1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4  * Copyright (C) 2010 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_extent_busy.h"
15 #include "xfs_quota.h"
16 #include "xfs_trans.h"
17 #include "xfs_trans_priv.h"
18 #include "xfs_log.h"
19 #include "xfs_log_priv.h"
20 #include "xfs_trace.h"
21 #include "xfs_error.h"
22 #include "xfs_defer.h"
23 #include "xfs_inode.h"
24 #include "xfs_dquot_item.h"
25 #include "xfs_dquot.h"
26 #include "xfs_icache.h"
27 
28 struct kmem_cache	*xfs_trans_cache;
29 
30 #if defined(CONFIG_TRACEPOINTS)
31 static void
xfs_trans_trace_reservations(struct xfs_mount * mp)32 xfs_trans_trace_reservations(
33 	struct xfs_mount	*mp)
34 {
35 	struct xfs_trans_res	*res;
36 	struct xfs_trans_res	*end_res;
37 	int			i;
38 
39 	res = (struct xfs_trans_res *)M_RES(mp);
40 	end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
41 	for (i = 0; res < end_res; i++, res++)
42 		trace_xfs_trans_resv_calc(mp, i, res);
43 }
44 #else
45 # define xfs_trans_trace_reservations(mp)
46 #endif
47 
48 /*
49  * Initialize the precomputed transaction reservation values
50  * in the mount structure.
51  */
52 void
xfs_trans_init(struct xfs_mount * mp)53 xfs_trans_init(
54 	struct xfs_mount	*mp)
55 {
56 	xfs_trans_resv_calc(mp, M_RES(mp));
57 	xfs_trans_trace_reservations(mp);
58 }
59 
60 /*
61  * Free the transaction structure.  If there is more clean up
62  * to do when the structure is freed, add it here.
63  */
64 STATIC void
xfs_trans_free(struct xfs_trans * tp)65 xfs_trans_free(
66 	struct xfs_trans	*tp)
67 {
68 	xfs_extent_busy_sort(&tp->t_busy);
69 	xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
70 
71 	trace_xfs_trans_free(tp, _RET_IP_);
72 	xfs_trans_clear_context(tp);
73 	if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
74 		sb_end_intwrite(tp->t_mountp->m_super);
75 	xfs_trans_free_dqinfo(tp);
76 	kmem_cache_free(xfs_trans_cache, tp);
77 }
78 
79 /*
80  * This is called to create a new transaction which will share the
81  * permanent log reservation of the given transaction.  The remaining
82  * unused block and rt extent reservations are also inherited.  This
83  * implies that the original transaction is no longer allowed to allocate
84  * blocks.  Locks and log items, however, are no inherited.  They must
85  * be added to the new transaction explicitly.
86  */
87 STATIC struct xfs_trans *
xfs_trans_dup(struct xfs_trans * tp)88 xfs_trans_dup(
89 	struct xfs_trans	*tp)
90 {
91 	struct xfs_trans	*ntp;
92 
93 	trace_xfs_trans_dup(tp, _RET_IP_);
94 
95 	ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
96 
97 	/*
98 	 * Initialize the new transaction structure.
99 	 */
100 	ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
101 	ntp->t_mountp = tp->t_mountp;
102 	INIT_LIST_HEAD(&ntp->t_items);
103 	INIT_LIST_HEAD(&ntp->t_busy);
104 	INIT_LIST_HEAD(&ntp->t_dfops);
105 	ntp->t_firstblock = NULLFSBLOCK;
106 
107 	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
108 	ASSERT(tp->t_ticket != NULL);
109 
110 	ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
111 		       (tp->t_flags & XFS_TRANS_RESERVE) |
112 		       (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
113 		       (tp->t_flags & XFS_TRANS_RES_FDBLKS);
114 	/* We gave our writer reference to the new transaction */
115 	tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
116 	ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
117 
118 	ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
119 	ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
120 	tp->t_blk_res = tp->t_blk_res_used;
121 
122 	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
123 	tp->t_rtx_res = tp->t_rtx_res_used;
124 
125 	xfs_trans_switch_context(tp, ntp);
126 
127 	/* move deferred ops over to the new tp */
128 	xfs_defer_move(ntp, tp);
129 
130 	xfs_trans_dup_dqinfo(tp, ntp);
131 	return ntp;
132 }
133 
134 /*
135  * This is called to reserve free disk blocks and log space for the
136  * given transaction.  This must be done before allocating any resources
137  * within the transaction.
138  *
139  * This will return ENOSPC if there are not enough blocks available.
140  * It will sleep waiting for available log space.
141  * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
142  * is used by long running transactions.  If any one of the reservations
143  * fails then they will all be backed out.
144  *
145  * This does not do quota reservations. That typically is done by the
146  * caller afterwards.
147  */
148 static int
xfs_trans_reserve(struct xfs_trans * tp,struct xfs_trans_res * resp,uint blocks,uint rtextents)149 xfs_trans_reserve(
150 	struct xfs_trans	*tp,
151 	struct xfs_trans_res	*resp,
152 	uint			blocks,
153 	uint			rtextents)
154 {
155 	struct xfs_mount	*mp = tp->t_mountp;
156 	int			error = 0;
157 	bool			rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
158 
159 	/*
160 	 * Attempt to reserve the needed disk blocks by decrementing
161 	 * the number needed from the number available.  This will
162 	 * fail if the count would go below zero.
163 	 */
164 	if (blocks > 0) {
165 		error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
166 		if (error != 0)
167 			return -ENOSPC;
168 		tp->t_blk_res += blocks;
169 	}
170 
171 	/*
172 	 * Reserve the log space needed for this transaction.
173 	 */
174 	if (resp->tr_logres > 0) {
175 		bool	permanent = false;
176 
177 		ASSERT(tp->t_log_res == 0 ||
178 		       tp->t_log_res == resp->tr_logres);
179 		ASSERT(tp->t_log_count == 0 ||
180 		       tp->t_log_count == resp->tr_logcount);
181 
182 		if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
183 			tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
184 			permanent = true;
185 		} else {
186 			ASSERT(tp->t_ticket == NULL);
187 			ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
188 		}
189 
190 		if (tp->t_ticket != NULL) {
191 			ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
192 			error = xfs_log_regrant(mp, tp->t_ticket);
193 		} else {
194 			error = xfs_log_reserve(mp, resp->tr_logres,
195 						resp->tr_logcount,
196 						&tp->t_ticket, permanent);
197 		}
198 
199 		if (error)
200 			goto undo_blocks;
201 
202 		tp->t_log_res = resp->tr_logres;
203 		tp->t_log_count = resp->tr_logcount;
204 	}
205 
206 	/*
207 	 * Attempt to reserve the needed realtime extents by decrementing
208 	 * the number needed from the number available.  This will
209 	 * fail if the count would go below zero.
210 	 */
211 	if (rtextents > 0) {
212 		error = xfs_mod_frextents(mp, -((int64_t)rtextents));
213 		if (error) {
214 			error = -ENOSPC;
215 			goto undo_log;
216 		}
217 		tp->t_rtx_res += rtextents;
218 	}
219 
220 	return 0;
221 
222 	/*
223 	 * Error cases jump to one of these labels to undo any
224 	 * reservations which have already been performed.
225 	 */
226 undo_log:
227 	if (resp->tr_logres > 0) {
228 		xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
229 		tp->t_ticket = NULL;
230 		tp->t_log_res = 0;
231 		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
232 	}
233 
234 undo_blocks:
235 	if (blocks > 0) {
236 		xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
237 		tp->t_blk_res = 0;
238 	}
239 	return error;
240 }
241 
242 int
xfs_trans_alloc(struct xfs_mount * mp,struct xfs_trans_res * resp,uint blocks,uint rtextents,uint flags,struct xfs_trans ** tpp)243 xfs_trans_alloc(
244 	struct xfs_mount	*mp,
245 	struct xfs_trans_res	*resp,
246 	uint			blocks,
247 	uint			rtextents,
248 	uint			flags,
249 	struct xfs_trans	**tpp)
250 {
251 	struct xfs_trans	*tp;
252 	bool			want_retry = true;
253 	int			error;
254 
255 	/*
256 	 * Allocate the handle before we do our freeze accounting and setting up
257 	 * GFP_NOFS allocation context so that we avoid lockdep false positives
258 	 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
259 	 */
260 retry:
261 	tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
262 	if (!(flags & XFS_TRANS_NO_WRITECOUNT))
263 		sb_start_intwrite(mp->m_super);
264 	xfs_trans_set_context(tp);
265 
266 	/*
267 	 * Zero-reservation ("empty") transactions can't modify anything, so
268 	 * they're allowed to run while we're frozen.
269 	 */
270 	WARN_ON(resp->tr_logres > 0 &&
271 		mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
272 	ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
273 	       xfs_has_lazysbcount(mp));
274 
275 	tp->t_magic = XFS_TRANS_HEADER_MAGIC;
276 	tp->t_flags = flags;
277 	tp->t_mountp = mp;
278 	INIT_LIST_HEAD(&tp->t_items);
279 	INIT_LIST_HEAD(&tp->t_busy);
280 	INIT_LIST_HEAD(&tp->t_dfops);
281 	tp->t_firstblock = NULLFSBLOCK;
282 
283 	error = xfs_trans_reserve(tp, resp, blocks, rtextents);
284 	if (error == -ENOSPC && want_retry) {
285 		xfs_trans_cancel(tp);
286 
287 		/*
288 		 * We weren't able to reserve enough space for the transaction.
289 		 * Flush the other speculative space allocations to free space.
290 		 * Do not perform a synchronous scan because callers can hold
291 		 * other locks.
292 		 */
293 		xfs_blockgc_flush_all(mp);
294 		want_retry = false;
295 		goto retry;
296 	}
297 	if (error) {
298 		xfs_trans_cancel(tp);
299 		return error;
300 	}
301 
302 	trace_xfs_trans_alloc(tp, _RET_IP_);
303 
304 	*tpp = tp;
305 	return 0;
306 }
307 
308 /*
309  * Create an empty transaction with no reservation.  This is a defensive
310  * mechanism for routines that query metadata without actually modifying them --
311  * if the metadata being queried is somehow cross-linked (think a btree block
312  * pointer that points higher in the tree), we risk deadlock.  However, blocks
313  * grabbed as part of a transaction can be re-grabbed.  The verifiers will
314  * notice the corrupt block and the operation will fail back to userspace
315  * without deadlocking.
316  *
317  * Note the zero-length reservation; this transaction MUST be cancelled without
318  * any dirty data.
319  *
320  * Callers should obtain freeze protection to avoid a conflict with fs freezing
321  * where we can be grabbing buffers at the same time that freeze is trying to
322  * drain the buffer LRU list.
323  */
324 int
xfs_trans_alloc_empty(struct xfs_mount * mp,struct xfs_trans ** tpp)325 xfs_trans_alloc_empty(
326 	struct xfs_mount		*mp,
327 	struct xfs_trans		**tpp)
328 {
329 	struct xfs_trans_res		resv = {0};
330 
331 	return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
332 }
333 
334 /*
335  * Record the indicated change to the given field for application
336  * to the file system's superblock when the transaction commits.
337  * For now, just store the change in the transaction structure.
338  *
339  * Mark the transaction structure to indicate that the superblock
340  * needs to be updated before committing.
341  *
342  * Because we may not be keeping track of allocated/free inodes and
343  * used filesystem blocks in the superblock, we do not mark the
344  * superblock dirty in this transaction if we modify these fields.
345  * We still need to update the transaction deltas so that they get
346  * applied to the incore superblock, but we don't want them to
347  * cause the superblock to get locked and logged if these are the
348  * only fields in the superblock that the transaction modifies.
349  */
350 void
xfs_trans_mod_sb(xfs_trans_t * tp,uint field,int64_t delta)351 xfs_trans_mod_sb(
352 	xfs_trans_t	*tp,
353 	uint		field,
354 	int64_t		delta)
355 {
356 	uint32_t	flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
357 	xfs_mount_t	*mp = tp->t_mountp;
358 
359 	switch (field) {
360 	case XFS_TRANS_SB_ICOUNT:
361 		tp->t_icount_delta += delta;
362 		if (xfs_has_lazysbcount(mp))
363 			flags &= ~XFS_TRANS_SB_DIRTY;
364 		break;
365 	case XFS_TRANS_SB_IFREE:
366 		tp->t_ifree_delta += delta;
367 		if (xfs_has_lazysbcount(mp))
368 			flags &= ~XFS_TRANS_SB_DIRTY;
369 		break;
370 	case XFS_TRANS_SB_FDBLOCKS:
371 		/*
372 		 * Track the number of blocks allocated in the transaction.
373 		 * Make sure it does not exceed the number reserved. If so,
374 		 * shutdown as this can lead to accounting inconsistency.
375 		 */
376 		if (delta < 0) {
377 			tp->t_blk_res_used += (uint)-delta;
378 			if (tp->t_blk_res_used > tp->t_blk_res)
379 				xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
380 		} else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
381 			int64_t	blkres_delta;
382 
383 			/*
384 			 * Return freed blocks directly to the reservation
385 			 * instead of the global pool, being careful not to
386 			 * overflow the trans counter. This is used to preserve
387 			 * reservation across chains of transaction rolls that
388 			 * repeatedly free and allocate blocks.
389 			 */
390 			blkres_delta = min_t(int64_t, delta,
391 					     UINT_MAX - tp->t_blk_res);
392 			tp->t_blk_res += blkres_delta;
393 			delta -= blkres_delta;
394 		}
395 		tp->t_fdblocks_delta += delta;
396 		if (xfs_has_lazysbcount(mp))
397 			flags &= ~XFS_TRANS_SB_DIRTY;
398 		break;
399 	case XFS_TRANS_SB_RES_FDBLOCKS:
400 		/*
401 		 * The allocation has already been applied to the
402 		 * in-core superblock's counter.  This should only
403 		 * be applied to the on-disk superblock.
404 		 */
405 		tp->t_res_fdblocks_delta += delta;
406 		if (xfs_has_lazysbcount(mp))
407 			flags &= ~XFS_TRANS_SB_DIRTY;
408 		break;
409 	case XFS_TRANS_SB_FREXTENTS:
410 		/*
411 		 * Track the number of blocks allocated in the
412 		 * transaction.  Make sure it does not exceed the
413 		 * number reserved.
414 		 */
415 		if (delta < 0) {
416 			tp->t_rtx_res_used += (uint)-delta;
417 			ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
418 		}
419 		tp->t_frextents_delta += delta;
420 		break;
421 	case XFS_TRANS_SB_RES_FREXTENTS:
422 		/*
423 		 * The allocation has already been applied to the
424 		 * in-core superblock's counter.  This should only
425 		 * be applied to the on-disk superblock.
426 		 */
427 		ASSERT(delta < 0);
428 		tp->t_res_frextents_delta += delta;
429 		break;
430 	case XFS_TRANS_SB_DBLOCKS:
431 		tp->t_dblocks_delta += delta;
432 		break;
433 	case XFS_TRANS_SB_AGCOUNT:
434 		ASSERT(delta > 0);
435 		tp->t_agcount_delta += delta;
436 		break;
437 	case XFS_TRANS_SB_IMAXPCT:
438 		tp->t_imaxpct_delta += delta;
439 		break;
440 	case XFS_TRANS_SB_REXTSIZE:
441 		tp->t_rextsize_delta += delta;
442 		break;
443 	case XFS_TRANS_SB_RBMBLOCKS:
444 		tp->t_rbmblocks_delta += delta;
445 		break;
446 	case XFS_TRANS_SB_RBLOCKS:
447 		tp->t_rblocks_delta += delta;
448 		break;
449 	case XFS_TRANS_SB_REXTENTS:
450 		tp->t_rextents_delta += delta;
451 		break;
452 	case XFS_TRANS_SB_REXTSLOG:
453 		tp->t_rextslog_delta += delta;
454 		break;
455 	default:
456 		ASSERT(0);
457 		return;
458 	}
459 
460 	tp->t_flags |= flags;
461 }
462 
463 /*
464  * xfs_trans_apply_sb_deltas() is called from the commit code
465  * to bring the superblock buffer into the current transaction
466  * and modify it as requested by earlier calls to xfs_trans_mod_sb().
467  *
468  * For now we just look at each field allowed to change and change
469  * it if necessary.
470  */
471 STATIC void
xfs_trans_apply_sb_deltas(xfs_trans_t * tp)472 xfs_trans_apply_sb_deltas(
473 	xfs_trans_t	*tp)
474 {
475 	struct xfs_dsb	*sbp;
476 	struct xfs_buf	*bp;
477 	int		whole = 0;
478 
479 	bp = xfs_trans_getsb(tp);
480 	sbp = bp->b_addr;
481 
482 	/*
483 	 * Only update the superblock counters if we are logging them
484 	 */
485 	if (!xfs_has_lazysbcount((tp->t_mountp))) {
486 		if (tp->t_icount_delta)
487 			be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
488 		if (tp->t_ifree_delta)
489 			be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
490 		if (tp->t_fdblocks_delta)
491 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
492 		if (tp->t_res_fdblocks_delta)
493 			be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
494 	}
495 
496 	/*
497 	 * Updating frextents requires careful handling because it does not
498 	 * behave like the lazysb counters because we cannot rely on log
499 	 * recovery in older kenels to recompute the value from the rtbitmap.
500 	 * This means that the ondisk frextents must be consistent with the
501 	 * rtbitmap.
502 	 *
503 	 * Therefore, log the frextents change to the ondisk superblock and
504 	 * update the incore superblock so that future calls to xfs_log_sb
505 	 * write the correct value ondisk.
506 	 *
507 	 * Don't touch m_frextents because it includes incore reservations,
508 	 * and those are handled by the unreserve function.
509 	 */
510 	if (tp->t_frextents_delta || tp->t_res_frextents_delta) {
511 		struct xfs_mount	*mp = tp->t_mountp;
512 		int64_t			rtxdelta;
513 
514 		rtxdelta = tp->t_frextents_delta + tp->t_res_frextents_delta;
515 
516 		spin_lock(&mp->m_sb_lock);
517 		be64_add_cpu(&sbp->sb_frextents, rtxdelta);
518 		mp->m_sb.sb_frextents += rtxdelta;
519 		spin_unlock(&mp->m_sb_lock);
520 	}
521 
522 	if (tp->t_dblocks_delta) {
523 		be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
524 		whole = 1;
525 	}
526 	if (tp->t_agcount_delta) {
527 		be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
528 		whole = 1;
529 	}
530 	if (tp->t_imaxpct_delta) {
531 		sbp->sb_imax_pct += tp->t_imaxpct_delta;
532 		whole = 1;
533 	}
534 	if (tp->t_rextsize_delta) {
535 		be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
536 		whole = 1;
537 	}
538 	if (tp->t_rbmblocks_delta) {
539 		be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
540 		whole = 1;
541 	}
542 	if (tp->t_rblocks_delta) {
543 		be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
544 		whole = 1;
545 	}
546 	if (tp->t_rextents_delta) {
547 		be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
548 		whole = 1;
549 	}
550 	if (tp->t_rextslog_delta) {
551 		sbp->sb_rextslog += tp->t_rextslog_delta;
552 		whole = 1;
553 	}
554 
555 	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
556 	if (whole)
557 		/*
558 		 * Log the whole thing, the fields are noncontiguous.
559 		 */
560 		xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
561 	else
562 		/*
563 		 * Since all the modifiable fields are contiguous, we
564 		 * can get away with this.
565 		 */
566 		xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
567 				  offsetof(struct xfs_dsb, sb_frextents) +
568 				  sizeof(sbp->sb_frextents) - 1);
569 }
570 
571 /*
572  * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
573  * apply superblock counter changes to the in-core superblock.  The
574  * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
575  * applied to the in-core superblock.  The idea is that that has already been
576  * done.
577  *
578  * If we are not logging superblock counters, then the inode allocated/free and
579  * used block counts are not updated in the on disk superblock. In this case,
580  * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
581  * still need to update the incore superblock with the changes.
582  *
583  * Deltas for the inode count are +/-64, hence we use a large batch size of 128
584  * so we don't need to take the counter lock on every update.
585  */
586 #define XFS_ICOUNT_BATCH	128
587 
588 void
xfs_trans_unreserve_and_mod_sb(struct xfs_trans * tp)589 xfs_trans_unreserve_and_mod_sb(
590 	struct xfs_trans	*tp)
591 {
592 	struct xfs_mount	*mp = tp->t_mountp;
593 	bool			rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
594 	int64_t			blkdelta = 0;
595 	int64_t			rtxdelta = 0;
596 	int64_t			idelta = 0;
597 	int64_t			ifreedelta = 0;
598 	int			error;
599 
600 	/* calculate deltas */
601 	if (tp->t_blk_res > 0)
602 		blkdelta = tp->t_blk_res;
603 	if ((tp->t_fdblocks_delta != 0) &&
604 	    (xfs_has_lazysbcount(mp) ||
605 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)))
606 	        blkdelta += tp->t_fdblocks_delta;
607 
608 	if (tp->t_rtx_res > 0)
609 		rtxdelta = tp->t_rtx_res;
610 	if ((tp->t_frextents_delta != 0) &&
611 	    (tp->t_flags & XFS_TRANS_SB_DIRTY))
612 		rtxdelta += tp->t_frextents_delta;
613 
614 	if (xfs_has_lazysbcount(mp) ||
615 	     (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
616 		idelta = tp->t_icount_delta;
617 		ifreedelta = tp->t_ifree_delta;
618 	}
619 
620 	/* apply the per-cpu counters */
621 	if (blkdelta) {
622 		error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
623 		ASSERT(!error);
624 	}
625 
626 	if (idelta)
627 		percpu_counter_add_batch(&mp->m_icount, idelta,
628 					 XFS_ICOUNT_BATCH);
629 
630 	if (ifreedelta)
631 		percpu_counter_add(&mp->m_ifree, ifreedelta);
632 
633 	if (rtxdelta) {
634 		error = xfs_mod_frextents(mp, rtxdelta);
635 		ASSERT(!error);
636 	}
637 
638 	if (!(tp->t_flags & XFS_TRANS_SB_DIRTY))
639 		return;
640 
641 	/* apply remaining deltas */
642 	spin_lock(&mp->m_sb_lock);
643 	mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
644 	mp->m_sb.sb_icount += idelta;
645 	mp->m_sb.sb_ifree += ifreedelta;
646 	/*
647 	 * Do not touch sb_frextents here because we are dealing with incore
648 	 * reservation.  sb_frextents is not part of the lazy sb counters so it
649 	 * must be consistent with the ondisk rtbitmap and must never include
650 	 * incore reservations.
651 	 */
652 	mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
653 	mp->m_sb.sb_agcount += tp->t_agcount_delta;
654 	mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
655 	mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
656 	mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
657 	mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
658 	mp->m_sb.sb_rextents += tp->t_rextents_delta;
659 	mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
660 	spin_unlock(&mp->m_sb_lock);
661 
662 	/*
663 	 * Debug checks outside of the spinlock so they don't lock up the
664 	 * machine if they fail.
665 	 */
666 	ASSERT(mp->m_sb.sb_imax_pct >= 0);
667 	ASSERT(mp->m_sb.sb_rextslog >= 0);
668 	return;
669 }
670 
671 /* Add the given log item to the transaction's list of log items. */
672 void
xfs_trans_add_item(struct xfs_trans * tp,struct xfs_log_item * lip)673 xfs_trans_add_item(
674 	struct xfs_trans	*tp,
675 	struct xfs_log_item	*lip)
676 {
677 	ASSERT(lip->li_log == tp->t_mountp->m_log);
678 	ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
679 	ASSERT(list_empty(&lip->li_trans));
680 	ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
681 
682 	list_add_tail(&lip->li_trans, &tp->t_items);
683 	trace_xfs_trans_add_item(tp, _RET_IP_);
684 }
685 
686 /*
687  * Unlink the log item from the transaction. the log item is no longer
688  * considered dirty in this transaction, as the linked transaction has
689  * finished, either by abort or commit completion.
690  */
691 void
xfs_trans_del_item(struct xfs_log_item * lip)692 xfs_trans_del_item(
693 	struct xfs_log_item	*lip)
694 {
695 	clear_bit(XFS_LI_DIRTY, &lip->li_flags);
696 	list_del_init(&lip->li_trans);
697 }
698 
699 /* Detach and unlock all of the items in a transaction */
700 static void
xfs_trans_free_items(struct xfs_trans * tp,bool abort)701 xfs_trans_free_items(
702 	struct xfs_trans	*tp,
703 	bool			abort)
704 {
705 	struct xfs_log_item	*lip, *next;
706 
707 	trace_xfs_trans_free_items(tp, _RET_IP_);
708 
709 	list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
710 		xfs_trans_del_item(lip);
711 		if (abort)
712 			set_bit(XFS_LI_ABORTED, &lip->li_flags);
713 		if (lip->li_ops->iop_release)
714 			lip->li_ops->iop_release(lip);
715 	}
716 }
717 
718 static inline void
xfs_log_item_batch_insert(struct xfs_ail * ailp,struct xfs_ail_cursor * cur,struct xfs_log_item ** log_items,int nr_items,xfs_lsn_t commit_lsn)719 xfs_log_item_batch_insert(
720 	struct xfs_ail		*ailp,
721 	struct xfs_ail_cursor	*cur,
722 	struct xfs_log_item	**log_items,
723 	int			nr_items,
724 	xfs_lsn_t		commit_lsn)
725 {
726 	int	i;
727 
728 	spin_lock(&ailp->ail_lock);
729 	/* xfs_trans_ail_update_bulk drops ailp->ail_lock */
730 	xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
731 
732 	for (i = 0; i < nr_items; i++) {
733 		struct xfs_log_item *lip = log_items[i];
734 
735 		if (lip->li_ops->iop_unpin)
736 			lip->li_ops->iop_unpin(lip, 0);
737 	}
738 }
739 
740 /*
741  * Bulk operation version of xfs_trans_committed that takes a log vector of
742  * items to insert into the AIL. This uses bulk AIL insertion techniques to
743  * minimise lock traffic.
744  *
745  * If we are called with the aborted flag set, it is because a log write during
746  * a CIL checkpoint commit has failed. In this case, all the items in the
747  * checkpoint have already gone through iop_committed and iop_committing, which
748  * means that checkpoint commit abort handling is treated exactly the same
749  * as an iclog write error even though we haven't started any IO yet. Hence in
750  * this case all we need to do is iop_committed processing, followed by an
751  * iop_unpin(aborted) call.
752  *
753  * The AIL cursor is used to optimise the insert process. If commit_lsn is not
754  * at the end of the AIL, the insert cursor avoids the need to walk
755  * the AIL to find the insertion point on every xfs_log_item_batch_insert()
756  * call. This saves a lot of needless list walking and is a net win, even
757  * though it slightly increases that amount of AIL lock traffic to set it up
758  * and tear it down.
759  */
760 void
xfs_trans_committed_bulk(struct xfs_ail * ailp,struct xfs_log_vec * log_vector,xfs_lsn_t commit_lsn,bool aborted)761 xfs_trans_committed_bulk(
762 	struct xfs_ail		*ailp,
763 	struct xfs_log_vec	*log_vector,
764 	xfs_lsn_t		commit_lsn,
765 	bool			aborted)
766 {
767 #define LOG_ITEM_BATCH_SIZE	32
768 	struct xfs_log_item	*log_items[LOG_ITEM_BATCH_SIZE];
769 	struct xfs_log_vec	*lv;
770 	struct xfs_ail_cursor	cur;
771 	int			i = 0;
772 
773 	spin_lock(&ailp->ail_lock);
774 	xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
775 	spin_unlock(&ailp->ail_lock);
776 
777 	/* unpin all the log items */
778 	for (lv = log_vector; lv; lv = lv->lv_next ) {
779 		struct xfs_log_item	*lip = lv->lv_item;
780 		xfs_lsn_t		item_lsn;
781 
782 		if (aborted)
783 			set_bit(XFS_LI_ABORTED, &lip->li_flags);
784 
785 		if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
786 			lip->li_ops->iop_release(lip);
787 			continue;
788 		}
789 
790 		if (lip->li_ops->iop_committed)
791 			item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
792 		else
793 			item_lsn = commit_lsn;
794 
795 		/* item_lsn of -1 means the item needs no further processing */
796 		if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
797 			continue;
798 
799 		/*
800 		 * if we are aborting the operation, no point in inserting the
801 		 * object into the AIL as we are in a shutdown situation.
802 		 */
803 		if (aborted) {
804 			ASSERT(xlog_is_shutdown(ailp->ail_log));
805 			if (lip->li_ops->iop_unpin)
806 				lip->li_ops->iop_unpin(lip, 1);
807 			continue;
808 		}
809 
810 		if (item_lsn != commit_lsn) {
811 
812 			/*
813 			 * Not a bulk update option due to unusual item_lsn.
814 			 * Push into AIL immediately, rechecking the lsn once
815 			 * we have the ail lock. Then unpin the item. This does
816 			 * not affect the AIL cursor the bulk insert path is
817 			 * using.
818 			 */
819 			spin_lock(&ailp->ail_lock);
820 			if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
821 				xfs_trans_ail_update(ailp, lip, item_lsn);
822 			else
823 				spin_unlock(&ailp->ail_lock);
824 			if (lip->li_ops->iop_unpin)
825 				lip->li_ops->iop_unpin(lip, 0);
826 			continue;
827 		}
828 
829 		/* Item is a candidate for bulk AIL insert.  */
830 		log_items[i++] = lv->lv_item;
831 		if (i >= LOG_ITEM_BATCH_SIZE) {
832 			xfs_log_item_batch_insert(ailp, &cur, log_items,
833 					LOG_ITEM_BATCH_SIZE, commit_lsn);
834 			i = 0;
835 		}
836 	}
837 
838 	/* make sure we insert the remainder! */
839 	if (i)
840 		xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
841 
842 	spin_lock(&ailp->ail_lock);
843 	xfs_trans_ail_cursor_done(&cur);
844 	spin_unlock(&ailp->ail_lock);
845 }
846 
847 /*
848  * Commit the given transaction to the log.
849  *
850  * XFS disk error handling mechanism is not based on a typical
851  * transaction abort mechanism. Logically after the filesystem
852  * gets marked 'SHUTDOWN', we can't let any new transactions
853  * be durable - ie. committed to disk - because some metadata might
854  * be inconsistent. In such cases, this returns an error, and the
855  * caller may assume that all locked objects joined to the transaction
856  * have already been unlocked as if the commit had succeeded.
857  * Do not reference the transaction structure after this call.
858  */
859 static int
__xfs_trans_commit(struct xfs_trans * tp,bool regrant)860 __xfs_trans_commit(
861 	struct xfs_trans	*tp,
862 	bool			regrant)
863 {
864 	struct xfs_mount	*mp = tp->t_mountp;
865 	struct xlog		*log = mp->m_log;
866 	xfs_csn_t		commit_seq = 0;
867 	int			error = 0;
868 	int			sync = tp->t_flags & XFS_TRANS_SYNC;
869 
870 	trace_xfs_trans_commit(tp, _RET_IP_);
871 
872 	/*
873 	 * Finish deferred items on final commit. Only permanent transactions
874 	 * should ever have deferred ops.
875 	 */
876 	WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
877 		     !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
878 	if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
879 		error = xfs_defer_finish_noroll(&tp);
880 		if (error)
881 			goto out_unreserve;
882 	}
883 
884 	/*
885 	 * If there is nothing to be logged by the transaction,
886 	 * then unlock all of the items associated with the
887 	 * transaction and free the transaction structure.
888 	 * Also make sure to return any reserved blocks to
889 	 * the free pool.
890 	 */
891 	if (!(tp->t_flags & XFS_TRANS_DIRTY))
892 		goto out_unreserve;
893 
894 	/*
895 	 * We must check against log shutdown here because we cannot abort log
896 	 * items and leave them dirty, inconsistent and unpinned in memory while
897 	 * the log is active. This leaves them open to being written back to
898 	 * disk, and that will lead to on-disk corruption.
899 	 */
900 	if (xlog_is_shutdown(log)) {
901 		error = -EIO;
902 		goto out_unreserve;
903 	}
904 
905 	ASSERT(tp->t_ticket != NULL);
906 
907 	/*
908 	 * If we need to update the superblock, then do it now.
909 	 */
910 	if (tp->t_flags & XFS_TRANS_SB_DIRTY)
911 		xfs_trans_apply_sb_deltas(tp);
912 	xfs_trans_apply_dquot_deltas(tp);
913 
914 	xlog_cil_commit(log, tp, &commit_seq, regrant);
915 
916 	xfs_trans_free(tp);
917 
918 	/*
919 	 * If the transaction needs to be synchronous, then force the
920 	 * log out now and wait for it.
921 	 */
922 	if (sync) {
923 		error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
924 		XFS_STATS_INC(mp, xs_trans_sync);
925 	} else {
926 		XFS_STATS_INC(mp, xs_trans_async);
927 	}
928 
929 	return error;
930 
931 out_unreserve:
932 	xfs_trans_unreserve_and_mod_sb(tp);
933 
934 	/*
935 	 * It is indeed possible for the transaction to be not dirty but
936 	 * the dqinfo portion to be.  All that means is that we have some
937 	 * (non-persistent) quota reservations that need to be unreserved.
938 	 */
939 	xfs_trans_unreserve_and_mod_dquots(tp);
940 	if (tp->t_ticket) {
941 		if (regrant && !xlog_is_shutdown(log))
942 			xfs_log_ticket_regrant(log, tp->t_ticket);
943 		else
944 			xfs_log_ticket_ungrant(log, tp->t_ticket);
945 		tp->t_ticket = NULL;
946 	}
947 	xfs_trans_free_items(tp, !!error);
948 	xfs_trans_free(tp);
949 
950 	XFS_STATS_INC(mp, xs_trans_empty);
951 	return error;
952 }
953 
954 int
xfs_trans_commit(struct xfs_trans * tp)955 xfs_trans_commit(
956 	struct xfs_trans	*tp)
957 {
958 	return __xfs_trans_commit(tp, false);
959 }
960 
961 /*
962  * Unlock all of the transaction's items and free the transaction.  If the
963  * transaction is dirty, we must shut down the filesystem because there is no
964  * way to restore them to their previous state.
965  *
966  * If the transaction has made a log reservation, make sure to release it as
967  * well.
968  *
969  * This is a high level function (equivalent to xfs_trans_commit()) and so can
970  * be called after the transaction has effectively been aborted due to the mount
971  * being shut down. However, if the mount has not been shut down and the
972  * transaction is dirty we will shut the mount down and, in doing so, that
973  * guarantees that the log is shut down, too. Hence we don't need to be as
974  * careful with shutdown state and dirty items here as we need to be in
975  * xfs_trans_commit().
976  */
977 void
xfs_trans_cancel(struct xfs_trans * tp)978 xfs_trans_cancel(
979 	struct xfs_trans	*tp)
980 {
981 	struct xfs_mount	*mp = tp->t_mountp;
982 	struct xlog		*log = mp->m_log;
983 	bool			dirty = (tp->t_flags & XFS_TRANS_DIRTY);
984 
985 	trace_xfs_trans_cancel(tp, _RET_IP_);
986 
987 	/*
988 	 * It's never valid to cancel a transaction with deferred ops attached,
989 	 * because the transaction is effectively dirty.  Complain about this
990 	 * loudly before freeing the in-memory defer items.
991 	 */
992 	if (!list_empty(&tp->t_dfops)) {
993 		ASSERT(xfs_is_shutdown(mp) || list_empty(&tp->t_dfops));
994 		ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
995 		dirty = true;
996 		xfs_defer_cancel(tp);
997 	}
998 
999 	/*
1000 	 * See if the caller is relying on us to shut down the filesystem. We
1001 	 * only want an error report if there isn't already a shutdown in
1002 	 * progress, so we only need to check against the mount shutdown state
1003 	 * here.
1004 	 */
1005 	if (dirty && !xfs_is_shutdown(mp)) {
1006 		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1007 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1008 	}
1009 #ifdef DEBUG
1010 	/* Log items need to be consistent until the log is shut down. */
1011 	if (!dirty && !xlog_is_shutdown(log)) {
1012 		struct xfs_log_item *lip;
1013 
1014 		list_for_each_entry(lip, &tp->t_items, li_trans)
1015 			ASSERT(!xlog_item_is_intent_done(lip));
1016 	}
1017 #endif
1018 	xfs_trans_unreserve_and_mod_sb(tp);
1019 	xfs_trans_unreserve_and_mod_dquots(tp);
1020 
1021 	if (tp->t_ticket) {
1022 		xfs_log_ticket_ungrant(log, tp->t_ticket);
1023 		tp->t_ticket = NULL;
1024 	}
1025 
1026 	xfs_trans_free_items(tp, dirty);
1027 	xfs_trans_free(tp);
1028 }
1029 
1030 /*
1031  * Roll from one trans in the sequence of PERMANENT transactions to
1032  * the next: permanent transactions are only flushed out when
1033  * committed with xfs_trans_commit(), but we still want as soon
1034  * as possible to let chunks of it go to the log. So we commit the
1035  * chunk we've been working on and get a new transaction to continue.
1036  */
1037 int
xfs_trans_roll(struct xfs_trans ** tpp)1038 xfs_trans_roll(
1039 	struct xfs_trans	**tpp)
1040 {
1041 	struct xfs_trans	*trans = *tpp;
1042 	struct xfs_trans_res	tres;
1043 	int			error;
1044 
1045 	trace_xfs_trans_roll(trans, _RET_IP_);
1046 
1047 	/*
1048 	 * Copy the critical parameters from one trans to the next.
1049 	 */
1050 	tres.tr_logres = trans->t_log_res;
1051 	tres.tr_logcount = trans->t_log_count;
1052 
1053 	*tpp = xfs_trans_dup(trans);
1054 
1055 	/*
1056 	 * Commit the current transaction.
1057 	 * If this commit failed, then it'd just unlock those items that
1058 	 * are not marked ihold. That also means that a filesystem shutdown
1059 	 * is in progress. The caller takes the responsibility to cancel
1060 	 * the duplicate transaction that gets returned.
1061 	 */
1062 	error = __xfs_trans_commit(trans, true);
1063 	if (error)
1064 		return error;
1065 
1066 	/*
1067 	 * Reserve space in the log for the next transaction.
1068 	 * This also pushes items in the "AIL", the list of logged items,
1069 	 * out to disk if they are taking up space at the tail of the log
1070 	 * that we want to use.  This requires that either nothing be locked
1071 	 * across this call, or that anything that is locked be logged in
1072 	 * the prior and the next transactions.
1073 	 */
1074 	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1075 	return xfs_trans_reserve(*tpp, &tres, 0, 0);
1076 }
1077 
1078 /*
1079  * Allocate an transaction, lock and join the inode to it, and reserve quota.
1080  *
1081  * The caller must ensure that the on-disk dquots attached to this inode have
1082  * already been allocated and initialized.  The caller is responsible for
1083  * releasing ILOCK_EXCL if a new transaction is returned.
1084  */
1085 int
xfs_trans_alloc_inode(struct xfs_inode * ip,struct xfs_trans_res * resv,unsigned int dblocks,unsigned int rblocks,bool force,struct xfs_trans ** tpp)1086 xfs_trans_alloc_inode(
1087 	struct xfs_inode	*ip,
1088 	struct xfs_trans_res	*resv,
1089 	unsigned int		dblocks,
1090 	unsigned int		rblocks,
1091 	bool			force,
1092 	struct xfs_trans	**tpp)
1093 {
1094 	struct xfs_trans	*tp;
1095 	struct xfs_mount	*mp = ip->i_mount;
1096 	bool			retried = false;
1097 	int			error;
1098 
1099 retry:
1100 	error = xfs_trans_alloc(mp, resv, dblocks,
1101 			rblocks / mp->m_sb.sb_rextsize,
1102 			force ? XFS_TRANS_RESERVE : 0, &tp);
1103 	if (error)
1104 		return error;
1105 
1106 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1107 	xfs_trans_ijoin(tp, ip, 0);
1108 
1109 	error = xfs_qm_dqattach_locked(ip, false);
1110 	if (error) {
1111 		/* Caller should have allocated the dquots! */
1112 		ASSERT(error != -ENOENT);
1113 		goto out_cancel;
1114 	}
1115 
1116 	error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1117 	if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1118 		xfs_trans_cancel(tp);
1119 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1120 		xfs_blockgc_free_quota(ip, 0);
1121 		retried = true;
1122 		goto retry;
1123 	}
1124 	if (error)
1125 		goto out_cancel;
1126 
1127 	*tpp = tp;
1128 	return 0;
1129 
1130 out_cancel:
1131 	xfs_trans_cancel(tp);
1132 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1133 	return error;
1134 }
1135 
1136 /*
1137  * Allocate an transaction in preparation for inode creation by reserving quota
1138  * against the given dquots.  Callers are not required to hold any inode locks.
1139  */
1140 int
xfs_trans_alloc_icreate(struct xfs_mount * mp,struct xfs_trans_res * resv,struct xfs_dquot * udqp,struct xfs_dquot * gdqp,struct xfs_dquot * pdqp,unsigned int dblocks,struct xfs_trans ** tpp)1141 xfs_trans_alloc_icreate(
1142 	struct xfs_mount	*mp,
1143 	struct xfs_trans_res	*resv,
1144 	struct xfs_dquot	*udqp,
1145 	struct xfs_dquot	*gdqp,
1146 	struct xfs_dquot	*pdqp,
1147 	unsigned int		dblocks,
1148 	struct xfs_trans	**tpp)
1149 {
1150 	struct xfs_trans	*tp;
1151 	bool			retried = false;
1152 	int			error;
1153 
1154 retry:
1155 	error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1156 	if (error)
1157 		return error;
1158 
1159 	error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1160 	if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1161 		xfs_trans_cancel(tp);
1162 		xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1163 		retried = true;
1164 		goto retry;
1165 	}
1166 	if (error) {
1167 		xfs_trans_cancel(tp);
1168 		return error;
1169 	}
1170 
1171 	*tpp = tp;
1172 	return 0;
1173 }
1174 
1175 /*
1176  * Allocate an transaction, lock and join the inode to it, and reserve quota
1177  * in preparation for inode attribute changes that include uid, gid, or prid
1178  * changes.
1179  *
1180  * The caller must ensure that the on-disk dquots attached to this inode have
1181  * already been allocated and initialized.  The ILOCK will be dropped when the
1182  * transaction is committed or cancelled.
1183  */
1184 int
xfs_trans_alloc_ichange(struct xfs_inode * ip,struct xfs_dquot * new_udqp,struct xfs_dquot * new_gdqp,struct xfs_dquot * new_pdqp,bool force,struct xfs_trans ** tpp)1185 xfs_trans_alloc_ichange(
1186 	struct xfs_inode	*ip,
1187 	struct xfs_dquot	*new_udqp,
1188 	struct xfs_dquot	*new_gdqp,
1189 	struct xfs_dquot	*new_pdqp,
1190 	bool			force,
1191 	struct xfs_trans	**tpp)
1192 {
1193 	struct xfs_trans	*tp;
1194 	struct xfs_mount	*mp = ip->i_mount;
1195 	struct xfs_dquot	*udqp;
1196 	struct xfs_dquot	*gdqp;
1197 	struct xfs_dquot	*pdqp;
1198 	bool			retried = false;
1199 	int			error;
1200 
1201 retry:
1202 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1203 	if (error)
1204 		return error;
1205 
1206 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1207 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1208 
1209 	error = xfs_qm_dqattach_locked(ip, false);
1210 	if (error) {
1211 		/* Caller should have allocated the dquots! */
1212 		ASSERT(error != -ENOENT);
1213 		goto out_cancel;
1214 	}
1215 
1216 	/*
1217 	 * For each quota type, skip quota reservations if the inode's dquots
1218 	 * now match the ones that came from the caller, or the caller didn't
1219 	 * pass one in.  The inode's dquots can change if we drop the ILOCK to
1220 	 * perform a blockgc scan, so we must preserve the caller's arguments.
1221 	 */
1222 	udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1223 	gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1224 	pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1225 	if (udqp || gdqp || pdqp) {
1226 		unsigned int	qflags = XFS_QMOPT_RES_REGBLKS;
1227 
1228 		if (force)
1229 			qflags |= XFS_QMOPT_FORCE_RES;
1230 
1231 		/*
1232 		 * Reserve enough quota to handle blocks on disk and reserved
1233 		 * for a delayed allocation.  We'll actually transfer the
1234 		 * delalloc reservation between dquots at chown time, even
1235 		 * though that part is only semi-transactional.
1236 		 */
1237 		error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1238 				pdqp, ip->i_nblocks + ip->i_delayed_blks,
1239 				1, qflags);
1240 		if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1241 			xfs_trans_cancel(tp);
1242 			xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1243 			retried = true;
1244 			goto retry;
1245 		}
1246 		if (error)
1247 			goto out_cancel;
1248 	}
1249 
1250 	*tpp = tp;
1251 	return 0;
1252 
1253 out_cancel:
1254 	xfs_trans_cancel(tp);
1255 	return error;
1256 }
1257 
1258 /*
1259  * Allocate an transaction, lock and join the directory and child inodes to it,
1260  * and reserve quota for a directory update.  If there isn't sufficient space,
1261  * @dblocks will be set to zero for a reservationless directory update and
1262  * @nospace_error will be set to a negative errno describing the space
1263  * constraint we hit.
1264  *
1265  * The caller must ensure that the on-disk dquots attached to this inode have
1266  * already been allocated and initialized.  The ILOCKs will be dropped when the
1267  * transaction is committed or cancelled.
1268  */
1269 int
xfs_trans_alloc_dir(struct xfs_inode * dp,struct xfs_trans_res * resv,struct xfs_inode * ip,unsigned int * dblocks,struct xfs_trans ** tpp,int * nospace_error)1270 xfs_trans_alloc_dir(
1271 	struct xfs_inode	*dp,
1272 	struct xfs_trans_res	*resv,
1273 	struct xfs_inode	*ip,
1274 	unsigned int		*dblocks,
1275 	struct xfs_trans	**tpp,
1276 	int			*nospace_error)
1277 {
1278 	struct xfs_trans	*tp;
1279 	struct xfs_mount	*mp = ip->i_mount;
1280 	unsigned int		resblks;
1281 	bool			retried = false;
1282 	int			error;
1283 
1284 retry:
1285 	*nospace_error = 0;
1286 	resblks = *dblocks;
1287 	error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1288 	if (error == -ENOSPC) {
1289 		*nospace_error = error;
1290 		resblks = 0;
1291 		error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1292 	}
1293 	if (error)
1294 		return error;
1295 
1296 	xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
1297 
1298 	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1299 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1300 
1301 	error = xfs_qm_dqattach_locked(dp, false);
1302 	if (error) {
1303 		/* Caller should have allocated the dquots! */
1304 		ASSERT(error != -ENOENT);
1305 		goto out_cancel;
1306 	}
1307 
1308 	error = xfs_qm_dqattach_locked(ip, false);
1309 	if (error) {
1310 		/* Caller should have allocated the dquots! */
1311 		ASSERT(error != -ENOENT);
1312 		goto out_cancel;
1313 	}
1314 
1315 	if (resblks == 0)
1316 		goto done;
1317 
1318 	error = xfs_trans_reserve_quota_nblks(tp, dp, resblks, 0, false);
1319 	if (error == -EDQUOT || error == -ENOSPC) {
1320 		if (!retried) {
1321 			xfs_trans_cancel(tp);
1322 			xfs_blockgc_free_quota(dp, 0);
1323 			retried = true;
1324 			goto retry;
1325 		}
1326 
1327 		*nospace_error = error;
1328 		resblks = 0;
1329 		error = 0;
1330 	}
1331 	if (error)
1332 		goto out_cancel;
1333 
1334 done:
1335 	*tpp = tp;
1336 	*dblocks = resblks;
1337 	return 0;
1338 
1339 out_cancel:
1340 	xfs_trans_cancel(tp);
1341 	return error;
1342 }
1343