blob: 3d04a9f386acc51744aad2c098b7ee3b9abeb518 [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>
11#include <linux/blkdev.h>
12#include <linux/completion.h>
13#include <linux/kernel.h>
14#include <linux/mempool.h>
15#include <linux/slab.h>
16#include <linux/init.h>
17#include <linux/pci.h>
18#include <linux/delay.h>
James Bottomleyfaead262006-02-14 10:42:07 -060019#include <linux/hardirq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
21#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020022#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <scsi/scsi_dbg.h>
24#include <scsi/scsi_device.h>
25#include <scsi/scsi_driver.h>
26#include <scsi/scsi_eh.h>
27#include <scsi/scsi_host.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29#include "scsi_priv.h"
30#include "scsi_logging.h"
31
32
Tobias Klauser6391a112006-06-08 22:23:48 -070033#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#define SG_MEMPOOL_SIZE 32
35
36struct scsi_host_sg_pool {
37 size_t size;
38 char *name;
39 kmem_cache_t *slab;
40 mempool_t *pool;
41};
42
43#if (SCSI_MAX_PHYS_SEGMENTS < 32)
44#error SCSI_MAX_PHYS_SEGMENTS is too small
45#endif
46
47#define SP(x) { x, "sgpool-" #x }
Adrian Bunk52c1da32005-06-23 22:05:33 -070048static struct scsi_host_sg_pool scsi_sg_pools[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 SP(8),
50 SP(16),
51 SP(32),
52#if (SCSI_MAX_PHYS_SEGMENTS > 32)
53 SP(64),
54#if (SCSI_MAX_PHYS_SEGMENTS > 64)
55 SP(128),
56#if (SCSI_MAX_PHYS_SEGMENTS > 128)
57 SP(256),
58#if (SCSI_MAX_PHYS_SEGMENTS > 256)
59#error SCSI_MAX_PHYS_SEGMENTS is too large
60#endif
61#endif
62#endif
63#endif
64};
65#undef SP
66
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -050067static void scsi_run_queue(struct request_queue *q);
James Bottomleye91442b2005-09-09 10:44:16 -050068
69/*
70 * Function: scsi_unprep_request()
71 *
72 * Purpose: Remove all preparation done for a request, including its
73 * associated scsi_cmnd, so that it can be requeued.
74 *
75 * Arguments: req - request to unprepare
76 *
77 * Lock status: Assumed that no locks are held upon entry.
78 *
79 * Returns: Nothing.
80 */
81static void scsi_unprep_request(struct request *req)
82{
83 struct scsi_cmnd *cmd = req->special;
84
85 req->flags &= ~REQ_DONTPREP;
Christoph Hellwigbeb40482006-06-10 18:01:03 +020086 req->special = NULL;
James Bottomleye91442b2005-09-09 10:44:16 -050087
James Bottomleye91442b2005-09-09 10:44:16 -050088 scsi_put_command(cmd);
89}
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -050090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * Function: scsi_queue_insert()
93 *
94 * Purpose: Insert a command in the midlevel queue.
95 *
96 * Arguments: cmd - command that we are adding to queue.
97 * reason - why we are inserting command to queue.
98 *
99 * Lock status: Assumed that lock is not held upon entry.
100 *
101 * Returns: Nothing.
102 *
103 * Notes: We do this for one of two cases. Either the host is busy
104 * and it cannot accept any more commands for the time being,
105 * or the device returned QUEUE_FULL and can accept no more
106 * commands.
107 * Notes: This could be called either from an interrupt context or a
108 * normal process context.
109 */
110int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
111{
112 struct Scsi_Host *host = cmd->device->host;
113 struct scsi_device *device = cmd->device;
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500114 struct request_queue *q = device->request_queue;
115 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117 SCSI_LOG_MLQUEUE(1,
118 printk("Inserting command %p into mlqueue\n", cmd));
119
120 /*
Tejun Heo d8c37e72005-05-14 00:46:08 +0900121 * Set the appropriate busy bit for the device/host.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 *
123 * If the host/device isn't busy, assume that something actually
124 * completed, and that we should be able to queue a command now.
125 *
126 * Note that the prior mid-layer assumption that any host could
127 * always queue at least one command is now broken. The mid-layer
128 * will implement a user specifiable stall (see
129 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
130 * if a command is requeued with no other commands outstanding
131 * either for the device or for the host.
132 */
133 if (reason == SCSI_MLQUEUE_HOST_BUSY)
134 host->host_blocked = host->max_host_blocked;
135 else if (reason == SCSI_MLQUEUE_DEVICE_BUSY)
136 device->device_blocked = device->max_device_blocked;
137
138 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 * Decrement the counters, since these commands are no longer
140 * active on the host/device.
141 */
142 scsi_device_unbusy(device);
143
144 /*
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500145 * Requeue this command. It will go before all other commands
146 * that are already in the queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 *
148 * NOTE: there is magic here about the way the queue is plugged if
149 * we have no outstanding commands.
150 *
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500151 * Although we *don't* plug the queue, we call the request
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 * function. The SCSI request function detects the blocked condition
153 * and plugs the queue appropriately.
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500154 */
155 spin_lock_irqsave(q->queue_lock, flags);
James Bottomley59897da2005-09-14 12:57:42 -0400156 blk_requeue_request(q, cmd->request);
Tejun Heo a1bf9d1d2005-04-24 02:08:52 -0500157 spin_unlock_irqrestore(q->queue_lock, flags);
158
159 scsi_run_queue(q);
160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 return 0;
162}
163
James Bottomley39216032005-06-15 18:48:29 -0500164/**
James Bottomley33aa6872005-08-28 11:31:14 -0500165 * scsi_execute - insert request and wait for the result
James Bottomley39216032005-06-15 18:48:29 -0500166 * @sdev: scsi device
167 * @cmd: scsi command
168 * @data_direction: data direction
169 * @buffer: data buffer
170 * @bufflen: len of buffer
171 * @sense: optional sense buffer
172 * @timeout: request timeout in seconds
173 * @retries: number of times to retry request
James Bottomley33aa6872005-08-28 11:31:14 -0500174 * @flags: or into request flags;
James Bottomley39216032005-06-15 18:48:29 -0500175 *
James Bottomleyea73a9f2005-08-28 11:33:52 -0500176 * returns the req->errors value which is the the scsi_cmnd result
177 * field.
James Bottomley39216032005-06-15 18:48:29 -0500178 **/
James Bottomley33aa6872005-08-28 11:31:14 -0500179int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
180 int data_direction, void *buffer, unsigned bufflen,
181 unsigned char *sense, int timeout, int retries, int flags)
James Bottomley39216032005-06-15 18:48:29 -0500182{
183 struct request *req;
184 int write = (data_direction == DMA_TO_DEVICE);
185 int ret = DRIVER_ERROR << 24;
186
187 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
188
189 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
190 buffer, bufflen, __GFP_WAIT))
191 goto out;
192
193 req->cmd_len = COMMAND_SIZE(cmd[0]);
194 memcpy(req->cmd, cmd, req->cmd_len);
195 req->sense = sense;
196 req->sense_len = 0;
Mike Christie17e01f22005-11-11 05:31:37 -0600197 req->retries = retries;
James Bottomley39216032005-06-15 18:48:29 -0500198 req->timeout = timeout;
James Bottomley3173d8c2005-09-04 11:32:05 -0500199 req->flags |= flags | REQ_BLOCK_PC | REQ_SPECIAL | REQ_QUIET;
James Bottomley39216032005-06-15 18:48:29 -0500200
201 /*
202 * head injection *required* here otherwise quiesce won't work
203 */
204 blk_execute_rq(req->q, NULL, req, 1);
205
206 ret = req->errors;
207 out:
208 blk_put_request(req);
209
210 return ret;
211}
James Bottomley33aa6872005-08-28 11:31:14 -0500212EXPORT_SYMBOL(scsi_execute);
James Bottomley39216032005-06-15 18:48:29 -0500213
James Bottomleyea73a9f2005-08-28 11:33:52 -0500214
215int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
216 int data_direction, void *buffer, unsigned bufflen,
217 struct scsi_sense_hdr *sshdr, int timeout, int retries)
218{
219 char *sense = NULL;
akpm@osdl.org1ccb48b2005-06-26 00:12:51 -0700220 int result;
221
James Bottomleyea73a9f2005-08-28 11:33:52 -0500222 if (sshdr) {
Jes Sorensen24669f752006-01-16 10:31:18 -0500223 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
James Bottomleyea73a9f2005-08-28 11:33:52 -0500224 if (!sense)
225 return DRIVER_ERROR << 24;
James Bottomleyea73a9f2005-08-28 11:33:52 -0500226 }
akpm@osdl.org1ccb48b2005-06-26 00:12:51 -0700227 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
Jes Sorensen24669f752006-01-16 10:31:18 -0500228 sense, timeout, retries, 0);
James Bottomleyea73a9f2005-08-28 11:33:52 -0500229 if (sshdr)
James Bottomleye5143852005-08-09 11:55:36 -0500230 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
James Bottomleyea73a9f2005-08-28 11:33:52 -0500231
232 kfree(sense);
233 return result;
234}
235EXPORT_SYMBOL(scsi_execute_req);
236
Mike Christie6e68af62005-11-11 05:30:27 -0600237struct scsi_io_context {
238 void *data;
239 void (*done)(void *data, char *sense, int result, int resid);
240 char sense[SCSI_SENSE_BUFFERSIZE];
241};
242
Mike Christieaa7b5cd2005-11-11 05:31:41 -0600243static kmem_cache_t *scsi_io_context_cache;
244
Jens Axboee650c302006-01-06 12:38:30 +0100245static void scsi_end_async(struct request *req, int uptodate)
Mike Christie6e68af62005-11-11 05:30:27 -0600246{
247 struct scsi_io_context *sioc = req->end_io_data;
248
249 if (sioc->done)
250 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len);
251
Mike Christieaa7b5cd2005-11-11 05:31:41 -0600252 kmem_cache_free(scsi_io_context_cache, sioc);
Mike Christie6e68af62005-11-11 05:30:27 -0600253 __blk_put_request(req->q, req);
254}
255
256static int scsi_merge_bio(struct request *rq, struct bio *bio)
257{
258 struct request_queue *q = rq->q;
259
260 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
261 if (rq_data_dir(rq) == WRITE)
262 bio->bi_rw |= (1 << BIO_RW);
263 blk_queue_bounce(q, &bio);
264
265 if (!rq->bio)
266 blk_rq_bio_prep(q, rq, bio);
267 else if (!q->back_merge_fn(q, rq, bio))
268 return -EINVAL;
269 else {
270 rq->biotail->bi_next = bio;
271 rq->biotail = bio;
272 rq->hard_nr_sectors += bio_sectors(bio);
273 rq->nr_sectors = rq->hard_nr_sectors;
274 }
275
276 return 0;
277}
278
279static int scsi_bi_endio(struct bio *bio, unsigned int bytes_done, int error)
280{
281 if (bio->bi_size)
282 return 1;
283
284 bio_put(bio);
285 return 0;
286}
287
288/**
289 * scsi_req_map_sg - map a scatterlist into a request
290 * @rq: request to fill
291 * @sg: scatterlist
292 * @nsegs: number of elements
293 * @bufflen: len of buffer
294 * @gfp: memory allocation flags
295 *
296 * scsi_req_map_sg maps a scatterlist into a request so that the
297 * request can be sent to the block layer. We do not trust the scatterlist
298 * sent to use, as some ULDs use that struct to only organize the pages.
299 */
300static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl,
301 int nsegs, unsigned bufflen, gfp_t gfp)
302{
303 struct request_queue *q = rq->q;
Bryan Holtyf5235962006-03-22 06:35:39 -0600304 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
Mike Christie6e68af62005-11-11 05:30:27 -0600305 unsigned int data_len = 0, len, bytes, off;
306 struct page *page;
307 struct bio *bio = NULL;
308 int i, err, nr_vecs = 0;
309
310 for (i = 0; i < nsegs; i++) {
311 page = sgl[i].page;
312 off = sgl[i].offset;
313 len = sgl[i].length;
314 data_len += len;
315
316 while (len > 0) {
317 bytes = min_t(unsigned int, len, PAGE_SIZE - off);
318
319 if (!bio) {
320 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages);
321 nr_pages -= nr_vecs;
322
323 bio = bio_alloc(gfp, nr_vecs);
324 if (!bio) {
325 err = -ENOMEM;
326 goto free_bios;
327 }
328 bio->bi_end_io = scsi_bi_endio;
329 }
330
331 if (bio_add_pc_page(q, bio, page, bytes, off) !=
332 bytes) {
333 bio_put(bio);
334 err = -EINVAL;
335 goto free_bios;
336 }
337
338 if (bio->bi_vcnt >= nr_vecs) {
339 err = scsi_merge_bio(rq, bio);
340 if (err) {
341 bio_endio(bio, bio->bi_size, 0);
342 goto free_bios;
343 }
344 bio = NULL;
345 }
346
347 page++;
348 len -= bytes;
349 off = 0;
350 }
351 }
352
353 rq->buffer = rq->data = NULL;
354 rq->data_len = data_len;
355 return 0;
356
357free_bios:
358 while ((bio = rq->bio) != NULL) {
359 rq->bio = bio->bi_next;
360 /*
361 * call endio instead of bio_put incase it was bounced
362 */
363 bio_endio(bio, bio->bi_size, 0);
364 }
365
366 return err;
367}
368
369/**
370 * scsi_execute_async - insert request
371 * @sdev: scsi device
372 * @cmd: scsi command
brking@us.ibm.combb1d1072006-01-23 15:03:22 -0600373 * @cmd_len: length of scsi cdb
Mike Christie6e68af62005-11-11 05:30:27 -0600374 * @data_direction: data direction
375 * @buffer: data buffer (this can be a kernel buffer or scatterlist)
376 * @bufflen: len of buffer
377 * @use_sg: if buffer is a scatterlist this is the number of elements
378 * @timeout: request timeout in seconds
379 * @retries: number of times to retry request
380 * @flags: or into request flags
381 **/
382int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd,
brking@us.ibm.combb1d1072006-01-23 15:03:22 -0600383 int cmd_len, int data_direction, void *buffer, unsigned bufflen,
Mike Christie6e68af62005-11-11 05:30:27 -0600384 int use_sg, int timeout, int retries, void *privdata,
385 void (*done)(void *, char *, int, int), gfp_t gfp)
386{
387 struct request *req;
388 struct scsi_io_context *sioc;
389 int err = 0;
390 int write = (data_direction == DMA_TO_DEVICE);
391
Mike Christieaa7b5cd2005-11-11 05:31:41 -0600392 sioc = kmem_cache_alloc(scsi_io_context_cache, gfp);
Mike Christie6e68af62005-11-11 05:30:27 -0600393 if (!sioc)
394 return DRIVER_ERROR << 24;
Mike Christieaa7b5cd2005-11-11 05:31:41 -0600395 memset(sioc, 0, sizeof(*sioc));
Mike Christie6e68af62005-11-11 05:30:27 -0600396
397 req = blk_get_request(sdev->request_queue, write, gfp);
398 if (!req)
399 goto free_sense;
Mike Christiedefd94b2005-12-05 02:37:06 -0600400 req->flags |= REQ_BLOCK_PC | REQ_QUIET;
Mike Christie6e68af62005-11-11 05:30:27 -0600401
402 if (use_sg)
403 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp);
404 else if (bufflen)
405 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp);
406
407 if (err)
408 goto free_req;
409
brking@us.ibm.combb1d1072006-01-23 15:03:22 -0600410 req->cmd_len = cmd_len;
Mike Christie6e68af62005-11-11 05:30:27 -0600411 memcpy(req->cmd, cmd, req->cmd_len);
412 req->sense = sioc->sense;
413 req->sense_len = 0;
414 req->timeout = timeout;
Mike Christie17e01f22005-11-11 05:31:37 -0600415 req->retries = retries;
Mike Christie6e68af62005-11-11 05:30:27 -0600416 req->end_io_data = sioc;
417
418 sioc->data = privdata;
419 sioc->done = done;
420
421 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async);
422 return 0;
423
424free_req:
425 blk_put_request(req);
426free_sense:
427 kfree(sioc);
428 return DRIVER_ERROR << 24;
429}
430EXPORT_SYMBOL_GPL(scsi_execute_async);
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432/*
433 * Function: scsi_init_cmd_errh()
434 *
435 * Purpose: Initialize cmd fields related to error handling.
436 *
437 * Arguments: cmd - command that is ready to be queued.
438 *
439 * Returns: Nothing
440 *
441 * Notes: This function has the job of initializing a number of
442 * fields related to error handling. Typically this will
443 * be called once for each command, as required.
444 */
445static int scsi_init_cmd_errh(struct scsi_cmnd *cmd)
446{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 cmd->serial_number = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 memset(cmd->sense_buffer, 0, sizeof cmd->sense_buffer);
450
451 if (cmd->cmd_len == 0)
452 cmd->cmd_len = COMMAND_SIZE(cmd->cmnd[0]);
453
454 /*
455 * We need saved copies of a number of fields - this is because
456 * error handling may need to overwrite these with different values
457 * to run different commands, and once error handling is complete,
458 * we will need to restore these values prior to running the actual
459 * command.
460 */
461 cmd->old_use_sg = cmd->use_sg;
462 cmd->old_cmd_len = cmd->cmd_len;
463 cmd->sc_old_data_direction = cmd->sc_data_direction;
464 cmd->old_underflow = cmd->underflow;
465 memcpy(cmd->data_cmnd, cmd->cmnd, sizeof(cmd->cmnd));
466 cmd->buffer = cmd->request_buffer;
467 cmd->bufflen = cmd->request_bufflen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 return 1;
470}
471
472/*
473 * Function: scsi_setup_cmd_retry()
474 *
475 * Purpose: Restore the command state for a retry
476 *
477 * Arguments: cmd - command to be restored
478 *
479 * Returns: Nothing
480 *
481 * Notes: Immediately prior to retrying a command, we need
482 * to restore certain fields that we saved above.
483 */
484void scsi_setup_cmd_retry(struct scsi_cmnd *cmd)
485{
486 memcpy(cmd->cmnd, cmd->data_cmnd, sizeof(cmd->data_cmnd));
487 cmd->request_buffer = cmd->buffer;
488 cmd->request_bufflen = cmd->bufflen;
489 cmd->use_sg = cmd->old_use_sg;
490 cmd->cmd_len = cmd->old_cmd_len;
491 cmd->sc_data_direction = cmd->sc_old_data_direction;
492 cmd->underflow = cmd->old_underflow;
493}
494
495void scsi_device_unbusy(struct scsi_device *sdev)
496{
497 struct Scsi_Host *shost = sdev->host;
498 unsigned long flags;
499
500 spin_lock_irqsave(shost->host_lock, flags);
501 shost->host_busy--;
James Bottomley939647e2005-09-18 15:05:20 -0500502 if (unlikely(scsi_host_in_recovery(shost) &&
Tejun Heoee7863b2006-05-15 20:57:20 +0900503 (shost->host_failed || shost->host_eh_scheduled)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 scsi_eh_wakeup(shost);
505 spin_unlock(shost->host_lock);
152587d2005-04-12 16:22:06 -0500506 spin_lock(sdev->request_queue->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 sdev->device_busy--;
152587d2005-04-12 16:22:06 -0500508 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509}
510
511/*
512 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
513 * and call blk_run_queue for all the scsi_devices on the target -
514 * including current_sdev first.
515 *
516 * Called with *no* scsi locks held.
517 */
518static void scsi_single_lun_run(struct scsi_device *current_sdev)
519{
520 struct Scsi_Host *shost = current_sdev->host;
521 struct scsi_device *sdev, *tmp;
522 struct scsi_target *starget = scsi_target(current_sdev);
523 unsigned long flags;
524
525 spin_lock_irqsave(shost->host_lock, flags);
526 starget->starget_sdev_user = NULL;
527 spin_unlock_irqrestore(shost->host_lock, flags);
528
529 /*
530 * Call blk_run_queue for all LUNs on the target, starting with
531 * current_sdev. We race with others (to set starget_sdev_user),
532 * but in most cases, we will be first. Ideally, each LU on the
533 * target would get some limited time or requests on the target.
534 */
535 blk_run_queue(current_sdev->request_queue);
536
537 spin_lock_irqsave(shost->host_lock, flags);
538 if (starget->starget_sdev_user)
539 goto out;
540 list_for_each_entry_safe(sdev, tmp, &starget->devices,
541 same_target_siblings) {
542 if (sdev == current_sdev)
543 continue;
544 if (scsi_device_get(sdev))
545 continue;
546
547 spin_unlock_irqrestore(shost->host_lock, flags);
548 blk_run_queue(sdev->request_queue);
549 spin_lock_irqsave(shost->host_lock, flags);
550
551 scsi_device_put(sdev);
552 }
553 out:
554 spin_unlock_irqrestore(shost->host_lock, flags);
555}
556
557/*
558 * Function: scsi_run_queue()
559 *
560 * Purpose: Select a proper request queue to serve next
561 *
562 * Arguments: q - last request's queue
563 *
564 * Returns: Nothing
565 *
566 * Notes: The previous command was completely finished, start
567 * a new one if possible.
568 */
569static void scsi_run_queue(struct request_queue *q)
570{
571 struct scsi_device *sdev = q->queuedata;
572 struct Scsi_Host *shost = sdev->host;
573 unsigned long flags;
574
575 if (sdev->single_lun)
576 scsi_single_lun_run(sdev);
577
578 spin_lock_irqsave(shost->host_lock, flags);
579 while (!list_empty(&shost->starved_list) &&
580 !shost->host_blocked && !shost->host_self_blocked &&
581 !((shost->can_queue > 0) &&
582 (shost->host_busy >= shost->can_queue))) {
583 /*
584 * As long as shost is accepting commands and we have
585 * starved queues, call blk_run_queue. scsi_request_fn
586 * drops the queue_lock and can add us back to the
587 * starved_list.
588 *
589 * host_lock protects the starved_list and starved_entry.
590 * scsi_request_fn must get the host_lock before checking
591 * or modifying starved_list or starved_entry.
592 */
593 sdev = list_entry(shost->starved_list.next,
594 struct scsi_device, starved_entry);
595 list_del_init(&sdev->starved_entry);
596 spin_unlock_irqrestore(shost->host_lock, flags);
597
598 blk_run_queue(sdev->request_queue);
599
600 spin_lock_irqsave(shost->host_lock, flags);
601 if (unlikely(!list_empty(&sdev->starved_entry)))
602 /*
603 * sdev lost a race, and was put back on the
604 * starved list. This is unlikely but without this
605 * in theory we could loop forever.
606 */
607 break;
608 }
609 spin_unlock_irqrestore(shost->host_lock, flags);
610
611 blk_run_queue(q);
612}
613
614/*
615 * Function: scsi_requeue_command()
616 *
617 * Purpose: Handle post-processing of completed commands.
618 *
619 * Arguments: q - queue to operate on
620 * cmd - command that may need to be requeued.
621 *
622 * Returns: Nothing
623 *
624 * Notes: After command completion, there may be blocks left
625 * over which weren't finished by the previous command
626 * this can be for a number of reasons - the main one is
627 * I/O errors in the middle of the request, in which case
628 * we need to request the blocks that come after the bad
629 * sector.
James Bottomleye91442b2005-09-09 10:44:16 -0500630 * Notes: Upon return, cmd is a stale pointer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 */
632static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
633{
James Bottomleye91442b2005-09-09 10:44:16 -0500634 struct request *req = cmd->request;
Tejun Heo 283369c2005-04-24 02:06:36 -0500635 unsigned long flags;
636
James Bottomleye91442b2005-09-09 10:44:16 -0500637 scsi_unprep_request(req);
Tejun Heo 283369c2005-04-24 02:06:36 -0500638 spin_lock_irqsave(q->queue_lock, flags);
James Bottomleye91442b2005-09-09 10:44:16 -0500639 blk_requeue_request(q, req);
Tejun Heo 283369c2005-04-24 02:06:36 -0500640 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 scsi_run_queue(q);
643}
644
645void scsi_next_command(struct scsi_cmnd *cmd)
646{
Linus Torvalds49d7bc62005-12-12 11:25:04 -0800647 struct scsi_device *sdev = cmd->device;
648 struct request_queue *q = sdev->request_queue;
649
650 /* need to hold a reference on the device before we let go of the cmd */
651 get_device(&sdev->sdev_gendev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653 scsi_put_command(cmd);
654 scsi_run_queue(q);
Linus Torvalds49d7bc62005-12-12 11:25:04 -0800655
656 /* ok to remove device now */
657 put_device(&sdev->sdev_gendev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
660void scsi_run_host_queues(struct Scsi_Host *shost)
661{
662 struct scsi_device *sdev;
663
664 shost_for_each_device(sdev, shost)
665 scsi_run_queue(sdev->request_queue);
666}
667
668/*
669 * Function: scsi_end_request()
670 *
671 * Purpose: Post-processing of completed commands (usually invoked at end
672 * of upper level post-processing and scsi_io_completion).
673 *
674 * Arguments: cmd - command that is complete.
675 * uptodate - 1 if I/O indicates success, <= 0 for I/O error.
676 * bytes - number of bytes of completed I/O
677 * requeue - indicates whether we should requeue leftovers.
678 *
679 * Lock status: Assumed that lock is not held upon entry.
680 *
James Bottomleye91442b2005-09-09 10:44:16 -0500681 * Returns: cmd if requeue required, NULL otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 *
683 * Notes: This is called for block device requests in order to
684 * mark some number of sectors as complete.
685 *
686 * We are guaranteeing that the request queue will be goosed
687 * at some point during this call.
James Bottomleye91442b2005-09-09 10:44:16 -0500688 * Notes: If cmd was requeued, upon return it will be a stale pointer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 */
690static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int uptodate,
691 int bytes, int requeue)
692{
693 request_queue_t *q = cmd->device->request_queue;
694 struct request *req = cmd->request;
695 unsigned long flags;
696
697 /*
698 * If there are blocks left over at the end, set up the command
699 * to queue the remainder of them.
700 */
701 if (end_that_request_chunk(req, uptodate, bytes)) {
702 int leftover = (req->hard_nr_sectors << 9);
703
704 if (blk_pc_request(req))
705 leftover = req->data_len;
706
707 /* kill remainder if no retrys */
708 if (!uptodate && blk_noretry_request(req))
709 end_that_request_chunk(req, 0, leftover);
710 else {
James Bottomleye91442b2005-09-09 10:44:16 -0500711 if (requeue) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 /*
713 * Bleah. Leftovers again. Stick the
714 * leftovers in the front of the
715 * queue, and goose the queue again.
716 */
717 scsi_requeue_command(q, cmd);
James Bottomleye91442b2005-09-09 10:44:16 -0500718 cmd = NULL;
719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 return cmd;
721 }
722 }
723
724 add_disk_randomness(req->rq_disk);
725
726 spin_lock_irqsave(q->queue_lock, flags);
727 if (blk_rq_tagged(req))
728 blk_queue_end_tag(q, req);
Tejun Heo8ffdc652006-01-06 09:49:03 +0100729 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 spin_unlock_irqrestore(q->queue_lock, flags);
731
732 /*
733 * This will goose the queue request function at the end, so we don't
734 * need to worry about launching another command.
735 */
736 scsi_next_command(cmd);
737 return NULL;
738}
739
Al Viroc53033f2005-10-21 03:22:08 -0400740static struct scatterlist *scsi_alloc_sgtable(struct scsi_cmnd *cmd, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 struct scsi_host_sg_pool *sgp;
743 struct scatterlist *sgl;
744
745 BUG_ON(!cmd->use_sg);
746
747 switch (cmd->use_sg) {
748 case 1 ... 8:
749 cmd->sglist_len = 0;
750 break;
751 case 9 ... 16:
752 cmd->sglist_len = 1;
753 break;
754 case 17 ... 32:
755 cmd->sglist_len = 2;
756 break;
757#if (SCSI_MAX_PHYS_SEGMENTS > 32)
758 case 33 ... 64:
759 cmd->sglist_len = 3;
760 break;
761#if (SCSI_MAX_PHYS_SEGMENTS > 64)
762 case 65 ... 128:
763 cmd->sglist_len = 4;
764 break;
765#if (SCSI_MAX_PHYS_SEGMENTS > 128)
766 case 129 ... 256:
767 cmd->sglist_len = 5;
768 break;
769#endif
770#endif
771#endif
772 default:
773 return NULL;
774 }
775
776 sgp = scsi_sg_pools + cmd->sglist_len;
777 sgl = mempool_alloc(sgp->pool, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 return sgl;
779}
780
781static void scsi_free_sgtable(struct scatterlist *sgl, int index)
782{
783 struct scsi_host_sg_pool *sgp;
784
KAMBAROV, ZAURa77e3362005-06-28 20:45:06 -0700785 BUG_ON(index >= SG_MEMPOOL_NR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787 sgp = scsi_sg_pools + index;
788 mempool_free(sgl, sgp->pool);
789}
790
791/*
792 * Function: scsi_release_buffers()
793 *
794 * Purpose: Completion processing for block device I/O requests.
795 *
796 * Arguments: cmd - command that we are bailing.
797 *
798 * Lock status: Assumed that no lock is held upon entry.
799 *
800 * Returns: Nothing
801 *
802 * Notes: In the event that an upper level driver rejects a
803 * command, we must release resources allocated during
804 * the __init_io() function. Primarily this would involve
805 * the scatter-gather table, and potentially any bounce
806 * buffers.
807 */
808static void scsi_release_buffers(struct scsi_cmnd *cmd)
809{
810 struct request *req = cmd->request;
811
812 /*
813 * Free up any indirection buffers we allocated for DMA purposes.
814 */
815 if (cmd->use_sg)
816 scsi_free_sgtable(cmd->request_buffer, cmd->sglist_len);
817 else if (cmd->request_buffer != req->buffer)
818 kfree(cmd->request_buffer);
819
820 /*
821 * Zero these out. They now point to freed memory, and it is
822 * dangerous to hang onto the pointers.
823 */
824 cmd->buffer = NULL;
825 cmd->bufflen = 0;
826 cmd->request_buffer = NULL;
827 cmd->request_bufflen = 0;
828}
829
830/*
831 * Function: scsi_io_completion()
832 *
833 * Purpose: Completion processing for block device I/O requests.
834 *
835 * Arguments: cmd - command that is finished.
836 *
837 * Lock status: Assumed that no lock is held upon entry.
838 *
839 * Returns: Nothing
840 *
841 * Notes: This function is matched in terms of capabilities to
842 * the function that created the scatter-gather list.
843 * In other words, if there are no bounce buffers
844 * (the normal case for most drivers), we don't need
845 * the logic to deal with cleaning up afterwards.
846 *
847 * We must do one of several things here:
848 *
849 * a) Call scsi_end_request. This will finish off the
850 * specified number of sectors. If we are done, the
851 * command block will be released, and the queue
852 * function will be goosed. If we are not done, then
853 * scsi_end_request will directly goose the queue.
854 *
855 * b) We can just use scsi_requeue_command() here. This would
856 * be used if we just wanted to retry, for example.
857 */
858void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes,
859 unsigned int block_bytes)
860{
861 int result = cmd->result;
862 int this_count = cmd->bufflen;
863 request_queue_t *q = cmd->device->request_queue;
864 struct request *req = cmd->request;
865 int clear_errors = 1;
866 struct scsi_sense_hdr sshdr;
867 int sense_valid = 0;
868 int sense_deferred = 0;
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 /*
871 * Free up any indirection buffers we allocated for DMA purposes.
872 * For the case of a READ, we need to copy the data out of the
873 * bounce buffer and into the real buffer.
874 */
875 if (cmd->use_sg)
876 scsi_free_sgtable(cmd->buffer, cmd->sglist_len);
877 else if (cmd->buffer != req->buffer) {
878 if (rq_data_dir(req) == READ) {
879 unsigned long flags;
880 char *to = bio_kmap_irq(req->bio, &flags);
881 memcpy(to, cmd->buffer, cmd->bufflen);
882 bio_kunmap_irq(to, &flags);
883 }
884 kfree(cmd->buffer);
885 }
886
887 if (result) {
888 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
889 if (sense_valid)
890 sense_deferred = scsi_sense_is_deferred(&sshdr);
891 }
892 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
893 req->errors = result;
894 if (result) {
895 clear_errors = 0;
896 if (sense_valid && req->sense) {
897 /*
898 * SG_IO wants current and deferred errors
899 */
900 int len = 8 + cmd->sense_buffer[7];
901
902 if (len > SCSI_SENSE_BUFFERSIZE)
903 len = SCSI_SENSE_BUFFERSIZE;
904 memcpy(req->sense, cmd->sense_buffer, len);
905 req->sense_len = len;
906 }
907 } else
908 req->data_len = cmd->resid;
909 }
910
911 /*
912 * Zero these out. They now point to freed memory, and it is
913 * dangerous to hang onto the pointers.
914 */
915 cmd->buffer = NULL;
916 cmd->bufflen = 0;
917 cmd->request_buffer = NULL;
918 cmd->request_bufflen = 0;
919
920 /*
921 * Next deal with any sectors which we were able to correctly
922 * handle.
923 */
924 if (good_bytes >= 0) {
925 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, %d bytes done.\n",
926 req->nr_sectors, good_bytes));
927 SCSI_LOG_HLCOMPLETE(1, printk("use_sg is %d\n", cmd->use_sg));
928
929 if (clear_errors)
930 req->errors = 0;
931 /*
932 * If multiple sectors are requested in one buffer, then
933 * they will have been finished off by the first command.
934 * If not, then we have a multi-buffer command.
935 *
936 * If block_bytes != 0, it means we had a medium error
937 * of some sort, and that we want to mark some number of
938 * sectors as not uptodate. Thus we want to inhibit
939 * requeueing right here - we will requeue down below
940 * when we handle the bad sectors.
941 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 /*
James Bottomleye91442b2005-09-09 10:44:16 -0500944 * If the command completed without error, then either
945 * finish off the rest of the command, or start a new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 */
James Bottomleye91442b2005-09-09 10:44:16 -0500947 if (scsi_end_request(cmd, 1, good_bytes, result == 0) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 }
950 /*
951 * Now, if we were good little boys and girls, Santa left us a request
952 * sense buffer. We can extract information from this, so we
953 * can choose a block to remap, etc.
954 */
955 if (sense_valid && !sense_deferred) {
956 switch (sshdr.sense_key) {
957 case UNIT_ATTENTION:
958 if (cmd->device->removable) {
959 /* detected disc change. set a bit
960 * and quietly refuse further access.
961 */
962 cmd->device->changed = 1;
James Bottomleye91442b2005-09-09 10:44:16 -0500963 scsi_end_request(cmd, 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 this_count, 1);
965 return;
966 } else {
967 /*
968 * Must have been a power glitch, or a
969 * bus reset. Could not have been a
970 * media change, so we just retry the
971 * request and see what happens.
972 */
973 scsi_requeue_command(q, cmd);
974 return;
975 }
976 break;
977 case ILLEGAL_REQUEST:
978 /*
979 * If we had an ILLEGAL REQUEST returned, then we may
980 * have performed an unsupported command. The only
981 * thing this should be would be a ten byte read where
982 * only a six byte read was supported. Also, on a
983 * system where READ CAPACITY failed, we may have read
984 * past the end of the disk.
985 */
Jens Axboe26a68012005-11-29 21:03:34 +0100986 if ((cmd->device->use_10_for_rw &&
987 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 (cmd->cmnd[0] == READ_10 ||
989 cmd->cmnd[0] == WRITE_10)) {
990 cmd->device->use_10_for_rw = 0;
991 /*
992 * This will cause a retry with a 6-byte
993 * command.
994 */
995 scsi_requeue_command(q, cmd);
996 result = 0;
997 } else {
James Bottomleye91442b2005-09-09 10:44:16 -0500998 scsi_end_request(cmd, 0, this_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 return;
1000 }
1001 break;
1002 case NOT_READY:
1003 /*
James Bottomleyf3e93f72006-04-25 16:48:30 -05001004 * If the device is in the process of becoming
1005 * ready, or has a temporary blockage, retry.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 */
James Bottomleyf3e93f72006-04-25 16:48:30 -05001007 if (sshdr.asc == 0x04) {
1008 switch (sshdr.ascq) {
1009 case 0x01: /* becoming ready */
1010 case 0x04: /* format in progress */
1011 case 0x05: /* rebuild in progress */
1012 case 0x06: /* recalculation in progress */
1013 case 0x07: /* operation in progress */
1014 case 0x08: /* Long write in progress */
1015 case 0x09: /* self test in progress */
1016 scsi_requeue_command(q, cmd);
1017 return;
1018 default:
1019 break;
1020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
James Bottomleyf3e93f72006-04-25 16:48:30 -05001022 if (!(req->flags & REQ_QUIET)) {
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001023 scmd_printk(KERN_INFO, cmd,
James Bottomleyf3e93f72006-04-25 16:48:30 -05001024 "Device not ready: ");
1025 scsi_print_sense_hdr("", &sshdr);
1026 }
James Bottomleye91442b2005-09-09 10:44:16 -05001027 scsi_end_request(cmd, 0, this_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 return;
1029 case VOLUME_OVERFLOW:
James Bottomley3173d8c2005-09-04 11:32:05 -05001030 if (!(req->flags & REQ_QUIET)) {
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001031 scmd_printk(KERN_INFO, cmd,
1032 "Volume overflow, CDB: ");
James Bottomley3173d8c2005-09-04 11:32:05 -05001033 __scsi_print_command(cmd->data_cmnd);
1034 scsi_print_sense("", cmd);
1035 }
James Bottomleye91442b2005-09-09 10:44:16 -05001036 scsi_end_request(cmd, 0, block_bytes, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 return;
1038 default:
1039 break;
1040 }
1041 } /* driver byte != 0 */
1042 if (host_byte(result) == DID_RESET) {
1043 /*
1044 * Third party bus reset or reset for error
1045 * recovery reasons. Just retry the request
1046 * and see what happens.
1047 */
1048 scsi_requeue_command(q, cmd);
1049 return;
1050 }
1051 if (result) {
James Bottomley3173d8c2005-09-04 11:32:05 -05001052 if (!(req->flags & REQ_QUIET)) {
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001053 scmd_printk(KERN_INFO, cmd,
1054 "SCSI error: return code = 0x%x\n", result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
James Bottomley3173d8c2005-09-04 11:32:05 -05001056 if (driver_byte(result) & DRIVER_SENSE)
1057 scsi_print_sense("", cmd);
1058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 /*
1060 * Mark a single buffer as not uptodate. Queue the remainder.
1061 * We sometimes get this cruft in the event that a medium error
1062 * isn't properly reported.
1063 */
1064 block_bytes = req->hard_cur_sectors << 9;
1065 if (!block_bytes)
1066 block_bytes = req->data_len;
James Bottomleye91442b2005-09-09 10:44:16 -05001067 scsi_end_request(cmd, 0, block_bytes, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 }
1069}
1070EXPORT_SYMBOL(scsi_io_completion);
1071
1072/*
1073 * Function: scsi_init_io()
1074 *
1075 * Purpose: SCSI I/O initialize function.
1076 *
1077 * Arguments: cmd - Command descriptor we wish to initialize
1078 *
1079 * Returns: 0 on success
1080 * BLKPREP_DEFER if the failure is retryable
1081 * BLKPREP_KILL if the failure is fatal
1082 */
1083static int scsi_init_io(struct scsi_cmnd *cmd)
1084{
1085 struct request *req = cmd->request;
1086 struct scatterlist *sgpnt;
1087 int count;
1088
1089 /*
1090 * if this is a rq->data based REQ_BLOCK_PC, setup for a non-sg xfer
1091 */
1092 if ((req->flags & REQ_BLOCK_PC) && !req->bio) {
1093 cmd->request_bufflen = req->data_len;
1094 cmd->request_buffer = req->data;
1095 req->buffer = req->data;
1096 cmd->use_sg = 0;
1097 return 0;
1098 }
1099
1100 /*
1101 * we used to not use scatter-gather for single segment request,
1102 * but now we do (it makes highmem I/O easier to support without
1103 * kmapping pages)
1104 */
1105 cmd->use_sg = req->nr_phys_segments;
1106
1107 /*
1108 * if sg table allocation fails, requeue request later.
1109 */
1110 sgpnt = scsi_alloc_sgtable(cmd, GFP_ATOMIC);
Alan Stern7c72ce82005-10-14 11:23:27 -04001111 if (unlikely(!sgpnt)) {
1112 scsi_unprep_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 return BLKPREP_DEFER;
Alan Stern7c72ce82005-10-14 11:23:27 -04001114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
1116 cmd->request_buffer = (char *) sgpnt;
1117 cmd->request_bufflen = req->nr_sectors << 9;
1118 if (blk_pc_request(req))
1119 cmd->request_bufflen = req->data_len;
1120 req->buffer = NULL;
1121
1122 /*
1123 * Next, walk the list, and fill in the addresses and sizes of
1124 * each segment.
1125 */
1126 count = blk_rq_map_sg(req->q, req, cmd->request_buffer);
1127
1128 /*
1129 * mapped well, send it off
1130 */
1131 if (likely(count <= cmd->use_sg)) {
1132 cmd->use_sg = count;
1133 return 0;
1134 }
1135
1136 printk(KERN_ERR "Incorrect number of segments after building list\n");
1137 printk(KERN_ERR "counted %d, received %d\n", count, cmd->use_sg);
1138 printk(KERN_ERR "req nr_sec %lu, cur_nr_sec %u\n", req->nr_sectors,
1139 req->current_nr_sectors);
1140
1141 /* release the command and kill it */
1142 scsi_release_buffers(cmd);
1143 scsi_put_command(cmd);
1144 return BLKPREP_KILL;
1145}
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147static int scsi_issue_flush_fn(request_queue_t *q, struct gendisk *disk,
1148 sector_t *error_sector)
1149{
1150 struct scsi_device *sdev = q->queuedata;
1151 struct scsi_driver *drv;
1152
1153 if (sdev->sdev_state != SDEV_RUNNING)
1154 return -ENXIO;
1155
1156 drv = *(struct scsi_driver **) disk->private_data;
1157 if (drv->issue_flush)
1158 return drv->issue_flush(&sdev->sdev_gendev, error_sector);
1159
1160 return -EOPNOTSUPP;
1161}
1162
Christoph Hellwig776b23a2006-01-06 18:34:07 +01001163static void scsi_blk_pc_done(struct scsi_cmnd *cmd)
James Bottomleye537a362005-06-05 02:07:14 -05001164{
1165 BUG_ON(!blk_pc_request(cmd->request));
Mike Christie0d957162005-11-11 05:31:40 -06001166 /*
1167 * This will complete the whole command with uptodate=1 so
1168 * as far as the block layer is concerned the command completed
1169 * successfully. Since this is a REQ_BLOCK_PC command the
1170 * caller should check the request's errors value
1171 */
1172 scsi_io_completion(cmd, cmd->bufflen, 0);
James Bottomleye537a362005-06-05 02:07:14 -05001173}
1174
Christoph Hellwig776b23a2006-01-06 18:34:07 +01001175static void scsi_setup_blk_pc_cmnd(struct scsi_cmnd *cmd)
James Bottomley7b163182005-12-15 20:17:02 -06001176{
1177 struct request *req = cmd->request;
1178
1179 BUG_ON(sizeof(req->cmd) > sizeof(cmd->cmnd));
1180 memcpy(cmd->cmnd, req->cmd, sizeof(cmd->cmnd));
1181 cmd->cmd_len = req->cmd_len;
1182 if (!req->data_len)
1183 cmd->sc_data_direction = DMA_NONE;
1184 else if (rq_data_dir(req) == WRITE)
1185 cmd->sc_data_direction = DMA_TO_DEVICE;
1186 else
1187 cmd->sc_data_direction = DMA_FROM_DEVICE;
1188
1189 cmd->transfersize = req->data_len;
1190 cmd->allowed = req->retries;
1191 cmd->timeout_per_command = req->timeout;
Christoph Hellwig776b23a2006-01-06 18:34:07 +01001192 cmd->done = scsi_blk_pc_done;
James Bottomley7b163182005-12-15 20:17:02 -06001193}
James Bottomley7b163182005-12-15 20:17:02 -06001194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195static int scsi_prep_fn(struct request_queue *q, struct request *req)
1196{
1197 struct scsi_device *sdev = q->queuedata;
1198 struct scsi_cmnd *cmd;
1199 int specials_only = 0;
1200
1201 /*
1202 * Just check to see if the device is online. If it isn't, we
1203 * refuse to process any commands. The device must be brought
1204 * online before trying any recovery commands
1205 */
1206 if (unlikely(!scsi_device_online(sdev))) {
James Bottomley9ccfc752005-10-02 11:45:08 -05001207 sdev_printk(KERN_ERR, sdev,
1208 "rejecting I/O to offline device\n");
Mike Christie6f16b532005-09-10 16:45:35 -05001209 goto kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 }
1211 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1212 /* OK, we're not in a running state don't prep
1213 * user commands */
1214 if (sdev->sdev_state == SDEV_DEL) {
1215 /* Device is fully deleted, no commands
1216 * at all allowed down */
James Bottomley9ccfc752005-10-02 11:45:08 -05001217 sdev_printk(KERN_ERR, sdev,
1218 "rejecting I/O to dead device\n");
Mike Christie6f16b532005-09-10 16:45:35 -05001219 goto kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
1221 /* OK, we only allow special commands (i.e. not
1222 * user initiated ones */
1223 specials_only = sdev->sdev_state;
1224 }
1225
1226 /*
1227 * Find the actual device driver associated with this command.
1228 * The SPECIAL requests are things like character device or
1229 * ioctls, which did not originate from ll_rw_blk. Note that
1230 * the special field is also used to indicate the cmd for
1231 * the remainder of a partially fulfilled request that can
1232 * come up when there is a medium error. We have to treat
1233 * these two cases differently. We differentiate by looking
1234 * at request->cmd, as this tells us the real story.
1235 */
James Bottomleye537a362005-06-05 02:07:14 -05001236 if (req->flags & REQ_SPECIAL && req->special) {
Christoph Hellwigbeb40482006-06-10 18:01:03 +02001237 cmd = req->special;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 } else if (req->flags & (REQ_CMD | REQ_BLOCK_PC)) {
1239
James Bottomleye537a362005-06-05 02:07:14 -05001240 if(unlikely(specials_only) && !(req->flags & REQ_SPECIAL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 if(specials_only == SDEV_QUIESCE ||
1242 specials_only == SDEV_BLOCK)
Mike Christie6f16b532005-09-10 16:45:35 -05001243 goto defer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
James Bottomley9ccfc752005-10-02 11:45:08 -05001245 sdev_printk(KERN_ERR, sdev,
1246 "rejecting I/O to device being removed\n");
Mike Christie6f16b532005-09-10 16:45:35 -05001247 goto kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 }
1249
1250
1251 /*
1252 * Now try and find a command block that we can use.
1253 */
1254 if (!req->special) {
1255 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1256 if (unlikely(!cmd))
1257 goto defer;
1258 } else
1259 cmd = req->special;
1260
1261 /* pull a tag out of the request if we have one */
1262 cmd->tag = req->tag;
1263 } else {
1264 blk_dump_rq_flags(req, "SCSI bad req");
Mike Christie6f16b532005-09-10 16:45:35 -05001265 goto kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 }
1267
1268 /* note the overloading of req->special. When the tag
1269 * is active it always means cmd. If the tag goes
1270 * back for re-queueing, it may be reset */
1271 req->special = cmd;
1272 cmd->request = req;
1273
1274 /*
1275 * FIXME: drop the lock here because the functions below
1276 * expect to be called without the queue lock held. Also,
1277 * previously, we dequeued the request before dropping the
1278 * lock. We hope REQ_STARTED prevents anything untoward from
1279 * happening now.
1280 */
1281 if (req->flags & (REQ_CMD | REQ_BLOCK_PC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 int ret;
1283
1284 /*
1285 * This will do a couple of things:
1286 * 1) Fill in the actual SCSI command.
1287 * 2) Fill in any other upper-level specific fields
1288 * (timeout).
1289 *
1290 * If this returns 0, it means that the request failed
1291 * (reading past end of disk, reading offline device,
1292 * etc). This won't actually talk to the device, but
1293 * some kinds of consistency checking may cause the
1294 * request to be rejected immediately.
1295 */
1296
1297 /*
1298 * This sets up the scatter-gather table (allocating if
1299 * required).
1300 */
1301 ret = scsi_init_io(cmd);
Mike Christie6f16b532005-09-10 16:45:35 -05001302 switch(ret) {
Alan Stern7c72ce82005-10-14 11:23:27 -04001303 /* For BLKPREP_KILL/DEFER the cmd was released */
Mike Christie6f16b532005-09-10 16:45:35 -05001304 case BLKPREP_KILL:
Mike Christie6f16b532005-09-10 16:45:35 -05001305 goto kill;
1306 case BLKPREP_DEFER:
1307 goto defer;
1308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 /*
1311 * Initialize the actual SCSI command for this request.
1312 */
Christoph Hellwig776b23a2006-01-06 18:34:07 +01001313 if (req->flags & REQ_BLOCK_PC) {
1314 scsi_setup_blk_pc_cmnd(cmd);
1315 } else if (req->rq_disk) {
1316 struct scsi_driver *drv;
1317
James Bottomleye537a362005-06-05 02:07:14 -05001318 drv = *(struct scsi_driver **)req->rq_disk->private_data;
1319 if (unlikely(!drv->init_command(cmd))) {
1320 scsi_release_buffers(cmd);
1321 scsi_put_command(cmd);
Mike Christie6f16b532005-09-10 16:45:35 -05001322 goto kill;
James Bottomleye537a362005-06-05 02:07:14 -05001323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 }
1325 }
1326
1327 /*
1328 * The request is now prepped, no need to come back here
1329 */
1330 req->flags |= REQ_DONTPREP;
1331 return BLKPREP_OK;
1332
1333 defer:
1334 /* If we defer, the elv_next_request() returns NULL, but the
1335 * queue must be restarted, so we plug here if no returning
1336 * command will automatically do that. */
1337 if (sdev->device_busy == 0)
1338 blk_plug_device(q);
1339 return BLKPREP_DEFER;
Mike Christie6f16b532005-09-10 16:45:35 -05001340 kill:
1341 req->errors = DID_NO_CONNECT << 16;
1342 return BLKPREP_KILL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343}
1344
1345/*
1346 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1347 * return 0.
1348 *
1349 * Called with the queue_lock held.
1350 */
1351static inline int scsi_dev_queue_ready(struct request_queue *q,
1352 struct scsi_device *sdev)
1353{
1354 if (sdev->device_busy >= sdev->queue_depth)
1355 return 0;
1356 if (sdev->device_busy == 0 && sdev->device_blocked) {
1357 /*
1358 * unblock after device_blocked iterates to zero
1359 */
1360 if (--sdev->device_blocked == 0) {
1361 SCSI_LOG_MLQUEUE(3,
James Bottomley9ccfc752005-10-02 11:45:08 -05001362 sdev_printk(KERN_INFO, sdev,
1363 "unblocking device at zero depth\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 } else {
1365 blk_plug_device(q);
1366 return 0;
1367 }
1368 }
1369 if (sdev->device_blocked)
1370 return 0;
1371
1372 return 1;
1373}
1374
1375/*
1376 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1377 * return 0. We must end up running the queue again whenever 0 is
1378 * returned, else IO can hang.
1379 *
1380 * Called with host_lock held.
1381 */
1382static inline int scsi_host_queue_ready(struct request_queue *q,
1383 struct Scsi_Host *shost,
1384 struct scsi_device *sdev)
1385{
James Bottomley939647e2005-09-18 15:05:20 -05001386 if (scsi_host_in_recovery(shost))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 return 0;
1388 if (shost->host_busy == 0 && shost->host_blocked) {
1389 /*
1390 * unblock after host_blocked iterates to zero
1391 */
1392 if (--shost->host_blocked == 0) {
1393 SCSI_LOG_MLQUEUE(3,
1394 printk("scsi%d unblocking host at zero depth\n",
1395 shost->host_no));
1396 } else {
1397 blk_plug_device(q);
1398 return 0;
1399 }
1400 }
1401 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
1402 shost->host_blocked || shost->host_self_blocked) {
1403 if (list_empty(&sdev->starved_entry))
1404 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1405 return 0;
1406 }
1407
1408 /* We're OK to process the command, so we can't be starved */
1409 if (!list_empty(&sdev->starved_entry))
1410 list_del_init(&sdev->starved_entry);
1411
1412 return 1;
1413}
1414
1415/*
James Bottomleye91442b2005-09-09 10:44:16 -05001416 * Kill a request for a dead device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 */
James Bottomleye91442b2005-09-09 10:44:16 -05001418static void scsi_kill_request(struct request *req, request_queue_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
James Bottomleye91442b2005-09-09 10:44:16 -05001420 struct scsi_cmnd *cmd = req->special;
Tejun Heoe36e0c82006-04-11 17:27:53 +09001421 struct scsi_device *sdev = cmd->device;
1422 struct Scsi_Host *shost = sdev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
James Bottomley788ce432005-09-09 13:40:23 -05001424 blkdev_dequeue_request(req);
1425
James Bottomleye91442b2005-09-09 10:44:16 -05001426 if (unlikely(cmd == NULL)) {
1427 printk(KERN_CRIT "impossible request in %s.\n",
1428 __FUNCTION__);
1429 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 }
James Bottomleye91442b2005-09-09 10:44:16 -05001431
1432 scsi_init_cmd_errh(cmd);
1433 cmd->result = DID_NO_CONNECT << 16;
1434 atomic_inc(&cmd->device->iorequest_cnt);
Tejun Heoe36e0c82006-04-11 17:27:53 +09001435
1436 /*
1437 * SCSI request completion path will do scsi_device_unbusy(),
1438 * bump busy counts. To bump the counters, we need to dance
1439 * with the locks as normal issue path does.
1440 */
1441 sdev->device_busy++;
1442 spin_unlock(sdev->request_queue->queue_lock);
1443 spin_lock(shost->host_lock);
1444 shost->host_busy++;
1445 spin_unlock(shost->host_lock);
1446 spin_lock(sdev->request_queue->queue_lock);
1447
James Bottomleye91442b2005-09-09 10:44:16 -05001448 __scsi_done(cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449}
1450
Jens Axboe1aea6432006-01-09 16:03:03 +01001451static void scsi_softirq_done(struct request *rq)
1452{
1453 struct scsi_cmnd *cmd = rq->completion_data;
Brian King8884efa2006-02-24 17:10:04 -06001454 unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;
Jens Axboe1aea6432006-01-09 16:03:03 +01001455 int disposition;
1456
1457 INIT_LIST_HEAD(&cmd->eh_entry);
1458
1459 disposition = scsi_decide_disposition(cmd);
1460 if (disposition != SUCCESS &&
1461 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1462 sdev_printk(KERN_ERR, cmd->device,
1463 "timing out command, waited %lus\n",
1464 wait_for/HZ);
1465 disposition = SUCCESS;
1466 }
1467
1468 scsi_log_completion(cmd, disposition);
1469
1470 switch (disposition) {
1471 case SUCCESS:
1472 scsi_finish_command(cmd);
1473 break;
1474 case NEEDS_RETRY:
1475 scsi_retry_command(cmd);
1476 break;
1477 case ADD_TO_MLQUEUE:
1478 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1479 break;
1480 default:
1481 if (!scsi_eh_scmd_add(cmd, 0))
1482 scsi_finish_command(cmd);
1483 }
1484}
1485
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486/*
1487 * Function: scsi_request_fn()
1488 *
1489 * Purpose: Main strategy routine for SCSI.
1490 *
1491 * Arguments: q - Pointer to actual queue.
1492 *
1493 * Returns: Nothing
1494 *
1495 * Lock status: IO request lock assumed to be held when called.
1496 */
1497static void scsi_request_fn(struct request_queue *q)
1498{
1499 struct scsi_device *sdev = q->queuedata;
1500 struct Scsi_Host *shost;
1501 struct scsi_cmnd *cmd;
1502 struct request *req;
1503
1504 if (!sdev) {
1505 printk("scsi: killing requests for dead queue\n");
James Bottomleye91442b2005-09-09 10:44:16 -05001506 while ((req = elv_next_request(q)) != NULL)
1507 scsi_kill_request(req, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return;
1509 }
1510
1511 if(!get_device(&sdev->sdev_gendev))
1512 /* We must be tearing the block queue down already */
1513 return;
1514
1515 /*
1516 * To start with, we keep looping until the queue is empty, or until
1517 * the host is no longer able to accept any more requests.
1518 */
1519 shost = sdev->host;
1520 while (!blk_queue_plugged(q)) {
1521 int rtn;
1522 /*
1523 * get next queueable request. We do this early to make sure
1524 * that the request is fully prepared even if we cannot
1525 * accept it.
1526 */
1527 req = elv_next_request(q);
1528 if (!req || !scsi_dev_queue_ready(q, sdev))
1529 break;
1530
1531 if (unlikely(!scsi_device_online(sdev))) {
James Bottomley9ccfc752005-10-02 11:45:08 -05001532 sdev_printk(KERN_ERR, sdev,
1533 "rejecting I/O to offline device\n");
James Bottomleye91442b2005-09-09 10:44:16 -05001534 scsi_kill_request(req, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 continue;
1536 }
1537
1538
1539 /*
1540 * Remove the request from the request list.
1541 */
1542 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1543 blkdev_dequeue_request(req);
1544 sdev->device_busy++;
1545
1546 spin_unlock(q->queue_lock);
James Bottomleye91442b2005-09-09 10:44:16 -05001547 cmd = req->special;
1548 if (unlikely(cmd == NULL)) {
1549 printk(KERN_CRIT "impossible request in %s.\n"
1550 "please mail a stack trace to "
1551 "linux-scsi@vger.kernel.org",
1552 __FUNCTION__);
1553 BUG();
1554 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 spin_lock(shost->host_lock);
1556
1557 if (!scsi_host_queue_ready(q, shost, sdev))
1558 goto not_ready;
1559 if (sdev->single_lun) {
1560 if (scsi_target(sdev)->starget_sdev_user &&
1561 scsi_target(sdev)->starget_sdev_user != sdev)
1562 goto not_ready;
1563 scsi_target(sdev)->starget_sdev_user = sdev;
1564 }
1565 shost->host_busy++;
1566
1567 /*
1568 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1569 * take the lock again.
1570 */
1571 spin_unlock_irq(shost->host_lock);
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 /*
1574 * Finally, initialize any error handling parameters, and set up
1575 * the timers for timeouts.
1576 */
1577 scsi_init_cmd_errh(cmd);
1578
1579 /*
1580 * Dispatch the command to the low-level driver.
1581 */
1582 rtn = scsi_dispatch_cmd(cmd);
1583 spin_lock_irq(q->queue_lock);
1584 if(rtn) {
1585 /* we're refusing the command; because of
1586 * the way locks get dropped, we need to
1587 * check here if plugging is required */
1588 if(sdev->device_busy == 0)
1589 blk_plug_device(q);
1590
1591 break;
1592 }
1593 }
1594
1595 goto out;
1596
1597 not_ready:
1598 spin_unlock_irq(shost->host_lock);
1599
1600 /*
1601 * lock q, handle tag, requeue req, and decrement device_busy. We
1602 * must return with queue_lock held.
1603 *
1604 * Decrementing device_busy without checking it is OK, as all such
1605 * cases (host limits or settings) should run the queue at some
1606 * later time.
1607 */
1608 spin_lock_irq(q->queue_lock);
1609 blk_requeue_request(q, req);
1610 sdev->device_busy--;
1611 if(sdev->device_busy == 0)
1612 blk_plug_device(q);
1613 out:
1614 /* must be careful here...if we trigger the ->remove() function
1615 * we cannot be holding the q lock */
1616 spin_unlock_irq(q->queue_lock);
1617 put_device(&sdev->sdev_gendev);
1618 spin_lock_irq(q->queue_lock);
1619}
1620
1621u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1622{
1623 struct device *host_dev;
1624 u64 bounce_limit = 0xffffffff;
1625
1626 if (shost->unchecked_isa_dma)
1627 return BLK_BOUNCE_ISA;
1628 /*
1629 * Platforms with virtual-DMA translation
1630 * hardware have no practical limit.
1631 */
1632 if (!PCI_DMA_BUS_IS_PHYS)
1633 return BLK_BOUNCE_ANY;
1634
1635 host_dev = scsi_get_device(shost);
1636 if (host_dev && host_dev->dma_mask)
1637 bounce_limit = *host_dev->dma_mask;
1638
1639 return bounce_limit;
1640}
1641EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1642
1643struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1644{
1645 struct Scsi_Host *shost = sdev->host;
1646 struct request_queue *q;
1647
152587d2005-04-12 16:22:06 -05001648 q = blk_init_queue(scsi_request_fn, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 if (!q)
1650 return NULL;
1651
1652 blk_queue_prep_rq(q, scsi_prep_fn);
1653
1654 blk_queue_max_hw_segments(q, shost->sg_tablesize);
1655 blk_queue_max_phys_segments(q, SCSI_MAX_PHYS_SEGMENTS);
1656 blk_queue_max_sectors(q, shost->max_sectors);
1657 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1658 blk_queue_segment_boundary(q, shost->dma_boundary);
1659 blk_queue_issue_flush_fn(q, scsi_issue_flush_fn);
Jens Axboe1aea6432006-01-09 16:03:03 +01001660 blk_queue_softirq_done(q, scsi_softirq_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 if (!shost->use_clustering)
1663 clear_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
1664 return q;
1665}
1666
1667void scsi_free_queue(struct request_queue *q)
1668{
1669 blk_cleanup_queue(q);
1670}
1671
1672/*
1673 * Function: scsi_block_requests()
1674 *
1675 * Purpose: Utility function used by low-level drivers to prevent further
1676 * commands from being queued to the device.
1677 *
1678 * Arguments: shost - Host in question
1679 *
1680 * Returns: Nothing
1681 *
1682 * Lock status: No locks are assumed held.
1683 *
1684 * Notes: There is no timer nor any other means by which the requests
1685 * get unblocked other than the low-level driver calling
1686 * scsi_unblock_requests().
1687 */
1688void scsi_block_requests(struct Scsi_Host *shost)
1689{
1690 shost->host_self_blocked = 1;
1691}
1692EXPORT_SYMBOL(scsi_block_requests);
1693
1694/*
1695 * Function: scsi_unblock_requests()
1696 *
1697 * Purpose: Utility function used by low-level drivers to allow further
1698 * commands from being queued to the device.
1699 *
1700 * Arguments: shost - Host in question
1701 *
1702 * Returns: Nothing
1703 *
1704 * Lock status: No locks are assumed held.
1705 *
1706 * Notes: There is no timer nor any other means by which the requests
1707 * get unblocked other than the low-level driver calling
1708 * scsi_unblock_requests().
1709 *
1710 * This is done as an API function so that changes to the
1711 * internals of the scsi mid-layer won't require wholesale
1712 * changes to drivers that use this feature.
1713 */
1714void scsi_unblock_requests(struct Scsi_Host *shost)
1715{
1716 shost->host_self_blocked = 0;
1717 scsi_run_host_queues(shost);
1718}
1719EXPORT_SYMBOL(scsi_unblock_requests);
1720
1721int __init scsi_init_queue(void)
1722{
1723 int i;
1724
Mike Christieaa7b5cd2005-11-11 05:31:41 -06001725 scsi_io_context_cache = kmem_cache_create("scsi_io_context",
1726 sizeof(struct scsi_io_context),
1727 0, 0, NULL, NULL);
1728 if (!scsi_io_context_cache) {
1729 printk(KERN_ERR "SCSI: can't init scsi io context cache\n");
1730 return -ENOMEM;
1731 }
1732
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1734 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1735 int size = sgp->size * sizeof(struct scatterlist);
1736
1737 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1738 SLAB_HWCACHE_ALIGN, NULL, NULL);
1739 if (!sgp->slab) {
1740 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1741 sgp->name);
1742 }
1743
Matthew Dobson93d23412006-03-26 01:37:50 -08001744 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1745 sgp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (!sgp->pool) {
1747 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1748 sgp->name);
1749 }
1750 }
1751
1752 return 0;
1753}
1754
1755void scsi_exit_queue(void)
1756{
1757 int i;
1758
Mike Christieaa7b5cd2005-11-11 05:31:41 -06001759 kmem_cache_destroy(scsi_io_context_cache);
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1762 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1763 mempool_destroy(sgp->pool);
1764 kmem_cache_destroy(sgp->slab);
1765 }
1766}
James Bottomley5baba832006-03-18 14:10:35 -06001767
1768/**
1769 * scsi_mode_select - issue a mode select
1770 * @sdev: SCSI device to be queried
1771 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1772 * @sp: Save page bit (0 == don't save, 1 == save)
1773 * @modepage: mode page being requested
1774 * @buffer: request buffer (may not be smaller than eight bytes)
1775 * @len: length of request buffer.
1776 * @timeout: command timeout
1777 * @retries: number of retries before failing
1778 * @data: returns a structure abstracting the mode header data
1779 * @sense: place to put sense data (or NULL if no sense to be collected).
1780 * must be SCSI_SENSE_BUFFERSIZE big.
1781 *
1782 * Returns zero if successful; negative error number or scsi
1783 * status on error
1784 *
1785 */
1786int
1787scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1788 unsigned char *buffer, int len, int timeout, int retries,
1789 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1790{
1791 unsigned char cmd[10];
1792 unsigned char *real_buffer;
1793 int ret;
1794
1795 memset(cmd, 0, sizeof(cmd));
1796 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1797
1798 if (sdev->use_10_for_ms) {
1799 if (len > 65535)
1800 return -EINVAL;
1801 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1802 if (!real_buffer)
1803 return -ENOMEM;
1804 memcpy(real_buffer + 8, buffer, len);
1805 len += 8;
1806 real_buffer[0] = 0;
1807 real_buffer[1] = 0;
1808 real_buffer[2] = data->medium_type;
1809 real_buffer[3] = data->device_specific;
1810 real_buffer[4] = data->longlba ? 0x01 : 0;
1811 real_buffer[5] = 0;
1812 real_buffer[6] = data->block_descriptor_length >> 8;
1813 real_buffer[7] = data->block_descriptor_length;
1814
1815 cmd[0] = MODE_SELECT_10;
1816 cmd[7] = len >> 8;
1817 cmd[8] = len;
1818 } else {
1819 if (len > 255 || data->block_descriptor_length > 255 ||
1820 data->longlba)
1821 return -EINVAL;
1822
1823 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1824 if (!real_buffer)
1825 return -ENOMEM;
1826 memcpy(real_buffer + 4, buffer, len);
1827 len += 4;
1828 real_buffer[0] = 0;
1829 real_buffer[1] = data->medium_type;
1830 real_buffer[2] = data->device_specific;
1831 real_buffer[3] = data->block_descriptor_length;
1832
1833
1834 cmd[0] = MODE_SELECT;
1835 cmd[4] = len;
1836 }
1837
1838 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1839 sshdr, timeout, retries);
1840 kfree(real_buffer);
1841 return ret;
1842}
1843EXPORT_SYMBOL_GPL(scsi_mode_select);
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845/**
James Bottomleyea73a9f2005-08-28 11:33:52 -05001846 * scsi_mode_sense - issue a mode sense, falling back from 10 to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * six bytes if necessary.
James Bottomley1cf72692005-08-28 11:27:01 -05001848 * @sdev: SCSI device to be queried
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 * @dbd: set if mode sense will allow block descriptors to be returned
1850 * @modepage: mode page being requested
1851 * @buffer: request buffer (may not be smaller than eight bytes)
1852 * @len: length of request buffer.
1853 * @timeout: command timeout
1854 * @retries: number of retries before failing
1855 * @data: returns a structure abstracting the mode header data
James Bottomley1cf72692005-08-28 11:27:01 -05001856 * @sense: place to put sense data (or NULL if no sense to be collected).
1857 * must be SCSI_SENSE_BUFFERSIZE big.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 *
1859 * Returns zero if unsuccessful, or the header offset (either 4
1860 * or 8 depending on whether a six or ten byte command was
1861 * issued) if successful.
1862 **/
1863int
James Bottomley1cf72692005-08-28 11:27:01 -05001864scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 unsigned char *buffer, int len, int timeout, int retries,
James Bottomley5baba832006-03-18 14:10:35 -06001866 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1867{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 unsigned char cmd[12];
1869 int use_10_for_ms;
1870 int header_length;
James Bottomley1cf72692005-08-28 11:27:01 -05001871 int result;
James Bottomleyea73a9f2005-08-28 11:33:52 -05001872 struct scsi_sense_hdr my_sshdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 memset(data, 0, sizeof(*data));
1875 memset(&cmd[0], 0, 12);
1876 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1877 cmd[2] = modepage;
1878
James Bottomleyea73a9f2005-08-28 11:33:52 -05001879 /* caller might not be interested in sense, but we need it */
1880 if (!sshdr)
1881 sshdr = &my_sshdr;
1882
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 retry:
James Bottomley1cf72692005-08-28 11:27:01 -05001884 use_10_for_ms = sdev->use_10_for_ms;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
1886 if (use_10_for_ms) {
1887 if (len < 8)
1888 len = 8;
1889
1890 cmd[0] = MODE_SENSE_10;
1891 cmd[8] = len;
1892 header_length = 8;
1893 } else {
1894 if (len < 4)
1895 len = 4;
1896
1897 cmd[0] = MODE_SENSE;
1898 cmd[4] = len;
1899 header_length = 4;
1900 }
1901
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 memset(buffer, 0, len);
1903
James Bottomley1cf72692005-08-28 11:27:01 -05001904 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
James Bottomleyea73a9f2005-08-28 11:33:52 -05001905 sshdr, timeout, retries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
1907 /* This code looks awful: what it's doing is making sure an
1908 * ILLEGAL REQUEST sense return identifies the actual command
1909 * byte as the problem. MODE_SENSE commands can return
1910 * ILLEGAL REQUEST if the code page isn't supported */
1911
James Bottomley1cf72692005-08-28 11:27:01 -05001912 if (use_10_for_ms && !scsi_status_is_good(result) &&
1913 (driver_byte(result) & DRIVER_SENSE)) {
James Bottomleyea73a9f2005-08-28 11:33:52 -05001914 if (scsi_sense_valid(sshdr)) {
1915 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1916 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 /*
1918 * Invalid command operation code
1919 */
James Bottomley1cf72692005-08-28 11:27:01 -05001920 sdev->use_10_for_ms = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 goto retry;
1922 }
1923 }
1924 }
1925
James Bottomley1cf72692005-08-28 11:27:01 -05001926 if(scsi_status_is_good(result)) {
Al Viro6d73c852006-02-23 02:03:16 +01001927 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1928 (modepage == 6 || modepage == 8))) {
1929 /* Initio breakage? */
1930 header_length = 0;
1931 data->length = 13;
1932 data->medium_type = 0;
1933 data->device_specific = 0;
1934 data->longlba = 0;
1935 data->block_descriptor_length = 0;
1936 } else if(use_10_for_ms) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 data->length = buffer[0]*256 + buffer[1] + 2;
1938 data->medium_type = buffer[2];
1939 data->device_specific = buffer[3];
1940 data->longlba = buffer[4] & 0x01;
1941 data->block_descriptor_length = buffer[6]*256
1942 + buffer[7];
1943 } else {
1944 data->length = buffer[0] + 1;
1945 data->medium_type = buffer[1];
1946 data->device_specific = buffer[2];
1947 data->block_descriptor_length = buffer[3];
1948 }
Al Viro6d73c852006-02-23 02:03:16 +01001949 data->header_length = header_length;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 }
1951
James Bottomley1cf72692005-08-28 11:27:01 -05001952 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953}
1954EXPORT_SYMBOL(scsi_mode_sense);
1955
1956int
1957scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries)
1958{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 char cmd[] = {
1960 TEST_UNIT_READY, 0, 0, 0, 0, 0,
1961 };
James Bottomleyea73a9f2005-08-28 11:33:52 -05001962 struct scsi_sense_hdr sshdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 int result;
1964
James Bottomleyea73a9f2005-08-28 11:33:52 -05001965 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, &sshdr,
James Bottomley1cf72692005-08-28 11:27:01 -05001966 timeout, retries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
James Bottomley1cf72692005-08-28 11:27:01 -05001968 if ((driver_byte(result) & DRIVER_SENSE) && sdev->removable) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
James Bottomleyea73a9f2005-08-28 11:33:52 -05001970 if ((scsi_sense_valid(&sshdr)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 ((sshdr.sense_key == UNIT_ATTENTION) ||
1972 (sshdr.sense_key == NOT_READY))) {
1973 sdev->changed = 1;
James Bottomley1cf72692005-08-28 11:27:01 -05001974 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
1976 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 return result;
1978}
1979EXPORT_SYMBOL(scsi_test_unit_ready);
1980
1981/**
1982 * scsi_device_set_state - Take the given device through the device
1983 * state model.
1984 * @sdev: scsi device to change the state of.
1985 * @state: state to change to.
1986 *
1987 * Returns zero if unsuccessful or an error if the requested
1988 * transition is illegal.
1989 **/
1990int
1991scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
1992{
1993 enum scsi_device_state oldstate = sdev->sdev_state;
1994
1995 if (state == oldstate)
1996 return 0;
1997
1998 switch (state) {
1999 case SDEV_CREATED:
2000 /* There are no legal states that come back to
2001 * created. This is the manually initialised start
2002 * state */
2003 goto illegal;
2004
2005 case SDEV_RUNNING:
2006 switch (oldstate) {
2007 case SDEV_CREATED:
2008 case SDEV_OFFLINE:
2009 case SDEV_QUIESCE:
2010 case SDEV_BLOCK:
2011 break;
2012 default:
2013 goto illegal;
2014 }
2015 break;
2016
2017 case SDEV_QUIESCE:
2018 switch (oldstate) {
2019 case SDEV_RUNNING:
2020 case SDEV_OFFLINE:
2021 break;
2022 default:
2023 goto illegal;
2024 }
2025 break;
2026
2027 case SDEV_OFFLINE:
2028 switch (oldstate) {
2029 case SDEV_CREATED:
2030 case SDEV_RUNNING:
2031 case SDEV_QUIESCE:
2032 case SDEV_BLOCK:
2033 break;
2034 default:
2035 goto illegal;
2036 }
2037 break;
2038
2039 case SDEV_BLOCK:
2040 switch (oldstate) {
2041 case SDEV_CREATED:
2042 case SDEV_RUNNING:
2043 break;
2044 default:
2045 goto illegal;
2046 }
2047 break;
2048
2049 case SDEV_CANCEL:
2050 switch (oldstate) {
2051 case SDEV_CREATED:
2052 case SDEV_RUNNING:
2053 case SDEV_OFFLINE:
2054 case SDEV_BLOCK:
2055 break;
2056 default:
2057 goto illegal;
2058 }
2059 break;
2060
2061 case SDEV_DEL:
2062 switch (oldstate) {
2063 case SDEV_CANCEL:
2064 break;
2065 default:
2066 goto illegal;
2067 }
2068 break;
2069
2070 }
2071 sdev->sdev_state = state;
2072 return 0;
2073
2074 illegal:
2075 SCSI_LOG_ERROR_RECOVERY(1,
James Bottomley9ccfc752005-10-02 11:45:08 -05002076 sdev_printk(KERN_ERR, sdev,
2077 "Illegal state transition %s->%s\n",
2078 scsi_device_state_name(oldstate),
2079 scsi_device_state_name(state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 );
2081 return -EINVAL;
2082}
2083EXPORT_SYMBOL(scsi_device_set_state);
2084
2085/**
2086 * scsi_device_quiesce - Block user issued commands.
2087 * @sdev: scsi device to quiesce.
2088 *
2089 * This works by trying to transition to the SDEV_QUIESCE state
2090 * (which must be a legal transition). When the device is in this
2091 * state, only special requests will be accepted, all others will
2092 * be deferred. Since special requests may also be requeued requests,
2093 * a successful return doesn't guarantee the device will be
2094 * totally quiescent.
2095 *
2096 * Must be called with user context, may sleep.
2097 *
2098 * Returns zero if unsuccessful or an error if not.
2099 **/
2100int
2101scsi_device_quiesce(struct scsi_device *sdev)
2102{
2103 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2104 if (err)
2105 return err;
2106
2107 scsi_run_queue(sdev->request_queue);
2108 while (sdev->device_busy) {
2109 msleep_interruptible(200);
2110 scsi_run_queue(sdev->request_queue);
2111 }
2112 return 0;
2113}
2114EXPORT_SYMBOL(scsi_device_quiesce);
2115
2116/**
2117 * scsi_device_resume - Restart user issued commands to a quiesced device.
2118 * @sdev: scsi device to resume.
2119 *
2120 * Moves the device from quiesced back to running and restarts the
2121 * queues.
2122 *
2123 * Must be called with user context, may sleep.
2124 **/
2125void
2126scsi_device_resume(struct scsi_device *sdev)
2127{
2128 if(scsi_device_set_state(sdev, SDEV_RUNNING))
2129 return;
2130 scsi_run_queue(sdev->request_queue);
2131}
2132EXPORT_SYMBOL(scsi_device_resume);
2133
2134static void
2135device_quiesce_fn(struct scsi_device *sdev, void *data)
2136{
2137 scsi_device_quiesce(sdev);
2138}
2139
2140void
2141scsi_target_quiesce(struct scsi_target *starget)
2142{
2143 starget_for_each_device(starget, NULL, device_quiesce_fn);
2144}
2145EXPORT_SYMBOL(scsi_target_quiesce);
2146
2147static void
2148device_resume_fn(struct scsi_device *sdev, void *data)
2149{
2150 scsi_device_resume(sdev);
2151}
2152
2153void
2154scsi_target_resume(struct scsi_target *starget)
2155{
2156 starget_for_each_device(starget, NULL, device_resume_fn);
2157}
2158EXPORT_SYMBOL(scsi_target_resume);
2159
2160/**
2161 * scsi_internal_device_block - internal function to put a device
2162 * temporarily into the SDEV_BLOCK state
2163 * @sdev: device to block
2164 *
2165 * Block request made by scsi lld's to temporarily stop all
2166 * scsi commands on the specified device. Called from interrupt
2167 * or normal process context.
2168 *
2169 * Returns zero if successful or error if not
2170 *
2171 * Notes:
2172 * This routine transitions the device to the SDEV_BLOCK state
2173 * (which must be a legal transition). When the device is in this
2174 * state, all commands are deferred until the scsi lld reenables
2175 * the device with scsi_device_unblock or device_block_tmo fires.
2176 * This routine assumes the host_lock is held on entry.
2177 **/
2178int
2179scsi_internal_device_block(struct scsi_device *sdev)
2180{
2181 request_queue_t *q = sdev->request_queue;
2182 unsigned long flags;
2183 int err = 0;
2184
2185 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2186 if (err)
2187 return err;
2188
2189 /*
2190 * The device has transitioned to SDEV_BLOCK. Stop the
2191 * block layer from calling the midlayer with this device's
2192 * request queue.
2193 */
2194 spin_lock_irqsave(q->queue_lock, flags);
2195 blk_stop_queue(q);
2196 spin_unlock_irqrestore(q->queue_lock, flags);
2197
2198 return 0;
2199}
2200EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2201
2202/**
2203 * scsi_internal_device_unblock - resume a device after a block request
2204 * @sdev: device to resume
2205 *
2206 * Called by scsi lld's or the midlayer to restart the device queue
2207 * for the previously suspended scsi device. Called from interrupt or
2208 * normal process context.
2209 *
2210 * Returns zero if successful or error if not.
2211 *
2212 * Notes:
2213 * This routine transitions the device to the SDEV_RUNNING state
2214 * (which must be a legal transition) allowing the midlayer to
2215 * goose the queue for this device. This routine assumes the
2216 * host_lock is held upon entry.
2217 **/
2218int
2219scsi_internal_device_unblock(struct scsi_device *sdev)
2220{
2221 request_queue_t *q = sdev->request_queue;
2222 int err;
2223 unsigned long flags;
2224
2225 /*
2226 * Try to transition the scsi device to SDEV_RUNNING
2227 * and goose the device queue if successful.
2228 */
2229 err = scsi_device_set_state(sdev, SDEV_RUNNING);
2230 if (err)
2231 return err;
2232
2233 spin_lock_irqsave(q->queue_lock, flags);
2234 blk_start_queue(q);
2235 spin_unlock_irqrestore(q->queue_lock, flags);
2236
2237 return 0;
2238}
2239EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2240
2241static void
2242device_block(struct scsi_device *sdev, void *data)
2243{
2244 scsi_internal_device_block(sdev);
2245}
2246
2247static int
2248target_block(struct device *dev, void *data)
2249{
2250 if (scsi_is_target_device(dev))
2251 starget_for_each_device(to_scsi_target(dev), NULL,
2252 device_block);
2253 return 0;
2254}
2255
2256void
2257scsi_target_block(struct device *dev)
2258{
2259 if (scsi_is_target_device(dev))
2260 starget_for_each_device(to_scsi_target(dev), NULL,
2261 device_block);
2262 else
2263 device_for_each_child(dev, NULL, target_block);
2264}
2265EXPORT_SYMBOL_GPL(scsi_target_block);
2266
2267static void
2268device_unblock(struct scsi_device *sdev, void *data)
2269{
2270 scsi_internal_device_unblock(sdev);
2271}
2272
2273static int
2274target_unblock(struct device *dev, void *data)
2275{
2276 if (scsi_is_target_device(dev))
2277 starget_for_each_device(to_scsi_target(dev), NULL,
2278 device_unblock);
2279 return 0;
2280}
2281
2282void
2283scsi_target_unblock(struct device *dev)
2284{
2285 if (scsi_is_target_device(dev))
2286 starget_for_each_device(to_scsi_target(dev), NULL,
2287 device_unblock);
2288 else
2289 device_for_each_child(dev, NULL, target_unblock);
2290}
2291EXPORT_SYMBOL_GPL(scsi_target_unblock);
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002292
2293/**
2294 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2295 * @sg: scatter-gather list
2296 * @sg_count: number of segments in sg
2297 * @offset: offset in bytes into sg, on return offset into the mapped area
2298 * @len: bytes to map, on return number of bytes mapped
2299 *
2300 * Returns virtual address of the start of the mapped page
2301 */
2302void *scsi_kmap_atomic_sg(struct scatterlist *sg, int sg_count,
2303 size_t *offset, size_t *len)
2304{
2305 int i;
2306 size_t sg_len = 0, len_complete = 0;
2307 struct page *page;
2308
2309 for (i = 0; i < sg_count; i++) {
2310 len_complete = sg_len; /* Complete sg-entries */
2311 sg_len += sg[i].length;
2312 if (sg_len > *offset)
2313 break;
2314 }
2315
2316 if (unlikely(i == sg_count)) {
Andrew Morton169e1a22006-04-18 21:09:08 -07002317 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2318 "elements %d\n",
Guennadi Liakhovetskicdb8c2a2006-04-02 21:57:43 +02002319 __FUNCTION__, sg_len, *offset, sg_count);
2320 WARN_ON(1);
2321 return NULL;
2322 }
2323
2324 /* Offset starting from the beginning of first page in this sg-entry */
2325 *offset = *offset - len_complete + sg[i].offset;
2326
2327 /* Assumption: contiguous pages can be accessed as "page + i" */
2328 page = nth_page(sg[i].page, (*offset >> PAGE_SHIFT));
2329 *offset &= ~PAGE_MASK;
2330
2331 /* Bytes in this sg-entry from *offset to the end of the page */
2332 sg_len = PAGE_SIZE - *offset;
2333 if (*len > sg_len)
2334 *len = sg_len;
2335
2336 return kmap_atomic(page, KM_BIO_SRC_IRQ);
2337}
2338EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2339
2340/**
2341 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously
2342 * mapped with scsi_kmap_atomic_sg
2343 * @virt: virtual address to be unmapped
2344 */
2345void scsi_kunmap_atomic_sg(void *virt)
2346{
2347 kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2348}
2349EXPORT_SYMBOL(scsi_kunmap_atomic_sg);