drivers/md/dm-emc.c

   1 /*
   2  * Copyright (C) 2004 SUSE LINUX Products GmbH. All rights reserved.
   3  * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  *
   7  * Multipath support for EMC CLARiiON AX/CX-series hardware.
   8  */
   9
  10 #include "dm.h"
  11 #include "dm-hw-handler.h"
  12 #include <scsi/scsi.h>
  13 #include <scsi/scsi_cmnd.h>
  14
  15 #define DM_MSG_PREFIX "multipath emc"
  16
  17 struct emc_handler {
  18         spinlock_t lock;
  19
  20         /* Whether we should send the short trespass command (FC-series)
  21          * or the long version (default for AX/CX CLARiiON arrays). */
  22         unsigned short_trespass;
  23         /* Whether or not to honor SCSI reservations when initiating a
  24          * switch-over. Default: Don't. */
  25         unsigned hr;
  26
  27         unsigned char sense[SCSI_SENSE_BUFFERSIZE];
  28 };
  29
  30 #define TRESPASS_PAGE 0x22
  31 #define EMC_FAILOVER_TIMEOUT (60 * HZ)
  32
  33 /* Code borrowed from dm-lsi-rdac by Mike Christie */
  34
  35 static inline void free_bio(struct bio *bio)
  36 {
  37         __free_page(bio->bi_io_vec[0].bv_page);
  38         bio_put(bio);
  39 }
  40
  41 static int emc_endio(struct bio *bio, unsigned int bytes_done, int error)
  42 {
  43         struct dm_path *path = bio->bi_private;
  44
  45         if (bio->bi_size)
  46                 return 1;
  47
  48         /* We also need to look at the sense keys here whether or not to
  49          * switch to the next PG etc.
  50          *
  51          * For now simple logic: either it works or it doesn't.
  52          */
  53         if (error)
  54                 dm_pg_init_complete(path, MP_FAIL_PATH);
  55         else
  56                 dm_pg_init_complete(path, 0);
  57
  58         /* request is freed in block layer */
  59         free_bio(bio);
  60
  61         return 0;
  62 }
  63
  64 static struct bio *get_failover_bio(struct dm_path *path, unsigned data_size)
  65 {
  66         struct bio *bio;
  67         struct page *page;
  68
  69         bio = bio_alloc(GFP_ATOMIC, 1);
  70         if (!bio) {
  71                 DMERR("get_failover_bio: bio_alloc() failed.");
  72                 return NULL;
  73         }
  74
  75         bio->bi_rw |= (1 << BIO_RW);
  76         bio->bi_bdev = path->dev->bdev;
  77         bio->bi_sector = 0;
  78         bio->bi_private = path;
  79         bio->bi_end_io = emc_endio;
  80
  81         page = alloc_page(GFP_ATOMIC);
  82         if (!page) {
  83                 DMERR("get_failover_bio: alloc_page() failed.");
  84                 bio_put(bio);
  85                 return NULL;
  86         }
  87
  88         if (bio_add_page(bio, page, data_size, 0) != data_size) {
  89                 DMERR("get_failover_bio: alloc_page() failed.");
  90                 __free_page(page);
  91                 bio_put(bio);
  92                 return NULL;
  93         }
  94
  95         return bio;
  96 }
  97
  98 static struct request *get_failover_req(struct emc_handler *h,
  99                                         struct bio *bio, struct dm_path *path)
 100 {
 101         struct request *rq;
 102         struct block_device *bdev = bio->bi_bdev;
 103         struct request_queue *q = bdev_get_queue(bdev);
 104
 105         /* FIXME: Figure out why it fails with GFP_ATOMIC. */
 106         rq = blk_get_request(q, WRITE, __GFP_WAIT);
 107         if (!rq) {
 108                 DMERR("get_failover_req: blk_get_request failed");
 109                 return NULL;
 110         }
 111
 112         blk_rq_append_bio(q, rq, bio);
 113
 114         rq->sense = h->sense;
 115         memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE);
 116         rq->sense_len = 0;
 117
 118         memset(&rq->cmd, 0, BLK_MAX_CDB);
 119
 120         rq->timeout = EMC_FAILOVER_TIMEOUT;
 121         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 122         rq->cmd_flags |= REQ_FAILFAST | REQ_NOMERGE;
 123
 124         return rq;
 125 }
 126
 127 static struct request *emc_trespass_get(struct emc_handler *h,
 128                                         struct dm_path *path)
 129 {
 130         struct bio *bio;
 131         struct request *rq;
 132         unsigned char *page22;
 133         unsigned char long_trespass_pg[] = {
 134                 0, 0, 0, 0,
 135                 TRESPASS_PAGE,        /* Page code */
 136                 0x09,                 /* Page length - 2 */
 137                 h->hr ? 0x01 : 0x81,  /* Trespass code + Honor reservation bit */
 138                 0xff, 0xff,           /* Trespass target */
 139                 0, 0, 0, 0, 0, 0      /* Reserved bytes / unknown */
 140                 };
 141         unsigned char short_trespass_pg[] = {
 142                 0, 0, 0, 0,
 143                 TRESPASS_PAGE,        /* Page code */
 144                 0x02,                 /* Page length - 2 */
 145                 h->hr ? 0x01 : 0x81,  /* Trespass code + Honor reservation bit */
 146                 0xff,                 /* Trespass target */
 147                 };
 148         unsigned data_size = h->short_trespass ? sizeof(short_trespass_pg) :
 149                                 sizeof(long_trespass_pg);
 150
 151         /* get bio backing */
 152         if (data_size > PAGE_SIZE)
 153                 /* this should never happen */
 154                 return NULL;
 155
 156         bio = get_failover_bio(path, data_size);
 157         if (!bio) {
 158                 DMERR("emc_trespass_get: no bio");
 159                 return NULL;
 160         }
 161
 162         page22 = (unsigned char *)bio_data(bio);
 163         memset(page22, 0, data_size);
 164
 165         memcpy(page22, h->short_trespass ?
 166                 short_trespass_pg : long_trespass_pg, data_size);
 167
 168         /* get request for block layer packet command */
 169         rq = get_failover_req(h, bio, path);
 170         if (!rq) {
 171                 DMERR("emc_trespass_get: no rq");
 172                 free_bio(bio);
 173                 return NULL;
 174         }
 175
 176         /* Prepare the command. */
 177         rq->cmd[0] = MODE_SELECT;
 178         rq->cmd[1] = 0x10;
 179         rq->cmd[4] = data_size;
 180         rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
 181
 182         return rq;
 183 }
 184
 185 static void emc_pg_init(struct hw_handler *hwh, unsigned bypassed,
 186                         struct dm_path *path)
 187 {
 188         struct request *rq;
 189         struct request_queue *q = bdev_get_queue(path->dev->bdev);
 190
 191         /*
 192          * We can either blindly init the pg (then look at the sense),
 193          * or we can send some commands to get the state here (then
 194          * possibly send the fo cmnd), or we can also have the
 195          * initial state passed into us and then get an update here.
 196          */
 197         if (!q) {
 198                 DMINFO("emc_pg_init: no queue");
 199                 goto fail_path;
 200         }
 201
 202         /* FIXME: The request should be pre-allocated. */
 203         rq = emc_trespass_get(hwh->context, path);
 204         if (!rq) {
 205                 DMERR("emc_pg_init: no rq");
 206                 goto fail_path;
 207         }
 208
 209         DMINFO("emc_pg_init: sending switch-over command");
 210         elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 1);
 211         return;
 212
 213 fail_path:
 214         dm_pg_init_complete(path, MP_FAIL_PATH);
 215 }
 216
 217 static struct emc_handler *alloc_emc_handler(void)
 218 {
 219         struct emc_handler *h = kmalloc(sizeof(*h), GFP_KERNEL);
 220
 221         if (h) {
 222                 memset(h, 0, sizeof(*h));
 223                 spin_lock_init(&h->lock);
 224         }
 225
 226         return h;
 227 }
 228
 229 static int emc_create(struct hw_handler *hwh, unsigned argc, char **argv)
 230 {
 231         struct emc_handler *h;
 232         unsigned hr, short_trespass;
 233
 234         if (argc == 0) {
 235                 /* No arguments: use defaults */
 236                 hr = 0;
 237                 short_trespass = 0;
 238         } else if (argc != 2) {
 239                 DMWARN("incorrect number of arguments");
 240                 return -EINVAL;
 241         } else {
 242                 if ((sscanf(argv[0], "%u", &short_trespass) != 1)
 243                         || (short_trespass > 1)) {
 244                         DMWARN("invalid trespass mode selected");
 245                         return -EINVAL;
 246                 }
 247
 248                 if ((sscanf(argv[1], "%u", &hr) != 1)
 249                         || (hr > 1)) {
 250                         DMWARN("invalid honor reservation flag selected");
 251                         return -EINVAL;
 252                 }
 253         }
 254
 255         h = alloc_emc_handler();
 256         if (!h)
 257                 return -ENOMEM;
 258
 259         hwh->context = h;
 260
 261         if ((h->short_trespass = short_trespass))
 262                 DMWARN("short trespass command will be send");
 263         else
 264                 DMWARN("long trespass command will be send");
 265
 266         if ((h->hr = hr))
 267                 DMWARN("honor reservation bit will be set");
 268         else
 269                 DMWARN("honor reservation bit will not be set (default)");
 270
 271         return 0;
 272 }
 273
 274 static void emc_destroy(struct hw_handler *hwh)
 275 {
 276         struct emc_handler *h = (struct emc_handler *) hwh->context;
 277
 278         kfree(h);
 279         hwh->context = NULL;
 280 }
 281
 282 static unsigned emc_error(struct hw_handler *hwh, struct bio *bio)
 283 {
 284         /* FIXME: Patch from axboe still missing */
 285 #if 0
 286         int sense;
 287
 288         if (bio->bi_error & BIO_SENSE) {
 289                 sense = bio->bi_error & 0xffffff; /* sense key / asc / ascq */
 290
 291                 if (sense == 0x020403) {
 292                         /* LUN Not Ready - Manual Intervention Required
 293                          * indicates this is a passive path.
 294                          *
 295                          * FIXME: However, if this is seen and EVPD C0
 296                          * indicates that this is due to a NDU in
 297                          * progress, we should set FAIL_PATH too.
 298                          * This indicates we might have to do a SCSI
 299                          * inquiry in the end_io path. Ugh. */
 300                         return MP_BYPASS_PG | MP_RETRY_IO;
 301                 } else if (sense == 0x052501) {
 302                         /* An array based copy is in progress. Do not
 303                          * fail the path, do not bypass to another PG,
 304                          * do not retry. Fail the IO immediately.
 305                          * (Actually this is the same conclusion as in
 306                          * the default handler, but lets make sure.) */
 307                         return 0;
 308                 } else if (sense == 0x062900) {
 309                         /* Unit Attention Code. This is the first IO
 310                          * to the new path, so just retry. */
 311                         return MP_RETRY_IO;
 312                 }
 313         }
 314 #endif
 315
 316         /* Try default handler */
 317         return dm_scsi_err_handler(hwh, bio);
 318 }
 319
 320 static struct hw_handler_type emc_hwh = {
 321         .name = "emc",
 322         .module = THIS_MODULE,
 323         .create = emc_create,
 324         .destroy = emc_destroy,
 325         .pg_init = emc_pg_init,
 326         .error = emc_error,
 327 };
 328
 329 static int __init dm_emc_init(void)
 330 {
 331         int r = dm_register_hw_handler(&emc_hwh);
 332
 333         if (r < 0)
 334                 DMERR("register failed %d", r);
 335
 336         DMINFO("version 0.0.3 loaded");
 337
 338         return r;
 339 }
 340
 341 static void __exit dm_emc_exit(void)
 342 {
 343         int r = dm_unregister_hw_handler(&emc_hwh);
 344
 345         if (r < 0)
 346                 DMERR("unregister failed %d", r);
 347 }
 348
 349 module_init(dm_emc_init);
 350 module_exit(dm_emc_exit);
 351
 352 MODULE_DESCRIPTION(DM_NAME " EMC CX/AX/FC-family multipath");
 353 MODULE_AUTHOR("Lars Marowsky-Bree <lmb@suse.de>");
 354 MODULE_LICENSE("GPL");