block/blk-integrity.c

   1 /*
   2  * blk-integrity.c - Block layer data integrity extensions
   3  *
   4  * Copyright (C) 2007, 2008 Oracle Corporation
   5  * Written by: Martin K. Petersen <martin.petersen@oracle.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version
   9  * 2 as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; see the file COPYING.  If not, write to
  18  * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
  19  * USA.
  20  *
  21  */
  22
  23 #include <linux/blkdev.h>
  24 #include <linux/backing-dev.h>
  25 #include <linux/mempool.h>
  26 #include <linux/bio.h>
  27 #include <linux/scatterlist.h>
  28 #include <linux/export.h>
  29 #include <linux/slab.h>
  30
  31 #include "blk.h"
  32
  33 static struct kmem_cache *integrity_cachep;
  34
  35 static const char *bi_unsupported_name = "unsupported";
  36
  37 /**
  38  * blk_rq_count_integrity_sg - Count number of integrity scatterlist elements
  39  * @q:          request queue
  40  * @bio:        bio with integrity metadata attached
  41  *
  42  * Description: Returns the number of elements required in a
  43  * scatterlist corresponding to the integrity metadata in a bio.
  44  */
  45 int blk_rq_count_integrity_sg(struct request_queue *q, struct bio *bio)
  46 {
  47         struct bio_vec iv, ivprv = { NULL };
  48         unsigned int segments = 0;
  49         unsigned int seg_size = 0;
  50         struct bvec_iter iter;
  51         int prev = 0;
  52
  53         bio_for_each_integrity_vec(iv, bio, iter) {
  54
  55                 if (prev) {
  56                         if (!BIOVEC_PHYS_MERGEABLE(&ivprv, &iv))
  57                                 goto new_segment;
  58
  59                         if (!BIOVEC_SEG_BOUNDARY(q, &ivprv, &iv))
  60                                 goto new_segment;
  61
  62                         if (seg_size + iv.bv_len > queue_max_segment_size(q))
  63                                 goto new_segment;
  64
  65                         seg_size += iv.bv_len;
  66                 } else {
  67 new_segment:
  68                         segments++;
  69                         seg_size = iv.bv_len;
  70                 }
  71
  72                 prev = 1;
  73                 ivprv = iv;
  74         }
  75
  76         return segments;
  77 }
  78 EXPORT_SYMBOL(blk_rq_count_integrity_sg);
  79
  80 /**
  81  * blk_rq_map_integrity_sg - Map integrity metadata into a scatterlist
  82  * @q:          request queue
  83  * @bio:        bio with integrity metadata attached
  84  * @sglist:     target scatterlist
  85  *
  86  * Description: Map the integrity vectors in request into a
  87  * scatterlist.  The scatterlist must be big enough to hold all
  88  * elements.  I.e. sized using blk_rq_count_integrity_sg().
  89  */
  90 int blk_rq_map_integrity_sg(struct request_queue *q, struct bio *bio,
  91                             struct scatterlist *sglist)
  92 {
  93         struct bio_vec iv, ivprv = { NULL };
  94         struct scatterlist *sg = NULL;
  95         unsigned int segments = 0;
  96         struct bvec_iter iter;
  97         int prev = 0;
  98
  99         bio_for_each_integrity_vec(iv, bio, iter) {
 100
 101                 if (prev) {
 102                         if (!BIOVEC_PHYS_MERGEABLE(&ivprv, &iv))
 103                                 goto new_segment;
 104
 105                         if (!BIOVEC_SEG_BOUNDARY(q, &ivprv, &iv))
 106                                 goto new_segment;
 107
 108                         if (sg->length + iv.bv_len > queue_max_segment_size(q))
 109                                 goto new_segment;
 110
 111                         sg->length += iv.bv_len;
 112                 } else {
 113 new_segment:
 114                         if (!sg)
 115                                 sg = sglist;
 116                         else {
 117                                 sg_unmark_end(sg);
 118                                 sg = sg_next(sg);
 119                         }
 120
 121                         sg_set_page(sg, iv.bv_page, iv.bv_len, iv.bv_offset);
 122                         segments++;
 123                 }
 124
 125                 prev = 1;
 126                 ivprv = iv;
 127         }
 128
 129         if (sg)
 130                 sg_mark_end(sg);
 131
 132         return segments;
 133 }
 134 EXPORT_SYMBOL(blk_rq_map_integrity_sg);
 135
 136 /**
 137  * blk_integrity_compare - Compare integrity profile of two disks
 138  * @gd1:        Disk to compare
 139  * @gd2:        Disk to compare
 140  *
 141  * Description: Meta-devices like DM and MD need to verify that all
 142  * sub-devices use the same integrity format before advertising to
 143  * upper layers that they can send/receive integrity metadata.  This
 144  * function can be used to check whether two gendisk devices have
 145  * compatible integrity formats.
 146  */
 147 int blk_integrity_compare(struct gendisk *gd1, struct gendisk *gd2)
 148 {
 149         struct blk_integrity *b1 = gd1->integrity;
 150         struct blk_integrity *b2 = gd2->integrity;
 151
 152         if (!b1 && !b2)
 153                 return 0;
 154
 155         if (!b1 || !b2)
 156                 return -1;
 157
 158         if (b1->interval != b2->interval) {
 159                 pr_err("%s: %s/%s protection interval %u != %u\n",
 160                        __func__, gd1->disk_name, gd2->disk_name,
 161                        b1->interval, b2->interval);
 162                 return -1;
 163         }
 164
 165         if (b1->tuple_size != b2->tuple_size) {
 166                 printk(KERN_ERR "%s: %s/%s tuple sz %u != %u\n", __func__,
 167                        gd1->disk_name, gd2->disk_name,
 168                        b1->tuple_size, b2->tuple_size);
 169                 return -1;
 170         }
 171
 172         if (b1->tag_size && b2->tag_size && (b1->tag_size != b2->tag_size)) {
 173                 printk(KERN_ERR "%s: %s/%s tag sz %u != %u\n", __func__,
 174                        gd1->disk_name, gd2->disk_name,
 175                        b1->tag_size, b2->tag_size);
 176                 return -1;
 177         }
 178
 179         if (strcmp(b1->name, b2->name)) {
 180                 printk(KERN_ERR "%s: %s/%s type %s != %s\n", __func__,
 181                        gd1->disk_name, gd2->disk_name,
 182                        b1->name, b2->name);
 183                 return -1;
 184         }
 185
 186         return 0;
 187 }
 188 EXPORT_SYMBOL(blk_integrity_compare);
 189
 190 bool blk_integrity_merge_rq(struct request_queue *q, struct request *req,
 191                             struct request *next)
 192 {
 193         if (blk_integrity_rq(req) == 0 && blk_integrity_rq(next) == 0)
 194                 return true;
 195
 196         if (blk_integrity_rq(req) == 0 || blk_integrity_rq(next) == 0)
 197                 return false;
 198
 199         if (bio_integrity(req->bio)->bip_flags !=
 200             bio_integrity(next->bio)->bip_flags)
 201                 return false;
 202
 203         if (req->nr_integrity_segments + next->nr_integrity_segments >
 204             q->limits.max_integrity_segments)
 205                 return false;
 206
 207         return true;
 208 }
 209 EXPORT_SYMBOL(blk_integrity_merge_rq);
 210
 211 bool blk_integrity_merge_bio(struct request_queue *q, struct request *req,
 212                              struct bio *bio)
 213 {
 214         int nr_integrity_segs;
 215         struct bio *next = bio->bi_next;
 216
 217         if (blk_integrity_rq(req) == 0 && bio_integrity(bio) == NULL)
 218                 return true;
 219
 220         if (blk_integrity_rq(req) == 0 || bio_integrity(bio) == NULL)
 221                 return false;
 222
 223         if (bio_integrity(req->bio)->bip_flags != bio_integrity(bio)->bip_flags)
 224                 return false;
 225
 226         bio->bi_next = NULL;
 227         nr_integrity_segs = blk_rq_count_integrity_sg(q, bio);
 228         bio->bi_next = next;
 229
 230         if (req->nr_integrity_segments + nr_integrity_segs >
 231             q->limits.max_integrity_segments)
 232                 return false;
 233
 234         req->nr_integrity_segments += nr_integrity_segs;
 235
 236         return true;
 237 }
 238 EXPORT_SYMBOL(blk_integrity_merge_bio);
 239
 240 struct integrity_sysfs_entry {
 241         struct attribute attr;
 242         ssize_t (*show)(struct blk_integrity *, char *);
 243         ssize_t (*store)(struct blk_integrity *, const char *, size_t);
 244 };
 245
 246 static ssize_t integrity_attr_show(struct kobject *kobj, struct attribute *attr,
 247                                    char *page)
 248 {
 249         struct blk_integrity *bi =
 250                 container_of(kobj, struct blk_integrity, kobj);
 251         struct integrity_sysfs_entry *entry =
 252                 container_of(attr, struct integrity_sysfs_entry, attr);
 253
 254         return entry->show(bi, page);
 255 }
 256
 257 static ssize_t integrity_attr_store(struct kobject *kobj,
 258                                     struct attribute *attr, const char *page,
 259                                     size_t count)
 260 {
 261         struct blk_integrity *bi =
 262                 container_of(kobj, struct blk_integrity, kobj);
 263         struct integrity_sysfs_entry *entry =
 264                 container_of(attr, struct integrity_sysfs_entry, attr);
 265         ssize_t ret = 0;
 266
 267         if (entry->store)
 268                 ret = entry->store(bi, page, count);
 269
 270         return ret;
 271 }
 272
 273 static ssize_t integrity_format_show(struct blk_integrity *bi, char *page)
 274 {
 275         if (bi != NULL && bi->name != NULL)
 276                 return sprintf(page, "%s\n", bi->name);
 277         else
 278                 return sprintf(page, "none\n");
 279 }
 280
 281 static ssize_t integrity_tag_size_show(struct blk_integrity *bi, char *page)
 282 {
 283         if (bi != NULL)
 284                 return sprintf(page, "%u\n", bi->tag_size);
 285         else
 286                 return sprintf(page, "0\n");
 287 }
 288
 289 static ssize_t integrity_verify_store(struct blk_integrity *bi,
 290                                       const char *page, size_t count)
 291 {
 292         char *p = (char *) page;
 293         unsigned long val = simple_strtoul(p, &p, 10);
 294
 295         if (val)
 296                 bi->flags |= BLK_INTEGRITY_VERIFY;
 297         else
 298                 bi->flags &= ~BLK_INTEGRITY_VERIFY;
 299
 300         return count;
 301 }
 302
 303 static ssize_t integrity_verify_show(struct blk_integrity *bi, char *page)
 304 {
 305         return sprintf(page, "%d\n", (bi->flags & BLK_INTEGRITY_VERIFY) != 0);
 306 }
 307
 308 static ssize_t integrity_generate_store(struct blk_integrity *bi,
 309                                         const char *page, size_t count)
 310 {
 311         char *p = (char *) page;
 312         unsigned long val = simple_strtoul(p, &p, 10);
 313
 314         if (val)
 315                 bi->flags |= BLK_INTEGRITY_GENERATE;
 316         else
 317                 bi->flags &= ~BLK_INTEGRITY_GENERATE;
 318
 319         return count;
 320 }
 321
 322 static ssize_t integrity_generate_show(struct blk_integrity *bi, char *page)
 323 {
 324         return sprintf(page, "%d\n", (bi->flags & BLK_INTEGRITY_GENERATE) != 0);
 325 }
 326
 327 static ssize_t integrity_device_show(struct blk_integrity *bi, char *page)
 328 {
 329         return sprintf(page, "%u\n",
 330                        (bi->flags & BLK_INTEGRITY_DEVICE_CAPABLE) != 0);
 331 }
 332
 333 static struct integrity_sysfs_entry integrity_format_entry = {
 334         .attr = { .name = "format", .mode = S_IRUGO },
 335         .show = integrity_format_show,
 336 };
 337
 338 static struct integrity_sysfs_entry integrity_tag_size_entry = {
 339         .attr = { .name = "tag_size", .mode = S_IRUGO },
 340         .show = integrity_tag_size_show,
 341 };
 342
 343 static struct integrity_sysfs_entry integrity_verify_entry = {
 344         .attr = { .name = "read_verify", .mode = S_IRUGO | S_IWUSR },
 345         .show = integrity_verify_show,
 346         .store = integrity_verify_store,
 347 };
 348
 349 static struct integrity_sysfs_entry integrity_generate_entry = {
 350         .attr = { .name = "write_generate", .mode = S_IRUGO | S_IWUSR },
 351         .show = integrity_generate_show,
 352         .store = integrity_generate_store,
 353 };
 354
 355 static struct integrity_sysfs_entry integrity_device_entry = {
 356         .attr = { .name = "device_is_integrity_capable", .mode = S_IRUGO },
 357         .show = integrity_device_show,
 358 };
 359
 360 static struct attribute *integrity_attrs[] = {
 361         &integrity_format_entry.attr,
 362         &integrity_tag_size_entry.attr,
 363         &integrity_verify_entry.attr,
 364         &integrity_generate_entry.attr,
 365         &integrity_device_entry.attr,
 366         NULL,
 367 };
 368
 369 static const struct sysfs_ops integrity_ops = {
 370         .show   = &integrity_attr_show,
 371         .store  = &integrity_attr_store,
 372 };
 373
 374 static int __init blk_dev_integrity_init(void)
 375 {
 376         integrity_cachep = kmem_cache_create("blkdev_integrity",
 377                                              sizeof(struct blk_integrity),
 378                                              0, SLAB_PANIC, NULL);
 379         return 0;
 380 }
 381 subsys_initcall(blk_dev_integrity_init);
 382
 383 static void blk_integrity_release(struct kobject *kobj)
 384 {
 385         struct blk_integrity *bi =
 386                 container_of(kobj, struct blk_integrity, kobj);
 387
 388         kmem_cache_free(integrity_cachep, bi);
 389 }
 390
 391 static struct kobj_type integrity_ktype = {
 392         .default_attrs  = integrity_attrs,
 393         .sysfs_ops      = &integrity_ops,
 394         .release        = blk_integrity_release,
 395 };
 396
 397 bool blk_integrity_is_initialized(struct gendisk *disk)
 398 {
 399         struct blk_integrity *bi = blk_get_integrity(disk);
 400
 401         return (bi && bi->name && strcmp(bi->name, bi_unsupported_name) != 0);
 402 }
 403 EXPORT_SYMBOL(blk_integrity_is_initialized);
 404
 405 /**
 406  * blk_integrity_register - Register a gendisk as being integrity-capable
 407  * @disk:       struct gendisk pointer to make integrity-aware
 408  * @template:   optional integrity profile to register
 409  *
 410  * Description: When a device needs to advertise itself as being able
 411  * to send/receive integrity metadata it must use this function to
 412  * register the capability with the block layer.  The template is a
 413  * blk_integrity struct with values appropriate for the underlying
 414  * hardware.  If template is NULL the new profile is allocated but
 415  * not filled out. See Documentation/block/data-integrity.txt.
 416  */
 417 int blk_integrity_register(struct gendisk *disk, struct blk_integrity *template)
 418 {
 419         struct blk_integrity *bi;
 420
 421         BUG_ON(disk == NULL);
 422
 423         if (disk->integrity == NULL) {
 424                 bi = kmem_cache_alloc(integrity_cachep,
 425                                       GFP_KERNEL | __GFP_ZERO);
 426                 if (!bi)
 427                         return -1;
 428
 429                 if (kobject_init_and_add(&bi->kobj, &integrity_ktype,
 430                                          &disk_to_dev(disk)->kobj,
 431                                          "%s", "integrity")) {
 432                         kmem_cache_free(integrity_cachep, bi);
 433                         return -1;
 434                 }
 435
 436                 kobject_uevent(&bi->kobj, KOBJ_ADD);
 437
 438                 bi->flags |= BLK_INTEGRITY_VERIFY | BLK_INTEGRITY_GENERATE;
 439                 bi->interval = queue_logical_block_size(disk->queue);
 440                 disk->integrity = bi;
 441         } else
 442                 bi = disk->integrity;
 443
 444         /* Use the provided profile as template */
 445         if (template != NULL) {
 446                 bi->name = template->name;
 447                 bi->generate_fn = template->generate_fn;
 448                 bi->verify_fn = template->verify_fn;
 449                 bi->tuple_size = template->tuple_size;
 450                 bi->tag_size = template->tag_size;
 451                 bi->flags |= template->flags;
 452         } else
 453                 bi->name = bi_unsupported_name;
 454
 455         disk->queue->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES;
 456
 457         return 0;
 458 }
 459 EXPORT_SYMBOL(blk_integrity_register);
 460
 461 /**
 462  * blk_integrity_unregister - Remove block integrity profile
 463  * @disk:       disk whose integrity profile to deallocate
 464  *
 465  * Description: This function frees all memory used by the block
 466  * integrity profile.  To be called at device teardown.
 467  */
 468 void blk_integrity_unregister(struct gendisk *disk)
 469 {
 470         struct blk_integrity *bi;
 471
 472         if (!disk || !disk->integrity)
 473                 return;
 474
 475         disk->queue->backing_dev_info.capabilities &= ~BDI_CAP_STABLE_WRITES;
 476
 477         bi = disk->integrity;
 478
 479         kobject_uevent(&bi->kobj, KOBJ_REMOVE);
 480         kobject_del(&bi->kobj);
 481         kobject_put(&bi->kobj);
 482         disk->integrity = NULL;
 483 }
 484 EXPORT_SYMBOL(blk_integrity_unregister);