Push BKL down into ->remount_fs()
[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/blkdev.h>
66 #include <linux/raid/md_u.h>
67 #include "md.h"
68 #include <linux/seq_file.h>
69
70
71 static void faulty_fail(struct bio *bio, int error)
72 {
73         struct bio *b = bio->bi_private;
74
75         b->bi_size = bio->bi_size;
76         b->bi_sector = bio->bi_sector;
77
78         bio_put(bio);
79
80         bio_io_error(b);
81 }
82
83 typedef struct faulty_conf {
84         int period[Modes];
85         atomic_t counters[Modes];
86         sector_t faults[MaxFault];
87         int     modes[MaxFault];
88         int nfaults;
89         mdk_rdev_t *rdev;
90 } conf_t;
91
92 static int check_mode(conf_t *conf, int mode)
93 {
94         if (conf->period[mode] == 0 &&
95             atomic_read(&conf->counters[mode]) <= 0)
96                 return 0; /* no failure, no decrement */
97
98
99         if (atomic_dec_and_test(&conf->counters[mode])) {
100                 if (conf->period[mode])
101                         atomic_set(&conf->counters[mode], conf->period[mode]);
102                 return 1;
103         }
104         return 0;
105 }
106
107 static int check_sector(conf_t *conf, sector_t start, sector_t end, int dir)
108 {
109         /* If we find a ReadFixable sector, we fix it ... */
110         int i;
111         for (i=0; i<conf->nfaults; i++)
112                 if (conf->faults[i] >= start &&
113                     conf->faults[i] < end) {
114                         /* found it ... */
115                         switch (conf->modes[i] * 2 + dir) {
116                         case WritePersistent*2+WRITE: return 1;
117                         case ReadPersistent*2+READ: return 1;
118                         case ReadFixable*2+READ: return 1;
119                         case ReadFixable*2+WRITE:
120                                 conf->modes[i] = NoPersist;
121                                 return 0;
122                         case AllPersist*2+READ:
123                         case AllPersist*2+WRITE: return 1;
124                         default:
125                                 return 0;
126                         }
127                 }
128         return 0;
129 }
130
131 static void add_sector(conf_t *conf, sector_t start, int mode)
132 {
133         int i;
134         int n = conf->nfaults;
135         for (i=0; i<conf->nfaults; i++)
136                 if (conf->faults[i] == start) {
137                         switch(mode) {
138                         case NoPersist: conf->modes[i] = mode; return;
139                         case WritePersistent:
140                                 if (conf->modes[i] == ReadPersistent ||
141                                     conf->modes[i] == ReadFixable)
142                                         conf->modes[i] = AllPersist;
143                                 else
144                                         conf->modes[i] = WritePersistent;
145                                 return;
146                         case ReadPersistent:
147                                 if (conf->modes[i] == WritePersistent)
148                                         conf->modes[i] = AllPersist;
149                                 else
150                                         conf->modes[i] = ReadPersistent;
151                                 return;
152                         case ReadFixable:
153                                 if (conf->modes[i] == WritePersistent ||
154                                     conf->modes[i] == ReadPersistent)
155                                         conf->modes[i] = AllPersist;
156                                 else
157                                         conf->modes[i] = ReadFixable;
158                                 return;
159                         }
160                 } else if (conf->modes[i] == NoPersist)
161                         n = i;
162
163         if (n >= MaxFault)
164                 return;
165         conf->faults[n] = start;
166         conf->modes[n] = mode;
167         if (conf->nfaults == n)
168                 conf->nfaults = n+1;
169 }
170
171 static int make_request(struct request_queue *q, struct bio *bio)
172 {
173         mddev_t *mddev = q->queuedata;
174         conf_t *conf = (conf_t*)mddev->private;
175         int failit = 0;
176
177         if (bio_data_dir(bio) == WRITE) {
178                 /* write request */
179                 if (atomic_read(&conf->counters[WriteAll])) {
180                         /* special case - don't decrement, don't generic_make_request,
181                          * just fail immediately
182                          */
183                         bio_endio(bio, -EIO);
184                         return 0;
185                 }
186
187                 if (check_sector(conf, bio->bi_sector, bio->bi_sector+(bio->bi_size>>9),
188                                  WRITE))
189                         failit = 1;
190                 if (check_mode(conf, WritePersistent)) {
191                         add_sector(conf, bio->bi_sector, WritePersistent);
192                         failit = 1;
193                 }
194                 if (check_mode(conf, WriteTransient))
195                         failit = 1;
196         } else {
197                 /* read request */
198                 if (check_sector(conf, bio->bi_sector, bio->bi_sector + (bio->bi_size>>9),
199                                  READ))
200                         failit = 1;
201                 if (check_mode(conf, ReadTransient))
202                         failit = 1;
203                 if (check_mode(conf, ReadPersistent)) {
204                         add_sector(conf, bio->bi_sector, ReadPersistent);
205                         failit = 1;
206                 }
207                 if (check_mode(conf, ReadFixable)) {
208                         add_sector(conf, bio->bi_sector, ReadFixable);
209                         failit = 1;
210                 }
211         }
212         if (failit) {
213                 struct bio *b = bio_clone(bio, GFP_NOIO);
214                 b->bi_bdev = conf->rdev->bdev;
215                 b->bi_private = bio;
216                 b->bi_end_io = faulty_fail;
217                 generic_make_request(b);
218                 return 0;
219         } else {
220                 bio->bi_bdev = conf->rdev->bdev;
221                 return 1;
222         }
223 }
224
225 static void status(struct seq_file *seq, mddev_t *mddev)
226 {
227         conf_t *conf = (conf_t*)mddev->private;
228         int n;
229
230         if ((n=atomic_read(&conf->counters[WriteTransient])) != 0)
231                 seq_printf(seq, " WriteTransient=%d(%d)",
232                            n, conf->period[WriteTransient]);
233
234         if ((n=atomic_read(&conf->counters[ReadTransient])) != 0)
235                 seq_printf(seq, " ReadTransient=%d(%d)",
236                            n, conf->period[ReadTransient]);
237
238         if ((n=atomic_read(&conf->counters[WritePersistent])) != 0)
239                 seq_printf(seq, " WritePersistent=%d(%d)",
240                            n, conf->period[WritePersistent]);
241
242         if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0)
243                 seq_printf(seq, " ReadPersistent=%d(%d)",
244                            n, conf->period[ReadPersistent]);
245
246
247         if ((n=atomic_read(&conf->counters[ReadFixable])) != 0)
248                 seq_printf(seq, " ReadFixable=%d(%d)",
249                            n, conf->period[ReadFixable]);
250
251         if ((n=atomic_read(&conf->counters[WriteAll])) != 0)
252                 seq_printf(seq, " WriteAll");
253
254         seq_printf(seq, " nfaults=%d", conf->nfaults);
255 }
256
257
258 static int reconfig(mddev_t *mddev, int layout, int chunk_size)
259 {
260         int mode = layout & ModeMask;
261         int count = layout >> ModeShift;
262         conf_t *conf = mddev->private;
263
264         if (chunk_size != -1)
265                 return -EINVAL;
266
267         /* new layout */
268         if (mode == ClearFaults)
269                 conf->nfaults = 0;
270         else if (mode == ClearErrors) {
271                 int i;
272                 for (i=0 ; i < Modes ; i++) {
273                         conf->period[i] = 0;
274                         atomic_set(&conf->counters[i], 0);
275                 }
276         } else if (mode < Modes) {
277                 conf->period[mode] = count;
278                 if (!count) count++;
279                 atomic_set(&conf->counters[mode], count);
280         } else
281                 return -EINVAL;
282         mddev->layout = -1; /* makes sure further changes come through */
283         return 0;
284 }
285
286 static sector_t faulty_size(mddev_t *mddev, sector_t sectors, int raid_disks)
287 {
288         WARN_ONCE(raid_disks,
289                   "%s does not support generic reshape\n", __func__);
290
291         if (sectors == 0)
292                 return mddev->dev_sectors;
293
294         return sectors;
295 }
296
297 static int run(mddev_t *mddev)
298 {
299         mdk_rdev_t *rdev;
300         int i;
301
302         conf_t *conf = kmalloc(sizeof(*conf), GFP_KERNEL);
303         if (!conf)
304                 return -ENOMEM;
305
306         for (i=0; i<Modes; i++) {
307                 atomic_set(&conf->counters[i], 0);
308                 conf->period[i] = 0;
309         }
310         conf->nfaults = 0;
311
312         list_for_each_entry(rdev, &mddev->disks, same_set)
313                 conf->rdev = rdev;
314
315         md_set_array_sectors(mddev, faulty_size(mddev, 0, 0));
316         mddev->private = conf;
317
318         reconfig(mddev, mddev->layout, -1);
319
320         return 0;
321 }
322
323 static int stop(mddev_t *mddev)
324 {
325         conf_t *conf = (conf_t *)mddev->private;
326
327         kfree(conf);
328         mddev->private = NULL;
329         return 0;
330 }
331
332 static struct mdk_personality faulty_personality =
333 {
334         .name           = "faulty",
335         .level          = LEVEL_FAULTY,
336         .owner          = THIS_MODULE,
337         .make_request   = make_request,
338         .run            = run,
339         .stop           = stop,
340         .status         = status,
341         .reconfig       = reconfig,
342         .size           = faulty_size,
343 };
344
345 static int __init raid_init(void)
346 {
347         return register_md_personality(&faulty_personality);
348 }
349
350 static void raid_exit(void)
351 {
352         unregister_md_personality(&faulty_personality);
353 }
354
355 module_init(raid_init);
356 module_exit(raid_exit);
357 MODULE_LICENSE("GPL");
358 MODULE_ALIAS("md-personality-10"); /* faulty */
359 MODULE_ALIAS("md-faulty");
360 MODULE_ALIAS("md-level--5");