xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt) |
175 | { |
176 | xfs_efi_log_format_t *src_efi_fmt = buf->i_addr; |
177 | uint i; |
178 | uint len = xfs_efi_log_format_sizeof(src_efi_fmt->efi_nextents); |
179 | uint len32 = xfs_efi_log_format32_sizeof(src_efi_fmt->efi_nextents); |
180 | uint len64 = xfs_efi_log_format64_sizeof(src_efi_fmt->efi_nextents); |
181 | |
182 | if (buf->i_len == len) { |
183 | memcpy(dst_efi_fmt, src_efi_fmt, |
184 | offsetof(struct xfs_efi_log_format, efi_extents)); |
185 | for (i = 0; i < src_efi_fmt->efi_nextents; i++) |
186 | memcpy(&dst_efi_fmt->efi_extents[i], |
187 | &src_efi_fmt->efi_extents[i], |
188 | sizeof(struct xfs_extent)); |
189 | return 0; |
190 | } else if (buf->i_len == len32) { |
191 | xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr; |
192 | |
193 | dst_efi_fmt->efi_type = src_efi_fmt_32->efi_type; |
194 | dst_efi_fmt->efi_size = src_efi_fmt_32->efi_size; |
195 | dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents; |
196 | dst_efi_fmt->efi_id = src_efi_fmt_32->efi_id; |
197 | for (i = 0; i < dst_efi_fmt->efi_nextents; i++) { |
198 | dst_efi_fmt->efi_extents[i].ext_start = |
199 | src_efi_fmt_32->efi_extents[i].ext_start; |
200 | dst_efi_fmt->efi_extents[i].ext_len = |
201 | src_efi_fmt_32->efi_extents[i].ext_len; |
202 | } |
203 | return 0; |
204 | } else if (buf->i_len == len64) { |
205 | xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr; |
206 | |
207 | dst_efi_fmt->efi_type = src_efi_fmt_64->efi_type; |
208 | dst_efi_fmt->efi_size = src_efi_fmt_64->efi_size; |
209 | dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents; |
210 | dst_efi_fmt->efi_id = src_efi_fmt_64->efi_id; |
211 | for (i = 0; i < dst_efi_fmt->efi_nextents; i++) { |
212 | dst_efi_fmt->efi_extents[i].ext_start = |
213 | src_efi_fmt_64->efi_extents[i].ext_start; |
214 | dst_efi_fmt->efi_extents[i].ext_len = |
215 | src_efi_fmt_64->efi_extents[i].ext_len; |
216 | } |
217 | return 0; |
218 | } |
219 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, NULL, buf->i_addr, |
220 | buf->i_len); |
221 | return -EFSCORRUPTED; |
222 | } |
223 | |
224 | static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip) |
225 | { |
226 | return container_of(lip, struct xfs_efd_log_item, efd_item); |
227 | } |
228 | |
229 | STATIC void |
230 | xfs_efd_item_free(struct xfs_efd_log_item *efdp) |
231 | { |
232 | kmem_free(ptr: efdp->efd_item.li_lv_shadow); |
233 | if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS) |
234 | kmem_free(ptr: efdp); |
235 | else |
236 | kmem_cache_free(s: xfs_efd_cache, objp: efdp); |
237 | } |
238 | |
239 | STATIC void |
240 | xfs_efd_item_size( |
241 | struct xfs_log_item *lip, |
242 | int *nvecs, |
243 | int *nbytes) |
244 | { |
245 | struct xfs_efd_log_item *efdp = EFD_ITEM(lip); |
246 | |
247 | *nvecs += 1; |
248 | *nbytes += xfs_efd_log_format_sizeof(efdp->efd_format.efd_nextents); |
249 | } |
250 | |
251 | /* |
252 | * This is called to fill in the vector of log iovecs for the |
253 | * given efd log item. We use only 1 iovec, and we point that |
254 | * at the efd_log_format structure embedded in the efd item. |
255 | * It is at this point that we assert that all of the extent |
256 | * slots in the efd item have been filled. |
257 | */ |
258 | STATIC void |
259 | xfs_efd_item_format( |
260 | struct xfs_log_item *lip, |
261 | struct xfs_log_vec *lv) |
262 | { |
263 | struct xfs_efd_log_item *efdp = EFD_ITEM(lip); |
264 | struct xfs_log_iovec *vecp = NULL; |
265 | |
266 | ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents); |
267 | |
268 | efdp->efd_format.efd_type = XFS_LI_EFD; |
269 | efdp->efd_format.efd_size = 1; |
270 | |
271 | xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT, |
272 | &efdp->efd_format, |
273 | xfs_efd_log_format_sizeof(efdp->efd_format.efd_nextents)); |
274 | } |
275 | |
276 | /* |
277 | * The EFD is either committed or aborted if the transaction is cancelled. If |
278 | * the transaction is cancelled, drop our reference to the EFI and free the EFD. |
279 | */ |
280 | STATIC void |
281 | xfs_efd_item_release( |
282 | struct xfs_log_item *lip) |
283 | { |
284 | struct xfs_efd_log_item *efdp = EFD_ITEM(lip); |
285 | |
286 | xfs_efi_release(efip: efdp->efd_efip); |
287 | xfs_efd_item_free(efdp); |
288 | } |
289 | |
290 | static struct xfs_log_item * |
291 | xfs_efd_item_intent( |
292 | struct xfs_log_item *lip) |
293 | { |
294 | return &EFD_ITEM(lip)->efd_efip->efi_item; |
295 | } |
296 | |
297 | static const struct xfs_item_ops xfs_efd_item_ops = { |
298 | .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED | |
299 | XFS_ITEM_INTENT_DONE, |
300 | .iop_size = xfs_efd_item_size, |
301 | .iop_format = xfs_efd_item_format, |
302 | .iop_release = xfs_efd_item_release, |
303 | .iop_intent = xfs_efd_item_intent, |
304 | }; |
305 | |
306 | /* |
307 | * Allocate an "extent free done" log item that will hold nextents worth of |
308 | * extents. The caller must use all nextents extents, because we are not |
309 | * flexible about this at all. |
310 | */ |
311 | static struct xfs_efd_log_item * |
312 | xfs_trans_get_efd( |
313 | struct xfs_trans *tp, |
314 | struct xfs_efi_log_item *efip, |
315 | unsigned int nextents) |
316 | { |
317 | struct xfs_efd_log_item *efdp; |
318 | |
319 | ASSERT(nextents > 0); |
320 | |
321 | if (nextents > XFS_EFD_MAX_FAST_EXTENTS) { |
322 | efdp = kzalloc(size: xfs_efd_log_item_sizeof(nr: nextents), |
323 | GFP_KERNEL | __GFP_NOFAIL); |
324 | } else { |
325 | efdp = kmem_cache_zalloc(k: xfs_efd_cache, |
326 | GFP_KERNEL | __GFP_NOFAIL); |
327 | } |
328 | |
329 | xfs_log_item_init(tp->t_mountp, &efdp->efd_item, XFS_LI_EFD, |
330 | &xfs_efd_item_ops); |
331 | efdp->efd_efip = efip; |
332 | efdp->efd_format.efd_nextents = nextents; |
333 | efdp->efd_format.efd_efi_id = efip->efi_format.efi_id; |
334 | |
335 | xfs_trans_add_item(tp, &efdp->efd_item); |
336 | return efdp; |
337 | } |
338 | |
339 | /* |
340 | * Fill the EFD with all extents from the EFI when we need to roll the |
341 | * transaction and continue with a new EFI. |
342 | * |
343 | * This simply copies all the extents in the EFI to the EFD rather than make |
344 | * assumptions about which extents in the EFI have already been processed. We |
345 | * currently keep the xefi list in the same order as the EFI extent list, but |
346 | * that may not always be the case. Copying everything avoids leaving a landmine |
347 | * were we fail to cancel all the extents in an EFI if the xefi list is |
348 | * processed in a different order to the extents in the EFI. |
349 | */ |
350 | static void |
351 | xfs_efd_from_efi( |
352 | struct xfs_efd_log_item *efdp) |
353 | { |
354 | struct xfs_efi_log_item *efip = efdp->efd_efip; |
355 | uint i; |
356 | |
357 | ASSERT(efip->efi_format.efi_nextents > 0); |
358 | ASSERT(efdp->efd_next_extent < efip->efi_format.efi_nextents); |
359 | |
360 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { |
361 | efdp->efd_format.efd_extents[i] = |
362 | efip->efi_format.efi_extents[i]; |
363 | } |
364 | efdp->efd_next_extent = efip->efi_format.efi_nextents; |
365 | } |
366 | |
367 | /* |
368 | * Free an extent and log it to the EFD. Note that the transaction is marked |
369 | * dirty regardless of whether the extent free succeeds or fails to support the |
370 | * EFI/EFD lifecycle rules. |
371 | */ |
372 | static int |
373 | xfs_trans_free_extent( |
374 | struct xfs_trans *tp, |
375 | struct xfs_efd_log_item *efdp, |
376 | struct xfs_extent_free_item *xefi) |
377 | { |
378 | struct xfs_owner_info oinfo = { }; |
379 | struct xfs_mount *mp = tp->t_mountp; |
380 | struct xfs_extent *extp; |
381 | uint next_extent; |
382 | xfs_agblock_t agbno = XFS_FSB_TO_AGBNO(mp, |
383 | xefi->xefi_startblock); |
384 | int error; |
385 | |
386 | oinfo.oi_owner = xefi->xefi_owner; |
387 | if (xefi->xefi_flags & XFS_EFI_ATTR_FORK) |
388 | oinfo.oi_flags |= XFS_OWNER_INFO_ATTR_FORK; |
389 | if (xefi->xefi_flags & XFS_EFI_BMBT_BLOCK) |
390 | oinfo.oi_flags |= XFS_OWNER_INFO_BMBT_BLOCK; |
391 | |
392 | trace_xfs_bmap_free_deferred(tp->t_mountp, xefi->xefi_pag->pag_agno, 0, |
393 | agbno, xefi->xefi_blockcount); |
394 | |
395 | error = __xfs_free_extent(tp, xefi->xefi_pag, agbno, |
396 | xefi->xefi_blockcount, &oinfo, xefi->xefi_agresv, |
397 | xefi->xefi_flags & XFS_EFI_SKIP_DISCARD); |
398 | |
399 | /* |
400 | * Mark the transaction dirty, even on error. This ensures the |
401 | * transaction is aborted, which: |
402 | * |
403 | * 1.) releases the EFI and frees the EFD |
404 | * 2.) shuts down the filesystem |
405 | */ |
406 | tp->t_flags |= XFS_TRANS_DIRTY | XFS_TRANS_HAS_INTENT_DONE; |
407 | set_bit(XFS_LI_DIRTY, addr: &efdp->efd_item.li_flags); |
408 | |
409 | /* |
410 | * If we need a new transaction to make progress, the caller will log a |
411 | * new EFI with the current contents. It will also log an EFD to cancel |
412 | * the existing EFI, and so we need to copy all the unprocessed extents |
413 | * in this EFI to the EFD so this works correctly. |
414 | */ |
415 | if (error == -EAGAIN) { |
416 | xfs_efd_from_efi(efdp); |
417 | return error; |
418 | } |
419 | |
420 | next_extent = efdp->efd_next_extent; |
421 | ASSERT(next_extent < efdp->efd_format.efd_nextents); |
422 | extp = &(efdp->efd_format.efd_extents[next_extent]); |
423 | extp->ext_start = xefi->xefi_startblock; |
424 | extp->ext_len = xefi->xefi_blockcount; |
425 | efdp->efd_next_extent++; |
426 | |
427 | return error; |
428 | } |
429 | |
430 | /* Sort bmap items by AG. */ |
431 | static int |
432 | xfs_extent_free_diff_items( |
433 | void *priv, |
434 | const struct list_head *a, |
435 | const struct list_head *b) |
436 | { |
437 | struct xfs_extent_free_item *ra; |
438 | struct xfs_extent_free_item *rb; |
439 | |
440 | ra = container_of(a, struct xfs_extent_free_item, xefi_list); |
441 | rb = container_of(b, struct xfs_extent_free_item, xefi_list); |
442 | |
443 | return ra->xefi_pag->pag_agno - rb->xefi_pag->pag_agno; |
444 | } |
445 | |
446 | /* Log a free extent to the intent item. */ |
447 | STATIC void |
448 | xfs_extent_free_log_item( |
449 | struct xfs_trans *tp, |
450 | struct xfs_efi_log_item *efip, |
451 | struct xfs_extent_free_item *xefi) |
452 | { |
453 | uint next_extent; |
454 | struct xfs_extent *extp; |
455 | |
456 | tp->t_flags |= XFS_TRANS_DIRTY; |
457 | set_bit(XFS_LI_DIRTY, addr: &efip->efi_item.li_flags); |
458 | |
459 | /* |
460 | * atomic_inc_return gives us the value after the increment; |
461 | * we want to use it as an array index so we need to subtract 1 from |
462 | * it. |
463 | */ |
464 | next_extent = atomic_inc_return(v: &efip->efi_next_extent) - 1; |
465 | ASSERT(next_extent < efip->efi_format.efi_nextents); |
466 | extp = &efip->efi_format.efi_extents[next_extent]; |
467 | extp->ext_start = xefi->xefi_startblock; |
468 | extp->ext_len = xefi->xefi_blockcount; |
469 | } |
470 | |
471 | static struct xfs_log_item * |
472 | xfs_extent_free_create_intent( |
473 | struct xfs_trans *tp, |
474 | struct list_head *items, |
475 | unsigned int count, |
476 | bool sort) |
477 | { |
478 | struct xfs_mount *mp = tp->t_mountp; |
479 | struct xfs_efi_log_item *efip = xfs_efi_init(mp, nextents: count); |
480 | struct xfs_extent_free_item *xefi; |
481 | |
482 | ASSERT(count > 0); |
483 | |
484 | xfs_trans_add_item(tp, &efip->efi_item); |
485 | if (sort) |
486 | list_sort(priv: mp, head: items, cmp: xfs_extent_free_diff_items); |
487 | list_for_each_entry(xefi, items, xefi_list) |
488 | xfs_extent_free_log_item(tp, efip, xefi); |
489 | return &efip->efi_item; |
490 | } |
491 | |
492 | /* Get an EFD so we can process all the free extents. */ |
493 | static struct xfs_log_item * |
494 | xfs_extent_free_create_done( |
495 | struct xfs_trans *tp, |
496 | struct xfs_log_item *intent, |
497 | unsigned int count) |
498 | { |
499 | return &xfs_trans_get_efd(tp, efip: EFI_ITEM(lip: intent), nextents: count)->efd_item; |
500 | } |
501 | |
502 | /* Take a passive ref to the AG containing the space we're freeing. */ |
503 | void |
504 | xfs_extent_free_get_group( |
505 | struct xfs_mount *mp, |
506 | struct xfs_extent_free_item *xefi) |
507 | { |
508 | xfs_agnumber_t agno; |
509 | |
510 | agno = XFS_FSB_TO_AGNO(mp, xefi->xefi_startblock); |
511 | xefi->xefi_pag = xfs_perag_intent_get(mp, agno); |
512 | } |
513 | |
514 | /* Release a passive AG ref after some freeing work. */ |
515 | static inline void |
516 | xfs_extent_free_put_group( |
517 | struct xfs_extent_free_item *xefi) |
518 | { |
519 | xfs_perag_intent_put(pag: xefi->xefi_pag); |
520 | } |
521 | |
522 | /* Process a free extent. */ |
523 | STATIC int |
524 | xfs_extent_free_finish_item( |
525 | struct xfs_trans *tp, |
526 | struct xfs_log_item *done, |
527 | struct list_head *item, |
528 | struct xfs_btree_cur **state) |
529 | { |
530 | struct xfs_extent_free_item *xefi; |
531 | int error; |
532 | |
533 | xefi = container_of(item, struct xfs_extent_free_item, xefi_list); |
534 | |
535 | error = xfs_trans_free_extent(tp, efdp: EFD_ITEM(lip: done), xefi); |
536 | |
537 | /* |
538 | * Don't free the XEFI if we need a new transaction to complete |
539 | * processing of it. |
540 | */ |
541 | if (error == -EAGAIN) |
542 | return error; |
543 | |
544 | xfs_extent_free_put_group(xefi); |
545 | kmem_cache_free(xfs_extfree_item_cache, xefi); |
546 | return error; |
547 | } |
548 | |
549 | /* Abort all pending EFIs. */ |
550 | STATIC void |
551 | xfs_extent_free_abort_intent( |
552 | struct xfs_log_item *intent) |
553 | { |
554 | xfs_efi_release(efip: EFI_ITEM(lip: intent)); |
555 | } |
556 | |
557 | /* Cancel a free extent. */ |
558 | STATIC void |
559 | xfs_extent_free_cancel_item( |
560 | struct list_head *item) |
561 | { |
562 | struct xfs_extent_free_item *xefi; |
563 | |
564 | xefi = container_of(item, struct xfs_extent_free_item, xefi_list); |
565 | |
566 | xfs_extent_free_put_group(xefi); |
567 | kmem_cache_free(xfs_extfree_item_cache, xefi); |
568 | } |
569 | |
570 | const struct xfs_defer_op_type xfs_extent_free_defer_type = { |
571 | .max_items = XFS_EFI_MAX_FAST_EXTENTS, |
572 | .create_intent = xfs_extent_free_create_intent, |
573 | .abort_intent = xfs_extent_free_abort_intent, |
574 | .create_done = xfs_extent_free_create_done, |
575 | .finish_item = xfs_extent_free_finish_item, |
576 | .cancel_item = xfs_extent_free_cancel_item, |
577 | }; |
578 | |
579 | /* |
580 | * AGFL blocks are accounted differently in the reserve pools and are not |
581 | * inserted into the busy extent list. |
582 | */ |
583 | STATIC int |
584 | xfs_agfl_free_finish_item( |
585 | struct xfs_trans *tp, |
586 | struct xfs_log_item *done, |
587 | struct list_head *item, |
588 | struct xfs_btree_cur **state) |
589 | { |
590 | struct xfs_owner_info oinfo = { }; |
591 | struct xfs_mount *mp = tp->t_mountp; |
592 | struct xfs_efd_log_item *efdp = EFD_ITEM(lip: done); |
593 | struct xfs_extent_free_item *xefi; |
594 | struct xfs_extent *extp; |
595 | struct xfs_buf *agbp; |
596 | int error; |
597 | xfs_agblock_t agbno; |
598 | uint next_extent; |
599 | |
600 | xefi = container_of(item, struct xfs_extent_free_item, xefi_list); |
601 | ASSERT(xefi->xefi_blockcount == 1); |
602 | agbno = XFS_FSB_TO_AGBNO(mp, xefi->xefi_startblock); |
603 | oinfo.oi_owner = xefi->xefi_owner; |
604 | |
605 | trace_xfs_agfl_free_deferred(mp, xefi->xefi_pag->pag_agno, 0, agbno, |
606 | xefi->xefi_blockcount); |
607 | |
608 | error = xfs_alloc_read_agf(xefi->xefi_pag, tp, 0, &agbp); |
609 | if (!error) |
610 | error = xfs_free_agfl_block(tp, xefi->xefi_pag->pag_agno, |
611 | agbno, agbp, &oinfo); |
612 | |
613 | /* |
614 | * Mark the transaction dirty, even on error. This ensures the |
615 | * transaction is aborted, which: |
616 | * |
617 | * 1.) releases the EFI and frees the EFD |
618 | * 2.) shuts down the filesystem |
619 | */ |
620 | tp->t_flags |= XFS_TRANS_DIRTY; |
621 | set_bit(XFS_LI_DIRTY, addr: &efdp->efd_item.li_flags); |
622 | |
623 | next_extent = efdp->efd_next_extent; |
624 | ASSERT(next_extent < efdp->efd_format.efd_nextents); |
625 | extp = &(efdp->efd_format.efd_extents[next_extent]); |
626 | extp->ext_start = xefi->xefi_startblock; |
627 | extp->ext_len = xefi->xefi_blockcount; |
628 | efdp->efd_next_extent++; |
629 | |
630 | xfs_extent_free_put_group(xefi); |
631 | kmem_cache_free(xfs_extfree_item_cache, xefi); |
632 | return error; |
633 | } |
634 | |
635 | /* sub-type with special handling for AGFL deferred frees */ |
636 | const struct xfs_defer_op_type xfs_agfl_free_defer_type = { |
637 | .max_items = XFS_EFI_MAX_FAST_EXTENTS, |
638 | .create_intent = xfs_extent_free_create_intent, |
639 | .abort_intent = xfs_extent_free_abort_intent, |
640 | .create_done = xfs_extent_free_create_done, |
641 | .finish_item = xfs_agfl_free_finish_item, |
642 | .cancel_item = xfs_extent_free_cancel_item, |
643 | }; |
644 | |
645 | /* Is this recovered EFI ok? */ |
646 | static inline bool |
647 | xfs_efi_validate_ext( |
648 | struct xfs_mount *mp, |
649 | struct xfs_extent *extp) |
650 | { |
651 | return xfs_verify_fsbext(mp, extp->ext_start, extp->ext_len); |
652 | } |
653 | |
654 | /* |
655 | * Process an extent free intent item that was recovered from |
656 | * the log. We need to free the extents that it describes. |
657 | */ |
658 | STATIC int |
659 | xfs_efi_item_recover( |
660 | struct xfs_log_item *lip, |
661 | struct list_head *capture_list) |
662 | { |
663 | struct xfs_trans_res resv; |
664 | struct xfs_efi_log_item *efip = EFI_ITEM(lip); |
665 | struct xfs_mount *mp = lip->li_log->l_mp; |
666 | struct xfs_efd_log_item *efdp; |
667 | struct xfs_trans *tp; |
668 | int i; |
669 | int error = 0; |
670 | bool requeue_only = false; |
671 | |
672 | /* |
673 | * First check the validity of the extents described by the |
674 | * EFI. If any are bad, then assume that all are bad and |
675 | * just toss the EFI. |
676 | */ |
677 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { |
678 | if (!xfs_efi_validate_ext(mp, |
679 | extp: &efip->efi_format.efi_extents[i])) { |
680 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, |
681 | &efip->efi_format, |
682 | sizeof(efip->efi_format)); |
683 | return -EFSCORRUPTED; |
684 | } |
685 | } |
686 | |
687 | resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate); |
688 | error = xfs_trans_alloc(mp, resp: &resv, blocks: 0, rtextents: 0, flags: 0, tpp: &tp); |
689 | if (error) |
690 | return error; |
691 | efdp = xfs_trans_get_efd(tp, efip, nextents: efip->efi_format.efi_nextents); |
692 | |
693 | for (i = 0; i < efip->efi_format.efi_nextents; i++) { |
694 | struct xfs_extent_free_item fake = { |
695 | .xefi_owner = XFS_RMAP_OWN_UNKNOWN, |
696 | .xefi_agresv = XFS_AG_RESV_NONE, |
697 | }; |
698 | struct xfs_extent *extp; |
699 | |
700 | extp = &efip->efi_format.efi_extents[i]; |
701 | |
702 | fake.xefi_startblock = extp->ext_start; |
703 | fake.xefi_blockcount = extp->ext_len; |
704 | |
705 | if (!requeue_only) { |
706 | xfs_extent_free_get_group(mp, xefi: &fake); |
707 | error = xfs_trans_free_extent(tp, efdp, xefi: &fake); |
708 | xfs_extent_free_put_group(xefi: &fake); |
709 | } |
710 | |
711 | /* |
712 | * If we can't free the extent without potentially deadlocking, |
713 | * requeue the rest of the extents to a new so that they get |
714 | * run again later with a new transaction context. |
715 | */ |
716 | if (error == -EAGAIN || requeue_only) { |
717 | error = xfs_free_extent_later(tp, fake.xefi_startblock, |
718 | fake.xefi_blockcount, |
719 | &XFS_RMAP_OINFO_ANY_OWNER, |
720 | fake.xefi_agresv); |
721 | if (!error) { |
722 | requeue_only = true; |
723 | continue; |
724 | } |
725 | } |
726 | |
727 | if (error == -EFSCORRUPTED) |
728 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, |
729 | extp, sizeof(*extp)); |
730 | if (error) |
731 | goto abort_error; |
732 | |
733 | } |
734 | |
735 | return xfs_defer_ops_capture_and_commit(tp, capture_list); |
736 | |
737 | abort_error: |
738 | xfs_trans_cancel(tp); |
739 | return error; |
740 | } |
741 | |
742 | STATIC bool |
743 | xfs_efi_item_match( |
744 | struct xfs_log_item *lip, |
745 | uint64_t intent_id) |
746 | { |
747 | return EFI_ITEM(lip)->efi_format.efi_id == intent_id; |
748 | } |
749 | |
750 | /* Relog an intent item to push the log tail forward. */ |
751 | static struct xfs_log_item * |
752 | xfs_efi_item_relog( |
753 | struct xfs_log_item *intent, |
754 | struct xfs_trans *tp) |
755 | { |
756 | struct xfs_efd_log_item *efdp; |
757 | struct xfs_efi_log_item *efip; |
758 | struct xfs_extent *extp; |
759 | unsigned int count; |
760 | |
761 | count = EFI_ITEM(lip: intent)->efi_format.efi_nextents; |
762 | extp = EFI_ITEM(lip: intent)->efi_format.efi_extents; |
763 | |
764 | tp->t_flags |= XFS_TRANS_DIRTY; |
765 | efdp = xfs_trans_get_efd(tp, efip: EFI_ITEM(lip: intent), nextents: count); |
766 | efdp->efd_next_extent = count; |
767 | memcpy(efdp->efd_format.efd_extents, extp, count * sizeof(*extp)); |
768 | set_bit(XFS_LI_DIRTY, addr: &efdp->efd_item.li_flags); |
769 | |
770 | efip = xfs_efi_init(mp: tp->t_mountp, nextents: count); |
771 | memcpy(efip->efi_format.efi_extents, extp, count * sizeof(*extp)); |
772 | atomic_set(v: &efip->efi_next_extent, i: count); |
773 | xfs_trans_add_item(tp, &efip->efi_item); |
774 | set_bit(XFS_LI_DIRTY, addr: &efip->efi_item.li_flags); |
775 | return &efip->efi_item; |
776 | } |
777 | |
778 | static const struct xfs_item_ops xfs_efi_item_ops = { |
779 | .flags = XFS_ITEM_INTENT, |
780 | .iop_size = xfs_efi_item_size, |
781 | .iop_format = xfs_efi_item_format, |
782 | .iop_unpin = xfs_efi_item_unpin, |
783 | .iop_release = xfs_efi_item_release, |
784 | .iop_recover = xfs_efi_item_recover, |
785 | .iop_match = xfs_efi_item_match, |
786 | .iop_relog = xfs_efi_item_relog, |
787 | }; |
788 | |
789 | /* |
790 | * This routine is called to create an in-core extent free intent |
791 | * item from the efi format structure which was logged on disk. |
792 | * It allocates an in-core efi, copies the extents from the format |
793 | * structure into it, and adds the efi to the AIL with the given |
794 | * LSN. |
795 | */ |
796 | STATIC int |
797 | xlog_recover_efi_commit_pass2( |
798 | struct xlog *log, |
799 | struct list_head *buffer_list, |
800 | struct xlog_recover_item *item, |
801 | xfs_lsn_t lsn) |
802 | { |
803 | struct xfs_mount *mp = log->l_mp; |
804 | struct xfs_efi_log_item *efip; |
805 | struct xfs_efi_log_format *efi_formatp; |
806 | int error; |
807 | |
808 | efi_formatp = item->ri_buf[0].i_addr; |
809 | |
810 | if (item->ri_buf[0].i_len < xfs_efi_log_format_sizeof(0)) { |
811 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, |
812 | item->ri_buf[0].i_addr, item->ri_buf[0].i_len); |
813 | return -EFSCORRUPTED; |
814 | } |
815 | |
816 | efip = xfs_efi_init(mp, nextents: efi_formatp->efi_nextents); |
817 | error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format); |
818 | if (error) { |
819 | xfs_efi_item_free(efip); |
820 | return error; |
821 | } |
822 | atomic_set(v: &efip->efi_next_extent, i: efi_formatp->efi_nextents); |
823 | /* |
824 | * Insert the intent into the AIL directly and drop one reference so |
825 | * that finishing or canceling the work will drop the other. |
826 | */ |
827 | xfs_trans_ail_insert(log->l_ailp, &efip->efi_item, lsn); |
828 | xfs_efi_release(efip); |
829 | return 0; |
830 | } |
831 | |
832 | const struct xlog_recover_item_ops xlog_efi_item_ops = { |
833 | .item_type = XFS_LI_EFI, |
834 | .commit_pass2 = xlog_recover_efi_commit_pass2, |
835 | }; |
836 | |
837 | /* |
838 | * This routine is called when an EFD format structure is found in a committed |
839 | * transaction in the log. Its purpose is to cancel the corresponding EFI if it |
840 | * was still in the log. To do this it searches the AIL for the EFI with an id |
841 | * equal to that in the EFD format structure. If we find it we drop the EFD |
842 | * reference, which removes the EFI from the AIL and frees it. |
843 | */ |
844 | STATIC int |
845 | xlog_recover_efd_commit_pass2( |
846 | struct xlog *log, |
847 | struct list_head *buffer_list, |
848 | struct xlog_recover_item *item, |
849 | xfs_lsn_t lsn) |
850 | { |
851 | struct xfs_efd_log_format *efd_formatp; |
852 | int buflen = item->ri_buf[0].i_len; |
853 | |
854 | efd_formatp = item->ri_buf[0].i_addr; |
855 | |
856 | if (buflen < sizeof(struct xfs_efd_log_format)) { |
857 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp, |
858 | efd_formatp, buflen); |
859 | return -EFSCORRUPTED; |
860 | } |
861 | |
862 | if (item->ri_buf[0].i_len != xfs_efd_log_format32_sizeof( |
863 | efd_formatp->efd_nextents) && |
864 | item->ri_buf[0].i_len != xfs_efd_log_format64_sizeof( |
865 | efd_formatp->efd_nextents)) { |
866 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp, |
867 | efd_formatp, buflen); |
868 | return -EFSCORRUPTED; |
869 | } |
870 | |
871 | xlog_recover_release_intent(log, XFS_LI_EFI, efd_formatp->efd_efi_id); |
872 | return 0; |
873 | } |
874 | |
875 | const struct xlog_recover_item_ops xlog_efd_item_ops = { |
876 | .item_type = XFS_LI_EFD, |
877 | .commit_pass2 = xlog_recover_efd_commit_pass2, |
878 | }; |
879 | |