forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2024-05-10 23fa18eaa71266feff7ba8d83022d9e1cc83c65a 
 kernel/fs/jbd2/commit.c
....@@ -184,23 +184,51 @@
184184 /*
185185  * write the filemap data using writepage() address_space_operations.
186186  * We don't do block allocation here even for delalloc. We don't
187
- * use writepages() because with dealyed allocation we may be doing
187
+ * use writepages() because with delayed allocation we may be doing
188188  * block allocation in writepages().
189189  */
190
-static int journal_submit_inode_data_buffers(struct address_space *mapping,
191
-		loff_t dirty_start, loff_t dirty_end)
190
+int jbd2_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
192191 {
193
-	int ret;
192
+	struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
194193 	struct writeback_control wbc = {
195194 		.sync_mode =  WB_SYNC_ALL,
196195 		.nr_to_write = mapping->nrpages * 2,
197
-		.range_start = dirty_start,
198
-		.range_end = dirty_end,
196
+		.range_start = jinode->i_dirty_start,
197
+		.range_end = jinode->i_dirty_end,
199198 	};
200199 
201
-	ret = generic_writepages(mapping, &wbc);
202
-	return ret;
200
+	/*
201
+	 * submit the inode data buffers. We use writepage
202
+	 * instead of writepages. Because writepages can do
203
+	 * block allocation with delalloc. We need to write
204
+	 * only allocated blocks here.
205
+	 */
206
+	return generic_writepages(mapping, &wbc);
203207 }
208
+
209
+/* Send all the data buffers related to an inode */
210
+int jbd2_submit_inode_data(struct jbd2_inode *jinode)
211
+{
212
+
213
+	if (!jinode || !(jinode->i_flags & JI_WRITE_DATA))
214
+		return 0;
215
+
216
+	trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
217
+	return jbd2_journal_submit_inode_data_buffers(jinode);
218
+
219
+}
220
+EXPORT_SYMBOL(jbd2_submit_inode_data);
221
+
222
+int jbd2_wait_inode_data(journal_t *journal, struct jbd2_inode *jinode)
223
+{
224
+	if (!jinode || !(jinode->i_flags & JI_WAIT_DATA) ||
225
+		!jinode->i_vfs_inode || !jinode->i_vfs_inode->i_mapping)
226
+		return 0;
227
+	return filemap_fdatawait_range_keep_errors(
228
+		jinode->i_vfs_inode->i_mapping, jinode->i_dirty_start,
229
+		jinode->i_dirty_end);
230
+}
231
+EXPORT_SYMBOL(jbd2_wait_inode_data);
204232 
205233 /*
206234  * Submit all the data buffers of inode associated with the transaction to
....@@ -215,29 +243,20 @@
215243 {
216244 	struct jbd2_inode *jinode;
217245 	int err, ret = 0;
218
-	struct address_space *mapping;
219246 
220247 	spin_lock(&journal->j_list_lock);
221248 	list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
222
-		loff_t dirty_start = jinode->i_dirty_start;
223
-		loff_t dirty_end = jinode->i_dirty_end;
224
-
225249 		if (!(jinode->i_flags & JI_WRITE_DATA))
226250 			continue;
227
-		mapping = jinode->i_vfs_inode->i_mapping;
228251 		jinode->i_flags |= JI_COMMIT_RUNNING;
229252 		spin_unlock(&journal->j_list_lock);
230
-		/*
231
-		 * submit the inode data buffers. We use writepage
232
-		 * instead of writepages. Because writepages can do
233
-		 * block allocation  with delalloc. We need to write
234
-		 * only allocated blocks here.
235
-		 */
253
+		/* submit the inode data buffers. */
236254 		trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
237
-		err = journal_submit_inode_data_buffers(mapping, dirty_start,
238
-				dirty_end);
239
-		if (!ret)
240
-			ret = err;
255
+		if (journal->j_submit_inode_data_buffers) {
256
+			err = journal->j_submit_inode_data_buffers(jinode);
257
+			if (!ret)
258
+				ret = err;
259
+		}
241260 		spin_lock(&journal->j_list_lock);
242261 		J_ASSERT(jinode->i_transaction == commit_transaction);
243262 		jinode->i_flags &= ~JI_COMMIT_RUNNING;
....@@ -246,6 +265,15 @@
246265 	}
247266 	spin_unlock(&journal->j_list_lock);
248267 	return ret;
268
+}
269
+
270
+int jbd2_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
271
+{
272
+	struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
273
+
274
+	return filemap_fdatawait_range_keep_errors(mapping,
275
+						   jinode->i_dirty_start,
276
+						   jinode->i_dirty_end);
249277 }
250278 
251279 /*
....@@ -262,18 +290,16 @@
262290 	/* For locking, see the comment in journal_submit_data_buffers() */
263291 	spin_lock(&journal->j_list_lock);
264292 	list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
265
-		loff_t dirty_start = jinode->i_dirty_start;
266
-		loff_t dirty_end = jinode->i_dirty_end;
267
-
268293 		if (!(jinode->i_flags & JI_WAIT_DATA))
269294 			continue;
270295 		jinode->i_flags |= JI_COMMIT_RUNNING;
271296 		spin_unlock(&journal->j_list_lock);
272
-		err = filemap_fdatawait_range_keep_errors(
273
-				jinode->i_vfs_inode->i_mapping, dirty_start,
274
-				dirty_end);
275
-		if (!ret)
276
-			ret = err;
297
+		/* wait for the inode data buffers writeout. */
298
+		if (journal->j_finish_inode_data_buffers) {
299
+			err = journal->j_finish_inode_data_buffers(jinode);
300
+			if (!ret)
301
+				ret = err;
302
+		}
277303 		spin_lock(&journal->j_list_lock);
278304 		jinode->i_flags &= ~JI_COMMIT_RUNNING;
279305 		smp_mb();
....@@ -413,6 +439,29 @@
413439 	J_ASSERT(journal->j_running_transaction != NULL);
414440 	J_ASSERT(journal->j_committing_transaction == NULL);
415441 
442
+	write_lock(&journal->j_state_lock);
443
+	journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
444
+	while (journal->j_flags & JBD2_FAST_COMMIT_ONGOING) {
445
+		DEFINE_WAIT(wait);
446
+
447
+		prepare_to_wait(&journal->j_fc_wait, &wait,
448
+				TASK_UNINTERRUPTIBLE);
449
+		write_unlock(&journal->j_state_lock);
450
+		schedule();
451
+		write_lock(&journal->j_state_lock);
452
+		finish_wait(&journal->j_fc_wait, &wait);
453
+		/*
454
+		 * TODO: by blocking fast commits here, we are increasing
455
+		 * fsync() latency slightly. Strictly speaking, we don't need
456
+		 * to block fast commits until the transaction enters T_FLUSH
457
+		 * state. So an optimization is possible where we block new fast
458
+		 * commits here and wait for existing ones to complete
459
+		 * just before we enter T_FLUSH. That way, the existing fast
460
+		 * commits and this full commit can proceed parallely.
461
+		 */
462
+	}
463
+	write_unlock(&journal->j_state_lock);
464
+
416465 	commit_transaction = journal->j_running_transaction;
417466 
418467 	trace_jbd2_start_commit(journal, commit_transaction);
....@@ -420,6 +469,7 @@
420469 			commit_transaction->t_tid);
421470 
422471 	write_lock(&journal->j_state_lock);
472
+	journal->j_fc_off = 0;
423473 	J_ASSERT(commit_transaction->t_state == T_RUNNING);
424474 	commit_transaction->t_state = T_LOCKED;
425475 
....@@ -450,6 +500,7 @@
450500 		finish_wait(&journal->j_wait_updates, &wait);
451501 	}
452502 	spin_unlock(&commit_transaction->t_handle_lock);
503
+	commit_transaction->t_state = T_SWITCH;
453504 
454505 	J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
455506 			journal->j_max_transaction_buffers);
....@@ -469,6 +520,8 @@
469520 	 * has reserved.  This is consistent with the existing behaviour
470521 	 * that multiple jbd2_journal_get_write_access() calls to the same
471522 	 * buffer are perfectly permissible.
523
+	 * We use journal->j_state_lock here to serialize processing of
524
+	 * t_reserved_list with eviction of buffers from journal_unmap_buffer().
472525 	 */
473526 	while (commit_transaction->t_reserved_list) {
474527 		jh = commit_transaction->t_reserved_list;
....@@ -480,14 +533,15 @@
480533 		if (jh->b_committed_data) {
481534 			struct buffer_head *bh = jh2bh(jh);
482535 
483
-			jbd_lock_bh_state(bh);
536
+			spin_lock(&jh->b_state_lock);
484537 			jbd2_free(jh->b_committed_data, bh->b_size);
485538 			jh->b_committed_data = NULL;
486
-			jbd_unlock_bh_state(bh);
539
+			spin_unlock(&jh->b_state_lock);
487540 		}
488541 		jbd2_journal_refile_buffer(journal, jh);
489542 	}
490543 
544
+	write_unlock(&journal->j_state_lock);
491545 	/*
492546 	 * Now try to drop any written-back buffers from the journal's
493547 	 * checkpoint lists.  We do this *before* commit because it potentially
....@@ -510,6 +564,7 @@
510564 	 */
511565 	jbd2_journal_switch_revoke_table(journal);
512566 
567
+	write_lock(&journal->j_state_lock);
513568 	/*
514569 	 * Reserved credits cannot be claimed anymore, free them
515570 	 */
....@@ -526,7 +581,7 @@
526581 	journal->j_running_transaction = NULL;
527582 	start_time = ktime_get();
528583 	commit_transaction->t_log_start = journal->j_head;
529
-	wake_up(&journal->j_wait_transaction_locked);
584
+	wake_up_all(&journal->j_wait_transaction_locked);
530585 	write_unlock(&journal->j_state_lock);
531586 
532587 	jbd_debug(3, "JBD2: commit phase 2a\n");
....@@ -557,8 +612,7 @@
557612 	stats.run.rs_logging = jiffies;
558613 	stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
559614 					       stats.run.rs_logging);
560
-	stats.run.rs_blocks =
561
-		atomic_read(&commit_transaction->t_outstanding_credits);
615
+	stats.run.rs_blocks = commit_transaction->t_nr_buffers;
562616 	stats.run.rs_blocks_logged = 0;
563617 
564618 	J_ASSERT(commit_transaction->t_nr_buffers <=
....@@ -639,8 +693,7 @@
639693 
640694 		/*
641695 		 * start_this_handle() uses t_outstanding_credits to determine
642
-		 * the free space in the log, but this counter is changed
643
-		 * by jbd2_journal_next_log_block() also.
696
+		 * the free space in the log.
644697 		 */
645698 		atomic_dec(&commit_transaction->t_outstanding_credits);
646699 
....@@ -759,7 +812,7 @@
759812 		if (first_block < journal->j_tail)
760813 			freed += journal->j_last - journal->j_first;
761814 		/* Update tail only if we free significant amount of space */
762
-		if (freed < journal->j_maxlen / 4)
815
+		if (freed < jbd2_journal_get_max_txn_bufs(journal))
763816 			update_tail = 0;
764817 	}
765818 	J_ASSERT(commit_transaction->t_state == T_COMMIT);
....@@ -774,7 +827,7 @@
774827 	if (commit_transaction->t_need_data_flush &&
775828 	    (journal->j_fs_dev != journal->j_dev) &&
776829 	    (journal->j_flags & JBD2_BARRIER))
777
-		blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
830
+		blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS);
778831 
779832 	/* Done it all: now write the commit record asynchronously. */
780833 	if (jbd2_has_feature_async_commit(journal)) {
....@@ -881,11 +934,14 @@
881934 	stats.run.rs_blocks_logged++;
882935 	if (jbd2_has_feature_async_commit(journal) &&
883936 	    journal->j_flags & JBD2_BARRIER) {
884
-		blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
937
+		blkdev_issue_flush(journal->j_dev, GFP_NOFS);
885938 	}
886939 
887940 	if (err)
888941 		jbd2_journal_abort(journal, err);
942
+
943
+	WARN_ON_ONCE(
944
+		atomic_read(&commit_transaction->t_outstanding_credits) < 0);
889945 
890946 	/*
891947 	 * Now disk caches for filesystem device are flushed so we are safe to
....@@ -917,6 +973,7 @@
917973 		transaction_t *cp_transaction;
918974 		struct buffer_head *bh;
919975 		int try_to_free = 0;
976
+		bool drop_ref;
920977 
921978 		jh = commit_transaction->t_forget;
922979 		spin_unlock(&journal->j_list_lock);
....@@ -926,7 +983,7 @@
926983 		 * done with it.
927984 		 */
928985 		get_bh(bh);
929
-		jbd_lock_bh_state(bh);
986
+		spin_lock(&jh->b_state_lock);
930987 		J_ASSERT_JH(jh,	jh->b_transaction == commit_transaction);
931988 
932989 		/*
....@@ -1026,8 +1083,10 @@
10261083 				try_to_free = 1;
10271084 		}
10281085 		JBUFFER_TRACE(jh, "refile or unfile buffer");
1029
-		__jbd2_journal_refile_buffer(jh);
1030
-		jbd_unlock_bh_state(bh);
1086
+		drop_ref = __jbd2_journal_refile_buffer(jh);
1087
+		spin_unlock(&jh->b_state_lock);
1088
+		if (drop_ref)
1089
+			jbd2_journal_put_journal_head(jh);
10311090 		if (try_to_free)
10321091 			release_buffer_page(bh);	/* Drops bh reference */
10331092 		else
....@@ -1112,12 +1171,16 @@
11121171 
11131172 	if (journal->j_commit_callback)
11141173 		journal->j_commit_callback(journal, commit_transaction);
1174
+	if (journal->j_fc_cleanup_callback)
1175
+		journal->j_fc_cleanup_callback(journal, 1);
11151176 
11161177 	trace_jbd2_end_commit(journal, commit_transaction);
11171178 	jbd_debug(1, "JBD2: commit %d complete, head %d\n",
11181179 		  journal->j_commit_sequence, journal->j_tail_sequence);
11191180 
11201181 	write_lock(&journal->j_state_lock);
1182
+	journal->j_flags &= ~JBD2_FULL_COMMIT_ONGOING;
1183
+	journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
11211184 	spin_lock(&journal->j_list_lock);
11221185 	commit_transaction->t_state = T_FINISHED;
11231186 	/* Check if the transaction can be dropped now that we are finished */
....@@ -1129,6 +1192,7 @@
11291192 	spin_unlock(&journal->j_list_lock);
11301193 	write_unlock(&journal->j_state_lock);
11311194 	wake_up(&journal->j_wait_done_commit);
1195
+	wake_up(&journal->j_fc_wait);
11321196 
11331197 	/*
11341198 	 * Calculate overall stats