forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-02-20 ea08eeccae9297f7aabd2ef7f0c2517ac4549acc 
 kernel/fs/block_dev.c
....@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  *  linux/fs/block_dev.c
34  *
....@@ -18,22 +19,21 @@
1819 #include <linux/module.h>
1920 #include <linux/blkpg.h>
2021 #include <linux/magic.h>
21
-#include <linux/dax.h>
2222 #include <linux/buffer_head.h>
2323 #include <linux/swap.h>
2424 #include <linux/pagevec.h>
2525 #include <linux/writeback.h>
2626 #include <linux/mpage.h>
2727 #include <linux/mount.h>
28
+#include <linux/pseudo_fs.h>
2829 #include <linux/uio.h>
2930 #include <linux/namei.h>
3031 #include <linux/log2.h>
3132 #include <linux/cleancache.h>
32
-#include <linux/dax.h>
33
-#include <linux/badblocks.h>
3433 #include <linux/task_io_accounting_ops.h>
3534 #include <linux/falloc.h>
3635 #include <linux/uaccess.h>
36
+#include <linux/suspend.h>
3737 #include "internal.h"
3838 
3939 struct bdev_inode {
....@@ -75,7 +75,7 @@
7575 }
7676 
7777 /* Kill _all_ buffers and pagecache , dirty or not.. */
78
-void kill_bdev(struct block_device *bdev)
78
+static void kill_bdev(struct block_device *bdev)
7979 {
8080 	struct address_space *mapping = bdev->bd_inode->i_mapping;
8181 
....@@ -84,8 +84,7 @@
8484 
8585 	invalidate_bh_lrus();
8686 	truncate_inode_pages(mapping, 0);
87
-}	
88
-EXPORT_SYMBOL(kill_bdev);
87
+}
8988 
9089 /* Invalidate clean unused buffers and pagecache. */
9190 void invalidate_bdev(struct block_device *bdev)
....@@ -104,9 +103,47 @@
104103 }
105104 EXPORT_SYMBOL(invalidate_bdev);
106105 
106
+/*
107
+ * Drop all buffers & page cache for given bdev range. This function bails
108
+ * with error if bdev has other exclusive owner (such as filesystem).
109
+ */
110
+int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
111
+			loff_t lstart, loff_t lend)
112
+{
113
+	struct block_device *claimed_bdev = NULL;
114
+	int err;
115
+
116
+	/*
117
+	 * If we don't hold exclusive handle for the device, upgrade to it
118
+	 * while we discard the buffer cache to avoid discarding buffers
119
+	 * under live filesystem.
120
+	 */
121
+	if (!(mode & FMODE_EXCL)) {
122
+		claimed_bdev = bdev->bd_contains;
123
+		err = bd_prepare_to_claim(bdev, claimed_bdev,
124
+					  truncate_bdev_range);
125
+		if (err)
126
+			goto invalidate;
127
+	}
128
+	truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
129
+	if (claimed_bdev)
130
+		bd_abort_claiming(bdev, claimed_bdev, truncate_bdev_range);
131
+	return 0;
132
+
133
+invalidate:
134
+	/*
135
+	 * Someone else has handle exclusively open. Try invalidating instead.
136
+	 * The 'end' argument is inclusive so the rounding is safe.
137
+	 */
138
+	return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
139
+					     lstart >> PAGE_SHIFT,
140
+					     lend >> PAGE_SHIFT);
141
+}
142
+EXPORT_SYMBOL(truncate_bdev_range);
143
+
107144 static void set_init_blocksize(struct block_device *bdev)
108145 {
109
-	unsigned bsize = bdev_logical_block_size(bdev);
146
+	unsigned int bsize = bdev_logical_block_size(bdev);
110147 	loff_t size = i_size_read(bdev->bd_inode);
111148 
112149 	while (bsize < PAGE_SIZE) {
....@@ -114,7 +151,6 @@
114151 			break;
115152 		bsize <<= 1;
116153 	}
117
-	bdev->bd_block_size = bsize;
118154 	bdev->bd_inode->i_blkbits = blksize_bits(bsize);
119155 }
120156 
....@@ -129,9 +165,8 @@
129165 		return -EINVAL;
130166 
131167 	/* Don't change the size if it is same as current */
132
-	if (bdev->bd_block_size != size) {
168
+	if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
133169 		sync_blockdev(bdev);
134
-		bdev->bd_block_size = size;
135170 		bdev->bd_inode->i_blkbits = blksize_bits(size);
136171 		kill_bdev(bdev);
137172 	}
....@@ -151,7 +186,7 @@
151186 	return sb->s_blocksize;
152187 }
153188 
154
-EXPORT_SYMBOL(sb_set_blocksize);
189
+EXPORT_SYMBOL_NS(sb_set_blocksize, ANDROID_GKI_VFS_EXPORT_ONLY);
155190 
156191 int sb_min_blocksize(struct super_block *sb, int size)
157192 {
....@@ -161,7 +196,7 @@
161196 	return sb_set_blocksize(sb, size);
162197 }
163198 
164
-EXPORT_SYMBOL(sb_min_blocksize);
199
+EXPORT_SYMBOL_NS(sb_min_blocksize, ANDROID_GKI_VFS_EXPORT_ONLY);
165200 
166201 static int
167202 blkdev_get_block(struct inode *inode, sector_t iblock,
....@@ -195,7 +230,7 @@
195230 	struct task_struct *waiter = bio->bi_private;
196231 
197232 	WRITE_ONCE(bio->bi_private, NULL);
198
-	wake_up_process(waiter);
233
+	blk_wake_io_task(waiter);
199234 }
200235 
201236 static ssize_t
....@@ -204,13 +239,12 @@
204239 {
205240 	struct file *file = iocb->ki_filp;
206241 	struct block_device *bdev = I_BDEV(bdev_file_inode(file));
207
-	struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
242
+	struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
208243 	loff_t pos = iocb->ki_pos;
209244 	bool should_dirty = false;
210245 	struct bio bio;
211246 	ssize_t ret;
212247 	blk_qc_t qc;
213
-	int i;
214248 
215249 	if ((pos | iov_iter_alignment(iter)) &
216250 	    (bdev_logical_block_size(bdev) - 1))
....@@ -246,6 +280,10 @@
246280 		bio.bi_opf = dio_bio_write_op(iocb);
247281 		task_io_account_write(ret);
248282 	}
283
+	if (iocb->ki_flags & IOCB_NOWAIT)
284
+		bio.bi_opf |= REQ_NOWAIT;
285
+	if (iocb->ki_flags & IOCB_HIPRI)
286
+		bio_set_polled(&bio, iocb);
249287 
250288 	qc = submit_bio(&bio);
251289 	for (;;) {
....@@ -253,17 +291,12 @@
253291 		if (!READ_ONCE(bio.bi_private))
254292 			break;
255293 		if (!(iocb->ki_flags & IOCB_HIPRI) ||
256
-		    !blk_poll(bdev_get_queue(bdev), qc))
257
-			io_schedule();
294
+		    !blk_poll(bdev_get_queue(bdev), qc, true))
295
+			blk_io_schedule();
258296 	}
259297 	__set_current_state(TASK_RUNNING);
260298 
261
-	bio_for_each_segment_all(bvec, &bio, i) {
262
-		if (should_dirty && !PageCompound(bvec->bv_page))
263
-			set_page_dirty_lock(bvec->bv_page);
264
-		put_page(bvec->bv_page);
265
-	}
266
-
299
+	bio_release_pages(&bio, should_dirty);
267300 	if (unlikely(bio.bi_status))
268301 		ret = blk_status_to_errno(bio.bi_status);
269302 
....@@ -291,6 +324,14 @@
291324 
292325 static struct bio_set blkdev_dio_pool;
293326 
327
+static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
328
+{
329
+	struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
330
+	struct request_queue *q = bdev_get_queue(bdev);
331
+
332
+	return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
333
+}
334
+
294335 static void blkdev_bio_end_io(struct bio *bio)
295336 {
296337 	struct blkdev_dio *dio = bio->bi_private;
....@@ -312,23 +353,20 @@
312353 			}
313354 
314355 			dio->iocb->ki_complete(iocb, ret, 0);
315
-			bio_put(&dio->bio);
356
+			if (dio->multi_bio)
357
+				bio_put(&dio->bio);
316358 		} else {
317359 			struct task_struct *waiter = dio->waiter;
318360 
319361 			WRITE_ONCE(dio->waiter, NULL);
320
-			wake_up_process(waiter);
362
+			blk_wake_io_task(waiter);
321363 		}
322364 	}
323365 
324366 	if (should_dirty) {
325367 		bio_check_pages_dirty(bio);
326368 	} else {
327
-		struct bio_vec *bvec;
328
-		int i;
329
-
330
-		bio_for_each_segment_all(bvec, bio, i)
331
-			put_page(bvec->bv_page);
369
+		bio_release_pages(bio, false);
332370 		bio_put(bio);
333371 	}
334372 }
....@@ -342,6 +380,7 @@
342380 	struct blk_plug plug;
343381 	struct blkdev_dio *dio;
344382 	struct bio *bio;
383
+	bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
345384 	bool is_read = (iov_iter_rw(iter) == READ), is_sync;
346385 	loff_t pos = iocb->ki_pos;
347386 	blk_qc_t qc = BLK_QC_T_NONE;
....@@ -352,20 +391,27 @@
352391 		return -EINVAL;
353392 
354393 	bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
355
-	bio_get(bio); /* extra ref for the completion handler */
356394 
357395 	dio = container_of(bio, struct blkdev_dio, bio);
358396 	dio->is_sync = is_sync = is_sync_kiocb(iocb);
359
-	if (dio->is_sync)
397
+	if (dio->is_sync) {
360398 		dio->waiter = current;
361
-	else
399
+		bio_get(bio);
400
+	} else {
362401 		dio->iocb = iocb;
402
+	}
363403 
364404 	dio->size = 0;
365405 	dio->multi_bio = false;
366
-	dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
406
+	dio->should_dirty = is_read && iter_is_iovec(iter);
367407 
368
-	blk_start_plug(&plug);
408
+	/*
409
+	 * Don't plug for HIPRI/polled IO, as those should go straight
410
+	 * to issue
411
+	 */
412
+	if (!is_poll)
413
+		blk_start_plug(&plug);
414
+
369415 	for (;;) {
370416 		bio_set_dev(bio, bdev);
371417 		bio->bi_iter.bi_sector = pos >> 9;
....@@ -389,17 +435,36 @@
389435 			bio->bi_opf = dio_bio_write_op(iocb);
390436 			task_io_account_write(bio->bi_iter.bi_size);
391437 		}
438
+		if (iocb->ki_flags & IOCB_NOWAIT)
439
+			bio->bi_opf |= REQ_NOWAIT;
392440 
393441 		dio->size += bio->bi_iter.bi_size;
394442 		pos += bio->bi_iter.bi_size;
395443 
396444 		nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
397445 		if (!nr_pages) {
446
+			bool polled = false;
447
+
448
+			if (iocb->ki_flags & IOCB_HIPRI) {
449
+				bio_set_polled(bio, iocb);
450
+				polled = true;
451
+			}
452
+
398453 			qc = submit_bio(bio);
454
+
455
+			if (polled)
456
+				WRITE_ONCE(iocb->ki_cookie, qc);
399457 			break;
400458 		}
401459 
402460 		if (!dio->multi_bio) {
461
+			/*
462
+			 * AIO needs an extra reference to ensure the dio
463
+			 * structure which is embedded into the first bio
464
+			 * stays around.
465
+			 */
466
+			if (!is_sync)
467
+				bio_get(bio);
403468 			dio->multi_bio = true;
404469 			atomic_set(&dio->ref, 2);
405470 		} else {
....@@ -409,7 +474,9 @@
409474 		submit_bio(bio);
410475 		bio = bio_alloc(GFP_KERNEL, nr_pages);
411476 	}
412
-	blk_finish_plug(&plug);
477
+
478
+	if (!is_poll)
479
+		blk_finish_plug(&plug);
413480 
414481 	if (!is_sync)
415482 		return -EIOCBQUEUED;
....@@ -420,8 +487,8 @@
420487 			break;
421488 
422489 		if (!(iocb->ki_flags & IOCB_HIPRI) ||
423
-		    !blk_poll(bdev_get_queue(bdev), qc))
424
-			io_schedule();
490
+		    !blk_poll(bdev_get_queue(bdev), qc, true))
491
+			blk_io_schedule();
425492 	}
426493 	__set_current_state(TASK_RUNNING);
427494 
....@@ -502,55 +569,47 @@
502569  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
503570  * actually.
504571  */
505
-struct super_block *freeze_bdev(struct block_device *bdev)
572
+int freeze_bdev(struct block_device *bdev)
506573 {
507574 	struct super_block *sb;
508575 	int error = 0;
509576 
510577 	mutex_lock(&bdev->bd_fsfreeze_mutex);
511
-	if (++bdev->bd_fsfreeze_count > 1) {
512
-		/*
513
-		 * We don't even need to grab a reference - the first call
514
-		 * to freeze_bdev grab an active reference and only the last
515
-		 * thaw_bdev drops it.
516
-		 */
517
-		sb = get_super(bdev);
518
-		if (sb)
519
-			drop_super(sb);
520
-		mutex_unlock(&bdev->bd_fsfreeze_mutex);
521
-		return sb;
522
-	}
578
+	if (++bdev->bd_fsfreeze_count > 1)
579
+		goto done;
523580 
524581 	sb = get_active_super(bdev);
525582 	if (!sb)
526
-		goto out;
583
+		goto sync;
527584 	if (sb->s_op->freeze_super)
528585 		error = sb->s_op->freeze_super(sb);
529586 	else
530587 		error = freeze_super(sb);
531
-	if (error) {
532
-		deactivate_super(sb);
533
-		bdev->bd_fsfreeze_count--;
534
-		mutex_unlock(&bdev->bd_fsfreeze_mutex);
535
-		return ERR_PTR(error);
536
-	}
537588 	deactivate_super(sb);
538
- out:
589
+
590
+	if (error) {
591
+		bdev->bd_fsfreeze_count--;
592
+		goto done;
593
+	}
594
+	bdev->bd_fsfreeze_sb = sb;
595
+
596
+sync:
539597 	sync_blockdev(bdev);
598
+done:
540599 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
541
-	return sb;	/* thaw_bdev releases s->s_umount */
600
+	return error;
542601 }
543602 EXPORT_SYMBOL(freeze_bdev);
544603 
545604 /**
546605  * thaw_bdev  -- unlock filesystem
547606  * @bdev:	blockdevice to unlock
548
- * @sb:		associated superblock
549607  *
550608  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
551609  */
552
-int thaw_bdev(struct block_device *bdev, struct super_block *sb)
610
+int thaw_bdev(struct block_device *bdev)
553611 {
612
+	struct super_block *sb;
554613 	int error = -EINVAL;
555614 
556615 	mutex_lock(&bdev->bd_fsfreeze_mutex);
....@@ -561,6 +620,7 @@
561620 	if (--bdev->bd_fsfreeze_count > 0)
562621 		goto out;
563622 
623
+	sb = bdev->bd_fsfreeze_sb;
564624 	if (!sb)
565625 		goto out;
566626 
....@@ -586,10 +646,9 @@
586646 	return block_read_full_page(page, blkdev_get_block);
587647 }
588648 
589
-static int blkdev_readpages(struct file *file, struct address_space *mapping,
590
-			struct list_head *pages, unsigned nr_pages)
649
+static void blkdev_readahead(struct readahead_control *rac)
591650 {
592
-	return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
651
+	mpage_readahead(rac, blkdev_get_block);
593652 }
594653 
595654 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
....@@ -644,7 +703,7 @@
644703 	 * i_mutex and doing so causes performance issues with concurrent
645704 	 * O_SYNC writers to a block device.
646705 	 */
647
-	error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
706
+	error = blkdev_issue_flush(bdev, GFP_KERNEL);
648707 	if (error == -EOPNOTSUPP)
649708 		error = 0;
650709 
....@@ -677,15 +736,14 @@
677736 	if (!ops->rw_page || bdev_get_integrity(bdev))
678737 		return result;
679738 
680
-	result = blk_queue_enter(bdev->bd_queue, 0);
739
+	result = blk_queue_enter(bdev->bd_disk->queue, 0);
681740 	if (result)
682741 		return result;
683742 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
684743 			      REQ_OP_READ);
685
-	blk_queue_exit(bdev->bd_queue);
744
+	blk_queue_exit(bdev->bd_disk->queue);
686745 	return result;
687746 }
688
-EXPORT_SYMBOL_GPL(bdev_read_page);
689747 
690748 /**
691749  * bdev_write_page() - Start writing a page to a block device
....@@ -714,7 +772,7 @@
714772 
715773 	if (!ops->rw_page || bdev_get_integrity(bdev))
716774 		return -EOPNOTSUPP;
717
-	result = blk_queue_enter(bdev->bd_queue, 0);
775
+	result = blk_queue_enter(bdev->bd_disk->queue, 0);
718776 	if (result)
719777 		return result;
720778 
....@@ -727,10 +785,9 @@
727785 		clean_page_buffers(page);
728786 		unlock_page(page);
729787 	}
730
-	blk_queue_exit(bdev->bd_queue);
788
+	blk_queue_exit(bdev->bd_disk->queue);
731789 	return result;
732790 }
733
-EXPORT_SYMBOL_GPL(bdev_write_page);
734791 
735792 /*
736793  * pseudo-fs
....@@ -747,17 +804,9 @@
747804 	return &ei->vfs_inode;
748805 }
749806 
750
-static void bdev_i_callback(struct rcu_head *head)
807
+static void bdev_free_inode(struct inode *inode)
751808 {
752
-	struct inode *inode = container_of(head, struct inode, i_rcu);
753
-	struct bdev_inode *bdi = BDEV_I(inode);
754
-
755
-	kmem_cache_free(bdev_cachep, bdi);
756
-}
757
-
758
-static void bdev_destroy_inode(struct inode *inode)
759
-{
760
-	call_rcu(&inode->i_rcu, bdev_i_callback);
809
+	kmem_cache_free(bdev_cachep, BDEV_I(inode));
761810 }
762811 
763812 static void init_once(void *foo)
....@@ -767,7 +816,6 @@
767816 
768817 	memset(bdev, 0, sizeof(*bdev));
769818 	mutex_init(&bdev->bd_mutex);
770
-	INIT_LIST_HEAD(&bdev->bd_list);
771819 #ifdef CONFIG_SYSFS
772820 	INIT_LIST_HEAD(&bdev->bd_holder_disks);
773821 #endif
....@@ -783,9 +831,6 @@
783831 	truncate_inode_pages_final(&inode->i_data);
784832 	invalidate_inode_buffers(inode); /* is it needed here? */
785833 	clear_inode(inode);
786
-	spin_lock(&bdev_lock);
787
-	list_del_init(&bdev->bd_list);
788
-	spin_unlock(&bdev_lock);
789834 	/* Detach inode from wb early as bdi_put() may free bdi->wb */
790835 	inode_detach_wb(inode);
791836 	if (bdev->bd_bdi != &noop_backing_dev_info) {
....@@ -797,24 +842,24 @@
797842 static const struct super_operations bdev_sops = {
798843 	.statfs = simple_statfs,
799844 	.alloc_inode = bdev_alloc_inode,
800
-	.destroy_inode = bdev_destroy_inode,
845
+	.free_inode = bdev_free_inode,
801846 	.drop_inode = generic_delete_inode,
802847 	.evict_inode = bdev_evict_inode,
803848 };
804849 
805
-static struct dentry *bd_mount(struct file_system_type *fs_type,
806
-	int flags, const char *dev_name, void *data)
850
+static int bd_init_fs_context(struct fs_context *fc)
807851 {
808
-	struct dentry *dent;
809
-	dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
810
-	if (!IS_ERR(dent))
811
-		dent->d_sb->s_iflags |= SB_I_CGROUPWB;
812
-	return dent;
852
+	struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
853
+	if (!ctx)
854
+		return -ENOMEM;
855
+	fc->s_iflags |= SB_I_CGROUPWB;
856
+	ctx->ops = &bdev_sops;
857
+	return 0;
813858 }
814859 
815860 static struct file_system_type bd_type = {
816861 	.name		= "bdev",
817
-	.mount		= bd_mount,
862
+	.init_fs_context = bd_init_fs_context,
818863 	.kill_sb	= kill_anon_super,
819864 };
820865 
....@@ -860,24 +905,7 @@
860905 	return 0;
861906 }
862907 
863
-static LIST_HEAD(all_bdevs);
864
-
865
-/*
866
- * If there is a bdev inode for this device, unhash it so that it gets evicted
867
- * as soon as last inode reference is dropped.
868
- */
869
-void bdev_unhash_inode(dev_t dev)
870
-{
871
-	struct inode *inode;
872
-
873
-	inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
874
-	if (inode) {
875
-		remove_inode_hash(inode);
876
-		iput(inode);
877
-	}
878
-}
879
-
880
-struct block_device *bdget(dev_t dev)
908
+static struct block_device *bdget(dev_t dev)
881909 {
882910 	struct block_device *bdev;
883911 	struct inode *inode;
....@@ -891,26 +919,20 @@
891919 	bdev = &BDEV_I(inode)->bdev;
892920 
893921 	if (inode->i_state & I_NEW) {
922
+		spin_lock_init(&bdev->bd_size_lock);
894923 		bdev->bd_contains = NULL;
895924 		bdev->bd_super = NULL;
896925 		bdev->bd_inode = inode;
897
-		bdev->bd_block_size = i_blocksize(inode);
898926 		bdev->bd_part_count = 0;
899
-		bdev->bd_invalidated = 0;
900927 		inode->i_mode = S_IFBLK;
901928 		inode->i_rdev = dev;
902929 		inode->i_bdev = bdev;
903930 		inode->i_data.a_ops = &def_blk_aops;
904931 		mapping_set_gfp_mask(&inode->i_data, GFP_USER);
905
-		spin_lock(&bdev_lock);
906
-		list_add(&bdev->bd_list, &all_bdevs);
907
-		spin_unlock(&bdev_lock);
908932 		unlock_new_inode(inode);
909933 	}
910934 	return bdev;
911935 }
912
-
913
-EXPORT_SYMBOL(bdget);
914936 
915937 /**
916938  * bdgrab -- Grab a reference to an already referenced block device
....@@ -923,15 +945,21 @@
923945 }
924946 EXPORT_SYMBOL(bdgrab);
925947 
948
+struct block_device *bdget_part(struct hd_struct *part)
949
+{
950
+	return bdget(part_devt(part));
951
+}
952
+
926953 long nr_blockdev_pages(void)
927954 {
928
-	struct block_device *bdev;
955
+	struct inode *inode;
929956 	long ret = 0;
930
-	spin_lock(&bdev_lock);
931
-	list_for_each_entry(bdev, &all_bdevs, bd_list) {
932
-		ret += bdev->bd_inode->i_mapping->nrpages;
933
-	}
934
-	spin_unlock(&bdev_lock);
957
+
958
+	spin_lock(&blockdev_superblock->s_inode_list_lock);
959
+	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
960
+		ret += inode->i_mapping->nrpages;
961
+	spin_unlock(&blockdev_superblock->s_inode_list_lock);
962
+
935963 	return ret;
936964 }
937965 
....@@ -1033,30 +1061,28 @@
10331061 }
10341062 
10351063 /**
1036
- * bd_prepare_to_claim - prepare to claim a block device
1064
+ * bd_prepare_to_claim - claim a block device
10371065  * @bdev: block device of interest
10381066  * @whole: the whole device containing @bdev, may equal @bdev
10391067  * @holder: holder trying to claim @bdev
10401068  *
1041
- * Prepare to claim @bdev.  This function fails if @bdev is already
1042
- * claimed by another holder and waits if another claiming is in
1043
- * progress.  This function doesn't actually claim.  On successful
1044
- * return, the caller has ownership of bd_claiming and bd_holder[s].
1045
- *
1046
- * CONTEXT:
1047
- * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
1048
- * it multiple times.
1069
+ * Claim @bdev.  This function fails if @bdev is already claimed by another
1070
+ * holder and waits if another claiming is in progress. return, the caller
1071
+ * has ownership of bd_claiming and bd_holder[s].
10491072  *
10501073  * RETURNS:
10511074  * 0 if @bdev can be claimed, -EBUSY otherwise.
10521075  */
1053
-static int bd_prepare_to_claim(struct block_device *bdev,
1054
-			       struct block_device *whole, void *holder)
1076
+int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1077
+		void *holder)
10551078 {
10561079 retry:
1080
+	spin_lock(&bdev_lock);
10571081 	/* if someone else claimed, fail */
1058
-	if (!bd_may_claim(bdev, whole, holder))
1082
+	if (!bd_may_claim(bdev, whole, holder)) {
1083
+		spin_unlock(&bdev_lock);
10591084 		return -EBUSY;
1085
+	}
10601086 
10611087 	/* if claiming is already in progress, wait for it to finish */
10621088 	if (whole->bd_claiming) {
....@@ -1067,13 +1093,15 @@
10671093 		spin_unlock(&bdev_lock);
10681094 		schedule();
10691095 		finish_wait(wq, &wait);
1070
-		spin_lock(&bdev_lock);
10711096 		goto retry;
10721097 	}
10731098 
10741099 	/* yay, all mine */
1100
+	whole->bd_claiming = holder;
1101
+	spin_unlock(&bdev_lock);
10751102 	return 0;
10761103 }
1104
+EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
10771105 
10781106 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
10791107 {
....@@ -1096,77 +1124,59 @@
10961124 	return disk;
10971125 }
10981126 
1099
-/**
1100
- * bd_start_claiming - start claiming a block device
1101
- * @bdev: block device of interest
1102
- * @holder: holder trying to claim @bdev
1103
- *
1104
- * @bdev is about to be opened exclusively.  Check @bdev can be opened
1105
- * exclusively and mark that an exclusive open is in progress.  Each
1106
- * successful call to this function must be matched with a call to
1107
- * either bd_finish_claiming() or bd_abort_claiming() (which do not
1108
- * fail).
1109
- *
1110
- * This function is used to gain exclusive access to the block device
1111
- * without actually causing other exclusive open attempts to fail. It
1112
- * should be used when the open sequence itself requires exclusive
1113
- * access but may subsequently fail.
1114
- *
1115
- * CONTEXT:
1116
- * Might sleep.
1117
- *
1118
- * RETURNS:
1119
- * Pointer to the block device containing @bdev on success, ERR_PTR()
1120
- * value on failure.
1121
- */
1122
-static struct block_device *bd_start_claiming(struct block_device *bdev,
1123
-					      void *holder)
1127
+static void bd_clear_claiming(struct block_device *whole, void *holder)
11241128 {
1125
-	struct gendisk *disk;
1126
-	struct block_device *whole;
1127
-	int partno, err;
1128
-
1129
-	might_sleep();
1130
-
1131
-	/*
1132
-	 * @bdev might not have been initialized properly yet, look up
1133
-	 * and grab the outer block device the hard way.
1134
-	 */
1135
-	disk = bdev_get_gendisk(bdev, &partno);
1136
-	if (!disk)
1137
-		return ERR_PTR(-ENXIO);
1138
-
1139
-	/*
1140
-	 * Normally, @bdev should equal what's returned from bdget_disk()
1141
-	 * if partno is 0; however, some drivers (floppy) use multiple
1142
-	 * bdev's for the same physical device and @bdev may be one of the
1143
-	 * aliases.  Keep @bdev if partno is 0.  This means claimer
1144
-	 * tracking is broken for those devices but it has always been that
1145
-	 * way.
1146
-	 */
1147
-	if (partno)
1148
-		whole = bdget_disk(disk, 0);
1149
-	else
1150
-		whole = bdgrab(bdev);
1151
-
1152
-	put_disk_and_module(disk);
1153
-	if (!whole)
1154
-		return ERR_PTR(-ENOMEM);
1155
-
1156
-	/* prepare to claim, if successful, mark claiming in progress */
1157
-	spin_lock(&bdev_lock);
1158
-
1159
-	err = bd_prepare_to_claim(bdev, whole, holder);
1160
-	if (err == 0) {
1161
-		whole->bd_claiming = holder;
1162
-		spin_unlock(&bdev_lock);
1163
-		return whole;
1164
-	} else {
1165
-		spin_unlock(&bdev_lock);
1166
-		bdput(whole);
1167
-		return ERR_PTR(err);
1168
-	}
1129
+	lockdep_assert_held(&bdev_lock);
1130
+	/* tell others that we're done */
1131
+	BUG_ON(whole->bd_claiming != holder);
1132
+	whole->bd_claiming = NULL;
1133
+	wake_up_bit(&whole->bd_claiming, 0);
11691134 }
1135
+
1136
+/**
1137
+ * bd_finish_claiming - finish claiming of a block device
1138
+ * @bdev: block device of interest
1139
+ * @whole: whole block device
1140
+ * @holder: holder that has claimed @bdev
1141
+ *
1142
+ * Finish exclusive open of a block device. Mark the device as exlusively
1143
+ * open by the holder and wake up all waiters for exclusive open to finish.
1144
+ */
1145
+static void bd_finish_claiming(struct block_device *bdev,
1146
+		struct block_device *whole, void *holder)
1147
+{
1148
+	spin_lock(&bdev_lock);
1149
+	BUG_ON(!bd_may_claim(bdev, whole, holder));
1150
+	/*
1151
+	 * Note that for a whole device bd_holders will be incremented twice,
1152
+	 * and bd_holder will be set to bd_may_claim before being set to holder
1153
+	 */
1154
+	whole->bd_holders++;
1155
+	whole->bd_holder = bd_may_claim;
1156
+	bdev->bd_holders++;
1157
+	bdev->bd_holder = holder;
1158
+	bd_clear_claiming(whole, holder);
1159
+	spin_unlock(&bdev_lock);
1160
+}
1161
+
1162
+/**
1163
+ * bd_abort_claiming - abort claiming of a block device
1164
+ * @bdev: block device of interest
1165
+ * @whole: whole block device
1166
+ * @holder: holder that has claimed @bdev
1167
+ *
1168
+ * Abort claiming of a block device when the exclusive open failed. This can be
1169
+ * also used when exclusive open is not actually desired and we just needed
1170
+ * to block other exclusive openers for a while.
1171
+ */
1172
+void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1173
+		       void *holder)
1174
+{
1175
+	spin_lock(&bdev_lock);
1176
+	bd_clear_claiming(whole, holder);
1177
+	spin_unlock(&bdev_lock);
1178
+}
1179
+EXPORT_SYMBOL(bd_abort_claiming);
11701180 
11711181 #ifdef CONFIG_SYSFS
11721182 struct bd_holder_disk {
....@@ -1312,26 +1322,6 @@
13121322 #endif
13131323 
13141324 /**
1315
- * flush_disk - invalidates all buffer-cache entries on a disk
1316
- *
1317
- * @bdev:      struct block device to be flushed
1318
- * @kill_dirty: flag to guide handling of dirty inodes
1319
- *
1320
- * Invalidates all buffer-cache entries on a disk. It should be called
1321
- * when a disk has been changed -- either by a media change or online
1322
- * resize.
1323
- */
1324
-static void flush_disk(struct block_device *bdev, bool kill_dirty)
1325
-{
1326
-	if (__invalidate_device(bdev, kill_dirty)) {
1327
-		printk(KERN_WARNING "VFS: busy inodes on changed media or "
1328
-		       "resized disk %s\n",
1329
-		       bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1330
-	}
1331
-	bdev->bd_invalidated = 1;
1332
-}
1333
-
1334
-/**
13351325  * check_disk_size_change - checks for disk size change and adjusts bdev size.
13361326  * @disk: struct gendisk to check
13371327  * @bdev: struct bdev to adjust.
....@@ -1341,11 +1331,12 @@
13411331  * and adjusts it if it differs. When shrinking the bdev size, its all caches
13421332  * are freed.
13431333  */
1344
-void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1345
-		bool verbose)
1334
+static void check_disk_size_change(struct gendisk *disk,
1335
+		struct block_device *bdev, bool verbose)
13461336 {
13471337 	loff_t disk_size, bdev_size;
13481338 
1339
+	spin_lock(&bdev->bd_size_lock);
13491340 	disk_size = (loff_t)get_capacity(disk) << 9;
13501341 	bdev_size = i_size_read(bdev->bd_inode);
13511342 	if (disk_size != bdev_size) {
....@@ -1355,89 +1346,109 @@
13551346 			       disk->disk_name, bdev_size, disk_size);
13561347 		}
13571348 		i_size_write(bdev->bd_inode, disk_size);
1358
-		if (bdev_size > disk_size)
1359
-			flush_disk(bdev, false);
1349
+	}
1350
+	spin_unlock(&bdev->bd_size_lock);
1351
+
1352
+	if (bdev_size > disk_size) {
1353
+		if (__invalidate_device(bdev, false))
1354
+			pr_warn("VFS: busy inodes on resized disk %s\n",
1355
+				disk->disk_name);
13601356 	}
13611357 }
13621358 
13631359 /**
1364
- * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1365
- * @disk: struct gendisk to be revalidated
1360
+ * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1361
+ * @disk: struct gendisk to check
1362
+ * @verbose: if %true log a message about a size change if there is any
13661363  *
1367
- * This routine is a wrapper for lower-level driver's revalidate_disk
1368
- * call-backs.  It is used to do common pre and post operations needed
1369
- * for all revalidate_disk operations.
1364
+ * This routine checks to see if the bdev size does not match the disk size
1365
+ * and adjusts it if it differs. When shrinking the bdev size, its all caches
1366
+ * are freed.
13701367  */
1371
-int revalidate_disk(struct gendisk *disk)
1368
+void revalidate_disk_size(struct gendisk *disk, bool verbose)
13721369 {
13731370 	struct block_device *bdev;
1374
-	int ret = 0;
13751371 
1376
-	if (disk->fops->revalidate_disk)
1377
-		ret = disk->fops->revalidate_disk(disk);
1372
+	/*
1373
+	 * Hidden disks don't have associated bdev so there's no point in
1374
+	 * revalidating them.
1375
+	 */
1376
+	if (disk->flags & GENHD_FL_HIDDEN)
1377
+		return;
1378
+
13781379 	bdev = bdget_disk(disk, 0);
1379
-	if (!bdev)
1380
-		return ret;
1381
-
1382
-	mutex_lock(&bdev->bd_mutex);
1383
-	check_disk_size_change(disk, bdev, ret == 0);
1384
-	bdev->bd_invalidated = 0;
1385
-	mutex_unlock(&bdev->bd_mutex);
1386
-	bdput(bdev);
1387
-	return ret;
1380
+	if (bdev) {
1381
+		check_disk_size_change(disk, bdev, verbose);
1382
+		bdput(bdev);
1383
+	}
13881384 }
1389
-EXPORT_SYMBOL(revalidate_disk);
1385
+EXPORT_SYMBOL(revalidate_disk_size);
13901386 
1391
-/*
1392
- * This routine checks whether a removable media has been changed,
1393
- * and invalidates all buffer-cache-entries in that case. This
1394
- * is a relatively slow routine, so we have to try to minimize using
1395
- * it. Thus it is called only upon a 'mount' or 'open'. This
1396
- * is the best way of combining speed and utility, I think.
1397
- * People changing diskettes in the middle of an operation deserve
1398
- * to lose :-)
1399
- */
1400
-int check_disk_change(struct block_device *bdev)
1387
+void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
14011388 {
1402
-	struct gendisk *disk = bdev->bd_disk;
1403
-	const struct block_device_operations *bdops = disk->fops;
1404
-	unsigned int events;
1405
-
1406
-	events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1407
-				   DISK_EVENT_EJECT_REQUEST);
1408
-	if (!(events & DISK_EVENT_MEDIA_CHANGE))
1409
-		return 0;
1410
-
1411
-	flush_disk(bdev, true);
1412
-	if (bdops->revalidate_disk)
1413
-		bdops->revalidate_disk(bdev->bd_disk);
1414
-	return 1;
1389
+	spin_lock(&bdev->bd_size_lock);
1390
+	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1391
+	spin_unlock(&bdev->bd_size_lock);
14151392 }
1416
-
1417
-EXPORT_SYMBOL(check_disk_change);
1418
-
1419
-void bd_set_size(struct block_device *bdev, loff_t size)
1420
-{
1421
-	inode_lock(bdev->bd_inode);
1422
-	i_size_write(bdev->bd_inode, size);
1423
-	inode_unlock(bdev->bd_inode);
1424
-}
1425
-EXPORT_SYMBOL(bd_set_size);
1393
+EXPORT_SYMBOL(bd_set_nr_sectors);
14261394 
14271395 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
14281396 
1429
-static void bdev_disk_changed(struct block_device *bdev, bool invalidate)
1397
+int bdev_disk_changed(struct block_device *bdev, bool invalidate)
14301398 {
1431
-	if (disk_part_scan_enabled(bdev->bd_disk)) {
1432
-		if (invalidate)
1433
-			invalidate_partitions(bdev->bd_disk, bdev);
1434
-		else
1435
-			rescan_partitions(bdev->bd_disk, bdev);
1399
+	struct gendisk *disk = bdev->bd_disk;
1400
+	int ret;
1401
+
1402
+	lockdep_assert_held(&bdev->bd_mutex);
1403
+
1404
+	if (!(disk->flags & GENHD_FL_UP))
1405
+		return -ENXIO;
1406
+
1407
+rescan:
1408
+	ret = blk_drop_partitions(bdev);
1409
+	if (ret)
1410
+		return ret;
1411
+
1412
+	clear_bit(GD_NEED_PART_SCAN, &disk->state);
1413
+
1414
+	/*
1415
+	 * Historically we only set the capacity to zero for devices that
1416
+	 * support partitions (independ of actually having partitions created).
1417
+	 * Doing that is rather inconsistent, but changing it broke legacy
1418
+	 * udisks polling for legacy ide-cdrom devices.  Use the crude check
1419
+	 * below to get the sane behavior for most device while not breaking
1420
+	 * userspace for this particular setup.
1421
+	 */
1422
+	if (invalidate) {
1423
+		if (disk_part_scan_enabled(disk) ||
1424
+		    !(disk->flags & GENHD_FL_REMOVABLE))
1425
+			set_capacity(disk, 0);
14361426 	} else {
1437
-		check_disk_size_change(bdev->bd_disk, bdev, !invalidate);
1438
-		bdev->bd_invalidated = 0;
1427
+		if (disk->fops->revalidate_disk)
1428
+			disk->fops->revalidate_disk(disk);
14391429 	}
1430
+
1431
+	check_disk_size_change(disk, bdev, !invalidate);
1432
+
1433
+	if (get_capacity(disk)) {
1434
+		ret = blk_add_partitions(disk, bdev);
1435
+		if (ret == -EAGAIN)
1436
+			goto rescan;
1437
+	} else if (invalidate) {
1438
+		/*
1439
+		 * Tell userspace that the media / partition table may have
1440
+		 * changed.
1441
+		 */
1442
+		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1443
+	}
1444
+
1445
+	return ret;
14401446 }
1447
+/*
1448
+ * Only exported for for loop and dasd for historic reasons.  Don't use in new
1449
+ * code!
1450
+ */
1451
+EXPORT_SYMBOL_GPL(bdev_disk_changed);
14411452 
14421453 /*
14431454  * bd_mutex locking:
....@@ -1446,40 +1457,46 @@
14461457  *    mutex_lock_nested(whole->bd_mutex, 1)
14471458  */
14481459 
1449
-static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1460
+static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1461
+		int for_part)
14501462 {
1463
+	struct block_device *whole = NULL, *claiming = NULL;
14511464 	struct gendisk *disk;
14521465 	int ret;
14531466 	int partno;
1454
-	int perm = 0;
1455
-	bool first_open = false;
1456
-
1457
-	if (mode & FMODE_READ)
1458
-		perm |= MAY_READ;
1459
-	if (mode & FMODE_WRITE)
1460
-		perm |= MAY_WRITE;
1461
-	/*
1462
-	 * hooks: /n/, see "layering violations".
1463
-	 */
1464
-	if (!for_part) {
1465
-		ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1466
-		if (ret != 0)
1467
-			return ret;
1468
-	}
1467
+	bool first_open = false, unblock_events = true, need_restart;
14691468 
14701469  restart:
1471
-
1470
+	need_restart = false;
14721471 	ret = -ENXIO;
14731472 	disk = bdev_get_gendisk(bdev, &partno);
14741473 	if (!disk)
14751474 		goto out;
1475
+
1476
+	if (partno) {
1477
+		whole = bdget_disk(disk, 0);
1478
+		if (!whole) {
1479
+			ret = -ENOMEM;
1480
+			goto out_put_disk;
1481
+		}
1482
+	}
1483
+
1484
+	if (!for_part && (mode & FMODE_EXCL)) {
1485
+		WARN_ON_ONCE(!holder);
1486
+		if (whole)
1487
+			claiming = whole;
1488
+		else
1489
+			claiming = bdev;
1490
+		ret = bd_prepare_to_claim(bdev, claiming, holder);
1491
+		if (ret)
1492
+			goto out_put_whole;
1493
+	}
14761494 
14771495 	disk_block_events(disk);
14781496 	mutex_lock_nested(&bdev->bd_mutex, for_part);
14791497 	if (!bdev->bd_openers) {
14801498 		first_open = true;
14811499 		bdev->bd_disk = disk;
1482
-		bdev->bd_queue = disk->queue;
14831500 		bdev->bd_contains = bdev;
14841501 		bdev->bd_partno = partno;
14851502 
....@@ -1492,24 +1509,16 @@
14921509 			ret = 0;
14931510 			if (disk->fops->open) {
14941511 				ret = disk->fops->open(bdev, mode);
1495
-				if (ret == -ERESTARTSYS) {
1496
-					/* Lost a race with 'disk' being
1497
-					 * deleted, try again.
1498
-					 * See md.c
1499
-					 */
1500
-					disk_put_part(bdev->bd_part);
1501
-					bdev->bd_part = NULL;
1502
-					bdev->bd_disk = NULL;
1503
-					bdev->bd_queue = NULL;
1504
-					mutex_unlock(&bdev->bd_mutex);
1505
-					disk_unblock_events(disk);
1506
-					put_disk_and_module(disk);
1507
-					goto restart;
1508
-				}
1512
+				/*
1513
+				 * If we lost a race with 'disk' being deleted,
1514
+				 * try again.  See md.c
1515
+				 */
1516
+				if (ret == -ERESTARTSYS)
1517
+					need_restart = true;
15091518 			}
15101519 
15111520 			if (!ret) {
1512
-				bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1521
+				bd_set_nr_sectors(bdev, get_capacity(disk));
15131522 				set_init_blocksize(bdev);
15141523 			}
15151524 
....@@ -1519,32 +1528,25 @@
15191528 			 * The latter is necessary to prevent ghost
15201529 			 * partitions on a removed medium.
15211530 			 */
1522
-			if (bdev->bd_invalidated &&
1531
+			if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
15231532 			    (!ret || ret == -ENOMEDIUM))
15241533 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
15251534 
15261535 			if (ret)
15271536 				goto out_clear;
15281537 		} else {
1529
-			struct block_device *whole;
1530
-			whole = bdget_disk(disk, 0);
1531
-			ret = -ENOMEM;
1532
-			if (!whole)
1533
-				goto out_clear;
15341538 			BUG_ON(for_part);
1535
-			ret = __blkdev_get(whole, mode, 1);
1536
-			if (ret) {
1537
-				bdput(whole);
1539
+			ret = __blkdev_get(whole, mode, NULL, 1);
1540
+			if (ret)
15381541 				goto out_clear;
1539
-			}
1540
-			bdev->bd_contains = whole;
1542
+			bdev->bd_contains = bdgrab(whole);
15411543 			bdev->bd_part = disk_get_part(disk, partno);
15421544 			if (!(disk->flags & GENHD_FL_UP) ||
15431545 			    !bdev->bd_part || !bdev->bd_part->nr_sects) {
15441546 				ret = -ENXIO;
15451547 				goto out_clear;
15461548 			}
1547
-			bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1549
+			bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
15481550 			set_init_blocksize(bdev);
15491551 		}
15501552 
....@@ -1556,7 +1558,7 @@
15561558 			if (bdev->bd_disk->fops->open)
15571559 				ret = bdev->bd_disk->fops->open(bdev, mode);
15581560 			/* the same as first opener case, read comment there */
1559
-			if (bdev->bd_invalidated &&
1561
+			if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
15601562 			    (!ret || ret == -ENOMEDIUM))
15611563 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
15621564 			if (ret)
....@@ -1566,27 +1568,52 @@
15661568 	bdev->bd_openers++;
15671569 	if (for_part)
15681570 		bdev->bd_part_count++;
1571
+	if (claiming)
1572
+		bd_finish_claiming(bdev, claiming, holder);
1573
+
1574
+	/*
1575
+	 * Block event polling for write claims if requested.  Any write holder
1576
+	 * makes the write_holder state stick until all are released.  This is
1577
+	 * good enough and tracking individual writeable reference is too
1578
+	 * fragile given the way @mode is used in blkdev_get/put().
1579
+	 */
1580
+	if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1581
+	    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1582
+		bdev->bd_write_holder = true;
1583
+		unblock_events = false;
1584
+	}
15691585 	mutex_unlock(&bdev->bd_mutex);
1570
-	disk_unblock_events(disk);
1586
+
1587
+	if (unblock_events)
1588
+		disk_unblock_events(disk);
1589
+
15711590 	/* only one opener holds refs to the module and disk */
15721591 	if (!first_open)
15731592 		put_disk_and_module(disk);
1593
+	if (whole)
1594
+		bdput(whole);
15741595 	return 0;
15751596 
15761597  out_clear:
15771598 	disk_put_part(bdev->bd_part);
15781599 	bdev->bd_disk = NULL;
15791600 	bdev->bd_part = NULL;
1580
-	bdev->bd_queue = NULL;
15811601 	if (bdev != bdev->bd_contains)
15821602 		__blkdev_put(bdev->bd_contains, mode, 1);
15831603 	bdev->bd_contains = NULL;
15841604  out_unlock_bdev:
1605
+	if (claiming)
1606
+		bd_abort_claiming(bdev, claiming, holder);
15851607 	mutex_unlock(&bdev->bd_mutex);
15861608 	disk_unblock_events(disk);
1609
+ out_put_whole:
1610
+ 	if (whole)
1611
+		bdput(whole);
1612
+ out_put_disk:
15871613 	put_disk_and_module(disk);
1614
+	if (need_restart)
1615
+		goto restart;
15881616  out:
1589
-
15901617 	return ret;
15911618 }
15921619 
....@@ -1609,74 +1636,27 @@
16091636  * RETURNS:
16101637  * 0 on success, -errno on failure.
16111638  */
1612
-int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1639
+static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
16131640 {
1614
-	struct block_device *whole = NULL;
1615
-	int res;
1641
+	int ret, perm = 0;
16161642 
1617
-	WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1643
+	if (mode & FMODE_READ)
1644
+		perm |= MAY_READ;
1645
+	if (mode & FMODE_WRITE)
1646
+		perm |= MAY_WRITE;
1647
+	ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1648
+	if (ret)
1649
+		goto bdput;
16181650 
1619
-	if ((mode & FMODE_EXCL) && holder) {
1620
-		whole = bd_start_claiming(bdev, holder);
1621
-		if (IS_ERR(whole)) {
1622
-			bdput(bdev);
1623
-			return PTR_ERR(whole);
1624
-		}
1625
-	}
1651
+	ret =__blkdev_get(bdev, mode, holder, 0);
1652
+	if (ret)
1653
+		goto bdput;
1654
+	return 0;
16261655 
1627
-	res = __blkdev_get(bdev, mode, 0);
1628
-
1629
-	if (whole) {
1630
-		struct gendisk *disk = whole->bd_disk;
1631
-
1632
-		/* finish claiming */
1633
-		mutex_lock(&bdev->bd_mutex);
1634
-		spin_lock(&bdev_lock);
1635
-
1636
-		if (!res) {
1637
-			BUG_ON(!bd_may_claim(bdev, whole, holder));
1638
-			/*
1639
-			 * Note that for a whole device bd_holders
1640
-			 * will be incremented twice, and bd_holder
1641
-			 * will be set to bd_may_claim before being
1642
-			 * set to holder
1643
-			 */
1644
-			whole->bd_holders++;
1645
-			whole->bd_holder = bd_may_claim;
1646
-			bdev->bd_holders++;
1647
-			bdev->bd_holder = holder;
1648
-		}
1649
-
1650
-		/* tell others that we're done */
1651
-		BUG_ON(whole->bd_claiming != holder);
1652
-		whole->bd_claiming = NULL;
1653
-		wake_up_bit(&whole->bd_claiming, 0);
1654
-
1655
-		spin_unlock(&bdev_lock);
1656
-
1657
-		/*
1658
-		 * Block event polling for write claims if requested.  Any
1659
-		 * write holder makes the write_holder state stick until
1660
-		 * all are released.  This is good enough and tracking
1661
-		 * individual writeable reference is too fragile given the
1662
-		 * way @mode is used in blkdev_get/put().
1663
-		 */
1664
-		if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1665
-		    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1666
-			bdev->bd_write_holder = true;
1667
-			disk_block_events(disk);
1668
-		}
1669
-
1670
-		mutex_unlock(&bdev->bd_mutex);
1671
-		bdput(whole);
1672
-	}
1673
-
1674
-	if (res)
1675
-		bdput(bdev);
1676
-
1677
-	return res;
1656
+bdput:
1657
+	bdput(bdev);
1658
+	return ret;
16781659 }
1679
-EXPORT_SYMBOL(blkdev_get);
16801660 
16811661 /**
16821662  * blkdev_get_by_path - open a block device by name
....@@ -1769,7 +1749,7 @@
17691749 	 */
17701750 	filp->f_flags |= O_LARGEFILE;
17711751 
1772
-	filp->f_mode |= FMODE_NOWAIT;
1752
+	filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
17731753 
17741754 	if (filp->f_flags & O_NDELAY)
17751755 		filp->f_mode |= FMODE_NDELAY;
....@@ -1925,7 +1905,7 @@
19251905 	if (bdev_read_only(I_BDEV(bd_inode)))
19261906 		return -EPERM;
19271907 
1928
-	if (IS_SWAPFILE(bd_inode))
1908
+	if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
19291909 		return -ETXTBSY;
19301910 
19311911 	if (!iov_iter_count(from))
....@@ -1999,13 +1979,14 @@
19991979 
20001980 static const struct address_space_operations def_blk_aops = {
20011981 	.readpage	= blkdev_readpage,
2002
-	.readpages	= blkdev_readpages,
1982
+	.readahead	= blkdev_readahead,
20031983 	.writepage	= blkdev_writepage,
20041984 	.write_begin	= blkdev_write_begin,
20051985 	.write_end	= blkdev_write_end,
20061986 	.writepages	= blkdev_writepages,
20071987 	.releasepage	= blkdev_releasepage,
20081988 	.direct_IO	= blkdev_direct_IO,
1989
+	.migratepage	= buffer_migrate_page_norefs,
20091990 	.is_dirty_writeback = buffer_check_dirty_writeback,
20101991 };
20111992 
....@@ -2017,7 +1998,6 @@
20171998 			     loff_t len)
20181999 {
20192000 	struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2020
-	struct address_space *mapping;
20212001 	loff_t end = start + len - 1;
20222002 	loff_t isize;
20232003 	int error;
....@@ -2045,8 +2025,9 @@
20452025 		return -EINVAL;
20462026 
20472027 	/* Invalidate the page cache, including dirty pages. */
2048
-	mapping = bdev->bd_inode->i_mapping;
2049
-	truncate_inode_pages_range(mapping, start, end);
2028
+	error = truncate_bdev_range(bdev, file->f_mode, start, end);
2029
+	if (error)
2030
+		return error;
20502031 
20512032 	switch (mode) {
20522033 	case FALLOC_FL_ZERO_RANGE:
....@@ -2073,7 +2054,7 @@
20732054 	 * the caller will be given -EBUSY.  The third argument is
20742055 	 * inclusive, so the rounding here is safe.
20752056 	 */
2076
-	return invalidate_inode_pages2_range(mapping,
2057
+	return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
20772058 					     start >> PAGE_SHIFT,
20782059 					     end >> PAGE_SHIFT);
20792060 }
....@@ -2084,6 +2065,7 @@
20842065 	.llseek		= block_llseek,
20852066 	.read_iter	= blkdev_read_iter,
20862067 	.write_iter	= blkdev_write_iter,
2068
+	.iopoll		= blkdev_iopoll,
20872069 	.mmap		= generic_file_mmap,
20882070 	.fsync		= blkdev_fsync,
20892071 	.unlocked_ioctl	= block_ioctl,
....@@ -2094,18 +2076,6 @@
20942076 	.splice_write	= iter_file_splice_write,
20952077 	.fallocate	= blkdev_fallocate,
20962078 };
2097
-
2098
-int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2099
-{
2100
-	int res;
2101
-	mm_segment_t old_fs = get_fs();
2102
-	set_fs(KERNEL_DS);
2103
-	res = blkdev_ioctl(bdev, 0, cmd, arg);
2104
-	set_fs(old_fs);
2105
-	return res;
2106
-}
2107
-
2108
-EXPORT_SYMBOL(ioctl_by_bdev);
21092079 
21102080 /**
21112081  * lookup_bdev  - lookup a struct block_device by name