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