blob: a82d2fe80fb544da5afb86d835e6fbafb3a74533 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
3 *
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
8 */
9
10#include <linux/bio.h>
James Bottomleyd3f46f32008-01-15 11:11:46 -060011#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/blkdev.h>
13#include <linux/completion.h>
14#include <linux/kernel.h>
15#include <linux/mempool.h>
16#include <linux/slab.h>
17#include <linux/init.h>
18#include <linux/pci.h>
19#include <linux/delay.h>
James Bottomleyfaead262006-02-14 10:42:07 -060020#include <linux/hardirq.h>
Jens Axboec6132da2007-10-16 11:08:49 +020021#include <linux/scatterlist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020024#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <scsi/scsi_dbg.h>
26#include <scsi/scsi_device.h>
27#include <scsi/scsi_driver.h>
28#include <scsi/scsi_eh.h>
29#include <scsi/scsi_host.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31#include "scsi_priv.h"
32#include "scsi_logging.h"
33
34
Tobias Klauser6391a112006-06-08 22:23:48 -070035#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
Jens Axboe59725112007-04-02 10:06:42 +020036#define SG_MEMPOOL_SIZE 2
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38struct scsi_host_sg_pool {
39 size_t size;
Jens Axboea8474ce2007-08-07 09:02:51 +020040 char *name;
Christoph Lametere18b8902006-12-06 20:33:20 -080041 struct kmem_cache *slab;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 mempool_t *pool;
43};
44
James Bottomleyd3f46f32008-01-15 11:11:46 -060045#define SP(x) { x, "sgpool-" __stringify(x) }
46#if (SCSI_MAX_SG_SEGMENTS < 32)
47#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
48#endif
Adrian Bunk52c1da32005-06-23 22:05:33 -070049static struct scsi_host_sg_pool scsi_sg_pools[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 SP(8),
51 SP(16),
FUJITA Tomonorifd820f42007-09-18 12:14:37 +020052#if (SCSI_MAX_SG_SEGMENTS > 32)
James Bottomleyd3f46f32008-01-15 11:11:46 -060053 SP(32),
FUJITA Tomonorifd820f42007-09-18 12:14:37 +020054#if (SCSI_MAX_SG_SEGMENTS > 64)
James Bottomleyd3f46f32008-01-15 11:11:46 -060055 SP(64),
56#if (SCSI_MAX_SG_SEGMENTS > 128)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 SP(128),
James Bottomleyd3f46f32008-01-15 11:11:46 -060058#if (SCSI_MAX_SG_SEGMENTS > 256)
59#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
FUJITA Tomonorifd820f42007-09-18 12:14:37 +020060#endif
61#endif
62#endif
James Bottomleyd3f46f32008-01-15 11:11:46 -060063#endif
64 SP(SCSI_MAX_SG_SEGMENTS)
Jens Axboea8474ce2007-08-07 09:02:51 +020065};
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#undef SP
67
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +020068static struct kmem_cache *scsi_bidi_sdb_cache;
69
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -050070static void scsi_run_queue(struct request_queue *q);
James Bottomleye91442b2005-09-09 10:44:16 -050071
72/*
73 * Function: scsi_unprep_request()
74 *
75 * Purpose: Remove all preparation done for a request, including its
76 * associated scsi_cmnd, so that it can be requeued.
77 *
78 * Arguments: req - request to unprepare
79 *
80 * Lock status: Assumed that no locks are held upon entry.
81 *
82 * Returns: Nothing.
83 */
84static void scsi_unprep_request(struct request *req)
85{
86 struct scsi_cmnd *cmd = req->special;
87
Jens Axboe4aff5e22006-08-10 08:44:47 +020088 req->cmd_flags &= ~REQ_DONTPREP;
Christoph Hellwigbeb40482006-06-10 18:01:03 +020089 req->special = NULL;
James Bottomleye91442b2005-09-09 10:44:16 -050090
James Bottomleye91442b2005-09-09 10:44:16 -050091 scsi_put_command(cmd);
92}
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -050093
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/*
95 * Function: scsi_queue_insert()
96 *
97 * Purpose: Insert a command in the midlevel queue.
98 *
99 * Arguments: cmd - command that we are adding to queue.
100 * reason - why we are inserting command to queue.
101 *
102 * Lock status: Assumed that lock is not held upon entry.
103 *
104 * Returns: Nothing.
105 *
106 * Notes: We do this for one of two cases. Either the host is busy
107 * and it cannot accept any more commands for the time being,
108 * or the device returned QUEUE_FULL and can accept no more
109 * commands.
110 * Notes: This could be called either from an interrupt context or a
111 * normal process context.
112 */
113int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
114{
115 struct Scsi_Host *host = cmd->device->host;
116 struct scsi_device *device = cmd->device;
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500117 struct request_queue *q = device->request_queue;
118 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120 SCSI_LOG_MLQUEUE(1,
121 printk("Inserting command %p into mlqueue\n", cmd));
122
123 /*
Tejun Heo d8c37e72005-05-14 00:46:08 +0900124 * Set the appropriate busy bit for the device/host.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 *
126 * If the host/device isn't busy, assume that something actually
127 * completed, and that we should be able to queue a command now.
128 *
129 * Note that the prior mid-layer assumption that any host could
130 * always queue at least one command is now broken. The mid-layer
131 * will implement a user specifiable stall (see
132 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
133 * if a command is requeued with no other commands outstanding
134 * either for the device or for the host.
135 */
136 if (reason == SCSI_MLQUEUE_HOST_BUSY)
137 host->host_blocked = host->max_host_blocked;
138 else if (reason == SCSI_MLQUEUE_DEVICE_BUSY)
139 device->device_blocked = device->max_device_blocked;
140
141 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 * Decrement the counters, since these commands are no longer
143 * active on the host/device.
144 */
145 scsi_device_unbusy(device);
146
147 /*
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500148 * Requeue this command. It will go before all other commands
149 * that are already in the queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 *
151 * NOTE: there is magic here about the way the queue is plugged if
152 * we have no outstanding commands.
153 *
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500154 * Although we *don't* plug the queue, we call the request
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 * function. The SCSI request function detects the blocked condition
156 * and plugs the queue appropriately.
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500157 */
158 spin_lock_irqsave(q->queue_lock, flags);
James Bottomley59897da2005-09-14 12:57:42 -0400159 blk_requeue_request(q, cmd->request);
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500160 spin_unlock_irqrestore(q->queue_lock, flags);
161
162 scsi_run_queue(q);
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 return 0;
165}
166
James Bottomley39216032005-06-15 18:48:29 -0500167/**
James Bottomley33aa6872005-08-28 11:31:14 -0500168 * scsi_execute - insert request and wait for the result
James Bottomley39216032005-06-15 18:48:29 -0500169 * @sdev: scsi device
170 * @cmd: scsi command
171 * @data_direction: data direction
172 * @buffer: data buffer
173 * @bufflen: len of buffer
174 * @sense: optional sense buffer
175 * @timeout: request timeout in seconds
176 * @retries: number of times to retry request
James Bottomley33aa6872005-08-28 11:31:14 -0500177 * @flags: or into request flags;
James Bottomley39216032005-06-15 18:48:29 -0500178 *
Michael Opdenacker59c51592007-05-09 08:57:56 +0200179 * returns the req->errors value which is the scsi_cmnd result
James Bottomleyea73a9f2005-08-28 11:33:52 -0500180 * field.
Rob Landleyeb44820c2007-11-03 13:30:39 -0500181 */
James Bottomley33aa6872005-08-28 11:31:14 -0500182int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
183 int data_direction, void *buffer, unsigned bufflen,
184 unsigned char *sense, int timeout, int retries, int flags)
James Bottomley39216032005-06-15 18:48:29 -0500185{
186 struct request *req;
187 int write = (data_direction == DMA_TO_DEVICE);
188 int ret = DRIVER_ERROR << 24;
189
190 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
191
192 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
193 buffer, bufflen, __GFP_WAIT))
194 goto out;
195
196 req->cmd_len = COMMAND_SIZE(cmd[0]);
197 memcpy(req->cmd, cmd, req->cmd_len);
198 req->sense = sense;
199 req->sense_len = 0;
Mike Christie17e01f22005-11-11 05:31:37 -0600200 req->retries = retries;
James Bottomley39216032005-06-15 18:48:29 -0500201 req->timeout = timeout;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200202 req->cmd_type = REQ_TYPE_BLOCK_PC;
203 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
James Bottomley39216032005-06-15 18:48:29 -0500204
205 /*
206 * head injection *required* here otherwise quiesce won't work
207 */
208 blk_execute_rq(req->q, NULL, req, 1);
209
210 ret = req->errors;
211 out:
212 blk_put_request(req);
213
214 return ret;
215}
James Bottomley33aa6872005-08-28 11:31:14 -0500216EXPORT_SYMBOL(scsi_execute);
James Bottomley39216032005-06-15 18:48:29 -0500217
James Bottomleyea73a9f2005-08-28 11:33:52 -0500218
219int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
220 int data_direction, void *buffer, unsigned bufflen,
221 struct scsi_sense_hdr *sshdr, int timeout, int retries)
222{
223 char *sense = NULL;
akpm@osdl.org1ccb48b2005-06-26 00:12:51 -0700224 int result;
225
James Bottomleyea73a9f2005-08-28 11:33:52 -0500226 if (sshdr) {
Jes Sorensen24669f752006-01-16 10:31:18 -0500227 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
James Bottomleyea73a9f2005-08-28 11:33:52 -0500228 if (!sense)
229 return DRIVER_ERROR << 24;
James Bottomleyea73a9f2005-08-28 11:33:52 -0500230 }
akpm@osdl.org1ccb48b2005-06-26 00:12:51 -0700231 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
Jes Sorensen24669f752006-01-16 10:31:18 -0500232 sense, timeout, retries, 0);
James Bottomleyea73a9f2005-08-28 11:33:52 -0500233 if (sshdr)
James Bottomleye5143852005-08-09 11:55:36 -0500234 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
James Bottomleyea73a9f2005-08-28 11:33:52 -0500235
236 kfree(sense);
237 return result;
238}
239EXPORT_SYMBOL(scsi_execute_req);
240
Mike Christie6e68af62005-11-11 05:30:27 -0600241struct scsi_io_context {
242 void *data;
243 void (*done)(void *data, char *sense, int result, int resid);
244 char sense[SCSI_SENSE_BUFFERSIZE];
245};
246
Christoph Lametere18b8902006-12-06 20:33:20 -0800247static struct kmem_cache *scsi_io_context_cache;
Mike Christieaa7b5cd2005-11-11 05:31:41 -0600248
Jens Axboee650c302006-01-06 12:38:30 +0100249static void scsi_end_async(struct request *req, int uptodate)
Mike Christie6e68af62005-11-11 05:30:27 -0600250{
251 struct scsi_io_context *sioc = req->end_io_data;
252
253 if (sioc->done)
254 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len);
255
Mike Christieaa7b5cd2005-11-11 05:31:41 -0600256 kmem_cache_free(scsi_io_context_cache, sioc);
Mike Christie6e68af62005-11-11 05:30:27 -0600257 __blk_put_request(req->q, req);
258}
259
260static int scsi_merge_bio(struct request *rq, struct bio *bio)
261{
262 struct request_queue *q = rq->q;
263
264 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
265 if (rq_data_dir(rq) == WRITE)
266 bio->bi_rw |= (1 << BIO_RW);
267 blk_queue_bounce(q, &bio);
268
NeilBrown3001ca72007-08-16 13:31:27 +0200269 return blk_rq_append_bio(q, rq, bio);
Mike Christie6e68af62005-11-11 05:30:27 -0600270}
271
NeilBrown6712ecf2007-09-27 12:47:43 +0200272static void scsi_bi_endio(struct bio *bio, int error)
Mike Christie6e68af62005-11-11 05:30:27 -0600273{
Mike Christie6e68af62005-11-11 05:30:27 -0600274 bio_put(bio);
Mike Christie6e68af62005-11-11 05:30:27 -0600275}
276
277/**
278 * scsi_req_map_sg - map a scatterlist into a request
279 * @rq: request to fill
Rob Landleyeb44820c2007-11-03 13:30:39 -0500280 * @sgl: scatterlist
Mike Christie6e68af62005-11-11 05:30:27 -0600281 * @nsegs: number of elements
282 * @bufflen: len of buffer
283 * @gfp: memory allocation flags
284 *
285 * scsi_req_map_sg maps a scatterlist into a request so that the
286 * request can be sent to the block layer. We do not trust the scatterlist
287 * sent to use, as some ULDs use that struct to only organize the pages.
288 */
289static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl,
290 int nsegs, unsigned bufflen, gfp_t gfp)
291{
292 struct request_queue *q = rq->q;
Bryan Holtyf5235962006-03-22 06:35:39 -0600293 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
Mike Christiebd441de2007-03-13 12:52:29 -0500294 unsigned int data_len = bufflen, len, bytes, off;
Jens Axboec6132da2007-10-16 11:08:49 +0200295 struct scatterlist *sg;
Mike Christie6e68af62005-11-11 05:30:27 -0600296 struct page *page;
297 struct bio *bio = NULL;
298 int i, err, nr_vecs = 0;
299
Jens Axboec6132da2007-10-16 11:08:49 +0200300 for_each_sg(sgl, sg, nsegs, i) {
Jens Axboe45711f12007-10-22 21:19:53 +0200301 page = sg_page(sg);
Jens Axboec6132da2007-10-16 11:08:49 +0200302 off = sg->offset;
303 len = sg->length;
Mike Christie6e68af62005-11-11 05:30:27 -0600304
Mike Christiebd441de2007-03-13 12:52:29 -0500305 while (len > 0 && data_len > 0) {
306 /*
307 * sg sends a scatterlist that is larger than
308 * the data_len it wants transferred for certain
309 * IO sizes
310 */
Mike Christie6e68af62005-11-11 05:30:27 -0600311 bytes = min_t(unsigned int, len, PAGE_SIZE - off);
Mike Christiebd441de2007-03-13 12:52:29 -0500312 bytes = min(bytes, data_len);
Mike Christie6e68af62005-11-11 05:30:27 -0600313
314 if (!bio) {
315 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages);
316 nr_pages -= nr_vecs;
317
318 bio = bio_alloc(gfp, nr_vecs);
319 if (!bio) {
320 err = -ENOMEM;
321 goto free_bios;
322 }
323 bio->bi_end_io = scsi_bi_endio;
324 }
325
326 if (bio_add_pc_page(q, bio, page, bytes, off) !=
327 bytes) {
328 bio_put(bio);
329 err = -EINVAL;
330 goto free_bios;
331 }
332
333 if (bio->bi_vcnt >= nr_vecs) {
334 err = scsi_merge_bio(rq, bio);
335 if (err) {
NeilBrown6712ecf2007-09-27 12:47:43 +0200336 bio_endio(bio, 0);
Mike Christie6e68af62005-11-11 05:30:27 -0600337 goto free_bios;
338 }
339 bio = NULL;
340 }
341
342 page++;
343 len -= bytes;
Mike Christiebd441de2007-03-13 12:52:29 -0500344 data_len -=bytes;
Mike Christie6e68af62005-11-11 05:30:27 -0600345 off = 0;
346 }
347 }
348
349 rq->buffer = rq->data = NULL;
Mike Christiebd441de2007-03-13 12:52:29 -0500350 rq->data_len = bufflen;
Mike Christie6e68af62005-11-11 05:30:27 -0600351 return 0;
352
353free_bios:
354 while ((bio = rq->bio) != NULL) {
355 rq->bio = bio->bi_next;
356 /*
357 * call endio instead of bio_put incase it was bounced
358 */
NeilBrown6712ecf2007-09-27 12:47:43 +0200359 bio_endio(bio, 0);
Mike Christie6e68af62005-11-11 05:30:27 -0600360 }
361
362 return err;
363}
364
365/**
366 * scsi_execute_async - insert request
367 * @sdev: scsi device
368 * @cmd: scsi command
brking@us.ibm.combb1d1072006-01-23 15:03:22 -0600369 * @cmd_len: length of scsi cdb
Rob Landleyeb44820c2007-11-03 13:30:39 -0500370 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE
Mike Christie6e68af62005-11-11 05:30:27 -0600371 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
372 * @bufflen: len of buffer
373 * @use_sg: if buffer is a scatterlist this is the number of elements
374 * @timeout: request timeout in seconds
375 * @retries: number of times to retry request
Rob Landleyeb44820c2007-11-03 13:30:39 -0500376 * @privdata: data passed to done()
377 * @done: callback function when done
378 * @gfp: memory allocation flags
379 */
Mike Christie6e68af62005-11-11 05:30:27 -0600380int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd,
brking@us.ibm.combb1d1072006-01-23 15:03:22 -0600381 int cmd_len, int data_direction, void *buffer, unsigned bufflen,
Mike Christie6e68af62005-11-11 05:30:27 -0600382 int use_sg, int timeout, int retries, void *privdata,
383 void (*done)(void *, char *, int, int), gfp_t gfp)
384{
385 struct request *req;
386 struct scsi_io_context *sioc;
387 int err = 0;
388 int write = (data_direction == DMA_TO_DEVICE);
389
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800390 sioc = kmem_cache_zalloc(scsi_io_context_cache, gfp);
Mike Christie6e68af62005-11-11 05:30:27 -0600391 if (!sioc)
392 return DRIVER_ERROR << 24;
393
394 req = blk_get_request(sdev->request_queue, write, gfp);
395 if (!req)
396 goto free_sense;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200397 req->cmd_type = REQ_TYPE_BLOCK_PC;
398 req->cmd_flags |= REQ_QUIET;
Mike Christie6e68af62005-11-11 05:30:27 -0600399
400 if (use_sg)
401 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp);
402 else if (bufflen)
403 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp);
404
405 if (err)
406 goto free_req;
407
brking@us.ibm.combb1d1072006-01-23 15:03:22 -0600408 req->cmd_len = cmd_len;
Tejun Heo097b8452006-11-16 01:19:31 -0800409 memset(req->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
Mike Christie6e68af62005-11-11 05:30:27 -0600410 memcpy(req->cmd, cmd, req->cmd_len);
411 req->sense = sioc->sense;
412 req->sense_len = 0;
413 req->timeout = timeout;
Mike Christie17e01f22005-11-11 05:31:37 -0600414 req->retries = retries;
Mike Christie6e68af62005-11-11 05:30:27 -0600415 req->end_io_data = sioc;
416
417 sioc->data = privdata;
418 sioc->done = done;
419
420 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async);
421 return 0;
422
423free_req:
424 blk_put_request(req);
425free_sense:
Arne Redlich6470f2b2006-09-30 15:49:40 +0200426 kmem_cache_free(scsi_io_context_cache, sioc);
Mike Christie6e68af62005-11-11 05:30:27 -0600427 return DRIVER_ERROR << 24;
428}
429EXPORT_SYMBOL_GPL(scsi_execute_async);
430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431/*
432 * Function: scsi_init_cmd_errh()
433 *
434 * Purpose: Initialize cmd fields related to error handling.
435 *
436 * Arguments: cmd - command that is ready to be queued.
437 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 * Notes: This function has the job of initializing a number of
439 * fields related to error handling. Typically this will
440 * be called once for each command, as required.
441 */
Christoph Hellwig631c2282006-07-08 20:42:15 +0200442static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 cmd->serial_number = 0;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200445 scsi_set_resid(cmd, 0);
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900446 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (cmd->cmd_len == 0)
Boaz Harroshdb4742d2008-04-30 11:27:26 +0300448 cmd->cmd_len = scsi_command_size(cmd->cmnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449}
450
451void scsi_device_unbusy(struct scsi_device *sdev)
452{
453 struct Scsi_Host *shost = sdev->host;
454 unsigned long flags;
455
456 spin_lock_irqsave(shost->host_lock, flags);
457 shost->host_busy--;
James Bottomley939647e2005-09-18 15:05:20 -0500458 if (unlikely(scsi_host_in_recovery(shost) &&
Tejun Heoee7863b2006-05-15 20:57:20 +0900459 (shost->host_failed || shost->host_eh_scheduled)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 scsi_eh_wakeup(shost);
461 spin_unlock(shost->host_lock);
152587d2005-04-12 16:22:06 -0500462 spin_lock(sdev->request_queue->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 sdev->device_busy--;
152587d2005-04-12 16:22:06 -0500464 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
466
467/*
468 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
469 * and call blk_run_queue for all the scsi_devices on the target -
470 * including current_sdev first.
471 *
472 * Called with *no* scsi locks held.
473 */
474static void scsi_single_lun_run(struct scsi_device *current_sdev)
475{
476 struct Scsi_Host *shost = current_sdev->host;
477 struct scsi_device *sdev, *tmp;
478 struct scsi_target *starget = scsi_target(current_sdev);
479 unsigned long flags;
480
481 spin_lock_irqsave(shost->host_lock, flags);
482 starget->starget_sdev_user = NULL;
483 spin_unlock_irqrestore(shost->host_lock, flags);
484
485 /*
486 * Call blk_run_queue for all LUNs on the target, starting with
487 * current_sdev. We race with others (to set starget_sdev_user),
488 * but in most cases, we will be first. Ideally, each LU on the
489 * target would get some limited time or requests on the target.
490 */
491 blk_run_queue(current_sdev->request_queue);
492
493 spin_lock_irqsave(shost->host_lock, flags);
494 if (starget->starget_sdev_user)
495 goto out;
496 list_for_each_entry_safe(sdev, tmp, &starget->devices,
497 same_target_siblings) {
498 if (sdev == current_sdev)
499 continue;
500 if (scsi_device_get(sdev))
501 continue;
502
503 spin_unlock_irqrestore(shost->host_lock, flags);
504 blk_run_queue(sdev->request_queue);
505 spin_lock_irqsave(shost->host_lock, flags);
506
507 scsi_device_put(sdev);
508 }
509 out:
510 spin_unlock_irqrestore(shost->host_lock, flags);
511}
512
513/*
514 * Function: scsi_run_queue()
515 *
516 * Purpose: Select a proper request queue to serve next
517 *
518 * Arguments: q - last request's queue
519 *
520 * Returns: Nothing
521 *
522 * Notes: The previous command was completely finished, start
523 * a new one if possible.
524 */
525static void scsi_run_queue(struct request_queue *q)
526{
527 struct scsi_device *sdev = q->queuedata;
528 struct Scsi_Host *shost = sdev->host;
529 unsigned long flags;
530
Tony Battersby25d7c362007-11-12 10:00:44 -0500531 if (scsi_target(sdev)->single_lun)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 scsi_single_lun_run(sdev);
533
534 spin_lock_irqsave(shost->host_lock, flags);
535 while (!list_empty(&shost->starved_list) &&
536 !shost->host_blocked && !shost->host_self_blocked &&
537 !((shost->can_queue > 0) &&
538 (shost->host_busy >= shost->can_queue))) {
Nick Piggin75ad23b2008-04-29 14:48:33 +0200539
540 int flagset;
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 /*
543 * As long as shost is accepting commands and we have
544 * starved queues, call blk_run_queue. scsi_request_fn
545 * drops the queue_lock and can add us back to the
546 * starved_list.
547 *
548 * host_lock protects the starved_list and starved_entry.
549 * scsi_request_fn must get the host_lock before checking
550 * or modifying starved_list or starved_entry.
551 */
552 sdev = list_entry(shost->starved_list.next,
553 struct scsi_device, starved_entry);
554 list_del_init(&sdev->starved_entry);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200555 spin_unlock(shost->host_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Nick Piggin75ad23b2008-04-29 14:48:33 +0200557 spin_lock(sdev->request_queue->queue_lock);
558 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
559 !test_bit(QUEUE_FLAG_REENTER,
560 &sdev->request_queue->queue_flags);
561 if (flagset)
562 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue);
563 __blk_run_queue(sdev->request_queue);
564 if (flagset)
565 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue);
566 spin_unlock(sdev->request_queue->queue_lock);
Andreas Herrmann04846f22006-08-09 17:31:16 +0200567
Nick Piggin75ad23b2008-04-29 14:48:33 +0200568 spin_lock(shost->host_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 if (unlikely(!list_empty(&sdev->starved_entry)))
570 /*
571 * sdev lost a race, and was put back on the
572 * starved list. This is unlikely but without this
573 * in theory we could loop forever.
574 */
575 break;
576 }
577 spin_unlock_irqrestore(shost->host_lock, flags);
578
579 blk_run_queue(q);
580}
581
582/*
583 * Function: scsi_requeue_command()
584 *
585 * Purpose: Handle post-processing of completed commands.
586 *
587 * Arguments: q - queue to operate on
588 * cmd - command that may need to be requeued.
589 *
590 * Returns: Nothing
591 *
592 * Notes: After command completion, there may be blocks left
593 * over which weren't finished by the previous command
594 * this can be for a number of reasons - the main one is
595 * I/O errors in the middle of the request, in which case
596 * we need to request the blocks that come after the bad
597 * sector.
James Bottomleye91442b2005-09-09 10:44:16 -0500598 * Notes: Upon return, cmd is a stale pointer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 */
600static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
601{
James Bottomleye91442b2005-09-09 10:44:16 -0500602 struct request *req = cmd->request;
Tejun Heo 283369c2005-04-24 02:06:36 -0500603 unsigned long flags;
604
James Bottomleye91442b2005-09-09 10:44:16 -0500605 scsi_unprep_request(req);
Tejun Heo 283369c2005-04-24 02:06:36 -0500606 spin_lock_irqsave(q->queue_lock, flags);
James Bottomleye91442b2005-09-09 10:44:16 -0500607 blk_requeue_request(q, req);
Tejun Heo 283369c2005-04-24 02:06:36 -0500608 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 scsi_run_queue(q);
611}
612
613void scsi_next_command(struct scsi_cmnd *cmd)
614{
Linus Torvalds49d7bc62005-12-12 11:25:04 -0800615 struct scsi_device *sdev = cmd->device;
616 struct request_queue *q = sdev->request_queue;
617
618 /* need to hold a reference on the device before we let go of the cmd */
619 get_device(&sdev->sdev_gendev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 scsi_put_command(cmd);
622 scsi_run_queue(q);
Linus Torvalds49d7bc62005-12-12 11:25:04 -0800623
624 /* ok to remove device now */
625 put_device(&sdev->sdev_gendev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627
628void scsi_run_host_queues(struct Scsi_Host *shost)
629{
630 struct scsi_device *sdev;
631
632 shost_for_each_device(sdev, shost)
633 scsi_run_queue(sdev->request_queue);
634}
635
636/*
637 * Function: scsi_end_request()
638 *
639 * Purpose: Post-processing of completed commands (usually invoked at end
640 * of upper level post-processing and scsi_io_completion).
641 *
642 * Arguments: cmd - command that is complete.
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500643 * error - 0 if I/O indicates success, < 0 for I/O error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 * bytes - number of bytes of completed I/O
645 * requeue - indicates whether we should requeue leftovers.
646 *
647 * Lock status: Assumed that lock is not held upon entry.
648 *
James Bottomleye91442b2005-09-09 10:44:16 -0500649 * Returns: cmd if requeue required, NULL otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 *
651 * Notes: This is called for block device requests in order to
652 * mark some number of sectors as complete.
653 *
654 * We are guaranteeing that the request queue will be goosed
655 * at some point during this call.
James Bottomleye91442b2005-09-09 10:44:16 -0500656 * Notes: If cmd was requeued, upon return it will be a stale pointer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 */
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500658static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 int bytes, int requeue)
660{
Jens Axboe165125e2007-07-24 09:28:11 +0200661 struct request_queue *q = cmd->device->request_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 struct request *req = cmd->request;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 /*
665 * If there are blocks left over at the end, set up the command
666 * to queue the remainder of them.
667 */
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500668 if (blk_end_request(req, error, bytes)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 int leftover = (req->hard_nr_sectors << 9);
670
671 if (blk_pc_request(req))
672 leftover = req->data_len;
673
674 /* kill remainder if no retrys */
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500675 if (error && blk_noretry_request(req))
676 blk_end_request(req, error, leftover);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 else {
James Bottomleye91442b2005-09-09 10:44:16 -0500678 if (requeue) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 /*
680 * Bleah. Leftovers again. Stick the
681 * leftovers in the front of the
682 * queue, and goose the queue again.
683 */
684 scsi_requeue_command(q, cmd);
James Bottomleye91442b2005-09-09 10:44:16 -0500685 cmd = NULL;
686 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return cmd;
688 }
689 }
690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 /*
692 * This will goose the queue request function at the end, so we don't
693 * need to worry about launching another command.
694 */
695 scsi_next_command(cmd);
696 return NULL;
697}
698
Jens Axboea8474ce2007-08-07 09:02:51 +0200699static inline unsigned int scsi_sgtable_index(unsigned short nents)
700{
701 unsigned int index;
702
James Bottomleyd3f46f32008-01-15 11:11:46 -0600703 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
704
705 if (nents <= 8)
Jens Axboea8474ce2007-08-07 09:02:51 +0200706 index = 0;
James Bottomleyd3f46f32008-01-15 11:11:46 -0600707 else
708 index = get_count_order(nents) - 3;
Jens Axboea8474ce2007-08-07 09:02:51 +0200709
710 return index;
711}
712
Jens Axboe5ed79592007-11-15 09:13:11 +0100713static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
715 struct scsi_host_sg_pool *sgp;
Jens Axboe5ed79592007-11-15 09:13:11 +0100716
717 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
718 mempool_free(sgl, sgp->pool);
719}
720
721static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
722{
723 struct scsi_host_sg_pool *sgp;
724
725 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
726 return mempool_alloc(sgp->pool, gfp_mask);
727}
728
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200729static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
730 gfp_t gfp_mask)
Jens Axboe5ed79592007-11-15 09:13:11 +0100731{
732 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200734 BUG_ON(!nents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200736 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
737 gfp_mask, scsi_sg_alloc);
Jens Axboe5ed79592007-11-15 09:13:11 +0100738 if (unlikely(ret))
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200739 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
James Bottomley7cedb1f2008-01-13 14:15:28 -0600740 scsi_sg_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Jens Axboea8474ce2007-08-07 09:02:51 +0200742 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200745static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200747 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748}
749
750/*
751 * Function: scsi_release_buffers()
752 *
753 * Purpose: Completion processing for block device I/O requests.
754 *
755 * Arguments: cmd - command that we are bailing.
756 *
757 * Lock status: Assumed that no lock is held upon entry.
758 *
759 * Returns: Nothing
760 *
761 * Notes: In the event that an upper level driver rejects a
762 * command, we must release resources allocated during
763 * the __init_io() function. Primarily this would involve
764 * the scatter-gather table, and potentially any bounce
765 * buffers.
766 */
Boaz Harroshbb52d822007-12-13 16:14:27 -0800767void scsi_release_buffers(struct scsi_cmnd *cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200769 if (cmd->sdb.table.nents)
770 scsi_free_sgtable(&cmd->sdb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200772 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200773
774 if (scsi_bidi_cmnd(cmd)) {
775 struct scsi_data_buffer *bidi_sdb =
776 cmd->request->next_rq->special;
777 scsi_free_sgtable(bidi_sdb);
778 kmem_cache_free(scsi_bidi_sdb_cache, bidi_sdb);
779 cmd->request->next_rq->special = NULL;
780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
Boaz Harroshbb52d822007-12-13 16:14:27 -0800782EXPORT_SYMBOL(scsi_release_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
784/*
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200785 * Bidi commands Must be complete as a whole, both sides at once.
786 * If part of the bytes were written and lld returned
787 * scsi_in()->resid and/or scsi_out()->resid this information will be left
788 * in req->data_len and req->next_rq->data_len. The upper-layer driver can
789 * decide what to do with this information.
790 */
Adrian Bunk8c5e03d2008-03-31 01:53:55 +0300791static void scsi_end_bidi_request(struct scsi_cmnd *cmd)
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200792{
Kiyoshi Uedab8de1632008-01-18 12:02:15 -0500793 struct request *req = cmd->request;
794 unsigned int dlen = req->data_len;
795 unsigned int next_dlen = req->next_rq->data_len;
796
797 req->data_len = scsi_out(cmd)->resid;
798 req->next_rq->data_len = scsi_in(cmd)->resid;
799
800 /* The req and req->next_rq have not been completed */
801 BUG_ON(blk_end_bidi_request(req, 0, dlen, next_dlen));
802
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200803 scsi_release_buffers(cmd);
804
805 /*
806 * This will goose the queue request function at the end, so we don't
807 * need to worry about launching another command.
808 */
809 scsi_next_command(cmd);
810}
811
812/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 * Function: scsi_io_completion()
814 *
815 * Purpose: Completion processing for block device I/O requests.
816 *
817 * Arguments: cmd - command that is finished.
818 *
819 * Lock status: Assumed that no lock is held upon entry.
820 *
821 * Returns: Nothing
822 *
823 * Notes: This function is matched in terms of capabilities to
824 * the function that created the scatter-gather list.
825 * In other words, if there are no bounce buffers
826 * (the normal case for most drivers), we don't need
827 * the logic to deal with cleaning up afterwards.
828 *
829 * We must do one of several things here:
830 *
831 * a) Call scsi_end_request. This will finish off the
832 * specified number of sectors. If we are done, the
833 * command block will be released, and the queue
834 * function will be goosed. If we are not done, then
835 * scsi_end_request will directly goose the queue.
836 *
837 * b) We can just use scsi_requeue_command() here. This would
838 * be used if we just wanted to retry, for example.
839 */
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700840void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
842 int result = cmd->result;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200843 int this_count = scsi_bufflen(cmd);
Jens Axboe165125e2007-07-24 09:28:11 +0200844 struct request_queue *q = cmd->device->request_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 struct request *req = cmd->request;
James Bottomleyfa8e36c2008-04-02 18:11:52 -0500846 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 struct scsi_sense_hdr sshdr;
848 int sense_valid = 0;
849 int sense_deferred = 0;
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (result) {
852 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
853 if (sense_valid)
854 sense_deferred = scsi_sense_is_deferred(&sshdr);
855 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
858 req->errors = result;
859 if (result) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (sense_valid && req->sense) {
861 /*
862 * SG_IO wants current and deferred errors
863 */
864 int len = 8 + cmd->sense_buffer[7];
865
866 if (len > SCSI_SENSE_BUFFERSIZE)
867 len = SCSI_SENSE_BUFFERSIZE;
868 memcpy(req->sense, cmd->sense_buffer, len);
869 req->sense_len = len;
870 }
James Bottomleyfa8e36c2008-04-02 18:11:52 -0500871 if (!sense_deferred)
872 error = -EIO;
Pete Wyckoffb22f6872007-03-13 16:53:28 -0400873 }
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200874 if (scsi_bidi_cmnd(cmd)) {
875 /* will also release_buffers */
876 scsi_end_bidi_request(cmd);
877 return;
878 }
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200879 req->data_len = scsi_get_resid(cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 }
881
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200882 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200883 scsi_release_buffers(cmd);
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 * Next deal with any sectors which we were able to correctly
887 * handle.
888 */
James Bottomleyd6b0c532006-07-02 10:06:28 -0500889 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, "
890 "%d bytes done.\n",
891 req->nr_sectors, good_bytes));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
James Bottomleyd6b0c532006-07-02 10:06:28 -0500893 /* A number of bytes were successfully read. If there
894 * are leftovers and there is some kind of error
895 * (result != 0), retry the rest.
896 */
James Bottomleyfa8e36c2008-04-02 18:11:52 -0500897 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
James Bottomleyd6b0c532006-07-02 10:06:28 -0500898 return;
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700899
900 /* good_bytes = 0, or (inclusive) there were leftovers and
901 * result = 0, so scsi_end_request couldn't retry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 */
903 if (sense_valid && !sense_deferred) {
904 switch (sshdr.sense_key) {
905 case UNIT_ATTENTION:
906 if (cmd->device->removable) {
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700907 /* Detected disc change. Set a bit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 * and quietly refuse further access.
909 */
910 cmd->device->changed = 1;
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500911 scsi_end_request(cmd, -EIO, this_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 return;
913 } else {
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700914 /* Must have been a power glitch, or a
915 * bus reset. Could not have been a
916 * media change, so we just retry the
917 * request and see what happens.
918 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 scsi_requeue_command(q, cmd);
920 return;
921 }
922 break;
923 case ILLEGAL_REQUEST:
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700924 /* If we had an ILLEGAL REQUEST returned, then
925 * we may have performed an unsupported
926 * command. The only thing this should be
927 * would be a ten byte read where only a six
928 * byte read was supported. Also, on a system
929 * where READ CAPACITY failed, we may have
930 * read past the end of the disk.
931 */
Jens Axboe26a68012005-11-29 21:03:34 +0100932 if ((cmd->device->use_10_for_rw &&
933 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 (cmd->cmnd[0] == READ_10 ||
935 cmd->cmnd[0] == WRITE_10)) {
936 cmd->device->use_10_for_rw = 0;
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700937 /* This will cause a retry with a
938 * 6-byte command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 */
940 scsi_requeue_command(q, cmd);
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700941 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 } else {
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500943 scsi_end_request(cmd, -EIO, this_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 return;
945 }
946 break;
947 case NOT_READY:
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700948 /* If the device is in the process of becoming
James Bottomleyf3e93f72006-04-25 16:48:30 -0500949 * ready, or has a temporary blockage, retry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 */
James Bottomleyf3e93f72006-04-25 16:48:30 -0500951 if (sshdr.asc == 0x04) {
952 switch (sshdr.ascq) {
953 case 0x01: /* becoming ready */
954 case 0x04: /* format in progress */
955 case 0x05: /* rebuild in progress */
956 case 0x06: /* recalculation in progress */
957 case 0x07: /* operation in progress */
958 case 0x08: /* Long write in progress */
959 case 0x09: /* self test in progress */
960 scsi_requeue_command(q, cmd);
961 return;
962 default:
963 break;
964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 }
James Bottomley311b5812007-09-23 09:08:46 -0500966 if (!(req->cmd_flags & REQ_QUIET))
967 scsi_cmd_print_sense_hdr(cmd,
968 "Device not ready",
969 &sshdr);
970
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500971 scsi_end_request(cmd, -EIO, this_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return;
973 case VOLUME_OVERFLOW:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200974 if (!(req->cmd_flags & REQ_QUIET)) {
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400975 scmd_printk(KERN_INFO, cmd,
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700976 "Volume overflow, CDB: ");
Christoph Hellwig631c2282006-07-08 20:42:15 +0200977 __scsi_print_command(cmd->cmnd);
James Bottomley3173d8c2005-09-04 11:32:05 -0500978 scsi_print_sense("", cmd);
979 }
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700980 /* See SSC3rXX or current. */
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -0500981 scsi_end_request(cmd, -EIO, this_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 return;
983 default:
984 break;
985 }
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700986 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 if (host_byte(result) == DID_RESET) {
Luben Tuikov03aba2f2006-06-23 09:39:09 -0700988 /* Third party bus reset or reset for error recovery
989 * reasons. Just retry the request and see what
990 * happens.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 */
992 scsi_requeue_command(q, cmd);
993 return;
994 }
995 if (result) {
Jens Axboe4aff5e22006-08-10 08:44:47 +0200996 if (!(req->cmd_flags & REQ_QUIET)) {
Martin K. Petersena4d04a42007-02-27 22:40:27 -0500997 scsi_print_result(cmd);
James Bottomley3173d8c2005-09-04 11:32:05 -0500998 if (driver_byte(result) & DRIVER_SENSE)
999 scsi_print_sense("", cmd);
1000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 }
Kiyoshi Ueda610d8b02007-12-11 17:52:09 -05001002 scsi_end_request(cmd, -EIO, this_count, !result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +02001005static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1006 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007{
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +02001008 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 /*
Christoph Hellwig3b003152006-11-04 20:10:55 +01001011 * If sg table allocation fails, requeue request later.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 */
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001013 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1014 gfp_mask))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 return BLKPREP_DEFER;
Alan Stern7c72ce82005-10-14 11:23:27 -04001016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Christoph Hellwig3b003152006-11-04 20:10:55 +01001018 req->buffer = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 /*
1021 * Next, walk the list, and fill in the addresses and sizes of
1022 * each segment.
1023 */
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001024 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1025 BUG_ON(count > sdb->table.nents);
1026 sdb->table.nents = count;
Tejun Heo6b00769f2008-02-19 11:36:35 +01001027 if (blk_pc_request(req))
1028 sdb->length = req->data_len;
1029 else
1030 sdb->length = req->nr_sectors << 9;
Rusty Russell4a03d902007-11-19 11:28:48 +11001031 return BLKPREP_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032}
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +02001033
1034/*
1035 * Function: scsi_init_io()
1036 *
1037 * Purpose: SCSI I/O initialize function.
1038 *
1039 * Arguments: cmd - Command descriptor we wish to initialize
1040 *
1041 * Returns: 0 on success
1042 * BLKPREP_DEFER if the failure is retryable
1043 * BLKPREP_KILL if the failure is fatal
1044 */
1045int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1046{
1047 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
1048 if (error)
1049 goto err_exit;
1050
1051 if (blk_bidi_rq(cmd->request)) {
1052 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1053 scsi_bidi_sdb_cache, GFP_ATOMIC);
1054 if (!bidi_sdb) {
1055 error = BLKPREP_DEFER;
1056 goto err_exit;
1057 }
1058
1059 cmd->request->next_rq->special = bidi_sdb;
1060 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
1061 GFP_ATOMIC);
1062 if (error)
1063 goto err_exit;
1064 }
1065
1066 return BLKPREP_OK ;
1067
1068err_exit:
1069 scsi_release_buffers(cmd);
1070 if (error == BLKPREP_KILL)
1071 scsi_put_command(cmd);
1072 else /* BLKPREP_DEFER */
1073 scsi_unprep_request(cmd->request);
1074
1075 return error;
1076}
Boaz Harroshbb52d822007-12-13 16:14:27 -08001077EXPORT_SYMBOL(scsi_init_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Christoph Hellwig3b003152006-11-04 20:10:55 +01001079static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1080 struct request *req)
1081{
1082 struct scsi_cmnd *cmd;
1083
1084 if (!req->special) {
1085 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1086 if (unlikely(!cmd))
1087 return NULL;
1088 req->special = cmd;
1089 } else {
1090 cmd = req->special;
1091 }
1092
1093 /* pull a tag out of the request if we have one */
1094 cmd->tag = req->tag;
1095 cmd->request = req;
1096
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001097 cmd->cmnd = req->cmd;
1098
Christoph Hellwig3b003152006-11-04 20:10:55 +01001099 return cmd;
1100}
1101
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001102int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
James Bottomley7b163182005-12-15 20:17:02 -06001103{
Christoph Hellwig3b003152006-11-04 20:10:55 +01001104 struct scsi_cmnd *cmd;
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001105 int ret = scsi_prep_state_check(sdev, req);
1106
1107 if (ret != BLKPREP_OK)
1108 return ret;
Christoph Hellwig3b003152006-11-04 20:10:55 +01001109
1110 cmd = scsi_get_cmd_from_req(sdev, req);
1111 if (unlikely(!cmd))
1112 return BLKPREP_DEFER;
1113
1114 /*
1115 * BLOCK_PC requests may transfer data, in which case they must
1116 * a bio attached to them. Or they might contain a SCSI command
1117 * that does not transfer data, in which case they may optionally
1118 * submit a request without an attached bio.
1119 */
1120 if (req->bio) {
1121 int ret;
1122
1123 BUG_ON(!req->nr_phys_segments);
1124
Boaz Harroshbb52d822007-12-13 16:14:27 -08001125 ret = scsi_init_io(cmd, GFP_ATOMIC);
Christoph Hellwig3b003152006-11-04 20:10:55 +01001126 if (unlikely(ret))
1127 return ret;
1128 } else {
1129 BUG_ON(req->data_len);
1130 BUG_ON(req->data);
1131
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001132 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
Christoph Hellwig3b003152006-11-04 20:10:55 +01001133 req->buffer = NULL;
1134 }
James Bottomley7b163182005-12-15 20:17:02 -06001135
James Bottomley7b163182005-12-15 20:17:02 -06001136 cmd->cmd_len = req->cmd_len;
1137 if (!req->data_len)
1138 cmd->sc_data_direction = DMA_NONE;
1139 else if (rq_data_dir(req) == WRITE)
1140 cmd->sc_data_direction = DMA_TO_DEVICE;
1141 else
1142 cmd->sc_data_direction = DMA_FROM_DEVICE;
1143
1144 cmd->transfersize = req->data_len;
1145 cmd->allowed = req->retries;
1146 cmd->timeout_per_command = req->timeout;
Christoph Hellwig3b003152006-11-04 20:10:55 +01001147 return BLKPREP_OK;
1148}
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001149EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
Christoph Hellwig3b003152006-11-04 20:10:55 +01001150
1151/*
1152 * Setup a REQ_TYPE_FS command. These are simple read/write request
1153 * from filesystems that still need to be translated to SCSI CDBs from
1154 * the ULD.
1155 */
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001156int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
Christoph Hellwig3b003152006-11-04 20:10:55 +01001157{
1158 struct scsi_cmnd *cmd;
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001159 int ret = scsi_prep_state_check(sdev, req);
Christoph Hellwig3b003152006-11-04 20:10:55 +01001160
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001161 if (ret != BLKPREP_OK)
1162 return ret;
Christoph Hellwig3b003152006-11-04 20:10:55 +01001163 /*
1164 * Filesystem requests must transfer data.
1165 */
1166 BUG_ON(!req->nr_phys_segments);
1167
1168 cmd = scsi_get_cmd_from_req(sdev, req);
1169 if (unlikely(!cmd))
1170 return BLKPREP_DEFER;
1171
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001172 memset(cmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harroshbb52d822007-12-13 16:14:27 -08001173 return scsi_init_io(cmd, GFP_ATOMIC);
James Bottomley7b163182005-12-15 20:17:02 -06001174}
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001175EXPORT_SYMBOL(scsi_setup_fs_cmnd);
James Bottomley7b163182005-12-15 20:17:02 -06001176
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001177int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
Christoph Hellwig3b003152006-11-04 20:10:55 +01001179 int ret = BLKPREP_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
1181 /*
Christoph Hellwig3b003152006-11-04 20:10:55 +01001182 * If the device is not in running state we will reject some
1183 * or all commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
Christoph Hellwig3b003152006-11-04 20:10:55 +01001186 switch (sdev->sdev_state) {
1187 case SDEV_OFFLINE:
1188 /*
1189 * If the device is offline we refuse to process any
1190 * commands. The device must be brought online
1191 * before trying any recovery commands.
1192 */
1193 sdev_printk(KERN_ERR, sdev,
1194 "rejecting I/O to offline device\n");
1195 ret = BLKPREP_KILL;
1196 break;
1197 case SDEV_DEL:
1198 /*
1199 * If the device is fully deleted, we refuse to
1200 * process any commands as well.
1201 */
James Bottomley9ccfc752005-10-02 11:45:08 -05001202 sdev_printk(KERN_ERR, sdev,
1203 "rejecting I/O to dead device\n");
Christoph Hellwig3b003152006-11-04 20:10:55 +01001204 ret = BLKPREP_KILL;
1205 break;
1206 case SDEV_QUIESCE:
1207 case SDEV_BLOCK:
1208 /*
1209 * If the devices is blocked we defer normal commands.
1210 */
1211 if (!(req->cmd_flags & REQ_PREEMPT))
1212 ret = BLKPREP_DEFER;
1213 break;
1214 default:
1215 /*
1216 * For any other not fully online state we only allow
1217 * special commands. In particular any user initiated
1218 * command is not allowed.
1219 */
1220 if (!(req->cmd_flags & REQ_PREEMPT))
1221 ret = BLKPREP_KILL;
1222 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001225 return ret;
1226}
1227EXPORT_SYMBOL(scsi_prep_state_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001229int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1230{
1231 struct scsi_device *sdev = q->queuedata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Christoph Hellwig3b003152006-11-04 20:10:55 +01001233 switch (ret) {
1234 case BLKPREP_KILL:
1235 req->errors = DID_NO_CONNECT << 16;
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001236 /* release the command and kill it */
1237 if (req->special) {
1238 struct scsi_cmnd *cmd = req->special;
1239 scsi_release_buffers(cmd);
1240 scsi_put_command(cmd);
1241 req->special = NULL;
1242 }
Christoph Hellwig3b003152006-11-04 20:10:55 +01001243 break;
1244 case BLKPREP_DEFER:
1245 /*
1246 * If we defer, the elv_next_request() returns NULL, but the
1247 * queue must be restarted, so we plug here if no returning
1248 * command will automatically do that.
1249 */
1250 if (sdev->device_busy == 0)
1251 blk_plug_device(q);
1252 break;
1253 default:
1254 req->cmd_flags |= REQ_DONTPREP;
1255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Christoph Hellwig3b003152006-11-04 20:10:55 +01001257 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258}
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001259EXPORT_SYMBOL(scsi_prep_return);
1260
James Bottomley751bf4d2008-01-02 11:14:30 -06001261int scsi_prep_fn(struct request_queue *q, struct request *req)
James Bottomley7f9a6bc2007-08-04 10:06:25 -05001262{
1263 struct scsi_device *sdev = q->queuedata;
1264 int ret = BLKPREP_KILL;
1265
1266 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1267 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1268 return scsi_prep_return(q, req, ret);
1269}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271/*
1272 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1273 * return 0.
1274 *
1275 * Called with the queue_lock held.
1276 */
1277static inline int scsi_dev_queue_ready(struct request_queue *q,
1278 struct scsi_device *sdev)
1279{
1280 if (sdev->device_busy >= sdev->queue_depth)
1281 return 0;
1282 if (sdev->device_busy == 0 && sdev->device_blocked) {
1283 /*
1284 * unblock after device_blocked iterates to zero
1285 */
1286 if (--sdev->device_blocked == 0) {
1287 SCSI_LOG_MLQUEUE(3,
James Bottomley9ccfc752005-10-02 11:45:08 -05001288 sdev_printk(KERN_INFO, sdev,
1289 "unblocking device at zero depth\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 } else {
1291 blk_plug_device(q);
1292 return 0;
1293 }
1294 }
1295 if (sdev->device_blocked)
1296 return 0;
1297
1298 return 1;
1299}
1300
1301/*
1302 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1303 * return 0. We must end up running the queue again whenever 0 is
1304 * returned, else IO can hang.
1305 *
1306 * Called with host_lock held.
1307 */
1308static inline int scsi_host_queue_ready(struct request_queue *q,
1309 struct Scsi_Host *shost,
1310 struct scsi_device *sdev)
1311{
James Bottomley939647e2005-09-18 15:05:20 -05001312 if (scsi_host_in_recovery(shost))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 return 0;
1314 if (shost->host_busy == 0 && shost->host_blocked) {
1315 /*
1316 * unblock after host_blocked iterates to zero
1317 */
1318 if (--shost->host_blocked == 0) {
1319 SCSI_LOG_MLQUEUE(3,
1320 printk("scsi%d unblocking host at zero depth\n",
1321 shost->host_no));
1322 } else {
1323 blk_plug_device(q);
1324 return 0;
1325 }
1326 }
1327 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1328 shost->host_blocked || shost->host_self_blocked) {
1329 if (list_empty(&sdev->starved_entry))
1330 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1331 return 0;
1332 }
1333
1334 /* We're OK to process the command, so we can't be starved */
1335 if (!list_empty(&sdev->starved_entry))
1336 list_del_init(&sdev->starved_entry);
1337
1338 return 1;
1339}
1340
1341/*
James Bottomleye91442b2005-09-09 10:44:16 -05001342 * Kill a request for a dead device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 */
Jens Axboe165125e2007-07-24 09:28:11 +02001344static void scsi_kill_request(struct request *req, struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
James Bottomleye91442b2005-09-09 10:44:16 -05001346 struct scsi_cmnd *cmd = req->special;
Tejun Heoe36e0c82006-04-11 17:27:53 +09001347 struct scsi_device *sdev = cmd->device;
1348 struct Scsi_Host *shost = sdev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
James Bottomley788ce432005-09-09 13:40:23 -05001350 blkdev_dequeue_request(req);
1351
James Bottomleye91442b2005-09-09 10:44:16 -05001352 if (unlikely(cmd == NULL)) {
1353 printk(KERN_CRIT "impossible request in %s.\n",
1354 __FUNCTION__);
1355 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
James Bottomleye91442b2005-09-09 10:44:16 -05001357
1358 scsi_init_cmd_errh(cmd);
1359 cmd->result = DID_NO_CONNECT << 16;
1360 atomic_inc(&cmd->device->iorequest_cnt);
Tejun Heoe36e0c82006-04-11 17:27:53 +09001361
1362 /*
1363 * SCSI request completion path will do scsi_device_unbusy(),
1364 * bump busy counts. To bump the counters, we need to dance
1365 * with the locks as normal issue path does.
1366 */
1367 sdev->device_busy++;
1368 spin_unlock(sdev->request_queue->queue_lock);
1369 spin_lock(shost->host_lock);
1370 shost->host_busy++;
1371 spin_unlock(shost->host_lock);
1372 spin_lock(sdev->request_queue->queue_lock);
1373
James Bottomleye91442b2005-09-09 10:44:16 -05001374 __scsi_done(cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
Jens Axboe1aea6432006-01-09 16:03:03 +01001377static void scsi_softirq_done(struct request *rq)
1378{
1379 struct scsi_cmnd *cmd = rq->completion_data;
Brian King8884efa2006-02-24 17:10:04 -06001380 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
Jens Axboe1aea6432006-01-09 16:03:03 +01001381 int disposition;
1382
1383 INIT_LIST_HEAD(&cmd->eh_entry);
1384
1385 disposition = scsi_decide_disposition(cmd);
1386 if (disposition != SUCCESS &&
1387 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1388 sdev_printk(KERN_ERR, cmd->device,
1389 "timing out command, waited %lus\n",
1390 wait_for/HZ);
1391 disposition = SUCCESS;
1392 }
1393
1394 scsi_log_completion(cmd, disposition);
1395
1396 switch (disposition) {
1397 case SUCCESS:
1398 scsi_finish_command(cmd);
1399 break;
1400 case NEEDS_RETRY:
Christoph Hellwig596f4822007-01-02 12:56:00 +01001401 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
Jens Axboe1aea6432006-01-09 16:03:03 +01001402 break;
1403 case ADD_TO_MLQUEUE:
1404 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1405 break;
1406 default:
1407 if (!scsi_eh_scmd_add(cmd, 0))
1408 scsi_finish_command(cmd);
1409 }
1410}
1411
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412/*
1413 * Function: scsi_request_fn()
1414 *
1415 * Purpose: Main strategy routine for SCSI.
1416 *
1417 * Arguments: q - Pointer to actual queue.
1418 *
1419 * Returns: Nothing
1420 *
1421 * Lock status: IO request lock assumed to be held when called.
1422 */
1423static void scsi_request_fn(struct request_queue *q)
1424{
1425 struct scsi_device *sdev = q->queuedata;
1426 struct Scsi_Host *shost;
1427 struct scsi_cmnd *cmd;
1428 struct request *req;
1429
1430 if (!sdev) {
1431 printk("scsi: killing requests for dead queue\n");
James Bottomleye91442b2005-09-09 10:44:16 -05001432 while ((req = elv_next_request(q)) != NULL)
1433 scsi_kill_request(req, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 return;
1435 }
1436
1437 if(!get_device(&sdev->sdev_gendev))
1438 /* We must be tearing the block queue down already */
1439 return;
1440
1441 /*
1442 * To start with, we keep looping until the queue is empty, or until
1443 * the host is no longer able to accept any more requests.
1444 */
1445 shost = sdev->host;
1446 while (!blk_queue_plugged(q)) {
1447 int rtn;
1448 /*
1449 * get next queueable request. We do this early to make sure
1450 * that the request is fully prepared even if we cannot
1451 * accept it.
1452 */
1453 req = elv_next_request(q);
1454 if (!req || !scsi_dev_queue_ready(q, sdev))
1455 break;
1456
1457 if (unlikely(!scsi_device_online(sdev))) {
James Bottomley9ccfc752005-10-02 11:45:08 -05001458 sdev_printk(KERN_ERR, sdev,
1459 "rejecting I/O to offline device\n");
James Bottomleye91442b2005-09-09 10:44:16 -05001460 scsi_kill_request(req, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 continue;
1462 }
1463
1464
1465 /*
1466 * Remove the request from the request list.
1467 */
1468 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1469 blkdev_dequeue_request(req);
1470 sdev->device_busy++;
1471
1472 spin_unlock(q->queue_lock);
James Bottomleye91442b2005-09-09 10:44:16 -05001473 cmd = req->special;
1474 if (unlikely(cmd == NULL)) {
1475 printk(KERN_CRIT "impossible request in %s.\n"
1476 "please mail a stack trace to "
Jens Axboe4aff5e22006-08-10 08:44:47 +02001477 "linux-scsi@vger.kernel.org\n",
James Bottomleye91442b2005-09-09 10:44:16 -05001478 __FUNCTION__);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001479 blk_dump_rq_flags(req, "foo");
James Bottomleye91442b2005-09-09 10:44:16 -05001480 BUG();
1481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 spin_lock(shost->host_lock);
1483
1484 if (!scsi_host_queue_ready(q, shost, sdev))
1485 goto not_ready;
Tony Battersby25d7c362007-11-12 10:00:44 -05001486 if (scsi_target(sdev)->single_lun) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 if (scsi_target(sdev)->starget_sdev_user &&
1488 scsi_target(sdev)->starget_sdev_user != sdev)
1489 goto not_ready;
1490 scsi_target(sdev)->starget_sdev_user = sdev;
1491 }
1492 shost->host_busy++;
1493
1494 /*
1495 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1496 * take the lock again.
1497 */
1498 spin_unlock_irq(shost->host_lock);
1499
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 /*
1501 * Finally, initialize any error handling parameters, and set up
1502 * the timers for timeouts.
1503 */
1504 scsi_init_cmd_errh(cmd);
1505
1506 /*
1507 * Dispatch the command to the low-level driver.
1508 */
1509 rtn = scsi_dispatch_cmd(cmd);
1510 spin_lock_irq(q->queue_lock);
1511 if(rtn) {
1512 /* we're refusing the command; because of
1513 * the way locks get dropped, we need to
1514 * check here if plugging is required */
1515 if(sdev->device_busy == 0)
1516 blk_plug_device(q);
1517
1518 break;
1519 }
1520 }
1521
1522 goto out;
1523
1524 not_ready:
1525 spin_unlock_irq(shost->host_lock);
1526
1527 /*
1528 * lock q, handle tag, requeue req, and decrement device_busy. We
1529 * must return with queue_lock held.
1530 *
1531 * Decrementing device_busy without checking it is OK, as all such
1532 * cases (host limits or settings) should run the queue at some
1533 * later time.
1534 */
1535 spin_lock_irq(q->queue_lock);
1536 blk_requeue_request(q, req);
1537 sdev->device_busy--;
1538 if(sdev->device_busy == 0)
1539 blk_plug_device(q);
1540 out:
1541 /* must be careful here...if we trigger the ->remove() function
1542 * we cannot be holding the q lock */
1543 spin_unlock_irq(q->queue_lock);
1544 put_device(&sdev->sdev_gendev);
1545 spin_lock_irq(q->queue_lock);
1546}
1547
1548u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1549{
1550 struct device *host_dev;
1551 u64 bounce_limit = 0xffffffff;
1552
1553 if (shost->unchecked_isa_dma)
1554 return BLK_BOUNCE_ISA;
1555 /*
1556 * Platforms with virtual-DMA translation
1557 * hardware have no practical limit.
1558 */
1559 if (!PCI_DMA_BUS_IS_PHYS)
1560 return BLK_BOUNCE_ANY;
1561
1562 host_dev = scsi_get_device(shost);
1563 if (host_dev && host_dev->dma_mask)
1564 bounce_limit = *host_dev->dma_mask;
1565
1566 return bounce_limit;
1567}
1568EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1569
FUJITA Tomonorib58d9152006-11-16 19:24:10 +09001570struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1571 request_fn_proc *request_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 struct request_queue *q;
FUJITA Tomonori860ac562008-02-04 22:28:05 -08001574 struct device *dev = shost->shost_gendev.parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
FUJITA Tomonorib58d9152006-11-16 19:24:10 +09001576 q = blk_init_queue(request_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 if (!q)
1578 return NULL;
1579
Jens Axboea8474ce2007-08-07 09:02:51 +02001580 /*
1581 * this limit is imposed by hardware restrictions
1582 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 blk_queue_max_hw_segments(q, shost->sg_tablesize);
James Bottomleyd3f46f32008-01-15 11:11:46 -06001584 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS);
Jens Axboea8474ce2007-08-07 09:02:51 +02001585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 blk_queue_max_sectors(q, shost->max_sectors);
1587 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1588 blk_queue_segment_boundary(q, shost->dma_boundary);
FUJITA Tomonori99c84db2008-02-04 22:28:17 -08001589 dma_set_seg_boundary(dev, shost->dma_boundary);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
FUJITA Tomonori860ac562008-02-04 22:28:05 -08001591 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1592
Nick Piggin75ad23b2008-04-29 14:48:33 +02001593 /* New queue, no concurrency on queue_flags */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (!shost->use_clustering)
Nick Piggin75ad23b2008-04-29 14:48:33 +02001595 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
James Bottomley465ff312008-01-01 10:00:10 -06001596
1597 /*
1598 * set a reasonable default alignment on word boundaries: the
1599 * host and device may alter it using
1600 * blk_queue_update_dma_alignment() later.
1601 */
1602 blk_queue_dma_alignment(q, 0x03);
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 return q;
1605}
FUJITA Tomonorib58d9152006-11-16 19:24:10 +09001606EXPORT_SYMBOL(__scsi_alloc_queue);
1607
1608struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1609{
1610 struct request_queue *q;
1611
1612 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1613 if (!q)
1614 return NULL;
1615
1616 blk_queue_prep_rq(q, scsi_prep_fn);
FUJITA Tomonorib58d9152006-11-16 19:24:10 +09001617 blk_queue_softirq_done(q, scsi_softirq_done);
1618 return q;
1619}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
1621void scsi_free_queue(struct request_queue *q)
1622{
1623 blk_cleanup_queue(q);
1624}
1625
1626/*
1627 * Function: scsi_block_requests()
1628 *
1629 * Purpose: Utility function used by low-level drivers to prevent further
1630 * commands from being queued to the device.
1631 *
1632 * Arguments: shost - Host in question
1633 *
1634 * Returns: Nothing
1635 *
1636 * Lock status: No locks are assumed held.
1637 *
1638 * Notes: There is no timer nor any other means by which the requests
1639 * get unblocked other than the low-level driver calling
1640 * scsi_unblock_requests().
1641 */
1642void scsi_block_requests(struct Scsi_Host *shost)
1643{
1644 shost->host_self_blocked = 1;
1645}
1646EXPORT_SYMBOL(scsi_block_requests);
1647
1648/*
1649 * Function: scsi_unblock_requests()
1650 *
1651 * Purpose: Utility function used by low-level drivers to allow further
1652 * commands from being queued to the device.
1653 *
1654 * Arguments: shost - Host in question
1655 *
1656 * Returns: Nothing
1657 *
1658 * Lock status: No locks are assumed held.
1659 *
1660 * Notes: There is no timer nor any other means by which the requests
1661 * get unblocked other than the low-level driver calling
1662 * scsi_unblock_requests().
1663 *
1664 * This is done as an API function so that changes to the
1665 * internals of the scsi mid-layer won't require wholesale
1666 * changes to drivers that use this feature.
1667 */
1668void scsi_unblock_requests(struct Scsi_Host *shost)
1669{
1670 shost->host_self_blocked = 0;
1671 scsi_run_host_queues(shost);
1672}
1673EXPORT_SYMBOL(scsi_unblock_requests);
1674
1675int __init scsi_init_queue(void)
1676{
1677 int i;
1678
Mike Christieaa7b5cd2005-11-11 05:31:41 -06001679 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1680 sizeof(struct scsi_io_context),
Paul Mundt20c2df82007-07-20 10:11:58 +09001681 0, 0, NULL);
Mike Christieaa7b5cd2005-11-11 05:31:41 -06001682 if (!scsi_io_context_cache) {
1683 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1684 return -ENOMEM;
1685 }
1686
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +02001687 scsi_bidi_sdb_cache = kmem_cache_create("scsi_bidi_sdb",
1688 sizeof(struct scsi_data_buffer),
1689 0, 0, NULL);
1690 if (!scsi_bidi_sdb_cache) {
1691 printk(KERN_ERR "SCSI: can't init scsi bidi sdb cache\n");
FUJITA Tomonori3d9dd6e2008-01-25 23:25:14 +09001692 goto cleanup_io_context;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +02001693 }
1694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1696 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1697 int size = sgp->size * sizeof(struct scatterlist);
1698
1699 sgp->slab = kmem_cache_create(sgp->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001700 SLAB_HWCACHE_ALIGN, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 if (!sgp->slab) {
1702 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1703 sgp->name);
FUJITA Tomonori3d9dd6e2008-01-25 23:25:14 +09001704 goto cleanup_bidi_sdb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 }
1706
Matthew Dobson93d23412006-03-26 01:37:50 -08001707 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1708 sgp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 if (!sgp->pool) {
1710 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1711 sgp->name);
FUJITA Tomonori3d9dd6e2008-01-25 23:25:14 +09001712 goto cleanup_bidi_sdb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 }
1714 }
1715
1716 return 0;
FUJITA Tomonori3d9dd6e2008-01-25 23:25:14 +09001717
1718cleanup_bidi_sdb:
1719 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1720 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1721 if (sgp->pool)
1722 mempool_destroy(sgp->pool);
1723 if (sgp->slab)
1724 kmem_cache_destroy(sgp->slab);
1725 }
1726 kmem_cache_destroy(scsi_bidi_sdb_cache);
1727cleanup_io_context:
1728 kmem_cache_destroy(scsi_io_context_cache);
1729
1730 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731}
1732
1733void scsi_exit_queue(void)
1734{
1735 int i;
1736
Mike Christieaa7b5cd2005-11-11 05:31:41 -06001737 kmem_cache_destroy(scsi_io_context_cache);
FUJITA Tomonorib172b6e2008-01-25 23:25:13 +09001738 kmem_cache_destroy(scsi_bidi_sdb_cache);
Mike Christieaa7b5cd2005-11-11 05:31:41 -06001739
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1741 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1742 mempool_destroy(sgp->pool);
1743 kmem_cache_destroy(sgp->slab);
1744 }
1745}
James Bottomley5baba832006-03-18 14:10:35 -06001746
1747/**
1748 * scsi_mode_select - issue a mode select
1749 * @sdev: SCSI device to be queried
1750 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1751 * @sp: Save page bit (0 == don't save, 1 == save)
1752 * @modepage: mode page being requested
1753 * @buffer: request buffer (may not be smaller than eight bytes)
1754 * @len: length of request buffer.
1755 * @timeout: command timeout
1756 * @retries: number of retries before failing
1757 * @data: returns a structure abstracting the mode header data
Rob Landleyeb44820c2007-11-03 13:30:39 -05001758 * @sshdr: place to put sense data (or NULL if no sense to be collected).
James Bottomley5baba832006-03-18 14:10:35 -06001759 * must be SCSI_SENSE_BUFFERSIZE big.
1760 *
1761 * Returns zero if successful; negative error number or scsi
1762 * status on error
1763 *
1764 */
1765int
1766scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1767 unsigned char *buffer, int len, int timeout, int retries,
1768 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1769{
1770 unsigned char cmd[10];
1771 unsigned char *real_buffer;
1772 int ret;
1773
1774 memset(cmd, 0, sizeof(cmd));
1775 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1776
1777 if (sdev->use_10_for_ms) {
1778 if (len > 65535)
1779 return -EINVAL;
1780 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1781 if (!real_buffer)
1782 return -ENOMEM;
1783 memcpy(real_buffer + 8, buffer, len);
1784 len += 8;
1785 real_buffer[0] = 0;
1786 real_buffer[1] = 0;
1787 real_buffer[2] = data->medium_type;
1788 real_buffer[3] = data->device_specific;
1789 real_buffer[4] = data->longlba ? 0x01 : 0;
1790 real_buffer[5] = 0;
1791 real_buffer[6] = data->block_descriptor_length >> 8;
1792 real_buffer[7] = data->block_descriptor_length;
1793
1794 cmd[0] = MODE_SELECT_10;
1795 cmd[7] = len >> 8;
1796 cmd[8] = len;
1797 } else {
1798 if (len > 255 || data->block_descriptor_length > 255 ||
1799 data->longlba)
1800 return -EINVAL;
1801
1802 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1803 if (!real_buffer)
1804 return -ENOMEM;
1805 memcpy(real_buffer + 4, buffer, len);
1806 len += 4;
1807 real_buffer[0] = 0;
1808 real_buffer[1] = data->medium_type;
1809 real_buffer[2] = data->device_specific;
1810 real_buffer[3] = data->block_descriptor_length;
1811
1812
1813 cmd[0] = MODE_SELECT;
1814 cmd[4] = len;
1815 }
1816
1817 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1818 sshdr, timeout, retries);
1819 kfree(real_buffer);
1820 return ret;
1821}
1822EXPORT_SYMBOL_GPL(scsi_mode_select);
1823
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824/**
Rob Landleyeb44820c2007-11-03 13:30:39 -05001825 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
James Bottomley1cf72692005-08-28 11:27:01 -05001826 * @sdev: SCSI device to be queried
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 * @dbd: set if mode sense will allow block descriptors to be returned
1828 * @modepage: mode page being requested
1829 * @buffer: request buffer (may not be smaller than eight bytes)
1830 * @len: length of request buffer.
1831 * @timeout: command timeout
1832 * @retries: number of retries before failing
1833 * @data: returns a structure abstracting the mode header data
Rob Landleyeb44820c2007-11-03 13:30:39 -05001834 * @sshdr: place to put sense data (or NULL if no sense to be collected).
James Bottomley1cf72692005-08-28 11:27:01 -05001835 * must be SCSI_SENSE_BUFFERSIZE big.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 *
1837 * Returns zero if unsuccessful, or the header offset (either 4
1838 * or 8 depending on whether a six or ten byte command was
1839 * issued) if successful.
Rob Landleyeb44820c2007-11-03 13:30:39 -05001840 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841int
James Bottomley1cf72692005-08-28 11:27:01 -05001842scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 unsigned char *buffer, int len, int timeout, int retries,
James Bottomley5baba832006-03-18 14:10:35 -06001844 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1845{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 unsigned char cmd[12];
1847 int use_10_for_ms;
1848 int header_length;
James Bottomley1cf72692005-08-28 11:27:01 -05001849 int result;
James Bottomleyea73a9f2005-08-28 11:33:52 -05001850 struct scsi_sense_hdr my_sshdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
1852 memset(data, 0, sizeof(*data));
1853 memset(&cmd[0], 0, 12);
1854 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1855 cmd[2] = modepage;
1856
James Bottomleyea73a9f2005-08-28 11:33:52 -05001857 /* caller might not be interested in sense, but we need it */
1858 if (!sshdr)
1859 sshdr = &my_sshdr;
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 retry:
James Bottomley1cf72692005-08-28 11:27:01 -05001862 use_10_for_ms = sdev->use_10_for_ms;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
1864 if (use_10_for_ms) {
1865 if (len < 8)
1866 len = 8;
1867
1868 cmd[0] = MODE_SENSE_10;
1869 cmd[8] = len;
1870 header_length = 8;
1871 } else {
1872 if (len < 4)
1873 len = 4;
1874
1875 cmd[0] = MODE_SENSE;
1876 cmd[4] = len;
1877 header_length = 4;
1878 }
1879
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 memset(buffer, 0, len);
1881
James Bottomley1cf72692005-08-28 11:27:01 -05001882 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
James Bottomleyea73a9f2005-08-28 11:33:52 -05001883 sshdr, timeout, retries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
1885 /* This code looks awful: what it's doing is making sure an
1886 * ILLEGAL REQUEST sense return identifies the actual command
1887 * byte as the problem. MODE_SENSE commands can return
1888 * ILLEGAL REQUEST if the code page isn't supported */
1889
James Bottomley1cf72692005-08-28 11:27:01 -05001890 if (use_10_for_ms && !scsi_status_is_good(result) &&
1891 (driver_byte(result) & DRIVER_SENSE)) {
James Bottomleyea73a9f2005-08-28 11:33:52 -05001892 if (scsi_sense_valid(sshdr)) {
1893 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1894 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 /*
1896 * Invalid command operation code
1897 */
James Bottomley1cf72692005-08-28 11:27:01 -05001898 sdev->use_10_for_ms = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 goto retry;
1900 }
1901 }
1902 }
1903
James Bottomley1cf72692005-08-28 11:27:01 -05001904 if(scsi_status_is_good(result)) {
Al Viro6d73c852006-02-23 02:03:16 +01001905 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1906 (modepage == 6 || modepage == 8))) {
1907 /* Initio breakage? */
1908 header_length = 0;
1909 data->length = 13;
1910 data->medium_type = 0;
1911 data->device_specific = 0;
1912 data->longlba = 0;
1913 data->block_descriptor_length = 0;
1914 } else if(use_10_for_ms) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 data->length = buffer[0]*256 + buffer[1] + 2;
1916 data->medium_type = buffer[2];
1917 data->device_specific = buffer[3];
1918 data->longlba = buffer[4] & 0x01;
1919 data->block_descriptor_length = buffer[6]*256
1920 + buffer[7];
1921 } else {
1922 data->length = buffer[0] + 1;
1923 data->medium_type = buffer[1];
1924 data->device_specific = buffer[2];
1925 data->block_descriptor_length = buffer[3];
1926 }
Al Viro6d73c852006-02-23 02:03:16 +01001927 data->header_length = header_length;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 }
1929
James Bottomley1cf72692005-08-28 11:27:01 -05001930 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931}
1932EXPORT_SYMBOL(scsi_mode_sense);
1933
James Bottomley001aac22007-12-02 19:10:40 +02001934/**
1935 * scsi_test_unit_ready - test if unit is ready
1936 * @sdev: scsi device to change the state of.
1937 * @timeout: command timeout
1938 * @retries: number of retries before failing
1939 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
1940 * returning sense. Make sure that this is cleared before passing
1941 * in.
1942 *
1943 * Returns zero if unsuccessful or an error if TUR failed. For
1944 * removable media, a return of NOT_READY or UNIT_ATTENTION is
1945 * translated to success, with the ->changed flag updated.
1946 **/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947int
James Bottomley001aac22007-12-02 19:10:40 +02001948scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1949 struct scsi_sense_hdr *sshdr_external)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 char cmd[] = {
1952 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1953 };
James Bottomley001aac22007-12-02 19:10:40 +02001954 struct scsi_sense_hdr *sshdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 int result;
James Bottomley001aac22007-12-02 19:10:40 +02001956
1957 if (!sshdr_external)
1958 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1959 else
1960 sshdr = sshdr_external;
1961
1962 /* try to eat the UNIT_ATTENTION if there are enough retries */
1963 do {
1964 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
1965 timeout, retries);
1966 } while ((driver_byte(result) & DRIVER_SENSE) &&
1967 sshdr && sshdr->sense_key == UNIT_ATTENTION &&
1968 --retries);
1969
1970 if (!sshdr)
1971 /* could not allocate sense buffer, so can't process it */
1972 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
James Bottomley1cf72692005-08-28 11:27:01 -05001974 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
James Bottomley001aac22007-12-02 19:10:40 +02001976 if ((scsi_sense_valid(sshdr)) &&
1977 ((sshdr->sense_key == UNIT_ATTENTION) ||
1978 (sshdr->sense_key == NOT_READY))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 sdev->changed = 1;
James Bottomley1cf72692005-08-28 11:27:01 -05001980 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
1982 }
James Bottomley001aac22007-12-02 19:10:40 +02001983 if (!sshdr_external)
1984 kfree(sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 return result;
1986}
1987EXPORT_SYMBOL(scsi_test_unit_ready);
1988
1989/**
Rob Landleyeb44820c2007-11-03 13:30:39 -05001990 * scsi_device_set_state - Take the given device through the device state model.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 * @sdev: scsi device to change the state of.
1992 * @state: state to change to.
1993 *
1994 * Returns zero if unsuccessful or an error if the requested
1995 * transition is illegal.
Rob Landleyeb44820c2007-11-03 13:30:39 -05001996 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997int
1998scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
1999{
2000 enum scsi_device_state oldstate = sdev->sdev_state;
2001
2002 if (state == oldstate)
2003 return 0;
2004
2005 switch (state) {
2006 case SDEV_CREATED:
2007 /* There are no legal states that come back to
2008 * created. This is the manually initialised start
2009 * state */
2010 goto illegal;
2011
2012 case SDEV_RUNNING:
2013 switch (oldstate) {
2014 case SDEV_CREATED:
2015 case SDEV_OFFLINE:
2016 case SDEV_QUIESCE:
2017 case SDEV_BLOCK:
2018 break;
2019 default:
2020 goto illegal;
2021 }
2022 break;
2023
2024 case SDEV_QUIESCE:
2025 switch (oldstate) {
2026 case SDEV_RUNNING:
2027 case SDEV_OFFLINE:
2028 break;
2029 default:
2030 goto illegal;
2031 }
2032 break;
2033
2034 case SDEV_OFFLINE:
2035 switch (oldstate) {
2036 case SDEV_CREATED:
2037 case SDEV_RUNNING:
2038 case SDEV_QUIESCE:
2039 case SDEV_BLOCK:
2040 break;
2041 default:
2042 goto illegal;
2043 }
2044 break;
2045
2046 case SDEV_BLOCK:
2047 switch (oldstate) {
2048 case SDEV_CREATED:
2049 case SDEV_RUNNING:
2050 break;
2051 default:
2052 goto illegal;
2053 }
2054 break;
2055
2056 case SDEV_CANCEL:
2057 switch (oldstate) {
2058 case SDEV_CREATED:
2059 case SDEV_RUNNING:
Alan Stern9ea72902006-06-23 14:25:34 -04002060 case SDEV_QUIESCE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 case SDEV_OFFLINE:
2062 case SDEV_BLOCK:
2063 break;
2064 default:
2065 goto illegal;
2066 }
2067 break;
2068
2069 case SDEV_DEL:
2070 switch (oldstate) {
Brian King309bd272006-06-27 11:10:43 -05002071 case SDEV_CREATED:
2072 case SDEV_RUNNING:
2073 case SDEV_OFFLINE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 case SDEV_CANCEL:
2075 break;
2076 default:
2077 goto illegal;
2078 }
2079 break;
2080
2081 }
2082 sdev->sdev_state = state;
2083 return 0;
2084
2085 illegal:
2086 SCSI_LOG_ERROR_RECOVERY(1,
James Bottomley9ccfc752005-10-02 11:45:08 -05002087 sdev_printk(KERN_ERR, sdev,
2088 "Illegal state transition %s->%s\n",
2089 scsi_device_state_name(oldstate),
2090 scsi_device_state_name(state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 );
2092 return -EINVAL;
2093}
2094EXPORT_SYMBOL(scsi_device_set_state);
2095
2096/**
Jeff Garzika341cd0f2007-10-29 17:15:22 -04002097 * sdev_evt_emit - emit a single SCSI device uevent
2098 * @sdev: associated SCSI device
2099 * @evt: event to emit
2100 *
2101 * Send a single uevent (scsi_event) to the associated scsi_device.
2102 */
2103static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2104{
2105 int idx = 0;
2106 char *envp[3];
2107
2108 switch (evt->evt_type) {
2109 case SDEV_EVT_MEDIA_CHANGE:
2110 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2111 break;
2112
2113 default:
2114 /* do nothing */
2115 break;
2116 }
2117
2118 envp[idx++] = NULL;
2119
2120 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2121}
2122
2123/**
2124 * sdev_evt_thread - send a uevent for each scsi event
2125 * @work: work struct for scsi_device
2126 *
2127 * Dispatch queued events to their associated scsi_device kobjects
2128 * as uevents.
2129 */
2130void scsi_evt_thread(struct work_struct *work)
2131{
2132 struct scsi_device *sdev;
2133 LIST_HEAD(event_list);
2134
2135 sdev = container_of(work, struct scsi_device, event_work);
2136
2137 while (1) {
2138 struct scsi_event *evt;
2139 struct list_head *this, *tmp;
2140 unsigned long flags;
2141
2142 spin_lock_irqsave(&sdev->list_lock, flags);
2143 list_splice_init(&sdev->event_list, &event_list);
2144 spin_unlock_irqrestore(&sdev->list_lock, flags);
2145
2146 if (list_empty(&event_list))
2147 break;
2148
2149 list_for_each_safe(this, tmp, &event_list) {
2150 evt = list_entry(this, struct scsi_event, node);
2151 list_del(&evt->node);
2152 scsi_evt_emit(sdev, evt);
2153 kfree(evt);
2154 }
2155 }
2156}
2157
2158/**
2159 * sdev_evt_send - send asserted event to uevent thread
2160 * @sdev: scsi_device event occurred on
2161 * @evt: event to send
2162 *
2163 * Assert scsi device event asynchronously.
2164 */
2165void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2166{
2167 unsigned long flags;
2168
Kay Sievers4d1566e2008-03-19 13:04:47 +01002169#if 0
2170 /* FIXME: currently this check eliminates all media change events
2171 * for polled devices. Need to update to discriminate between AN
2172 * and polled events */
Jeff Garzika341cd0f2007-10-29 17:15:22 -04002173 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2174 kfree(evt);
2175 return;
2176 }
Kay Sievers4d1566e2008-03-19 13:04:47 +01002177#endif
Jeff Garzika341cd0f2007-10-29 17:15:22 -04002178
2179 spin_lock_irqsave(&sdev->list_lock, flags);
2180 list_add_tail(&evt->node, &sdev->event_list);
2181 schedule_work(&sdev->event_work);
2182 spin_unlock_irqrestore(&sdev->list_lock, flags);
2183}
2184EXPORT_SYMBOL_GPL(sdev_evt_send);
2185
2186/**
2187 * sdev_evt_alloc - allocate a new scsi event
2188 * @evt_type: type of event to allocate
2189 * @gfpflags: GFP flags for allocation
2190 *
2191 * Allocates and returns a new scsi_event.
2192 */
2193struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2194 gfp_t gfpflags)
2195{
2196 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2197 if (!evt)
2198 return NULL;
2199
2200 evt->evt_type = evt_type;
2201 INIT_LIST_HEAD(&evt->node);
2202
2203 /* evt_type-specific initialization, if any */
2204 switch (evt_type) {
2205 case SDEV_EVT_MEDIA_CHANGE:
2206 default:
2207 /* do nothing */
2208 break;
2209 }
2210
2211 return evt;
2212}
2213EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2214
2215/**
2216 * sdev_evt_send_simple - send asserted event to uevent thread
2217 * @sdev: scsi_device event occurred on
2218 * @evt_type: type of event to send
2219 * @gfpflags: GFP flags for allocation
2220 *
2221 * Assert scsi device event asynchronously, given an event type.
2222 */
2223void sdev_evt_send_simple(struct scsi_device *sdev,
2224 enum scsi_device_event evt_type, gfp_t gfpflags)
2225{
2226 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2227 if (!evt) {
2228 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2229 evt_type);
2230 return;
2231 }
2232
2233 sdev_evt_send(sdev, evt);
2234}
2235EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2236
2237/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 * scsi_device_quiesce - Block user issued commands.
2239 * @sdev: scsi device to quiesce.
2240 *
2241 * This works by trying to transition to the SDEV_QUIESCE state
2242 * (which must be a legal transition). When the device is in this
2243 * state, only special requests will be accepted, all others will
2244 * be deferred. Since special requests may also be requeued requests,
2245 * a successful return doesn't guarantee the device will be
2246 * totally quiescent.
2247 *
2248 * Must be called with user context, may sleep.
2249 *
2250 * Returns zero if unsuccessful or an error if not.
Rob Landleyeb44820c2007-11-03 13:30:39 -05002251 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252int
2253scsi_device_quiesce(struct scsi_device *sdev)
2254{
2255 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2256 if (err)
2257 return err;
2258
2259 scsi_run_queue(sdev->request_queue);
2260 while (sdev->device_busy) {
2261 msleep_interruptible(200);
2262 scsi_run_queue(sdev->request_queue);
2263 }
2264 return 0;
2265}
2266EXPORT_SYMBOL(scsi_device_quiesce);
2267
2268/**
2269 * scsi_device_resume - Restart user issued commands to a quiesced device.
2270 * @sdev: scsi device to resume.
2271 *
2272 * Moves the device from quiesced back to running and restarts the
2273 * queues.
2274 *
2275 * Must be called with user context, may sleep.
Rob Landleyeb44820c2007-11-03 13:30:39 -05002276 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277void
2278scsi_device_resume(struct scsi_device *sdev)
2279{
2280 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2281 return;
2282 scsi_run_queue(sdev->request_queue);
2283}
2284EXPORT_SYMBOL(scsi_device_resume);
2285
2286static void
2287device_quiesce_fn(struct scsi_device *sdev, void *data)
2288{
2289 scsi_device_quiesce(sdev);
2290}
2291
2292void
2293scsi_target_quiesce(struct scsi_target *starget)
2294{
2295 starget_for_each_device(starget, NULL, device_quiesce_fn);
2296}
2297EXPORT_SYMBOL(scsi_target_quiesce);
2298
2299static void
2300device_resume_fn(struct scsi_device *sdev, void *data)
2301{
2302 scsi_device_resume(sdev);
2303}
2304
2305void
2306scsi_target_resume(struct scsi_target *starget)
2307{
2308 starget_for_each_device(starget, NULL, device_resume_fn);
2309}
2310EXPORT_SYMBOL(scsi_target_resume);
2311
2312/**
Rob Landleyeb44820c2007-11-03 13:30:39 -05002313 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 * @sdev: device to block
2315 *
2316 * Block request made by scsi lld's to temporarily stop all
2317 * scsi commands on the specified device. Called from interrupt
2318 * or normal process context.
2319 *
2320 * Returns zero if successful or error if not
2321 *
2322 * Notes:
2323 * This routine transitions the device to the SDEV_BLOCK state
2324 * (which must be a legal transition). When the device is in this
2325 * state, all commands are deferred until the scsi lld reenables
2326 * the device with scsi_device_unblock or device_block_tmo fires.
2327 * This routine assumes the host_lock is held on entry.
Rob Landleyeb44820c2007-11-03 13:30:39 -05002328 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329int
2330scsi_internal_device_block(struct scsi_device *sdev)
2331{
Jens Axboe165125e2007-07-24 09:28:11 +02002332 struct request_queue *q = sdev->request_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 unsigned long flags;
2334 int err = 0;
2335
2336 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2337 if (err)
2338 return err;
2339
2340 /*
2341 * The device has transitioned to SDEV_BLOCK. Stop the
2342 * block layer from calling the midlayer with this device's
2343 * request queue.
2344 */
2345 spin_lock_irqsave(q->queue_lock, flags);
2346 blk_stop_queue(q);
2347 spin_unlock_irqrestore(q->queue_lock, flags);
2348
2349 return 0;
2350}
2351EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2352
2353/**
2354 * scsi_internal_device_unblock - resume a device after a block request
2355 * @sdev: device to resume
2356 *
2357 * Called by scsi lld's or the midlayer to restart the device queue
2358 * for the previously suspended scsi device. Called from interrupt or
2359 * normal process context.
2360 *
2361 * Returns zero if successful or error if not.
2362 *
2363 * Notes:
2364 * This routine transitions the device to the SDEV_RUNNING state
2365 * (which must be a legal transition) allowing the midlayer to
2366 * goose the queue for this device. This routine assumes the
2367 * host_lock is held upon entry.
Rob Landleyeb44820c2007-11-03 13:30:39 -05002368 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369int
2370scsi_internal_device_unblock(struct scsi_device *sdev)
2371{
Jens Axboe165125e2007-07-24 09:28:11 +02002372 struct request_queue *q = sdev->request_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 int err;
2374 unsigned long flags;
2375
2376 /*
2377 * Try to transition the scsi device to SDEV_RUNNING
2378 * and goose the device queue if successful.
2379 */
2380 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2381 if (err)
2382 return err;
2383
2384 spin_lock_irqsave(q->queue_lock, flags);
2385 blk_start_queue(q);
2386 spin_unlock_irqrestore(q->queue_lock, flags);
2387
2388 return 0;
2389}
2390EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2391
2392static void
2393device_block(struct scsi_device *sdev, void *data)
2394{
2395 scsi_internal_device_block(sdev);
2396}
2397
2398static int
2399target_block(struct device *dev, void *data)
2400{
2401 if (scsi_is_target_device(dev))
2402 starget_for_each_device(to_scsi_target(dev), NULL,
2403 device_block);
2404 return 0;
2405}
2406
2407void
2408scsi_target_block(struct device *dev)
2409{
2410 if (scsi_is_target_device(dev))
2411 starget_for_each_device(to_scsi_target(dev), NULL,
2412 device_block);
2413 else
2414 device_for_each_child(dev, NULL, target_block);
2415}
2416EXPORT_SYMBOL_GPL(scsi_target_block);
2417
2418static void
2419device_unblock(struct scsi_device *sdev, void *data)
2420{
2421 scsi_internal_device_unblock(sdev);
2422}
2423
2424static int
2425target_unblock(struct device *dev, void *data)
2426{
2427 if (scsi_is_target_device(dev))
2428 starget_for_each_device(to_scsi_target(dev), NULL,
2429 device_unblock);
2430 return 0;
2431}
2432
2433void
2434scsi_target_unblock(struct device *dev)
2435{
2436 if (scsi_is_target_device(dev))
2437 starget_for_each_device(to_scsi_target(dev), NULL,
2438 device_unblock);
2439 else
2440 device_for_each_child(dev, NULL, target_unblock);
2441}
2442EXPORT_SYMBOL_GPL(scsi_target_unblock);
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002443
2444/**
2445 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
Rob Landleyeb44820c2007-11-03 13:30:39 -05002446 * @sgl: scatter-gather list
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002447 * @sg_count: number of segments in sg
2448 * @offset: offset in bytes into sg, on return offset into the mapped area
2449 * @len: bytes to map, on return number of bytes mapped
2450 *
2451 * Returns virtual address of the start of the mapped page
2452 */
Jens Axboec6132da2007-10-16 11:08:49 +02002453void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002454 size_t *offset, size_t *len)
2455{
2456 int i;
2457 size_t sg_len = 0, len_complete = 0;
Jens Axboec6132da2007-10-16 11:08:49 +02002458 struct scatterlist *sg;
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002459 struct page *page;
2460
Andrew Morton22cfefb2007-02-05 16:39:03 -08002461 WARN_ON(!irqs_disabled());
2462
Jens Axboec6132da2007-10-16 11:08:49 +02002463 for_each_sg(sgl, sg, sg_count, i) {
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002464 len_complete = sg_len; /* Complete sg-entries */
Jens Axboec6132da2007-10-16 11:08:49 +02002465 sg_len += sg->length;
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002466 if (sg_len > *offset)
2467 break;
2468 }
2469
2470 if (unlikely(i == sg_count)) {
Andrew Morton169e1a22006-04-18 21:09:08 -07002471 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2472 "elements %d\n",
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002473 __FUNCTION__, sg_len, *offset, sg_count);
2474 WARN_ON(1);
2475 return NULL;
2476 }
2477
2478 /* Offset starting from the beginning of first page in this sg-entry */
Jens Axboec6132da2007-10-16 11:08:49 +02002479 *offset = *offset - len_complete + sg->offset;
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002480
2481 /* Assumption: contiguous pages can be accessed as "page + i" */
Jens Axboe45711f12007-10-22 21:19:53 +02002482 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002483 *offset &= ~PAGE_MASK;
2484
2485 /* Bytes in this sg-entry from *offset to the end of the page */
2486 sg_len = PAGE_SIZE - *offset;
2487 if (*len > sg_len)
2488 *len = sg_len;
2489
2490 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2491}
2492EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2493
2494/**
Rob Landleyeb44820c2007-11-03 13:30:39 -05002495 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002496 * @virt: virtual address to be unmapped
2497 */
2498void scsi_kunmap_atomic_sg(void *virt)
2499{
2500 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2501}
2502EXPORT_SYMBOL(scsi_kunmap_atomic_sg);