1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_mount.h"
14 #include "xfs_trans.h"
15 #include "xfs_trans_priv.h"
16 #include "xfs_buf_item.h"
17 #include "xfs_inode.h"
18 #include "xfs_inode_item.h"
19 #include "xfs_quota.h"
20 #include "xfs_dquot_item.h"
21 #include "xfs_dquot.h"
22 #include "xfs_trace.h"
23 #include "xfs_log.h"
24 #include "xfs_log_priv.h"
25 #include "xfs_error.h"
26 
27 
28 struct kmem_cache	*xfs_buf_item_cache;
29 
BUF_ITEM(struct xfs_log_item * lip)30 static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
31 {
32 	return container_of(lip, struct xfs_buf_log_item, bli_item);
33 }
34 
35 /* Is this log iovec plausibly large enough to contain the buffer log format? */
36 bool
xfs_buf_log_check_iovec(struct xfs_log_iovec * iovec)37 xfs_buf_log_check_iovec(
38 	struct xfs_log_iovec		*iovec)
39 {
40 	struct xfs_buf_log_format	*blfp = iovec->i_addr;
41 	char				*bmp_end;
42 	char				*item_end;
43 
44 	if (offsetof(struct xfs_buf_log_format, blf_data_map) > iovec->i_len)
45 		return false;
46 
47 	item_end = (char *)iovec->i_addr + iovec->i_len;
48 	bmp_end = (char *)&blfp->blf_data_map[blfp->blf_map_size];
49 	return bmp_end <= item_end;
50 }
51 
52 static inline int
xfs_buf_log_format_size(struct xfs_buf_log_format * blfp)53 xfs_buf_log_format_size(
54 	struct xfs_buf_log_format *blfp)
55 {
56 	return offsetof(struct xfs_buf_log_format, blf_data_map) +
57 			(blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
58 }
59 
60 /*
61  * Return the number of log iovecs and space needed to log the given buf log
62  * item segment.
63  *
64  * It calculates this as 1 iovec for the buf log format structure and 1 for each
65  * stretch of non-contiguous chunks to be logged.  Contiguous chunks are logged
66  * in a single iovec.
67  */
68 STATIC void
xfs_buf_item_size_segment(struct xfs_buf_log_item * bip,struct xfs_buf_log_format * blfp,uint offset,int * nvecs,int * nbytes)69 xfs_buf_item_size_segment(
70 	struct xfs_buf_log_item		*bip,
71 	struct xfs_buf_log_format	*blfp,
72 	uint				offset,
73 	int				*nvecs,
74 	int				*nbytes)
75 {
76 	int				first_bit;
77 	int				nbits;
78 
79 	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
80 	if (first_bit == -1)
81 		return;
82 
83 	(*nvecs)++;
84 	*nbytes += xfs_buf_log_format_size(blfp);
85 
86 	do {
87 		nbits = xfs_contig_bits(blfp->blf_data_map,
88 					blfp->blf_map_size, first_bit);
89 		ASSERT(nbits > 0);
90 		(*nvecs)++;
91 		*nbytes += nbits * XFS_BLF_CHUNK;
92 
93 		/*
94 		 * This takes the bit number to start looking from and
95 		 * returns the next set bit from there.  It returns -1
96 		 * if there are no more bits set or the start bit is
97 		 * beyond the end of the bitmap.
98 		 */
99 		first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
100 					(uint)first_bit + nbits + 1);
101 	} while (first_bit != -1);
102 
103 	return;
104 }
105 
106 /*
107  * Return the number of log iovecs and space needed to log the given buf log
108  * item.
109  *
110  * Discontiguous buffers need a format structure per region that is being
111  * logged. This makes the changes in the buffer appear to log recovery as though
112  * they came from separate buffers, just like would occur if multiple buffers
113  * were used instead of a single discontiguous buffer. This enables
114  * discontiguous buffers to be in-memory constructs, completely transparent to
115  * what ends up on disk.
116  *
117  * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
118  * format structures. If the item has previously been logged and has dirty
119  * regions, we do not relog them in stale buffers. This has the effect of
120  * reducing the size of the relogged item by the amount of dirty data tracked
121  * by the log item. This can result in the committing transaction reducing the
122  * amount of space being consumed by the CIL.
123  */
124 STATIC void
xfs_buf_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)125 xfs_buf_item_size(
126 	struct xfs_log_item	*lip,
127 	int			*nvecs,
128 	int			*nbytes)
129 {
130 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
131 	struct xfs_buf		*bp = bip->bli_buf;
132 	int			i;
133 	int			bytes;
134 	uint			offset = 0;
135 
136 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
137 	if (bip->bli_flags & XFS_BLI_STALE) {
138 		/*
139 		 * The buffer is stale, so all we need to log is the buf log
140 		 * format structure with the cancel flag in it as we are never
141 		 * going to replay the changes tracked in the log item.
142 		 */
143 		trace_xfs_buf_item_size_stale(bip);
144 		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
145 		*nvecs += bip->bli_format_count;
146 		for (i = 0; i < bip->bli_format_count; i++) {
147 			*nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
148 		}
149 		return;
150 	}
151 
152 	ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
153 
154 	if (bip->bli_flags & XFS_BLI_ORDERED) {
155 		/*
156 		 * The buffer has been logged just to order it. It is not being
157 		 * included in the transaction commit, so no vectors are used at
158 		 * all.
159 		 */
160 		trace_xfs_buf_item_size_ordered(bip);
161 		*nvecs = XFS_LOG_VEC_ORDERED;
162 		return;
163 	}
164 
165 	/*
166 	 * The vector count is based on the number of buffer vectors we have
167 	 * dirty bits in. This will only be greater than one when we have a
168 	 * compound buffer with more than one segment dirty. Hence for compound
169 	 * buffers we need to track which segment the dirty bits correspond to,
170 	 * and when we move from one segment to the next increment the vector
171 	 * count for the extra buf log format structure that will need to be
172 	 * written.
173 	 */
174 	bytes = 0;
175 	for (i = 0; i < bip->bli_format_count; i++) {
176 		xfs_buf_item_size_segment(bip, &bip->bli_formats[i], offset,
177 					  nvecs, &bytes);
178 		offset += BBTOB(bp->b_maps[i].bm_len);
179 	}
180 
181 	/*
182 	 * Round up the buffer size required to minimise the number of memory
183 	 * allocations that need to be done as this item grows when relogged by
184 	 * repeated modifications.
185 	 */
186 	*nbytes = round_up(bytes, 512);
187 	trace_xfs_buf_item_size(bip);
188 }
189 
190 static inline void
xfs_buf_item_copy_iovec(struct xfs_log_vec * lv,struct xfs_log_iovec ** vecp,struct xfs_buf * bp,uint offset,int first_bit,uint nbits)191 xfs_buf_item_copy_iovec(
192 	struct xfs_log_vec	*lv,
193 	struct xfs_log_iovec	**vecp,
194 	struct xfs_buf		*bp,
195 	uint			offset,
196 	int			first_bit,
197 	uint			nbits)
198 {
199 	offset += first_bit * XFS_BLF_CHUNK;
200 	xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
201 			xfs_buf_offset(bp, offset),
202 			nbits * XFS_BLF_CHUNK);
203 }
204 
205 static void
xfs_buf_item_format_segment(struct xfs_buf_log_item * bip,struct xfs_log_vec * lv,struct xfs_log_iovec ** vecp,uint offset,struct xfs_buf_log_format * blfp)206 xfs_buf_item_format_segment(
207 	struct xfs_buf_log_item	*bip,
208 	struct xfs_log_vec	*lv,
209 	struct xfs_log_iovec	**vecp,
210 	uint			offset,
211 	struct xfs_buf_log_format *blfp)
212 {
213 	struct xfs_buf		*bp = bip->bli_buf;
214 	uint			base_size;
215 	int			first_bit;
216 	uint			nbits;
217 
218 	/* copy the flags across from the base format item */
219 	blfp->blf_flags = bip->__bli_format.blf_flags;
220 
221 	/*
222 	 * Base size is the actual size of the ondisk structure - it reflects
223 	 * the actual size of the dirty bitmap rather than the size of the in
224 	 * memory structure.
225 	 */
226 	base_size = xfs_buf_log_format_size(blfp);
227 
228 	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
229 	if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
230 		/*
231 		 * If the map is not be dirty in the transaction, mark
232 		 * the size as zero and do not advance the vector pointer.
233 		 */
234 		return;
235 	}
236 
237 	blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
238 	blfp->blf_size = 1;
239 
240 	if (bip->bli_flags & XFS_BLI_STALE) {
241 		/*
242 		 * The buffer is stale, so all we need to log
243 		 * is the buf log format structure with the
244 		 * cancel flag in it.
245 		 */
246 		trace_xfs_buf_item_format_stale(bip);
247 		ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
248 		return;
249 	}
250 
251 
252 	/*
253 	 * Fill in an iovec for each set of contiguous chunks.
254 	 */
255 	do {
256 		ASSERT(first_bit >= 0);
257 		nbits = xfs_contig_bits(blfp->blf_data_map,
258 					blfp->blf_map_size, first_bit);
259 		ASSERT(nbits > 0);
260 		xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
261 					first_bit, nbits);
262 		blfp->blf_size++;
263 
264 		/*
265 		 * This takes the bit number to start looking from and
266 		 * returns the next set bit from there.  It returns -1
267 		 * if there are no more bits set or the start bit is
268 		 * beyond the end of the bitmap.
269 		 */
270 		first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
271 					(uint)first_bit + nbits + 1);
272 	} while (first_bit != -1);
273 
274 	return;
275 }
276 
277 /*
278  * This is called to fill in the vector of log iovecs for the
279  * given log buf item.  It fills the first entry with a buf log
280  * format structure, and the rest point to contiguous chunks
281  * within the buffer.
282  */
283 STATIC void
xfs_buf_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)284 xfs_buf_item_format(
285 	struct xfs_log_item	*lip,
286 	struct xfs_log_vec	*lv)
287 {
288 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
289 	struct xfs_buf		*bp = bip->bli_buf;
290 	struct xfs_log_iovec	*vecp = NULL;
291 	uint			offset = 0;
292 	int			i;
293 
294 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
295 	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
296 	       (bip->bli_flags & XFS_BLI_STALE));
297 	ASSERT((bip->bli_flags & XFS_BLI_STALE) ||
298 	       (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF
299 	        && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF));
300 	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) ||
301 	       (bip->bli_flags & XFS_BLI_STALE));
302 
303 
304 	/*
305 	 * If it is an inode buffer, transfer the in-memory state to the
306 	 * format flags and clear the in-memory state.
307 	 *
308 	 * For buffer based inode allocation, we do not transfer
309 	 * this state if the inode buffer allocation has not yet been committed
310 	 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
311 	 * correct replay of the inode allocation.
312 	 *
313 	 * For icreate item based inode allocation, the buffers aren't written
314 	 * to the journal during allocation, and hence we should always tag the
315 	 * buffer as an inode buffer so that the correct unlinked list replay
316 	 * occurs during recovery.
317 	 */
318 	if (bip->bli_flags & XFS_BLI_INODE_BUF) {
319 		if (xfs_has_v3inodes(lip->li_log->l_mp) ||
320 		    !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
321 		      xfs_log_item_in_current_chkpt(lip)))
322 			bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
323 		bip->bli_flags &= ~XFS_BLI_INODE_BUF;
324 	}
325 
326 	for (i = 0; i < bip->bli_format_count; i++) {
327 		xfs_buf_item_format_segment(bip, lv, &vecp, offset,
328 					    &bip->bli_formats[i]);
329 		offset += BBTOB(bp->b_maps[i].bm_len);
330 	}
331 
332 	/*
333 	 * Check to make sure everything is consistent.
334 	 */
335 	trace_xfs_buf_item_format(bip);
336 }
337 
338 /*
339  * This is called to pin the buffer associated with the buf log item in memory
340  * so it cannot be written out.
341  *
342  * We take a reference to the buffer log item here so that the BLI life cycle
343  * extends at least until the buffer is unpinned via xfs_buf_item_unpin() and
344  * inserted into the AIL.
345  *
346  * We also need to take a reference to the buffer itself as the BLI unpin
347  * processing requires accessing the buffer after the BLI has dropped the final
348  * BLI reference. See xfs_buf_item_unpin() for an explanation.
349  * If unpins race to drop the final BLI reference and only the
350  * BLI owns a reference to the buffer, then the loser of the race can have the
351  * buffer fgreed from under it (e.g. on shutdown). Taking a buffer reference per
352  * pin count ensures the life cycle of the buffer extends for as
353  * long as we hold the buffer pin reference in xfs_buf_item_unpin().
354  */
355 STATIC void
xfs_buf_item_pin(struct xfs_log_item * lip)356 xfs_buf_item_pin(
357 	struct xfs_log_item	*lip)
358 {
359 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
360 
361 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
362 	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
363 	       (bip->bli_flags & XFS_BLI_ORDERED) ||
364 	       (bip->bli_flags & XFS_BLI_STALE));
365 
366 	trace_xfs_buf_item_pin(bip);
367 
368 	xfs_buf_hold(bip->bli_buf);
369 	atomic_inc(&bip->bli_refcount);
370 	atomic_inc(&bip->bli_buf->b_pin_count);
371 }
372 
373 /*
374  * This is called to unpin the buffer associated with the buf log item which was
375  * previously pinned with a call to xfs_buf_item_pin().  We enter this function
376  * with a buffer pin count, a buffer reference and a BLI reference.
377  *
378  * We must drop the BLI reference before we unpin the buffer because the AIL
379  * doesn't acquire a BLI reference whenever it accesses it. Therefore if the
380  * refcount drops to zero, the bli could still be AIL resident and the buffer
381  * submitted for I/O at any point before we return. This can result in IO
382  * completion freeing the buffer while we are still trying to access it here.
383  * This race condition can also occur in shutdown situations where we abort and
384  * unpin buffers from contexts other that journal IO completion.
385  *
386  * Hence we have to hold a buffer reference per pin count to ensure that the
387  * buffer cannot be freed until we have finished processing the unpin operation.
388  * The reference is taken in xfs_buf_item_pin(), and we must hold it until we
389  * are done processing the buffer state. In the case of an abort (remove =
390  * true) then we re-use the current pin reference as the IO reference we hand
391  * off to IO failure handling.
392  */
393 STATIC void
xfs_buf_item_unpin(struct xfs_log_item * lip,int remove)394 xfs_buf_item_unpin(
395 	struct xfs_log_item	*lip,
396 	int			remove)
397 {
398 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
399 	struct xfs_buf		*bp = bip->bli_buf;
400 	int			stale = bip->bli_flags & XFS_BLI_STALE;
401 	int			freed;
402 
403 	ASSERT(bp->b_log_item == bip);
404 	ASSERT(atomic_read(&bip->bli_refcount) > 0);
405 
406 	trace_xfs_buf_item_unpin(bip);
407 
408 	freed = atomic_dec_and_test(&bip->bli_refcount);
409 	if (atomic_dec_and_test(&bp->b_pin_count))
410 		wake_up_all(&bp->b_waiters);
411 
412 	/*
413 	 * Nothing to do but drop the buffer pin reference if the BLI is
414 	 * still active.
415 	 */
416 	if (!freed) {
417 		xfs_buf_rele(bp);
418 		return;
419 	}
420 
421 	if (stale) {
422 		ASSERT(bip->bli_flags & XFS_BLI_STALE);
423 		ASSERT(xfs_buf_islocked(bp));
424 		ASSERT(bp->b_flags & XBF_STALE);
425 		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
426 		ASSERT(list_empty(&lip->li_trans));
427 		ASSERT(!bp->b_transp);
428 
429 		trace_xfs_buf_item_unpin_stale(bip);
430 
431 		/*
432 		 * The buffer has been locked and referenced since it was marked
433 		 * stale so we own both lock and reference exclusively here. We
434 		 * do not need the pin reference any more, so drop it now so
435 		 * that we only have one reference to drop once item completion
436 		 * processing is complete.
437 		 */
438 		xfs_buf_rele(bp);
439 
440 		/*
441 		 * If we get called here because of an IO error, we may or may
442 		 * not have the item on the AIL. xfs_trans_ail_delete() will
443 		 * take care of that situation. xfs_trans_ail_delete() drops
444 		 * the AIL lock.
445 		 */
446 		if (bip->bli_flags & XFS_BLI_STALE_INODE) {
447 			xfs_buf_item_done(bp);
448 			xfs_buf_inode_iodone(bp);
449 			ASSERT(list_empty(&bp->b_li_list));
450 		} else {
451 			xfs_trans_ail_delete(lip, SHUTDOWN_LOG_IO_ERROR);
452 			xfs_buf_item_relse(bp);
453 			ASSERT(bp->b_log_item == NULL);
454 		}
455 		xfs_buf_relse(bp);
456 		return;
457 	}
458 
459 	if (remove) {
460 		/*
461 		 * We need to simulate an async IO failures here to ensure that
462 		 * the correct error completion is run on this buffer. This
463 		 * requires a reference to the buffer and for the buffer to be
464 		 * locked. We can safely pass ownership of the pin reference to
465 		 * the IO to ensure that nothing can free the buffer while we
466 		 * wait for the lock and then run the IO failure completion.
467 		 */
468 		xfs_buf_lock(bp);
469 		bp->b_flags |= XBF_ASYNC;
470 		xfs_buf_ioend_fail(bp);
471 		return;
472 	}
473 
474 	/*
475 	 * BLI has no more active references - it will be moved to the AIL to
476 	 * manage the remaining BLI/buffer life cycle. There is nothing left for
477 	 * us to do here so drop the pin reference to the buffer.
478 	 */
479 	xfs_buf_rele(bp);
480 }
481 
482 STATIC uint
xfs_buf_item_push(struct xfs_log_item * lip,struct list_head * buffer_list)483 xfs_buf_item_push(
484 	struct xfs_log_item	*lip,
485 	struct list_head	*buffer_list)
486 {
487 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
488 	struct xfs_buf		*bp = bip->bli_buf;
489 	uint			rval = XFS_ITEM_SUCCESS;
490 
491 	if (xfs_buf_ispinned(bp))
492 		return XFS_ITEM_PINNED;
493 	if (!xfs_buf_trylock(bp)) {
494 		/*
495 		 * If we have just raced with a buffer being pinned and it has
496 		 * been marked stale, we could end up stalling until someone else
497 		 * issues a log force to unpin the stale buffer. Check for the
498 		 * race condition here so xfsaild recognizes the buffer is pinned
499 		 * and queues a log force to move it along.
500 		 */
501 		if (xfs_buf_ispinned(bp))
502 			return XFS_ITEM_PINNED;
503 		return XFS_ITEM_LOCKED;
504 	}
505 
506 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
507 
508 	trace_xfs_buf_item_push(bip);
509 
510 	/* has a previous flush failed due to IO errors? */
511 	if (bp->b_flags & XBF_WRITE_FAIL) {
512 		xfs_buf_alert_ratelimited(bp, "XFS: Failing async write",
513 	    "Failing async write on buffer block 0x%llx. Retrying async write.",
514 					  (long long)xfs_buf_daddr(bp));
515 	}
516 
517 	if (!xfs_buf_delwri_queue(bp, buffer_list))
518 		rval = XFS_ITEM_FLUSHING;
519 	xfs_buf_unlock(bp);
520 	return rval;
521 }
522 
523 /*
524  * Drop the buffer log item refcount and take appropriate action. This helper
525  * determines whether the bli must be freed or not, since a decrement to zero
526  * does not necessarily mean the bli is unused.
527  *
528  * Return true if the bli is freed, false otherwise.
529  */
530 bool
xfs_buf_item_put(struct xfs_buf_log_item * bip)531 xfs_buf_item_put(
532 	struct xfs_buf_log_item	*bip)
533 {
534 	struct xfs_log_item	*lip = &bip->bli_item;
535 	bool			aborted;
536 	bool			dirty;
537 
538 	/* drop the bli ref and return if it wasn't the last one */
539 	if (!atomic_dec_and_test(&bip->bli_refcount))
540 		return false;
541 
542 	/*
543 	 * We dropped the last ref and must free the item if clean or aborted.
544 	 * If the bli is dirty and non-aborted, the buffer was clean in the
545 	 * transaction but still awaiting writeback from previous changes. In
546 	 * that case, the bli is freed on buffer writeback completion.
547 	 */
548 	aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) ||
549 			xlog_is_shutdown(lip->li_log);
550 	dirty = bip->bli_flags & XFS_BLI_DIRTY;
551 	if (dirty && !aborted)
552 		return false;
553 
554 	/*
555 	 * The bli is aborted or clean. An aborted item may be in the AIL
556 	 * regardless of dirty state.  For example, consider an aborted
557 	 * transaction that invalidated a dirty bli and cleared the dirty
558 	 * state.
559 	 */
560 	if (aborted)
561 		xfs_trans_ail_delete(lip, 0);
562 	xfs_buf_item_relse(bip->bli_buf);
563 	return true;
564 }
565 
566 /*
567  * Release the buffer associated with the buf log item.  If there is no dirty
568  * logged data associated with the buffer recorded in the buf log item, then
569  * free the buf log item and remove the reference to it in the buffer.
570  *
571  * This call ignores the recursion count.  It is only called when the buffer
572  * should REALLY be unlocked, regardless of the recursion count.
573  *
574  * We unconditionally drop the transaction's reference to the log item. If the
575  * item was logged, then another reference was taken when it was pinned, so we
576  * can safely drop the transaction reference now.  This also allows us to avoid
577  * potential races with the unpin code freeing the bli by not referencing the
578  * bli after we've dropped the reference count.
579  *
580  * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
581  * if necessary but do not unlock the buffer.  This is for support of
582  * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
583  * free the item.
584  */
585 STATIC void
xfs_buf_item_release(struct xfs_log_item * lip)586 xfs_buf_item_release(
587 	struct xfs_log_item	*lip)
588 {
589 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
590 	struct xfs_buf		*bp = bip->bli_buf;
591 	bool			released;
592 	bool			hold = bip->bli_flags & XFS_BLI_HOLD;
593 	bool			stale = bip->bli_flags & XFS_BLI_STALE;
594 #if defined(DEBUG) || defined(XFS_WARN)
595 	bool			ordered = bip->bli_flags & XFS_BLI_ORDERED;
596 	bool			dirty = bip->bli_flags & XFS_BLI_DIRTY;
597 	bool			aborted = test_bit(XFS_LI_ABORTED,
598 						   &lip->li_flags);
599 #endif
600 
601 	trace_xfs_buf_item_release(bip);
602 
603 	/*
604 	 * The bli dirty state should match whether the blf has logged segments
605 	 * except for ordered buffers, where only the bli should be dirty.
606 	 */
607 	ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) ||
608 	       (ordered && dirty && !xfs_buf_item_dirty_format(bip)));
609 	ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
610 
611 	/*
612 	 * Clear the buffer's association with this transaction and
613 	 * per-transaction state from the bli, which has been copied above.
614 	 */
615 	bp->b_transp = NULL;
616 	bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
617 
618 	/*
619 	 * Unref the item and unlock the buffer unless held or stale. Stale
620 	 * buffers remain locked until final unpin unless the bli is freed by
621 	 * the unref call. The latter implies shutdown because buffer
622 	 * invalidation dirties the bli and transaction.
623 	 */
624 	released = xfs_buf_item_put(bip);
625 	if (hold || (stale && !released))
626 		return;
627 	ASSERT(!stale || aborted);
628 	xfs_buf_relse(bp);
629 }
630 
631 STATIC void
xfs_buf_item_committing(struct xfs_log_item * lip,xfs_csn_t seq)632 xfs_buf_item_committing(
633 	struct xfs_log_item	*lip,
634 	xfs_csn_t		seq)
635 {
636 	return xfs_buf_item_release(lip);
637 }
638 
639 /*
640  * This is called to find out where the oldest active copy of the
641  * buf log item in the on disk log resides now that the last log
642  * write of it completed at the given lsn.
643  * We always re-log all the dirty data in a buffer, so usually the
644  * latest copy in the on disk log is the only one that matters.  For
645  * those cases we simply return the given lsn.
646  *
647  * The one exception to this is for buffers full of newly allocated
648  * inodes.  These buffers are only relogged with the XFS_BLI_INODE_BUF
649  * flag set, indicating that only the di_next_unlinked fields from the
650  * inodes in the buffers will be replayed during recovery.  If the
651  * original newly allocated inode images have not yet been flushed
652  * when the buffer is so relogged, then we need to make sure that we
653  * keep the old images in the 'active' portion of the log.  We do this
654  * by returning the original lsn of that transaction here rather than
655  * the current one.
656  */
657 STATIC xfs_lsn_t
xfs_buf_item_committed(struct xfs_log_item * lip,xfs_lsn_t lsn)658 xfs_buf_item_committed(
659 	struct xfs_log_item	*lip,
660 	xfs_lsn_t		lsn)
661 {
662 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
663 
664 	trace_xfs_buf_item_committed(bip);
665 
666 	if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
667 		return lip->li_lsn;
668 	return lsn;
669 }
670 
671 #ifdef DEBUG_EXPENSIVE
672 static int
xfs_buf_item_precommit(struct xfs_trans * tp,struct xfs_log_item * lip)673 xfs_buf_item_precommit(
674 	struct xfs_trans	*tp,
675 	struct xfs_log_item	*lip)
676 {
677 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
678 	struct xfs_buf		*bp = bip->bli_buf;
679 	struct xfs_mount	*mp = bp->b_mount;
680 	xfs_failaddr_t		fa;
681 
682 	if (!bp->b_ops || !bp->b_ops->verify_struct)
683 		return 0;
684 	if (bip->bli_flags & XFS_BLI_STALE)
685 		return 0;
686 
687 	fa = bp->b_ops->verify_struct(bp);
688 	if (fa) {
689 		xfs_buf_verifier_error(bp, -EFSCORRUPTED, bp->b_ops->name,
690 				bp->b_addr, BBTOB(bp->b_length), fa);
691 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
692 		ASSERT(fa == NULL);
693 	}
694 
695 	return 0;
696 }
697 #else
698 # define xfs_buf_item_precommit	NULL
699 #endif
700 
701 static const struct xfs_item_ops xfs_buf_item_ops = {
702 	.iop_size	= xfs_buf_item_size,
703 	.iop_precommit	= xfs_buf_item_precommit,
704 	.iop_format	= xfs_buf_item_format,
705 	.iop_pin	= xfs_buf_item_pin,
706 	.iop_unpin	= xfs_buf_item_unpin,
707 	.iop_release	= xfs_buf_item_release,
708 	.iop_committing	= xfs_buf_item_committing,
709 	.iop_committed	= xfs_buf_item_committed,
710 	.iop_push	= xfs_buf_item_push,
711 };
712 
713 STATIC void
xfs_buf_item_get_format(struct xfs_buf_log_item * bip,int count)714 xfs_buf_item_get_format(
715 	struct xfs_buf_log_item	*bip,
716 	int			count)
717 {
718 	ASSERT(bip->bli_formats == NULL);
719 	bip->bli_format_count = count;
720 
721 	if (count == 1) {
722 		bip->bli_formats = &bip->__bli_format;
723 		return;
724 	}
725 
726 	bip->bli_formats = kzalloc(count * sizeof(struct xfs_buf_log_format),
727 				GFP_KERNEL | __GFP_NOFAIL);
728 }
729 
730 STATIC void
xfs_buf_item_free_format(struct xfs_buf_log_item * bip)731 xfs_buf_item_free_format(
732 	struct xfs_buf_log_item	*bip)
733 {
734 	if (bip->bli_formats != &bip->__bli_format) {
735 		kfree(bip->bli_formats);
736 		bip->bli_formats = NULL;
737 	}
738 }
739 
740 /*
741  * Allocate a new buf log item to go with the given buffer.
742  * Set the buffer's b_log_item field to point to the new
743  * buf log item.
744  */
745 int
xfs_buf_item_init(struct xfs_buf * bp,struct xfs_mount * mp)746 xfs_buf_item_init(
747 	struct xfs_buf	*bp,
748 	struct xfs_mount *mp)
749 {
750 	struct xfs_buf_log_item	*bip = bp->b_log_item;
751 	int			chunks;
752 	int			map_size;
753 	int			i;
754 
755 	/*
756 	 * Check to see if there is already a buf log item for
757 	 * this buffer. If we do already have one, there is
758 	 * nothing to do here so return.
759 	 */
760 	ASSERT(bp->b_mount == mp);
761 	if (bip) {
762 		ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
763 		ASSERT(!bp->b_transp);
764 		ASSERT(bip->bli_buf == bp);
765 		return 0;
766 	}
767 
768 	bip = kmem_cache_zalloc(xfs_buf_item_cache, GFP_KERNEL | __GFP_NOFAIL);
769 	xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
770 	bip->bli_buf = bp;
771 
772 	/*
773 	 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
774 	 * can be divided into. Make sure not to truncate any pieces.
775 	 * map_size is the size of the bitmap needed to describe the
776 	 * chunks of the buffer.
777 	 *
778 	 * Discontiguous buffer support follows the layout of the underlying
779 	 * buffer. This makes the implementation as simple as possible.
780 	 */
781 	xfs_buf_item_get_format(bip, bp->b_map_count);
782 
783 	for (i = 0; i < bip->bli_format_count; i++) {
784 		chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
785 				      XFS_BLF_CHUNK);
786 		map_size = DIV_ROUND_UP(chunks, NBWORD);
787 
788 		if (map_size > XFS_BLF_DATAMAP_SIZE) {
789 			kmem_cache_free(xfs_buf_item_cache, bip);
790 			xfs_err(mp,
791 	"buffer item dirty bitmap (%u uints) too small to reflect %u bytes!",
792 					map_size,
793 					BBTOB(bp->b_maps[i].bm_len));
794 			return -EFSCORRUPTED;
795 		}
796 
797 		bip->bli_formats[i].blf_type = XFS_LI_BUF;
798 		bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
799 		bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
800 		bip->bli_formats[i].blf_map_size = map_size;
801 	}
802 
803 	bp->b_log_item = bip;
804 	xfs_buf_hold(bp);
805 	return 0;
806 }
807 
808 
809 /*
810  * Mark bytes first through last inclusive as dirty in the buf
811  * item's bitmap.
812  */
813 static void
xfs_buf_item_log_segment(uint first,uint last,uint * map)814 xfs_buf_item_log_segment(
815 	uint			first,
816 	uint			last,
817 	uint			*map)
818 {
819 	uint		first_bit;
820 	uint		last_bit;
821 	uint		bits_to_set;
822 	uint		bits_set;
823 	uint		word_num;
824 	uint		*wordp;
825 	uint		bit;
826 	uint		end_bit;
827 	uint		mask;
828 
829 	ASSERT(first < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD);
830 	ASSERT(last < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD);
831 
832 	/*
833 	 * Convert byte offsets to bit numbers.
834 	 */
835 	first_bit = first >> XFS_BLF_SHIFT;
836 	last_bit = last >> XFS_BLF_SHIFT;
837 
838 	/*
839 	 * Calculate the total number of bits to be set.
840 	 */
841 	bits_to_set = last_bit - first_bit + 1;
842 
843 	/*
844 	 * Get a pointer to the first word in the bitmap
845 	 * to set a bit in.
846 	 */
847 	word_num = first_bit >> BIT_TO_WORD_SHIFT;
848 	wordp = &map[word_num];
849 
850 	/*
851 	 * Calculate the starting bit in the first word.
852 	 */
853 	bit = first_bit & (uint)(NBWORD - 1);
854 
855 	/*
856 	 * First set any bits in the first word of our range.
857 	 * If it starts at bit 0 of the word, it will be
858 	 * set below rather than here.  That is what the variable
859 	 * bit tells us. The variable bits_set tracks the number
860 	 * of bits that have been set so far.  End_bit is the number
861 	 * of the last bit to be set in this word plus one.
862 	 */
863 	if (bit) {
864 		end_bit = min(bit + bits_to_set, (uint)NBWORD);
865 		mask = ((1U << (end_bit - bit)) - 1) << bit;
866 		*wordp |= mask;
867 		wordp++;
868 		bits_set = end_bit - bit;
869 	} else {
870 		bits_set = 0;
871 	}
872 
873 	/*
874 	 * Now set bits a whole word at a time that are between
875 	 * first_bit and last_bit.
876 	 */
877 	while ((bits_to_set - bits_set) >= NBWORD) {
878 		*wordp = 0xffffffff;
879 		bits_set += NBWORD;
880 		wordp++;
881 	}
882 
883 	/*
884 	 * Finally, set any bits left to be set in one last partial word.
885 	 */
886 	end_bit = bits_to_set - bits_set;
887 	if (end_bit) {
888 		mask = (1U << end_bit) - 1;
889 		*wordp |= mask;
890 	}
891 }
892 
893 /*
894  * Mark bytes first through last inclusive as dirty in the buf
895  * item's bitmap.
896  */
897 void
xfs_buf_item_log(struct xfs_buf_log_item * bip,uint first,uint last)898 xfs_buf_item_log(
899 	struct xfs_buf_log_item	*bip,
900 	uint			first,
901 	uint			last)
902 {
903 	int			i;
904 	uint			start;
905 	uint			end;
906 	struct xfs_buf		*bp = bip->bli_buf;
907 
908 	/*
909 	 * walk each buffer segment and mark them dirty appropriately.
910 	 */
911 	start = 0;
912 	for (i = 0; i < bip->bli_format_count; i++) {
913 		if (start > last)
914 			break;
915 		end = start + BBTOB(bp->b_maps[i].bm_len) - 1;
916 
917 		/* skip to the map that includes the first byte to log */
918 		if (first > end) {
919 			start += BBTOB(bp->b_maps[i].bm_len);
920 			continue;
921 		}
922 
923 		/*
924 		 * Trim the range to this segment and mark it in the bitmap.
925 		 * Note that we must convert buffer offsets to segment relative
926 		 * offsets (e.g., the first byte of each segment is byte 0 of
927 		 * that segment).
928 		 */
929 		if (first < start)
930 			first = start;
931 		if (end > last)
932 			end = last;
933 		xfs_buf_item_log_segment(first - start, end - start,
934 					 &bip->bli_formats[i].blf_data_map[0]);
935 
936 		start += BBTOB(bp->b_maps[i].bm_len);
937 	}
938 }
939 
940 
941 /*
942  * Return true if the buffer has any ranges logged/dirtied by a transaction,
943  * false otherwise.
944  */
945 bool
xfs_buf_item_dirty_format(struct xfs_buf_log_item * bip)946 xfs_buf_item_dirty_format(
947 	struct xfs_buf_log_item	*bip)
948 {
949 	int			i;
950 
951 	for (i = 0; i < bip->bli_format_count; i++) {
952 		if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
953 			     bip->bli_formats[i].blf_map_size))
954 			return true;
955 	}
956 
957 	return false;
958 }
959 
960 STATIC void
xfs_buf_item_free(struct xfs_buf_log_item * bip)961 xfs_buf_item_free(
962 	struct xfs_buf_log_item	*bip)
963 {
964 	xfs_buf_item_free_format(bip);
965 	kvfree(bip->bli_item.li_lv_shadow);
966 	kmem_cache_free(xfs_buf_item_cache, bip);
967 }
968 
969 /*
970  * xfs_buf_item_relse() is called when the buf log item is no longer needed.
971  */
972 void
xfs_buf_item_relse(struct xfs_buf * bp)973 xfs_buf_item_relse(
974 	struct xfs_buf	*bp)
975 {
976 	struct xfs_buf_log_item	*bip = bp->b_log_item;
977 
978 	trace_xfs_buf_item_relse(bp, _RET_IP_);
979 	ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags));
980 
981 	if (atomic_read(&bip->bli_refcount))
982 		return;
983 	bp->b_log_item = NULL;
984 	xfs_buf_rele(bp);
985 	xfs_buf_item_free(bip);
986 }
987 
988 void
xfs_buf_item_done(struct xfs_buf * bp)989 xfs_buf_item_done(
990 	struct xfs_buf		*bp)
991 {
992 	/*
993 	 * If we are forcibly shutting down, this may well be off the AIL
994 	 * already. That's because we simulate the log-committed callbacks to
995 	 * unpin these buffers. Or we may never have put this item on AIL
996 	 * because of the transaction was aborted forcibly.
997 	 * xfs_trans_ail_delete() takes care of these.
998 	 *
999 	 * Either way, AIL is useless if we're forcing a shutdown.
1000 	 *
1001 	 * Note that log recovery writes might have buffer items that are not on
1002 	 * the AIL even when the file system is not shut down.
1003 	 */
1004 	xfs_trans_ail_delete(&bp->b_log_item->bli_item,
1005 			     (bp->b_flags & _XBF_LOGRECOVERY) ? 0 :
1006 			     SHUTDOWN_CORRUPT_INCORE);
1007 	xfs_buf_item_relse(bp);
1008 }
1009