Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / md / faulty.c
diff --git a/kernel/drivers/md/faulty.c b/kernel/drivers/md/faulty.c
new file mode 100644 (file)
index 0000000..1277eb2
--- /dev/null
@@ -0,0 +1,371 @@
+/*
+ * faulty.c : Multiple Devices driver for Linux
+ *
+ * Copyright (C) 2004 Neil Brown
+ *
+ * fautly-device-simulator personality for md
+ *
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * You should have received a copy of the GNU General Public License
+ * (for example /usr/src/linux/COPYING); if not, write to the Free
+ * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+
+/*
+ * The "faulty" personality causes some requests to fail.
+ *
+ * Possible failure modes are:
+ *   reads fail "randomly" but succeed on retry
+ *   writes fail "randomly" but succeed on retry
+ *   reads for some address fail and then persist until a write
+ *   reads for some address fail and then persist irrespective of write
+ *   writes for some address fail and persist
+ *   all writes fail
+ *
+ * Different modes can be active at a time, but only
+ * one can be set at array creation.  Others can be added later.
+ * A mode can be one-shot or recurrent with the recurrence being
+ * once in every N requests.
+ * The bottom 5 bits of the "layout" indicate the mode.  The
+ * remainder indicate a period, or 0 for one-shot.
+ *
+ * There is an implementation limit on the number of concurrently
+ * persisting-faulty blocks. When a new fault is requested that would
+ * exceed the limit, it is ignored.
+ * All current faults can be clear using a layout of "0".
+ *
+ * Requests are always sent to the device.  If they are to fail,
+ * we clone the bio and insert a new b_end_io into the chain.
+ */
+
+#define        WriteTransient  0
+#define        ReadTransient   1
+#define        WritePersistent 2
+#define        ReadPersistent  3
+#define        WriteAll        4 /* doesn't go to device */
+#define        ReadFixable     5
+#define        Modes   6
+
+#define        ClearErrors     31
+#define        ClearFaults     30
+
+#define AllPersist     100 /* internal use only */
+#define        NoPersist       101
+
+#define        ModeMask        0x1f
+#define        ModeShift       5
+
+#define MaxFault       50
+#include <linux/blkdev.h>
+#include <linux/module.h>
+#include <linux/raid/md_u.h>
+#include <linux/slab.h>
+#include "md.h"
+#include <linux/seq_file.h>
+
+
+static void faulty_fail(struct bio *bio, int error)
+{
+       struct bio *b = bio->bi_private;
+
+       b->bi_iter.bi_size = bio->bi_iter.bi_size;
+       b->bi_iter.bi_sector = bio->bi_iter.bi_sector;
+
+       bio_put(bio);
+
+       bio_io_error(b);
+}
+
+struct faulty_conf {
+       int period[Modes];
+       atomic_t counters[Modes];
+       sector_t faults[MaxFault];
+       int     modes[MaxFault];
+       int nfaults;
+       struct md_rdev *rdev;
+};
+
+static int check_mode(struct faulty_conf *conf, int mode)
+{
+       if (conf->period[mode] == 0 &&
+           atomic_read(&conf->counters[mode]) <= 0)
+               return 0; /* no failure, no decrement */
+
+
+       if (atomic_dec_and_test(&conf->counters[mode])) {
+               if (conf->period[mode])
+                       atomic_set(&conf->counters[mode], conf->period[mode]);
+               return 1;
+       }
+       return 0;
+}
+
+static int check_sector(struct faulty_conf *conf, sector_t start, sector_t end, int dir)
+{
+       /* If we find a ReadFixable sector, we fix it ... */
+       int i;
+       for (i=0; i<conf->nfaults; i++)
+               if (conf->faults[i] >= start &&
+                   conf->faults[i] < end) {
+                       /* found it ... */
+                       switch (conf->modes[i] * 2 + dir) {
+                       case WritePersistent*2+WRITE: return 1;
+                       case ReadPersistent*2+READ: return 1;
+                       case ReadFixable*2+READ: return 1;
+                       case ReadFixable*2+WRITE:
+                               conf->modes[i] = NoPersist;
+                               return 0;
+                       case AllPersist*2+READ:
+                       case AllPersist*2+WRITE: return 1;
+                       default:
+                               return 0;
+                       }
+               }
+       return 0;
+}
+
+static void add_sector(struct faulty_conf *conf, sector_t start, int mode)
+{
+       int i;
+       int n = conf->nfaults;
+       for (i=0; i<conf->nfaults; i++)
+               if (conf->faults[i] == start) {
+                       switch(mode) {
+                       case NoPersist: conf->modes[i] = mode; return;
+                       case WritePersistent:
+                               if (conf->modes[i] == ReadPersistent ||
+                                   conf->modes[i] == ReadFixable)
+                                       conf->modes[i] = AllPersist;
+                               else
+                                       conf->modes[i] = WritePersistent;
+                               return;
+                       case ReadPersistent:
+                               if (conf->modes[i] == WritePersistent)
+                                       conf->modes[i] = AllPersist;
+                               else
+                                       conf->modes[i] = ReadPersistent;
+                               return;
+                       case ReadFixable:
+                               if (conf->modes[i] == WritePersistent ||
+                                   conf->modes[i] == ReadPersistent)
+                                       conf->modes[i] = AllPersist;
+                               else
+                                       conf->modes[i] = ReadFixable;
+                               return;
+                       }
+               } else if (conf->modes[i] == NoPersist)
+                       n = i;
+
+       if (n >= MaxFault)
+               return;
+       conf->faults[n] = start;
+       conf->modes[n] = mode;
+       if (conf->nfaults == n)
+               conf->nfaults = n+1;
+}
+
+static void make_request(struct mddev *mddev, struct bio *bio)
+{
+       struct faulty_conf *conf = mddev->private;
+       int failit = 0;
+
+       if (bio_data_dir(bio) == WRITE) {
+               /* write request */
+               if (atomic_read(&conf->counters[WriteAll])) {
+                       /* special case - don't decrement, don't generic_make_request,
+                        * just fail immediately
+                        */
+                       bio_endio(bio, -EIO);
+                       return;
+               }
+
+               if (check_sector(conf, bio->bi_iter.bi_sector,
+                                bio_end_sector(bio), WRITE))
+                       failit = 1;
+               if (check_mode(conf, WritePersistent)) {
+                       add_sector(conf, bio->bi_iter.bi_sector,
+                                  WritePersistent);
+                       failit = 1;
+               }
+               if (check_mode(conf, WriteTransient))
+                       failit = 1;
+       } else {
+               /* read request */
+               if (check_sector(conf, bio->bi_iter.bi_sector,
+                                bio_end_sector(bio), READ))
+                       failit = 1;
+               if (check_mode(conf, ReadTransient))
+                       failit = 1;
+               if (check_mode(conf, ReadPersistent)) {
+                       add_sector(conf, bio->bi_iter.bi_sector,
+                                  ReadPersistent);
+                       failit = 1;
+               }
+               if (check_mode(conf, ReadFixable)) {
+                       add_sector(conf, bio->bi_iter.bi_sector,
+                                  ReadFixable);
+                       failit = 1;
+               }
+       }
+       if (failit) {
+               struct bio *b = bio_clone_mddev(bio, GFP_NOIO, mddev);
+
+               b->bi_bdev = conf->rdev->bdev;
+               b->bi_private = bio;
+               b->bi_end_io = faulty_fail;
+               bio = b;
+       } else
+               bio->bi_bdev = conf->rdev->bdev;
+
+       generic_make_request(bio);
+}
+
+static void status(struct seq_file *seq, struct mddev *mddev)
+{
+       struct faulty_conf *conf = mddev->private;
+       int n;
+
+       if ((n=atomic_read(&conf->counters[WriteTransient])) != 0)
+               seq_printf(seq, " WriteTransient=%d(%d)",
+                          n, conf->period[WriteTransient]);
+
+       if ((n=atomic_read(&conf->counters[ReadTransient])) != 0)
+               seq_printf(seq, " ReadTransient=%d(%d)",
+                          n, conf->period[ReadTransient]);
+
+       if ((n=atomic_read(&conf->counters[WritePersistent])) != 0)
+               seq_printf(seq, " WritePersistent=%d(%d)",
+                          n, conf->period[WritePersistent]);
+
+       if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0)
+               seq_printf(seq, " ReadPersistent=%d(%d)",
+                          n, conf->period[ReadPersistent]);
+
+
+       if ((n=atomic_read(&conf->counters[ReadFixable])) != 0)
+               seq_printf(seq, " ReadFixable=%d(%d)",
+                          n, conf->period[ReadFixable]);
+
+       if ((n=atomic_read(&conf->counters[WriteAll])) != 0)
+               seq_printf(seq, " WriteAll");
+
+       seq_printf(seq, " nfaults=%d", conf->nfaults);
+}
+
+
+static int reshape(struct mddev *mddev)
+{
+       int mode = mddev->new_layout & ModeMask;
+       int count = mddev->new_layout >> ModeShift;
+       struct faulty_conf *conf = mddev->private;
+
+       if (mddev->new_layout < 0)
+               return 0;
+
+       /* new layout */
+       if (mode == ClearFaults)
+               conf->nfaults = 0;
+       else if (mode == ClearErrors) {
+               int i;
+               for (i=0 ; i < Modes ; i++) {
+                       conf->period[i] = 0;
+                       atomic_set(&conf->counters[i], 0);
+               }
+       } else if (mode < Modes) {
+               conf->period[mode] = count;
+               if (!count) count++;
+               atomic_set(&conf->counters[mode], count);
+       } else
+               return -EINVAL;
+       mddev->new_layout = -1;
+       mddev->layout = -1; /* makes sure further changes come through */
+       return 0;
+}
+
+static sector_t faulty_size(struct mddev *mddev, sector_t sectors, int raid_disks)
+{
+       WARN_ONCE(raid_disks,
+                 "%s does not support generic reshape\n", __func__);
+
+       if (sectors == 0)
+               return mddev->dev_sectors;
+
+       return sectors;
+}
+
+static int run(struct mddev *mddev)
+{
+       struct md_rdev *rdev;
+       int i;
+       struct faulty_conf *conf;
+
+       if (md_check_no_bitmap(mddev))
+               return -EINVAL;
+
+       conf = kmalloc(sizeof(*conf), GFP_KERNEL);
+       if (!conf)
+               return -ENOMEM;
+
+       for (i=0; i<Modes; i++) {
+               atomic_set(&conf->counters[i], 0);
+               conf->period[i] = 0;
+       }
+       conf->nfaults = 0;
+
+       rdev_for_each(rdev, mddev) {
+               conf->rdev = rdev;
+               disk_stack_limits(mddev->gendisk, rdev->bdev,
+                                 rdev->data_offset << 9);
+       }
+
+       md_set_array_sectors(mddev, faulty_size(mddev, 0, 0));
+       mddev->private = conf;
+
+       reshape(mddev);
+
+       return 0;
+}
+
+static void faulty_free(struct mddev *mddev, void *priv)
+{
+       struct faulty_conf *conf = priv;
+
+       kfree(conf);
+}
+
+static struct md_personality faulty_personality =
+{
+       .name           = "faulty",
+       .level          = LEVEL_FAULTY,
+       .owner          = THIS_MODULE,
+       .make_request   = make_request,
+       .run            = run,
+       .free           = faulty_free,
+       .status         = status,
+       .check_reshape  = reshape,
+       .size           = faulty_size,
+};
+
+static int __init raid_init(void)
+{
+       return register_md_personality(&faulty_personality);
+}
+
+static void raid_exit(void)
+{
+       unregister_md_personality(&faulty_personality);
+}
+
+module_init(raid_init);
+module_exit(raid_exit);
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Fault injection personality for MD");
+MODULE_ALIAS("md-personality-10"); /* faulty */
+MODULE_ALIAS("md-faulty");
+MODULE_ALIAS("md-level--5");