On Mon, Jul 28, 2025 at 10:55:42AM -0700, Joanne Koong wrote: > On Mon, Jul 28, 2025 at 10:34 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote: > > > > On Fri, Jul 25, 2025 at 06:16:15PM -0700, Joanne Koong wrote: > > > > > > > Also, I just noticed that apparently the blocksize can change > > > > > > > dynamically for an inode in fuse through getattr replies from the > > > > > > > server (see fuse_change_attributes_common()). This is a problem since > > > > > > > the iomap uses inode->i_blkbits for reading/writing to the bitmap. I > > > > > > > think we will have to cache the inode blkbits in the iomap_folio_state > > > > > > > struct unfortunately :( I'll think about this some more and send out a > > > > > > > patch for this. > > > > Does this actually happen in practice, once you've started _using_ the > > block device? Rather than all this complicated stuff to invalidate the For most block device filesystems? No. And as far as I can tell, none of the filesystems actually support changing i_blkbits on the fly; I think only block devices can do that: $ git grep 'i_blkbits\s=\s' block/bdev.c:150: BD_INODE(bdev)->i_blkbits = blksize_bits(bsize); block/bdev.c:209: inode->i_blkbits = blksize_bits(size); fs/ceph/inode.c:81: inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT; fs/ceph/inode.c:1071: inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT; fs/ceph/inode.c:1076: inode->i_blkbits = CEPH_BLOCK_SHIFT; fs/ceph/inode.c:1180: inode->i_blkbits = PAGE_SHIFT; fs/direct-io.c:612: unsigned int i_blkbits = sdio->blkbits + sdio->blkfactor; fs/direct-io.c:908: const unsigned i_blkbits = blkbits + sdio->blkfactor; fs/direct-io.c:1110: unsigned i_blkbits = READ_ONCE(inode->i_blkbits); fs/erofs/fscache.c:527: inode->i_blkbits = EROFS_SB(sb)->blkszbits; fs/fuse/file_iomap.c:2327: inode->i_blkbits = new_blkbits; fs/fuse/inode.c:304: inode->i_blkbits = new_blkbits; fs/inode.c:234: inode->i_blkbits = sb->s_blocksize_bits; fs/libfs.c:1761: inode->i_blkbits = PAGE_SHIFT; fs/ocfs2/aops.c:2123: unsigned int i_blkbits = inode->i_sb->s_blocksize_bits; fs/orangefs/orangefs-utils.c:320: inode->i_blkbits = ffs(new_op->downcall.resp.getattr. fs/smb/client/cifsfs.c:411: cifs_inode->netfs.inode.i_blkbits = 14; /* 2**14 = CIFS_MAX_MSGSIZE */ fs/stack.c:72: dest->i_blkbits = src->i_blkbits; fs/vboxsf/utils.c:123: inode->i_blkbits = 12; > > page cache based on the fuse server telling us something, maybe just > > declare the server to be misbehaving and shut the whole filesystem down? > > > > I don't think this case is likely at all but I guess one scenario > where the server might want to change the block size midway through is > if they send the data to some network filesystem on the backend and if > that backend shuts down or is at full capacity for whatever reason and > they need to migrate to another backend that uses a different block > size then I guess this would be useful for that. Maybe, but it would still be pretty extraordinary to change the block size on an open file -- any program that tries to do its IO in blocks (i.e. not a byte stream) has already stat'd the file and will be very confused. > fuse currently does allow the block size to be changed dynamically so > I'm not sure if we can change that behavior without breaking backwards > compatibility. <nod> For fuse+iomap I'm not going to allow it initially if there's anything in the pagecache ... but I could be talked into it. --D
