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_mount.h"
13 #include "xfs_trans.h"
14 #include "xfs_trans_priv.h"
15 #include "xfs_buf_item.h"
16 #include "xfs_extfree_item.h"
17 #include "xfs_log.h"
18 #include "xfs_btree.h"
19 #include "xfs_rmap.h"
20 
21 
22 kmem_zone_t	*xfs_efi_zone;
23 kmem_zone_t	*xfs_efd_zone;
24 
EFI_ITEM(struct xfs_log_item * lip)25 static inline struct xfs_efi_log_item *EFI_ITEM(struct xfs_log_item *lip)
26 {
27 	return container_of(lip, struct xfs_efi_log_item, efi_item);
28 }
29 
30 void
xfs_efi_item_free(struct xfs_efi_log_item * efip)31 xfs_efi_item_free(
32 	struct xfs_efi_log_item	*efip)
33 {
34 	kmem_free(efip->efi_item.li_lv_shadow);
35 	if (efip->efi_format.efi_nextents > XFS_EFI_MAX_FAST_EXTENTS)
36 		kmem_free(efip);
37 	else
38 		kmem_zone_free(xfs_efi_zone, efip);
39 }
40 
41 /*
42  * Freeing the efi requires that we remove it from the AIL if it has already
43  * been placed there. However, the EFI may not yet have been placed in the AIL
44  * when called by xfs_efi_release() from EFD 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 EFI.
47  */
48 void
xfs_efi_release(struct xfs_efi_log_item * efip)49 xfs_efi_release(
50 	struct xfs_efi_log_item	*efip)
51 {
52 	ASSERT(atomic_read(&efip->efi_refcount) > 0);
53 	if (atomic_dec_and_test(&efip->efi_refcount)) {
54 		xfs_trans_ail_remove(&efip->efi_item, SHUTDOWN_LOG_IO_ERROR);
55 		xfs_efi_item_free(efip);
56 	}
57 }
58 
59 /*
60  * This returns the number of iovecs needed to log the given efi item.
61  * We only need 1 iovec for an efi item.  It just logs the efi_log_format
62  * structure.
63  */
64 static inline int
xfs_efi_item_sizeof(struct xfs_efi_log_item * efip)65 xfs_efi_item_sizeof(
66 	struct xfs_efi_log_item *efip)
67 {
68 	return sizeof(struct xfs_efi_log_format) +
69 	       (efip->efi_format.efi_nextents - 1) * sizeof(xfs_extent_t);
70 }
71 
72 STATIC void
xfs_efi_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)73 xfs_efi_item_size(
74 	struct xfs_log_item	*lip,
75 	int			*nvecs,
76 	int			*nbytes)
77 {
78 	*nvecs += 1;
79 	*nbytes += xfs_efi_item_sizeof(EFI_ITEM(lip));
80 }
81 
82 /*
83  * This is called to fill in the vector of log iovecs for the
84  * given efi log item. We use only 1 iovec, and we point that
85  * at the efi_log_format structure embedded in the efi item.
86  * It is at this point that we assert that all of the extent
87  * slots in the efi item have been filled.
88  */
89 STATIC void
xfs_efi_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)90 xfs_efi_item_format(
91 	struct xfs_log_item	*lip,
92 	struct xfs_log_vec	*lv)
93 {
94 	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
95 	struct xfs_log_iovec	*vecp = NULL;
96 
97 	ASSERT(atomic_read(&efip->efi_next_extent) ==
98 				efip->efi_format.efi_nextents);
99 
100 	efip->efi_format.efi_type = XFS_LI_EFI;
101 	efip->efi_format.efi_size = 1;
102 
103 	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
104 			&efip->efi_format,
105 			xfs_efi_item_sizeof(efip));
106 }
107 
108 
109 /*
110  * Pinning has no meaning for an efi item, so just return.
111  */
112 STATIC void
xfs_efi_item_pin(struct xfs_log_item * lip)113 xfs_efi_item_pin(
114 	struct xfs_log_item	*lip)
115 {
116 }
117 
118 /*
119  * The unpin operation is the last place an EFI is manipulated in the log. It is
120  * either inserted in the AIL or aborted in the event of a log I/O error. In
121  * either case, the EFI transaction has been successfully committed to make it
122  * this far. Therefore, we expect whoever committed the EFI to either construct
123  * and commit the EFD or drop the EFD's reference in the event of error. Simply
124  * drop the log's EFI reference now that the log is done with it.
125  */
126 STATIC void
xfs_efi_item_unpin(struct xfs_log_item * lip,int remove)127 xfs_efi_item_unpin(
128 	struct xfs_log_item	*lip,
129 	int			remove)
130 {
131 	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
132 	xfs_efi_release(efip);
133 }
134 
135 /*
136  * Efi items have no locking or pushing.  However, since EFIs are pulled from
137  * the AIL when their corresponding EFDs are committed to disk, their situation
138  * is very similar to being pinned.  Return XFS_ITEM_PINNED so that the caller
139  * will eventually flush the log.  This should help in getting the EFI out of
140  * the AIL.
141  */
142 STATIC uint
xfs_efi_item_push(struct xfs_log_item * lip,struct list_head * buffer_list)143 xfs_efi_item_push(
144 	struct xfs_log_item	*lip,
145 	struct list_head	*buffer_list)
146 {
147 	return XFS_ITEM_PINNED;
148 }
149 
150 /*
151  * The EFI has been either committed or aborted if the transaction has been
152  * cancelled. If the transaction was cancelled, an EFD isn't going to be
153  * constructed and thus we free the EFI here directly.
154  */
155 STATIC void
xfs_efi_item_unlock(struct xfs_log_item * lip)156 xfs_efi_item_unlock(
157 	struct xfs_log_item	*lip)
158 {
159 	if (test_bit(XFS_LI_ABORTED, &lip->li_flags))
160 		xfs_efi_release(EFI_ITEM(lip));
161 }
162 
163 /*
164  * The EFI is logged only once and cannot be moved in the log, so simply return
165  * the lsn at which it's been logged.
166  */
167 STATIC xfs_lsn_t
xfs_efi_item_committed(struct xfs_log_item * lip,xfs_lsn_t lsn)168 xfs_efi_item_committed(
169 	struct xfs_log_item	*lip,
170 	xfs_lsn_t		lsn)
171 {
172 	return lsn;
173 }
174 
175 /*
176  * The EFI dependency tracking op doesn't do squat.  It can't because
177  * it doesn't know where the free extent is coming from.  The dependency
178  * tracking has to be handled by the "enclosing" metadata object.  For
179  * example, for inodes, the inode is locked throughout the extent freeing
180  * so the dependency should be recorded there.
181  */
182 STATIC void
xfs_efi_item_committing(struct xfs_log_item * lip,xfs_lsn_t lsn)183 xfs_efi_item_committing(
184 	struct xfs_log_item	*lip,
185 	xfs_lsn_t		lsn)
186 {
187 }
188 
189 /*
190  * This is the ops vector shared by all efi log items.
191  */
192 static const struct xfs_item_ops xfs_efi_item_ops = {
193 	.iop_size	= xfs_efi_item_size,
194 	.iop_format	= xfs_efi_item_format,
195 	.iop_pin	= xfs_efi_item_pin,
196 	.iop_unpin	= xfs_efi_item_unpin,
197 	.iop_unlock	= xfs_efi_item_unlock,
198 	.iop_committed	= xfs_efi_item_committed,
199 	.iop_push	= xfs_efi_item_push,
200 	.iop_committing = xfs_efi_item_committing
201 };
202 
203 
204 /*
205  * Allocate and initialize an efi item with the given number of extents.
206  */
207 struct xfs_efi_log_item *
xfs_efi_init(struct xfs_mount * mp,uint nextents)208 xfs_efi_init(
209 	struct xfs_mount	*mp,
210 	uint			nextents)
211 
212 {
213 	struct xfs_efi_log_item	*efip;
214 	uint			size;
215 
216 	ASSERT(nextents > 0);
217 	if (nextents > XFS_EFI_MAX_FAST_EXTENTS) {
218 		size = (uint)(sizeof(xfs_efi_log_item_t) +
219 			((nextents - 1) * sizeof(xfs_extent_t)));
220 		efip = kmem_zalloc(size, KM_SLEEP);
221 	} else {
222 		efip = kmem_zone_zalloc(xfs_efi_zone, KM_SLEEP);
223 	}
224 
225 	xfs_log_item_init(mp, &efip->efi_item, XFS_LI_EFI, &xfs_efi_item_ops);
226 	efip->efi_format.efi_nextents = nextents;
227 	efip->efi_format.efi_id = (uintptr_t)(void *)efip;
228 	atomic_set(&efip->efi_next_extent, 0);
229 	atomic_set(&efip->efi_refcount, 2);
230 
231 	return efip;
232 }
233 
234 /*
235  * Copy an EFI format buffer from the given buf, and into the destination
236  * EFI format structure.
237  * The given buffer can be in 32 bit or 64 bit form (which has different padding),
238  * one of which will be the native format for this kernel.
239  * It will handle the conversion of formats if necessary.
240  */
241 int
xfs_efi_copy_format(xfs_log_iovec_t * buf,xfs_efi_log_format_t * dst_efi_fmt)242 xfs_efi_copy_format(xfs_log_iovec_t *buf, xfs_efi_log_format_t *dst_efi_fmt)
243 {
244 	xfs_efi_log_format_t *src_efi_fmt = buf->i_addr;
245 	uint i;
246 	uint len = sizeof(xfs_efi_log_format_t) +
247 		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_t);
248 	uint len32 = sizeof(xfs_efi_log_format_32_t) +
249 		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_32_t);
250 	uint len64 = sizeof(xfs_efi_log_format_64_t) +
251 		(src_efi_fmt->efi_nextents - 1) * sizeof(xfs_extent_64_t);
252 
253 	if (buf->i_len == len) {
254 		memcpy((char *)dst_efi_fmt, (char*)src_efi_fmt, len);
255 		return 0;
256 	} else if (buf->i_len == len32) {
257 		xfs_efi_log_format_32_t *src_efi_fmt_32 = buf->i_addr;
258 
259 		dst_efi_fmt->efi_type     = src_efi_fmt_32->efi_type;
260 		dst_efi_fmt->efi_size     = src_efi_fmt_32->efi_size;
261 		dst_efi_fmt->efi_nextents = src_efi_fmt_32->efi_nextents;
262 		dst_efi_fmt->efi_id       = src_efi_fmt_32->efi_id;
263 		for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
264 			dst_efi_fmt->efi_extents[i].ext_start =
265 				src_efi_fmt_32->efi_extents[i].ext_start;
266 			dst_efi_fmt->efi_extents[i].ext_len =
267 				src_efi_fmt_32->efi_extents[i].ext_len;
268 		}
269 		return 0;
270 	} else if (buf->i_len == len64) {
271 		xfs_efi_log_format_64_t *src_efi_fmt_64 = buf->i_addr;
272 
273 		dst_efi_fmt->efi_type     = src_efi_fmt_64->efi_type;
274 		dst_efi_fmt->efi_size     = src_efi_fmt_64->efi_size;
275 		dst_efi_fmt->efi_nextents = src_efi_fmt_64->efi_nextents;
276 		dst_efi_fmt->efi_id       = src_efi_fmt_64->efi_id;
277 		for (i = 0; i < dst_efi_fmt->efi_nextents; i++) {
278 			dst_efi_fmt->efi_extents[i].ext_start =
279 				src_efi_fmt_64->efi_extents[i].ext_start;
280 			dst_efi_fmt->efi_extents[i].ext_len =
281 				src_efi_fmt_64->efi_extents[i].ext_len;
282 		}
283 		return 0;
284 	}
285 	return -EFSCORRUPTED;
286 }
287 
EFD_ITEM(struct xfs_log_item * lip)288 static inline struct xfs_efd_log_item *EFD_ITEM(struct xfs_log_item *lip)
289 {
290 	return container_of(lip, struct xfs_efd_log_item, efd_item);
291 }
292 
293 STATIC void
xfs_efd_item_free(struct xfs_efd_log_item * efdp)294 xfs_efd_item_free(struct xfs_efd_log_item *efdp)
295 {
296 	kmem_free(efdp->efd_item.li_lv_shadow);
297 	if (efdp->efd_format.efd_nextents > XFS_EFD_MAX_FAST_EXTENTS)
298 		kmem_free(efdp);
299 	else
300 		kmem_zone_free(xfs_efd_zone, efdp);
301 }
302 
303 /*
304  * This returns the number of iovecs needed to log the given efd item.
305  * We only need 1 iovec for an efd item.  It just logs the efd_log_format
306  * structure.
307  */
308 static inline int
xfs_efd_item_sizeof(struct xfs_efd_log_item * efdp)309 xfs_efd_item_sizeof(
310 	struct xfs_efd_log_item *efdp)
311 {
312 	return sizeof(xfs_efd_log_format_t) +
313 	       (efdp->efd_format.efd_nextents - 1) * sizeof(xfs_extent_t);
314 }
315 
316 STATIC void
xfs_efd_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)317 xfs_efd_item_size(
318 	struct xfs_log_item	*lip,
319 	int			*nvecs,
320 	int			*nbytes)
321 {
322 	*nvecs += 1;
323 	*nbytes += xfs_efd_item_sizeof(EFD_ITEM(lip));
324 }
325 
326 /*
327  * This is called to fill in the vector of log iovecs for the
328  * given efd log item. We use only 1 iovec, and we point that
329  * at the efd_log_format structure embedded in the efd item.
330  * It is at this point that we assert that all of the extent
331  * slots in the efd item have been filled.
332  */
333 STATIC void
xfs_efd_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)334 xfs_efd_item_format(
335 	struct xfs_log_item	*lip,
336 	struct xfs_log_vec	*lv)
337 {
338 	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
339 	struct xfs_log_iovec	*vecp = NULL;
340 
341 	ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
342 
343 	efdp->efd_format.efd_type = XFS_LI_EFD;
344 	efdp->efd_format.efd_size = 1;
345 
346 	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
347 			&efdp->efd_format,
348 			xfs_efd_item_sizeof(efdp));
349 }
350 
351 /*
352  * Pinning has no meaning for an efd item, so just return.
353  */
354 STATIC void
xfs_efd_item_pin(struct xfs_log_item * lip)355 xfs_efd_item_pin(
356 	struct xfs_log_item	*lip)
357 {
358 }
359 
360 /*
361  * Since pinning has no meaning for an efd item, unpinning does
362  * not either.
363  */
364 STATIC void
xfs_efd_item_unpin(struct xfs_log_item * lip,int remove)365 xfs_efd_item_unpin(
366 	struct xfs_log_item	*lip,
367 	int			remove)
368 {
369 }
370 
371 /*
372  * There isn't much you can do to push on an efd item.  It is simply stuck
373  * waiting for the log to be flushed to disk.
374  */
375 STATIC uint
xfs_efd_item_push(struct xfs_log_item * lip,struct list_head * buffer_list)376 xfs_efd_item_push(
377 	struct xfs_log_item	*lip,
378 	struct list_head	*buffer_list)
379 {
380 	return XFS_ITEM_PINNED;
381 }
382 
383 /*
384  * The EFD is either committed or aborted if the transaction is cancelled. If
385  * the transaction is cancelled, drop our reference to the EFI and free the EFD.
386  */
387 STATIC void
xfs_efd_item_unlock(struct xfs_log_item * lip)388 xfs_efd_item_unlock(
389 	struct xfs_log_item	*lip)
390 {
391 	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
392 
393 	if (test_bit(XFS_LI_ABORTED, &lip->li_flags)) {
394 		xfs_efi_release(efdp->efd_efip);
395 		xfs_efd_item_free(efdp);
396 	}
397 }
398 
399 /*
400  * When the efd item is committed to disk, all we need to do is delete our
401  * reference to our partner efi item and then free ourselves. Since we're
402  * freeing ourselves we must return -1 to keep the transaction code from further
403  * referencing this item.
404  */
405 STATIC xfs_lsn_t
xfs_efd_item_committed(struct xfs_log_item * lip,xfs_lsn_t lsn)406 xfs_efd_item_committed(
407 	struct xfs_log_item	*lip,
408 	xfs_lsn_t		lsn)
409 {
410 	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
411 
412 	/*
413 	 * Drop the EFI reference regardless of whether the EFD has been
414 	 * aborted. Once the EFD transaction is constructed, it is the sole
415 	 * responsibility of the EFD to release the EFI (even if the EFI is
416 	 * aborted due to log I/O error).
417 	 */
418 	xfs_efi_release(efdp->efd_efip);
419 	xfs_efd_item_free(efdp);
420 
421 	return (xfs_lsn_t)-1;
422 }
423 
424 /*
425  * The EFD dependency tracking op doesn't do squat.  It can't because
426  * it doesn't know where the free extent is coming from.  The dependency
427  * tracking has to be handled by the "enclosing" metadata object.  For
428  * example, for inodes, the inode is locked throughout the extent freeing
429  * so the dependency should be recorded there.
430  */
431 STATIC void
xfs_efd_item_committing(struct xfs_log_item * lip,xfs_lsn_t lsn)432 xfs_efd_item_committing(
433 	struct xfs_log_item	*lip,
434 	xfs_lsn_t		lsn)
435 {
436 }
437 
438 /*
439  * This is the ops vector shared by all efd log items.
440  */
441 static const struct xfs_item_ops xfs_efd_item_ops = {
442 	.iop_size	= xfs_efd_item_size,
443 	.iop_format	= xfs_efd_item_format,
444 	.iop_pin	= xfs_efd_item_pin,
445 	.iop_unpin	= xfs_efd_item_unpin,
446 	.iop_unlock	= xfs_efd_item_unlock,
447 	.iop_committed	= xfs_efd_item_committed,
448 	.iop_push	= xfs_efd_item_push,
449 	.iop_committing = xfs_efd_item_committing
450 };
451 
452 /*
453  * Allocate and initialize an efd item with the given number of extents.
454  */
455 struct xfs_efd_log_item *
xfs_efd_init(struct xfs_mount * mp,struct xfs_efi_log_item * efip,uint nextents)456 xfs_efd_init(
457 	struct xfs_mount	*mp,
458 	struct xfs_efi_log_item	*efip,
459 	uint			nextents)
460 
461 {
462 	struct xfs_efd_log_item	*efdp;
463 	uint			size;
464 
465 	ASSERT(nextents > 0);
466 	if (nextents > XFS_EFD_MAX_FAST_EXTENTS) {
467 		size = (uint)(sizeof(xfs_efd_log_item_t) +
468 			((nextents - 1) * sizeof(xfs_extent_t)));
469 		efdp = kmem_zalloc(size, KM_SLEEP);
470 	} else {
471 		efdp = kmem_zone_zalloc(xfs_efd_zone, KM_SLEEP);
472 	}
473 
474 	xfs_log_item_init(mp, &efdp->efd_item, XFS_LI_EFD, &xfs_efd_item_ops);
475 	efdp->efd_efip = efip;
476 	efdp->efd_format.efd_nextents = nextents;
477 	efdp->efd_format.efd_efi_id = efip->efi_format.efi_id;
478 
479 	return efdp;
480 }
481 
482 /*
483  * Process an extent free intent item that was recovered from
484  * the log.  We need to free the extents that it describes.
485  */
486 int
xfs_efi_recover(struct xfs_mount * mp,struct xfs_efi_log_item * efip)487 xfs_efi_recover(
488 	struct xfs_mount	*mp,
489 	struct xfs_efi_log_item	*efip)
490 {
491 	struct xfs_efd_log_item	*efdp;
492 	struct xfs_trans	*tp;
493 	int			i;
494 	int			error = 0;
495 	xfs_extent_t		*extp;
496 	xfs_fsblock_t		startblock_fsb;
497 	struct xfs_owner_info	oinfo;
498 
499 	ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
500 
501 	/*
502 	 * First check the validity of the extents described by the
503 	 * EFI.  If any are bad, then assume that all are bad and
504 	 * just toss the EFI.
505 	 */
506 	for (i = 0; i < efip->efi_format.efi_nextents; i++) {
507 		extp = &efip->efi_format.efi_extents[i];
508 		startblock_fsb = XFS_BB_TO_FSB(mp,
509 				   XFS_FSB_TO_DADDR(mp, extp->ext_start));
510 		if (startblock_fsb == 0 ||
511 		    extp->ext_len == 0 ||
512 		    startblock_fsb >= mp->m_sb.sb_dblocks ||
513 		    extp->ext_len >= mp->m_sb.sb_agblocks) {
514 			/*
515 			 * This will pull the EFI from the AIL and
516 			 * free the memory associated with it.
517 			 */
518 			set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
519 			xfs_efi_release(efip);
520 			return -EIO;
521 		}
522 	}
523 
524 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
525 	if (error)
526 		return error;
527 	efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
528 
529 	xfs_rmap_any_owner_update(&oinfo);
530 	for (i = 0; i < efip->efi_format.efi_nextents; i++) {
531 		extp = &efip->efi_format.efi_extents[i];
532 		error = xfs_trans_free_extent(tp, efdp, extp->ext_start,
533 					      extp->ext_len, &oinfo, false);
534 		if (error)
535 			goto abort_error;
536 
537 	}
538 
539 	set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
540 	error = xfs_trans_commit(tp);
541 	return error;
542 
543 abort_error:
544 	xfs_trans_cancel(tp);
545 	return error;
546 }
547