Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
[linux-2.6] / drivers / md / faulty.c
1 /*
2  * faulty.c : Multiple Devices driver for Linux
3  *
4  * Copyright (C) 2004 Neil Brown
5  *
6  * fautly-device-simulator personality for md
7  *
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2, or (at your option)
12  * any later version.
13  *
14  * You should have received a copy of the GNU General Public License
15  * (for example /usr/src/linux/COPYING); if not, write to the Free
16  * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
17  */
18
19
20 /*
21  * The "faulty" personality causes some requests to fail.
22  *
23  * Possible failure modes are:
24  *   reads fail "randomly" but succeed on retry
25  *   writes fail "randomly" but succeed on retry
26  *   reads for some address fail and then persist until a write
27  *   reads for some address fail and then persist irrespective of write
28  *   writes for some address fail and persist
29  *   all writes fail
30  *
31  * Different modes can be active at a time, but only
32  * one can be set at array creation.  Others can be added later.
33  * A mode can be one-shot or recurrent with the recurrance being
34  * once in every N requests.
35  * The bottom 5 bits of the "layout" indicate the mode.  The
36  * remainder indicate a period, or 0 for one-shot.
37  *
38  * There is an implementation limit on the number of concurrently
39  * persisting-faulty blocks. When a new fault is requested that would
40  * exceed the limit, it is ignored.
41  * All current faults can be clear using a layout of "0".
42  *
43  * Requests are always sent to the device.  If they are to fail,
44  * we clone the bio and insert a new b_end_io into the chain.
45  */
46
47 #define WriteTransient  0
48 #define ReadTransient   1
49 #define WritePersistent 2
50 #define ReadPersistent  3
51 #define WriteAll        4 /* doesn't go to device */
52 #define ReadFixable     5
53 #define Modes   6
54
55 #define ClearErrors     31
56 #define ClearFaults     30
57
58 #define AllPersist      100 /* internal use only */
59 #define NoPersist       101
60
61 #define ModeMask        0x1f
62 #define ModeShift       5
63
64 #define MaxFault        50
65 #include <linux/raid/md.h>
66
67
68 static void faulty_fail(struct bio *bio, int error)
69 {
70         struct bio *b = bio->bi_private;
71
72         b->bi_size = bio->bi_size;
73         b->bi_sector = bio->bi_sector;
74
75         bio_put(bio);
76
77         bio_io_error(b);
78 }
79
80 typedef struct faulty_conf {
81         int period[Modes];
82         atomic_t counters[Modes];
83         sector_t faults[MaxFault];
84         int     modes[MaxFault];
85         int nfaults;
86         mdk_rdev_t *rdev;
87 } conf_t;
88
89 static int check_mode(conf_t *conf, int mode)
90 {
91         if (conf->period[mode] == 0 &&
92             atomic_read(&conf->counters[mode]) <= 0)
93                 return 0; /* no failure, no decrement */
94
95
96         if (atomic_dec_and_test(&conf->counters[mode])) {
97                 if (conf->period[mode])
98                         atomic_set(&conf->counters[mode], conf->period[mode]);
99                 return 1;
100         }
101         return 0;
102 }
103
104 static int check_sector(conf_t *conf, sector_t start, sector_t end, int dir)
105 {
106         /* If we find a ReadFixable sector, we fix it ... */
107         int i;
108         for (i=0; i<conf->nfaults; i++)
109                 if (conf->faults[i] >= start &&
110                     conf->faults[i] < end) {
111                         /* found it ... */
112                         switch (conf->modes[i] * 2 + dir) {
113                         case WritePersistent*2+WRITE: return 1;
114                         case ReadPersistent*2+READ: return 1;
115                         case ReadFixable*2+READ: return 1;
116                         case ReadFixable*2+WRITE:
117                                 conf->modes[i] = NoPersist;
118                                 return 0;
119                         case AllPersist*2+READ:
120                         case AllPersist*2+WRITE: return 1;
121                         default:
122                                 return 0;
123                         }
124                 }
125         return 0;
126 }
127
128 static void add_sector(conf_t *conf, sector_t start, int mode)
129 {
130         int i;
131         int n = conf->nfaults;
132         for (i=0; i<conf->nfaults; i++)
133                 if (conf->faults[i] == start) {
134                         switch(mode) {
135                         case NoPersist: conf->modes[i] = mode; return;
136                         case WritePersistent:
137                                 if (conf->modes[i] == ReadPersistent ||
138                                     conf->modes[i] == ReadFixable)
139                                         conf->modes[i] = AllPersist;
140                                 else
141                                         conf->modes[i] = WritePersistent;
142                                 return;
143                         case ReadPersistent:
144                                 if (conf->modes[i] == WritePersistent)
145                                         conf->modes[i] = AllPersist;
146                                 else
147                                         conf->modes[i] = ReadPersistent;
148                                 return;
149                         case ReadFixable:
150                                 if (conf->modes[i] == WritePersistent ||
151                                     conf->modes[i] == ReadPersistent)
152                                         conf->modes[i] = AllPersist;
153                                 else
154                                         conf->modes[i] = ReadFixable;
155                                 return;
156                         }
157                 } else if (conf->modes[i] == NoPersist)
158                         n = i;
159
160         if (n >= MaxFault)
161                 return;
162         conf->faults[n] = start;
163         conf->modes[n] = mode;
164         if (conf->nfaults == n)
165                 conf->nfaults = n+1;
166 }
167
168 static int make_request(struct request_queue *q, struct bio *bio)
169 {
170         mddev_t *mddev = q->queuedata;
171         conf_t *conf = (conf_t*)mddev->private;
172         int failit = 0;
173
174         if (bio_data_dir(bio) == WRITE) {
175                 /* write request */
176                 if (atomic_read(&conf->counters[WriteAll])) {
177                         /* special case - don't decrement, don't generic_make_request,
178                          * just fail immediately
179                          */
180                         bio_endio(bio, -EIO);
181                         return 0;
182                 }
183
184                 if (check_sector(conf, bio->bi_sector, bio->bi_sector+(bio->bi_size>>9),
185                                  WRITE))
186                         failit = 1;
187                 if (check_mode(conf, WritePersistent)) {
188                         add_sector(conf, bio->bi_sector, WritePersistent);
189                         failit = 1;
190                 }
191                 if (check_mode(conf, WriteTransient))
192                         failit = 1;
193         } else {
194                 /* read request */
195                 if (check_sector(conf, bio->bi_sector, bio->bi_sector + (bio->bi_size>>9),
196                                  READ))
197                         failit = 1;
198                 if (check_mode(conf, ReadTransient))
199                         failit = 1;
200                 if (check_mode(conf, ReadPersistent)) {
201                         add_sector(conf, bio->bi_sector, ReadPersistent);
202                         failit = 1;
203                 }
204                 if (check_mode(conf, ReadFixable)) {
205                         add_sector(conf, bio->bi_sector, ReadFixable);
206                         failit = 1;
207                 }
208         }
209         if (failit) {
210                 struct bio *b = bio_clone(bio, GFP_NOIO);
211                 b->bi_bdev = conf->rdev->bdev;
212                 b->bi_private = bio;
213                 b->bi_end_io = faulty_fail;
214                 generic_make_request(b);
215                 return 0;
216         } else {
217                 bio->bi_bdev = conf->rdev->bdev;
218                 return 1;
219         }
220 }
221
222 static void status(struct seq_file *seq, mddev_t *mddev)
223 {
224         conf_t *conf = (conf_t*)mddev->private;
225         int n;
226
227         if ((n=atomic_read(&conf->counters[WriteTransient])) != 0)
228                 seq_printf(seq, " WriteTransient=%d(%d)",
229                            n, conf->period[WriteTransient]);
230
231         if ((n=atomic_read(&conf->counters[ReadTransient])) != 0)
232                 seq_printf(seq, " ReadTransient=%d(%d)",
233                            n, conf->period[ReadTransient]);
234
235         if ((n=atomic_read(&conf->counters[WritePersistent])) != 0)
236                 seq_printf(seq, " WritePersistent=%d(%d)",
237                            n, conf->period[WritePersistent]);
238
239         if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0)
240                 seq_printf(seq, " ReadPersistent=%d(%d)",
241                            n, conf->period[ReadPersistent]);
242
243
244         if ((n=atomic_read(&conf->counters[ReadFixable])) != 0)
245                 seq_printf(seq, " ReadFixable=%d(%d)",
246                            n, conf->period[ReadFixable]);
247
248         if ((n=atomic_read(&conf->counters[WriteAll])) != 0)
249                 seq_printf(seq, " WriteAll");
250
251         seq_printf(seq, " nfaults=%d", conf->nfaults);
252 }
253
254
255 static int reconfig(mddev_t *mddev, int layout, int chunk_size)
256 {
257         int mode = layout & ModeMask;
258         int count = layout >> ModeShift;
259         conf_t *conf = mddev->private;
260
261         if (chunk_size != -1)
262                 return -EINVAL;
263
264         /* new layout */
265         if (mode == ClearFaults)
266                 conf->nfaults = 0;
267         else if (mode == ClearErrors) {
268                 int i;
269                 for (i=0 ; i < Modes ; i++) {
270                         conf->period[i] = 0;
271                         atomic_set(&conf->counters[i], 0);
272                 }
273         } else if (mode < Modes) {
274                 conf->period[mode] = count;
275                 if (!count) count++;
276                 atomic_set(&conf->counters[mode], count);
277         } else
278                 return -EINVAL;
279         mddev->layout = -1; /* makes sure further changes come through */
280         return 0;
281 }
282
283 static int run(mddev_t *mddev)
284 {
285         mdk_rdev_t *rdev;
286         struct list_head *tmp;
287         int i;
288
289         conf_t *conf = kmalloc(sizeof(*conf), GFP_KERNEL);
290
291         for (i=0; i<Modes; i++) {
292                 atomic_set(&conf->counters[i], 0);
293                 conf->period[i] = 0;
294         }
295         conf->nfaults = 0;
296
297         rdev_for_each(rdev, tmp, mddev)
298                 conf->rdev = rdev;
299
300         mddev->array_size = mddev->size;
301         mddev->private = conf;
302
303         reconfig(mddev, mddev->layout, -1);
304
305         return 0;
306 }
307
308 static int stop(mddev_t *mddev)
309 {
310         conf_t *conf = (conf_t *)mddev->private;
311
312         kfree(conf);
313         mddev->private = NULL;
314         return 0;
315 }
316
317 static struct mdk_personality faulty_personality =
318 {
319         .name           = "faulty",
320         .level          = LEVEL_FAULTY,
321         .owner          = THIS_MODULE,
322         .make_request   = make_request,
323         .run            = run,
324         .stop           = stop,
325         .status         = status,
326         .reconfig       = reconfig,
327 };
328
329 static int __init raid_init(void)
330 {
331         return register_md_personality(&faulty_personality);
332 }
333
334 static void raid_exit(void)
335 {
336         unregister_md_personality(&faulty_personality);
337 }
338
339 module_init(raid_init);
340 module_exit(raid_exit);
341 MODULE_LICENSE("GPL");
342 MODULE_ALIAS("md-personality-10"); /* faulty */
343 MODULE_ALIAS("md-faulty");
344 MODULE_ALIAS("md-level--5");