virtio_blk: set the zone write granularitygrafted

virtblk_read_zoned_limits() reads the write granularity that the device
reports in virtio_blk_zoned_characteristics and assigns it to the
physical block size and to io_min, but never to the limit that is named
after it. queue_limits.zone_write_granularity is left at zero, so
blk_validate_zoned_limits() raises it to the logical block size:

	if (lim->zone_write_granularity < lim->logical_block_size)
		lim->zone_write_granularity = lim->logical_block_size;

A device that reports a granularity coarser than its logical block size,
which is what the field exists to express, therefore has it silently
reduced. A 512e host managed disk passed through to a guest reports a
logical block size of 512 and a write granularity of 4096, and the guest
ends up with a zone write granularity of 512.

bio_split_alignment() returns lim->zone_write_granularity if it is non-zero
and bio_split_io_at() may split a bio with as per bio_split_alignment().
This can real to the write getting rejected by the host drive, as the write
is not aligned to the physical block size.

zonefs also takes its block size from bdev_zone_write_granularity(), so it
would incorrectly use 512 on a disk that requires 4096.

sd_zbc_read_zones() sets the limit from the physical block size for the
same reason. NVMe ZNS and null_blk leave it unset, but the fallback
gives the right answer for them, as their write granularity is the
logical block size. virtio carries a separate value that may exceed it.

Set the zone write granularity from the value that the device reports.

Fixes: 95bfec41bd3d ("virtio-blk: add support for zoned block devices")
Signed-off-by: Niklas Cassel <cassel@kernel.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Link: https://patch.msgid.link/20260918140641.2031075-2-cassel@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

1 files changed, 356 insertions, 0 deletions

diff --git a/fs/nilfs2/btnode.c b/fs/nilfs2/btnode.c
new file mode 100644
index 000000000..680e4009a
--- /dev/null
+++ b/fs/nilfs2/btnode.c
@@ -0,0 +1,356 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * NILFS B-tree node cache
+ *
+ * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
+ *
+ * Originally written by Seiji Kihara.
+ * Fully revised by Ryusuke Konishi for stabilization and simplification.
+ *
+ */
+
+#include <linux/types.h>
+#include <linux/buffer_head.h>
+#include <linux/mm.h>
+#include <linux/backing-dev.h>
+#include <linux/gfp.h>
+#include "nilfs.h"
+#include "mdt.h"
+#include "dat.h"
+#include "page.h"
+#include "btnode.h"
+
+
+/**
+ * nilfs_init_btnc_inode - initialize B-tree node cache inode
+ * @btnc_inode: inode to be initialized
+ *
+ * nilfs_init_btnc_inode() sets up an inode for B-tree node cache.
+ */
+void nilfs_init_btnc_inode(struct inode *btnc_inode)
+{
+	struct nilfs_inode_info *ii = NILFS_I(btnc_inode);
+
+	btnc_inode->i_mode = S_IFREG;
+	ii->i_flags = 0;
+	memset(&ii->i_bmap_data, 0, sizeof(struct nilfs_bmap));
+	mapping_set_gfp_mask(btnc_inode->i_mapping, GFP_NOFS);
+	btnc_inode->i_mapping->a_ops = &nilfs_buffer_cache_aops;
+}
+
+void nilfs_btnode_cache_clear(struct address_space *btnc)
+{
+	invalidate_mapping_pages(btnc, 0, -1);
+	truncate_inode_pages(btnc, 0);
+}
+
+struct buffer_head *
+nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
+{
+	struct inode *inode = btnc->host;
+	struct buffer_head *bh;
+
+	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
+	if (unlikely(!bh))
+		return ERR_PTR(-ENOMEM);
+
+	if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
+		     buffer_dirty(bh))) {
+		/*
+		 * The block buffer at the specified new address was already
+		 * in use.  This can happen if it is a virtual block number
+		 * and has been reallocated due to corruption of the bitmap
+		 * used to manage its allocation state (if not, the buffer
+		 * clearing of an abandoned b-tree node is missing somewhere).
+		 */
+		nilfs_error(inode->i_sb,
+			    "state inconsistency probably due to duplicate use of b-tree node block address %llu (ino=%llu)",
+			    (unsigned long long)blocknr, inode->i_ino);
+		goto failed;
+	}
+	memset(bh->b_data, 0, i_blocksize(inode));
+	bh->b_blocknr = blocknr;
+	set_buffer_mapped(bh);
+	set_buffer_uptodate(bh);
+
+	folio_unlock(bh->b_folio);
+	folio_put(bh->b_folio);
+	return bh;
+
+failed:
+	folio_unlock(bh->b_folio);
+	folio_put(bh->b_folio);
+	brelse(bh);
+	return ERR_PTR(-EIO);
+}
+
+int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
+			      sector_t pblocknr, blk_opf_t opf,
+			      struct buffer_head **pbh, sector_t *submit_ptr)
+{
+	struct buffer_head *bh;
+	struct inode *inode = btnc->host;
+	struct folio *folio;
+	int err;
+
+	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
+	if (unlikely(!bh))
+		return -ENOMEM;
+
+	err = -EEXIST; /* internal code */
+	folio = bh->b_folio;
+
+	if (buffer_uptodate(bh) || buffer_dirty(bh))
+		goto found;
+
+	if (pblocknr == 0) {
+		pblocknr = blocknr;
+		if (inode->i_ino != NILFS_DAT_INO) {
+			struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
+
+			/* blocknr is a virtual block number */
+			err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
+						  &pblocknr);
+			if (unlikely(err)) {
+				brelse(bh);
+				goto out_locked;
+			}
+		}
+	}
+
+	if (opf & REQ_RAHEAD) {
+		if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
+			err = -EBUSY; /* internal code */
+			brelse(bh);
+			goto out_locked;
+		}
+	} else { /* opf == REQ_OP_READ */
+		lock_buffer(bh);
+	}
+	if (buffer_uptodate(bh)) {
+		unlock_buffer(bh);
+		err = -EEXIST; /* internal code */
+		goto found;
+	}
+	set_buffer_mapped(bh);
+	bh->b_blocknr = pblocknr; /* set block address for read */
+	bh_submit(bh, opf, bh_end_read);
+	bh->b_blocknr = blocknr; /* set back to the given block address */
+	*submit_ptr = pblocknr;
+	err = 0;
+found:
+	*pbh = bh;
+
+out_locked:
+	folio_unlock(folio);
+	folio_put(folio);
+	return err;
+}
+
+/**
+ * nilfs_btnode_delete - delete B-tree node buffer
+ * @bh: buffer to be deleted
+ *
+ * nilfs_btnode_delete() invalidates the specified buffer and delete the page
+ * including the buffer if the page gets unbusy.
+ */
+void nilfs_btnode_delete(struct buffer_head *bh)
+{
+	struct address_space *mapping;
+	struct folio *folio = bh->b_folio;
+	pgoff_t index = folio->index;
+	int still_dirty;
+
+	folio_get(folio);
+	folio_lock(folio);
+	folio_wait_writeback(folio);
+
+	nilfs_forget_buffer(bh);
+	still_dirty = folio_test_dirty(folio);
+	mapping = folio->mapping;
+	folio_unlock(folio);
+	folio_put(folio);
+
+	if (!still_dirty && mapping)
+		invalidate_inode_pages2_range(mapping, index, index);
+}
+
+/**
+ * nilfs_btnode_prepare_change_key - prepare to change the search key of a
+ *                                   b-tree node block
+ * @btnc: page cache in which the b-tree node block is buffered
+ * @ctxt: structure for exchanging context information for key change
+ *
+ * nilfs_btnode_prepare_change_key() prepares to move the contents of the
+ * b-tree node block of the old key given in the "oldkey" member of @ctxt to
+ * the position of the new key given in the "newkey" member of @ctxt in the
+ * page cache @btnc.  Here, the key of the block is an index in units of
+ * blocks, and if the page and block sizes match, it matches the page index
+ * in the page cache.
+ *
+ * If the page size and block size match, this function attempts to move the
+ * entire folio, and in preparation for this, inserts the original folio into
+ * the new index of the cache.  If this insertion fails or if the page size
+ * and block size are different, it falls back to a copy preparation using
+ * nilfs_btnode_create_block(), inserts a new block at the position
+ * corresponding to "newkey", and stores the buffer head pointer in the
+ * "newbh" member of @ctxt.
+ *
+ * Note that the current implementation does not support folio sizes larger
+ * than the page size.
+ *
+ * Return: 0 on success, or one of the following negative error codes on
+ * failure:
+ * * %-EIO	- I/O error (metadata corruption).
+ * * %-ENOMEM	- Insufficient memory available.
+ */
+int nilfs_btnode_prepare_change_key(struct address_space *btnc,
+				    struct nilfs_btnode_chkey_ctxt *ctxt)
+{
+	struct buffer_head *obh, *nbh;
+	struct inode *inode = btnc->host;
+	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
+	int err;
+
+	if (oldkey == newkey)
+		return 0;
+
+	obh = ctxt->bh;
+	ctxt->newbh = NULL;
+
+	if (inode->i_blkbits == PAGE_SHIFT) {
+		struct folio *ofolio = obh->b_folio;
+		folio_lock(ofolio);
+retry:
+		/* BUG_ON(oldkey != obh->b_folio->index); */
+		if (unlikely(oldkey != ofolio->index))
+			NILFS_FOLIO_BUG(ofolio,
+				       "invalid oldkey %lld (newkey=%lld)",
+				       (unsigned long long)oldkey,
+				       (unsigned long long)newkey);
+
+		xa_lock_irq(&btnc->i_pages);
+		err = __xa_insert(&btnc->i_pages, newkey, ofolio, GFP_NOFS);
+		xa_unlock_irq(&btnc->i_pages);
+		/*
+		 * Note: folio->index will not change to newkey until
+		 * nilfs_btnode_commit_change_key() will be called.
+		 * To protect the folio in intermediate state, the folio lock
+		 * is held.
+		 */
+		if (!err)
+			return 0;
+		else if (err != -EBUSY)
+			goto failed_unlock;
+
+		err = invalidate_inode_pages2_range(btnc, newkey, newkey);
+		if (!err)
+			goto retry;
+		/* fallback to copy mode */
+		folio_unlock(ofolio);
+	}
+
+	nbh = nilfs_btnode_create_block(btnc, newkey);
+	if (IS_ERR(nbh))
+		return PTR_ERR(nbh);
+
+	BUG_ON(nbh == obh);
+	ctxt->newbh = nbh;
+	return 0;
+
+ failed_unlock:
+	folio_unlock(obh->b_folio);
+	return err;
+}
+
+/**
+ * nilfs_btnode_commit_change_key - commit the change of the search key of
+ *                                  a b-tree node block
+ * @btnc: page cache in which the b-tree node block is buffered
+ * @ctxt: structure for exchanging context information for key change
+ *
+ * nilfs_btnode_commit_change_key() executes the key change based on the
+ * context @ctxt prepared by nilfs_btnode_prepare_change_key().  If no valid
+ * block buffer is prepared in "newbh" of @ctxt (i.e., a full folio move),
+ * this function removes the folio from the old index and completes the move.
+ * Otherwise, it copies the block data and inherited flag states of "oldbh"
+ * to "newbh" and clears the "oldbh" from the cache.  In either case, the
+ * relocated buffer is marked as dirty.
+ *
+ * As with nilfs_btnode_prepare_change_key(), the current implementation does
+ * not support folio sizes larger than the page size.
+ */
+void nilfs_btnode_commit_change_key(struct address_space *btnc,
+				    struct nilfs_btnode_chkey_ctxt *ctxt)
+{
+	struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
+	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
+	struct folio *ofolio;
+
+	if (oldkey == newkey)
+		return;
+
+	if (nbh == NULL) {	/* blocksize == pagesize */
+		ofolio = obh->b_folio;
+		if (unlikely(oldkey != ofolio->index))
+			NILFS_FOLIO_BUG(ofolio,
+				       "invalid oldkey %lld (newkey=%lld)",
+				       (unsigned long long)oldkey,
+				       (unsigned long long)newkey);
+		mark_buffer_dirty(obh);
+
+		xa_lock_irq(&btnc->i_pages);
+		__xa_erase(&btnc->i_pages, oldkey);
+		__xa_set_mark(&btnc->i_pages, newkey, PAGECACHE_TAG_DIRTY);
+		xa_unlock_irq(&btnc->i_pages);
+
+		ofolio->index = obh->b_blocknr = newkey;
+		folio_unlock(ofolio);
+	} else {
+		nilfs_copy_buffer(nbh, obh);
+		mark_buffer_dirty(nbh);
+
+		nbh->b_blocknr = newkey;
+		ctxt->bh = nbh;
+		nilfs_btnode_delete(obh); /* will decrement bh->b_count */
+	}
+}
+
+/**
+ * nilfs_btnode_abort_change_key - abort the change of the search key of a
+ *                                 b-tree node block
+ * @btnc: page cache in which the b-tree node block is buffered
+ * @ctxt: structure for exchanging context information for key change
+ *
+ * nilfs_btnode_abort_change_key() cancels the key change associated with the
+ * context @ctxt prepared via nilfs_btnode_prepare_change_key() and performs
+ * any necessary cleanup.  If no valid block buffer is prepared in "newbh" of
+ * @ctxt, this function removes the folio from the destination index and aborts
+ * the move.  Otherwise, it clears "newbh" from the cache.
+ *
+ * As with nilfs_btnode_prepare_change_key(), the current implementation does
+ * not support folio sizes larger than the page size.
+ */
+void nilfs_btnode_abort_change_key(struct address_space *btnc,
+				   struct nilfs_btnode_chkey_ctxt *ctxt)
+{
+	struct buffer_head *nbh = ctxt->newbh;
+	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
+
+	if (oldkey == newkey)
+		return;
+
+	if (nbh == NULL) {	/* blocksize == pagesize */
+		xa_erase_irq(&btnc->i_pages, newkey);
+		folio_unlock(ctxt->bh->b_folio);
+	} else {
+		/*
+		 * When canceling a buffer that a prepare operation has
+		 * allocated to copy a node block to another location, use
+		 * nilfs_btnode_delete() to initialize and release the buffer
+		 * so that the buffer flags will not be in an inconsistent
+		 * state when it is reallocated.
+		 */
+		nilfs_btnode_delete(nbh);
+	}
+}