projects / thirdparty / kernel / stable-queue.git / blame
| Commit | Line | Data |
|---|---|---|
| e2b52abc GKH |
1 | From a8733c7baf457b071528e385a0b7d4aaec79287c Mon Sep 17 00:00:00 2001 |
| 2 | From: James Bottomley <James.Bottomley@suse.de> | |
| 3 | Date: Fri, 17 Dec 2010 15:36:34 -0500 | |
| 4 | Subject: [SCSI] fix medium error problems with some arrays which can cause data corruption | |
| 5 | ||
| 6 | From: James Bottomley <James.Bottomley@suse.de> | |
| 7 | ||
| 8 | commit a8733c7baf457b071528e385a0b7d4aaec79287c upstream. | |
| 9 | ||
| 10 | Our current handling of medium error assumes that data is returned up | |
| 11 | to the bad sector. This assumption holds good for all disk devices, | |
| 12 | all DIF arrays and most ordinary arrays. However, an LSI array engine | |
| 13 | was recently discovered which reports a medium error without returning | |
| 14 | any data. This means that when we report good data up to the medium | |
| 15 | error, we've reported junk originally in the buffer as good. Worse, | |
| 16 | if the read consists of requested data plus a readahead, and the error | |
| 17 | occurs in readahead, we'll just strip off the readahead and report | |
| 18 | junk up to userspace as good data with no error. | |
| 19 | ||
| 20 | The fix for this is to have the error position computation take into | |
| 21 | account the amount of data returned by the driver using the scsi | |
| 22 | residual data. Unfortunately, not every driver fills in this data, | |
| 23 | but for those who don't, it's set to zero, which means we'll think a | |
| 24 | full set of data was transferred and the behaviour will be identical | |
| 25 | to the prior behaviour of the code (believe the buffer up to the error | |
| 26 | sector). All modern drivers seem to set the residual, so that should | |
| 27 | fix up the LSI failure/corruption case. | |
| 28 | ||
| 29 | Reported-by: Douglas Gilbert <dgilbert@interlog.com> | |
| 30 | Signed-off-by: James Bottomley <James.Bottomley@suse.de> | |
| 31 | Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> | |
| 32 | ||
| 33 | --- | |
| 34 | drivers/scsi/sd.c | 9 ++++++++- | |
| 35 | 1 file changed, 8 insertions(+), 1 deletion(-) | |
| 36 | ||
| 37 | --- a/drivers/scsi/sd.c | |
| 38 | +++ b/drivers/scsi/sd.c | |
| 39 | @@ -1153,6 +1153,12 @@ static unsigned int sd_completed_bytes(s | |
| 40 | u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512); | |
| 41 | u64 bad_lba; | |
| 42 | int info_valid; | |
| 43 | + /* | |
| 44 | + * resid is optional but mostly filled in. When it's unused, | |
| 45 | + * its value is zero, so we assume the whole buffer transferred | |
| 46 | + */ | |
| 47 | + unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd); | |
| 48 | + unsigned int good_bytes; | |
| 49 | ||
| 50 | if (scmd->request->cmd_type != REQ_TYPE_FS) | |
| 51 | return 0; | |
| 52 | @@ -1186,7 +1192,8 @@ static unsigned int sd_completed_bytes(s | |
| 53 | /* This computation should always be done in terms of | |
| 54 | * the resolution of the device's medium. | |
| 55 | */ | |
| 56 | - return (bad_lba - start_lba) * scmd->device->sector_size; | |
| 57 | + good_bytes = (bad_lba - start_lba) * scmd->device->sector_size; | |
| 58 | + return min(good_bytes, transferred); | |
| 59 | } | |
| 60 | ||
| 61 | /** |