blk-flush: fix invalid BUG_ON in blk_insert_flush
[linux-2.6.git] / 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/slab.h>
28
29 #include "blk.h"
30
31 static struct kmem_cache *integrity_cachep;
32
33 static const char *bi_unsupported_name = "unsupported";
34
35 /**
36  * blk_rq_count_integrity_sg - Count number of integrity scatterlist elements
37  * @q:          request queue
38  * @bio:        bio with integrity metadata attached
39  *
40  * Description: Returns the number of elements required in a
41  * scatterlist corresponding to the integrity metadata in a bio.
42  */
43 int blk_rq_count_integrity_sg(struct request_queue *q, struct bio *bio)
44 {
45         struct bio_vec *iv, *ivprv = NULL;
46         unsigned int segments = 0;
47         unsigned int seg_size = 0;
48         unsigned int i = 0;
49
50         bio_for_each_integrity_vec(iv, bio, i) {
51
52                 if (ivprv) {
53                         if (!BIOVEC_PHYS_MERGEABLE(ivprv, iv))
54                                 goto new_segment;
55
56                         if (!BIOVEC_SEG_BOUNDARY(q, ivprv, iv))
57                                 goto new_segment;
58
59                         if (seg_size + iv->bv_len > queue_max_segment_size(q))
60                                 goto new_segment;
61
62                         seg_size += iv->bv_len;
63                 } else {
64 new_segment:
65                         segments++;
66                         seg_size = iv->bv_len;
67                 }
68
69                 ivprv = iv;
70         }
71
72         return segments;
73 }
74 EXPORT_SYMBOL(blk_rq_count_integrity_sg);
75
76 /**
77  * blk_rq_map_integrity_sg - Map integrity metadata into a scatterlist
78  * @q:          request queue
79  * @bio:        bio with integrity metadata attached
80  * @sglist:     target scatterlist
81  *
82  * Description: Map the integrity vectors in request into a
83  * scatterlist.  The scatterlist must be big enough to hold all
84  * elements.  I.e. sized using blk_rq_count_integrity_sg().
85  */
86 int blk_rq_map_integrity_sg(struct request_queue *q, struct bio *bio,
87                             struct scatterlist *sglist)
88 {
89         struct bio_vec *iv, *ivprv = NULL;
90         struct scatterlist *sg = NULL;
91         unsigned int segments = 0;
92         unsigned int i = 0;
93
94         bio_for_each_integrity_vec(iv, bio, i) {
95
96                 if (ivprv) {
97                         if (!BIOVEC_PHYS_MERGEABLE(ivprv, iv))
98                                 goto new_segment;
99
100                         if (!BIOVEC_SEG_BOUNDARY(q, ivprv, iv))
101                                 goto new_segment;
102
103                         if (sg->length + iv->bv_len > queue_max_segment_size(q))
104                                 goto new_segment;
105
106                         sg->length += iv->bv_len;
107                 } else {
108 new_segment:
109                         if (!sg)
110                                 sg = sglist;
111                         else {
112                                 sg->page_link &= ~0x02;
113                                 sg = sg_next(sg);
114                         }
115
116                         sg_set_page(sg, iv->bv_page, iv->bv_len, iv->bv_offset);
117                         segments++;
118                 }
119
120                 ivprv = iv;
121         }
122
123         if (sg)
124                 sg_mark_end(sg);
125
126         return segments;
127 }
128 EXPORT_SYMBOL(blk_rq_map_integrity_sg);
129
130 /**
131  * blk_integrity_compare - Compare integrity profile of two disks
132  * @gd1:        Disk to compare
133  * @gd2:        Disk to compare
134  *
135  * Description: Meta-devices like DM and MD need to verify that all
136  * sub-devices use the same integrity format before advertising to
137  * upper layers that they can send/receive integrity metadata.  This
138  * function can be used to check whether two gendisk devices have
139  * compatible integrity formats.
140  */
141 int blk_integrity_compare(struct gendisk *gd1, struct gendisk *gd2)
142 {
143         struct blk_integrity *b1 = gd1->integrity;
144         struct blk_integrity *b2 = gd2->integrity;
145
146         if (!b1 && !b2)
147                 return 0;
148
149         if (!b1 || !b2)
150                 return -1;
151
152         if (b1->sector_size != b2->sector_size) {
153                 printk(KERN_ERR "%s: %s/%s sector sz %u != %u\n", __func__,
154                        gd1->disk_name, gd2->disk_name,
155                        b1->sector_size, b2->sector_size);
156                 return -1;
157         }
158
159         if (b1->tuple_size != b2->tuple_size) {
160                 printk(KERN_ERR "%s: %s/%s tuple sz %u != %u\n", __func__,
161                        gd1->disk_name, gd2->disk_name,
162                        b1->tuple_size, b2->tuple_size);
163                 return -1;
164         }
165
166         if (b1->tag_size && b2->tag_size && (b1->tag_size != b2->tag_size)) {
167                 printk(KERN_ERR "%s: %s/%s tag sz %u != %u\n", __func__,
168                        gd1->disk_name, gd2->disk_name,
169                        b1->tag_size, b2->tag_size);
170                 return -1;
171         }
172
173         if (strcmp(b1->name, b2->name)) {
174                 printk(KERN_ERR "%s: %s/%s type %s != %s\n", __func__,
175                        gd1->disk_name, gd2->disk_name,
176                        b1->name, b2->name);
177                 return -1;
178         }
179
180         return 0;
181 }
182 EXPORT_SYMBOL(blk_integrity_compare);
183
184 int blk_integrity_merge_rq(struct request_queue *q, struct request *req,
185                            struct request *next)
186 {
187         if (blk_integrity_rq(req) != blk_integrity_rq(next))
188                 return -1;
189
190         if (req->nr_integrity_segments + next->nr_integrity_segments >
191             q->limits.max_integrity_segments)
192                 return -1;
193
194         return 0;
195 }
196 EXPORT_SYMBOL(blk_integrity_merge_rq);
197
198 int blk_integrity_merge_bio(struct request_queue *q, struct request *req,
199                             struct bio *bio)
200 {
201         int nr_integrity_segs;
202         struct bio *next = bio->bi_next;
203
204         bio->bi_next = NULL;
205         nr_integrity_segs = blk_rq_count_integrity_sg(q, bio);
206         bio->bi_next = next;
207
208         if (req->nr_integrity_segments + nr_integrity_segs >
209             q->limits.max_integrity_segments)
210                 return -1;
211
212         req->nr_integrity_segments += nr_integrity_segs;
213
214         return 0;
215 }
216 EXPORT_SYMBOL(blk_integrity_merge_bio);
217
218 struct integrity_sysfs_entry {
219         struct attribute attr;
220         ssize_t (*show)(struct blk_integrity *, char *);
221         ssize_t (*store)(struct blk_integrity *, const char *, size_t);
222 };
223
224 static ssize_t integrity_attr_show(struct kobject *kobj, struct attribute *attr,
225                                    char *page)
226 {
227         struct blk_integrity *bi =
228                 container_of(kobj, struct blk_integrity, kobj);
229         struct integrity_sysfs_entry *entry =
230                 container_of(attr, struct integrity_sysfs_entry, attr);
231
232         return entry->show(bi, page);
233 }
234
235 static ssize_t integrity_attr_store(struct kobject *kobj,
236                                     struct attribute *attr, const char *page,
237                                     size_t count)
238 {
239         struct blk_integrity *bi =
240                 container_of(kobj, struct blk_integrity, kobj);
241         struct integrity_sysfs_entry *entry =
242                 container_of(attr, struct integrity_sysfs_entry, attr);
243         ssize_t ret = 0;
244
245         if (entry->store)
246                 ret = entry->store(bi, page, count);
247
248         return ret;
249 }
250
251 static ssize_t integrity_format_show(struct blk_integrity *bi, char *page)
252 {
253         if (bi != NULL && bi->name != NULL)
254                 return sprintf(page, "%s\n", bi->name);
255         else
256                 return sprintf(page, "none\n");
257 }
258
259 static ssize_t integrity_tag_size_show(struct blk_integrity *bi, char *page)
260 {
261         if (bi != NULL)
262                 return sprintf(page, "%u\n", bi->tag_size);
263         else
264                 return sprintf(page, "0\n");
265 }
266
267 static ssize_t integrity_read_store(struct blk_integrity *bi,
268                                     const char *page, size_t count)
269 {
270         char *p = (char *) page;
271         unsigned long val = simple_strtoul(p, &p, 10);
272
273         if (val)
274                 bi->flags |= INTEGRITY_FLAG_READ;
275         else
276                 bi->flags &= ~INTEGRITY_FLAG_READ;
277
278         return count;
279 }
280
281 static ssize_t integrity_read_show(struct blk_integrity *bi, char *page)
282 {
283         return sprintf(page, "%d\n", (bi->flags & INTEGRITY_FLAG_READ) != 0);
284 }
285
286 static ssize_t integrity_write_store(struct blk_integrity *bi,
287                                      const char *page, size_t count)
288 {
289         char *p = (char *) page;
290         unsigned long val = simple_strtoul(p, &p, 10);
291
292         if (val)
293                 bi->flags |= INTEGRITY_FLAG_WRITE;
294         else
295                 bi->flags &= ~INTEGRITY_FLAG_WRITE;
296
297         return count;
298 }
299
300 static ssize_t integrity_write_show(struct blk_integrity *bi, char *page)
301 {
302         return sprintf(page, "%d\n", (bi->flags & INTEGRITY_FLAG_WRITE) != 0);
303 }
304
305 static struct integrity_sysfs_entry integrity_format_entry = {
306         .attr = { .name = "format", .mode = S_IRUGO },
307         .show = integrity_format_show,
308 };
309
310 static struct integrity_sysfs_entry integrity_tag_size_entry = {
311         .attr = { .name = "tag_size", .mode = S_IRUGO },
312         .show = integrity_tag_size_show,
313 };
314
315 static struct integrity_sysfs_entry integrity_read_entry = {
316         .attr = { .name = "read_verify", .mode = S_IRUGO | S_IWUSR },
317         .show = integrity_read_show,
318         .store = integrity_read_store,
319 };
320
321 static struct integrity_sysfs_entry integrity_write_entry = {
322         .attr = { .name = "write_generate", .mode = S_IRUGO | S_IWUSR },
323         .show = integrity_write_show,
324         .store = integrity_write_store,
325 };
326
327 static struct attribute *integrity_attrs[] = {
328         &integrity_format_entry.attr,
329         &integrity_tag_size_entry.attr,
330         &integrity_read_entry.attr,
331         &integrity_write_entry.attr,
332         NULL,
333 };
334
335 static const struct sysfs_ops integrity_ops = {
336         .show   = &integrity_attr_show,
337         .store  = &integrity_attr_store,
338 };
339
340 static int __init blk_dev_integrity_init(void)
341 {
342         integrity_cachep = kmem_cache_create("blkdev_integrity",
343                                              sizeof(struct blk_integrity),
344                                              0, SLAB_PANIC, NULL);
345         return 0;
346 }
347 subsys_initcall(blk_dev_integrity_init);
348
349 static void blk_integrity_release(struct kobject *kobj)
350 {
351         struct blk_integrity *bi =
352                 container_of(kobj, struct blk_integrity, kobj);
353
354         kmem_cache_free(integrity_cachep, bi);
355 }
356
357 static struct kobj_type integrity_ktype = {
358         .default_attrs  = integrity_attrs,
359         .sysfs_ops      = &integrity_ops,
360         .release        = blk_integrity_release,
361 };
362
363 bool blk_integrity_is_initialized(struct gendisk *disk)
364 {
365         struct blk_integrity *bi = blk_get_integrity(disk);
366
367         return (bi && bi->name && strcmp(bi->name, bi_unsupported_name) != 0);
368 }
369 EXPORT_SYMBOL(blk_integrity_is_initialized);
370
371 /**
372  * blk_integrity_register - Register a gendisk as being integrity-capable
373  * @disk:       struct gendisk pointer to make integrity-aware
374  * @template:   optional integrity profile to register
375  *
376  * Description: When a device needs to advertise itself as being able
377  * to send/receive integrity metadata it must use this function to
378  * register the capability with the block layer.  The template is a
379  * blk_integrity struct with values appropriate for the underlying
380  * hardware.  If template is NULL the new profile is allocated but
381  * not filled out. See Documentation/block/data-integrity.txt.
382  */
383 int blk_integrity_register(struct gendisk *disk, struct blk_integrity *template)
384 {
385         struct blk_integrity *bi;
386
387         BUG_ON(disk == NULL);
388
389         if (disk->integrity == NULL) {
390                 bi = kmem_cache_alloc(integrity_cachep,
391                                       GFP_KERNEL | __GFP_ZERO);
392                 if (!bi)
393                         return -1;
394
395                 if (kobject_init_and_add(&bi->kobj, &integrity_ktype,
396                                          &disk_to_dev(disk)->kobj,
397                                          "%s", "integrity")) {
398                         kmem_cache_free(integrity_cachep, bi);
399                         return -1;
400                 }
401
402                 kobject_uevent(&bi->kobj, KOBJ_ADD);
403
404                 bi->flags |= INTEGRITY_FLAG_READ | INTEGRITY_FLAG_WRITE;
405                 bi->sector_size = queue_logical_block_size(disk->queue);
406                 disk->integrity = bi;
407         } else
408                 bi = disk->integrity;
409
410         /* Use the provided profile as template */
411         if (template != NULL) {
412                 bi->name = template->name;
413                 bi->generate_fn = template->generate_fn;
414                 bi->verify_fn = template->verify_fn;
415                 bi->tuple_size = template->tuple_size;
416                 bi->set_tag_fn = template->set_tag_fn;
417                 bi->get_tag_fn = template->get_tag_fn;
418                 bi->tag_size = template->tag_size;
419         } else
420                 bi->name = bi_unsupported_name;
421
422         return 0;
423 }
424 EXPORT_SYMBOL(blk_integrity_register);
425
426 /**
427  * blk_integrity_unregister - Remove block integrity profile
428  * @disk:       disk whose integrity profile to deallocate
429  *
430  * Description: This function frees all memory used by the block
431  * integrity profile.  To be called at device teardown.
432  */
433 void blk_integrity_unregister(struct gendisk *disk)
434 {
435         struct blk_integrity *bi;
436
437         if (!disk || !disk->integrity)
438                 return;
439
440         bi = disk->integrity;
441
442         kobject_uevent(&bi->kobj, KOBJ_REMOVE);
443         kobject_del(&bi->kobj);
444         kobject_put(&bi->kobj);
445         disk->integrity = NULL;
446 }
447 EXPORT_SYMBOL(blk_integrity_unregister);