On 9/3/18 4:05 PM, Mikulas Patocka wrote: > > > On Wed, 22 Aug 2018, Jens Axboe wrote: > >> On 8/22/18 10:23 AM, Mikulas Patocka wrote: >>> It is possible to call fsync on a read-only handle (for example, fsck.ext2 >>> does it when doing read-only check), and this call results in kernel >>> warning. This bug was introduced by the commit 721c7fc701c7 "block: fail >>> op_is_write() requests to read-only partitions". >> >> Similar patch is already queued up, it'll go into Linus's tree before >> -rc1. >> >> -- >> Jens Axboe > > I think you mean the patch b089cfd95d32638335c551651a8e00fd2c4edb0b, but > it doesn't work. > > The first problem is that bio_op returns the opcode, but op_is_flush > expects flags as its parameter. bio_op strips off the flags, and so the > condition op_is_flush is never true and the warning is not shut off. > > Anoher problem is that the flags REQ_FUA and REQ_PREFLUSH may be > superimposed on regular write bio and in that case the warning should be > triggered. We want to shut the warning off only if bio_sectors is zero > (i.e. if the flush request is not superimposed on regular write). Thanks for fixing that up, that was somewhat of a disaster. Applied. -- Jens Axboe
