src/patches/suse-2.6.27.31/patches.fixes/block-suppress-buffer-IO-errors

   1 Subject: block: Supress Buffer I/O errors when SCSI REQ_QUIET flag set
   2 From: Keith Mannthey <kmannth@us.ibm.com>
   3 Date: Mon Dec 29 08:28:44 2008 +0100:
   4 Git: 08bafc0341f2f7920e9045bc32c40299cac8c21b
   5 References: bnc#464155
   6
   7 Allow the scsi request REQ_QUIET flag to be propagated to the buffer
   8 file system layer. The basic ideas is to pass the flag from the scsi
   9 request to the bio (block IO) and then to the buffer layer.  The buffer
  10 layer can then suppress needless printks.
  11
  12 This patch declutters the kernel log by removed the 40-50 (per lun)
  13 buffer io error messages seen during a boot in my multipath setup . It
  14 is a good chance any real errors will be missed in the "noise" it the
  15 logs without this patch.
  16
  17 During boot I see blocks of messages like
  18 "
  19 __ratelimit: 211 callbacks suppressed
  20 Buffer I/O error on device sdm, logical block 5242879
  21 Buffer I/O error on device sdm, logical block 5242879
  22 Buffer I/O error on device sdm, logical block 5242847
  23 Buffer I/O error on device sdm, logical block 1
  24 Buffer I/O error on device sdm, logical block 5242878
  25 Buffer I/O error on device sdm, logical block 5242879
  26 Buffer I/O error on device sdm, logical block 5242879
  27 Buffer I/O error on device sdm, logical block 5242879
  28 Buffer I/O error on device sdm, logical block 5242879
  29 Buffer I/O error on device sdm, logical block 5242872
  30 "
  31 in my logs.
  32
  33 My disk environment is multipath fiber channel using the SCSI_DH_RDAC
  34 code and multipathd.  This topology includes an "active" and "ghost"
  35 path for each lun. IO's to the "ghost" path will never complete and the
  36 SCSI layer, via the scsi device handler rdac code, quick returns the IOs
  37 to theses paths and sets the REQ_QUIET scsi flag to suppress the scsi
  38 layer messages.
  39
  40  I am wanting to extend the QUIET behavior to include the buffer file
  41 system layer to deal with these errors as well. I have been running this
  42 patch for a while now on several boxes without issue.  A few runs of
  43 bonnie++ show no noticeable difference in performance in my setup.
  44
  45 Thanks for John Stultz for the quiet_error finalization.
  46
  47 Submitted-by:  Keith Mannthey <kmannth@us.ibm.com>
  48 Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
  49 Signed-off-by: Hannes Reinecke <hare@suse.de>
  50
  51 ---
  52  block/blk-core.c            |    3 +++
  53  fs/buffer.c                 |   19 +++++++++++++++----
  54  include/linux/bio.h         |    1 +
  55  include/linux/buffer_head.h |    1 +
  56  4 files changed, 20 insertions(+), 4 deletions(-)
  57
  58 --- a/block/blk-core.c
  59 +++ b/block/blk-core.c
  60 @@ -138,6 +138,9 @@ static void req_bio_endio(struct request
  61                         nbytes = bio->bi_size;
  62                 }
  63
  64 +               if (unlikely(rq->cmd_flags & REQ_QUIET))
  65 +                       set_bit(BIO_QUIET, &bio->bi_flags);
  66 +
  67                 bio->bi_size -= nbytes;
  68                 bio->bi_sector += (nbytes >> 9);
  69
  70 --- a/fs/buffer.c
  71 +++ b/fs/buffer.c
  72 @@ -100,10 +100,18 @@ __clear_page_buffers(struct page *page)
  73         page_cache_release(page);
  74  }
  75
  76 +
  77 +static int quiet_error(struct buffer_head *bh)
  78 +{
  79 +       if (!test_bit(BH_Quiet, &bh->b_state) && printk_ratelimit())
  80 +               return 0;
  81 +       return 1;
  82 +}
  83 +
  84 +
  85  static void buffer_io_error(struct buffer_head *bh)
  86  {
  87         char b[BDEVNAME_SIZE];
  88 -
  89         printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
  90                         bdevname(bh->b_bdev, b),
  91                         (unsigned long long)bh->b_blocknr);
  92 @@ -145,7 +153,7 @@ void end_buffer_write_sync(struct buffer
  93         if (uptodate) {
  94                 set_buffer_uptodate(bh);
  95         } else {
  96 -               if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
  97 +               if (!buffer_eopnotsupp(bh) && !quiet_error(bh)) {
  98                         buffer_io_error(bh);
  99                         printk(KERN_WARNING "lost page write due to "
 100                                         "I/O error on %s\n",
 101 @@ -395,7 +403,7 @@ static void end_buffer_async_read(struct
 102                 set_buffer_uptodate(bh);
 103         } else {
 104                 clear_buffer_uptodate(bh);
 105 -               if (printk_ratelimit())
 106 +               if (!quiet_error(bh))
 107                         buffer_io_error(bh);
 108                 SetPageError(page);
 109         }
 110 @@ -456,7 +464,7 @@ static void end_buffer_async_write(struc
 111         if (uptodate) {
 112                 set_buffer_uptodate(bh);
 113         } else {
 114 -               if (printk_ratelimit()) {
 115 +               if (!quiet_error(bh)) {
 116                         buffer_io_error(bh);
 117                         printk(KERN_WARNING "lost page write due to "
 118                                         "I/O error on %s\n",
 119 @@ -2958,6 +2966,9 @@ static void end_bio_bh_io_sync(struct bi
 120                 set_bit(BH_Eopnotsupp, &bh->b_state);
 121         }
 122
 123 +       if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags)))
 124 +               set_bit(BH_Quiet, &bh->b_state);
 125 +
 126         bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags));
 127         bio_put(bio);
 128  }
 129 --- a/include/linux/bio.h
 130 +++ b/include/linux/bio.h
 131 @@ -116,6 +116,7 @@ struct bio {
 132  #define BIO_EOPNOTSUPP 7       /* not supported */
 133  #define BIO_CPU_AFFINE 8       /* complete bio on same CPU as submitted */
 134  #define BIO_FS_INTEGRITY 10    /* fs owns integrity data, not block layer */
 135 +#define BIO_QUIET      11      /* Make BIO Quiet */
 136  #define bio_flagged(bio, flag) ((bio)->bi_flags & (1 << (flag)))
 137
 138  /*
 139 --- a/include/linux/buffer_head.h
 140 +++ b/include/linux/buffer_head.h
 141 @@ -35,6 +35,7 @@ enum bh_state_bits {
 142         BH_Ordered,     /* ordered write */
 143         BH_Eopnotsupp,  /* operation not supported (barrier) */
 144         BH_Unwritten,   /* Buffer is allocated on disk but not written */
 145 +       BH_Quiet,       /* Buffer Error Prinks to be quiet */
 146
 147         BH_PrivateStart,/* not a state bit, but the first bit available
 148                          * for private allocation by other entities