xref: /linux/drivers/scsi/sg.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
1 /*
2  *  History:
3  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
4  *           to allow user process control of SCSI devices.
5  *  Development Sponsored by Killy Corp. NY NY
6  *
7  * Original driver (sg.c):
8  *        Copyright (C) 1992 Lawrence Foard
9  * Version 2 and 3 extensions to driver:
10  *        Copyright (C) 1998 - 2014 Douglas Gilbert
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License as published by
14  * the Free Software Foundation; either version 2, or (at your option)
15  * any later version.
16  *
17  */
18 
19 static int sg_version_num = 30536;	/* 2 digits for each component */
20 #define SG_VERSION_STR "3.5.36"
21 
22 /*
23  *  D. P. Gilbert (dgilbert@interlog.com), notes:
24  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
25  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
26  *        (otherwise the macros compile to empty statements).
27  *
28  */
29 #include <linux/module.h>
30 
31 #include <linux/fs.h>
32 #include <linux/kernel.h>
33 #include <linux/sched.h>
34 #include <linux/string.h>
35 #include <linux/mm.h>
36 #include <linux/errno.h>
37 #include <linux/mtio.h>
38 #include <linux/ioctl.h>
39 #include <linux/slab.h>
40 #include <linux/fcntl.h>
41 #include <linux/init.h>
42 #include <linux/poll.h>
43 #include <linux/moduleparam.h>
44 #include <linux/cdev.h>
45 #include <linux/idr.h>
46 #include <linux/seq_file.h>
47 #include <linux/blkdev.h>
48 #include <linux/delay.h>
49 #include <linux/blktrace_api.h>
50 #include <linux/mutex.h>
51 #include <linux/atomic.h>
52 #include <linux/ratelimit.h>
53 #include <linux/uio.h>
54 
55 #include "scsi.h"
56 #include <scsi/scsi_dbg.h>
57 #include <scsi/scsi_host.h>
58 #include <scsi/scsi_driver.h>
59 #include <scsi/scsi_ioctl.h>
60 #include <scsi/sg.h>
61 
62 #include "scsi_logging.h"
63 
64 #ifdef CONFIG_SCSI_PROC_FS
65 #include <linux/proc_fs.h>
66 static char *sg_version_date = "20140603";
67 
68 static int sg_proc_init(void);
69 static void sg_proc_cleanup(void);
70 #endif
71 
72 #define SG_ALLOW_DIO_DEF 0
73 
74 #define SG_MAX_DEVS 32768
75 
76 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
77  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
78  * than 16 bytes are "variable length" whose length is a multiple of 4
79  */
80 #define SG_MAX_CDB_SIZE 252
81 
82 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
83 
84 int sg_big_buff = SG_DEF_RESERVED_SIZE;
85 /* N.B. This variable is readable and writeable via
86    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
87    of this size (or less if there is not enough memory) will be reserved
88    for use by this file descriptor. [Deprecated usage: this variable is also
89    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
90    the kernel (i.e. it is not a module).] */
91 static int def_reserved_size = -1;	/* picks up init parameter */
92 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
93 
94 static int scatter_elem_sz = SG_SCATTER_SZ;
95 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
96 
97 #define SG_SECTOR_SZ 512
98 
99 static int sg_add_device(struct device *, struct class_interface *);
100 static void sg_remove_device(struct device *, struct class_interface *);
101 
102 static DEFINE_IDR(sg_index_idr);
103 static DEFINE_RWLOCK(sg_index_lock);	/* Also used to lock
104 							   file descriptor list for device */
105 
106 static struct class_interface sg_interface = {
107 	.add_dev        = sg_add_device,
108 	.remove_dev     = sg_remove_device,
109 };
110 
111 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
112 	unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
113 	unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
114 	unsigned bufflen;	/* Size of (aggregate) data buffer */
115 	struct page **pages;
116 	int page_order;
117 	char dio_in_use;	/* 0->indirect IO (or mmap), 1->dio */
118 	unsigned char cmd_opcode; /* first byte of command */
119 } Sg_scatter_hold;
120 
121 struct sg_device;		/* forward declarations */
122 struct sg_fd;
123 
124 typedef struct sg_request {	/* SG_MAX_QUEUE requests outstanding per file */
125 	struct list_head entry;	/* list entry */
126 	struct sg_fd *parentfp;	/* NULL -> not in use */
127 	Sg_scatter_hold data;	/* hold buffer, perhaps scatter list */
128 	sg_io_hdr_t header;	/* scsi command+info, see <scsi/sg.h> */
129 	unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
130 	char res_used;		/* 1 -> using reserve buffer, 0 -> not ... */
131 	char orphan;		/* 1 -> drop on sight, 0 -> normal */
132 	char sg_io_owned;	/* 1 -> packet belongs to SG_IO */
133 	/* done protected by rq_list_lock */
134 	char done;		/* 0->before bh, 1->before read, 2->read */
135 	struct request *rq;
136 	struct bio *bio;
137 	struct execute_work ew;
138 } Sg_request;
139 
140 typedef struct sg_fd {		/* holds the state of a file descriptor */
141 	struct list_head sfd_siblings;  /* protected by device's sfd_lock */
142 	struct sg_device *parentdp;	/* owning device */
143 	wait_queue_head_t read_wait;	/* queue read until command done */
144 	rwlock_t rq_list_lock;	/* protect access to list in req_arr */
145 	struct mutex f_mutex;	/* protect against changes in this fd */
146 	int timeout;		/* defaults to SG_DEFAULT_TIMEOUT      */
147 	int timeout_user;	/* defaults to SG_DEFAULT_TIMEOUT_USER */
148 	Sg_scatter_hold reserve;	/* buffer held for this file descriptor */
149 	struct list_head rq_list; /* head of request list */
150 	struct fasync_struct *async_qp;	/* used by asynchronous notification */
151 	Sg_request req_arr[SG_MAX_QUEUE];	/* used as singly-linked list */
152 	char force_packid;	/* 1 -> pack_id input to read(), 0 -> ignored */
153 	char cmd_q;		/* 1 -> allow command queuing, 0 -> don't */
154 	unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
155 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
156 	char mmap_called;	/* 0 -> mmap() never called on this fd */
157 	char res_in_use;	/* 1 -> 'reserve' array in use */
158 	struct kref f_ref;
159 	struct execute_work ew;
160 } Sg_fd;
161 
162 typedef struct sg_device { /* holds the state of each scsi generic device */
163 	struct scsi_device *device;
164 	wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
165 	struct mutex open_rel_lock;     /* held when in open() or release() */
166 	int sg_tablesize;	/* adapter's max scatter-gather table size */
167 	u32 index;		/* device index number */
168 	struct list_head sfds;
169 	rwlock_t sfd_lock;      /* protect access to sfd list */
170 	atomic_t detaching;     /* 0->device usable, 1->device detaching */
171 	bool exclude;		/* 1->open(O_EXCL) succeeded and is active */
172 	int open_cnt;		/* count of opens (perhaps < num(sfds) ) */
173 	char sgdebug;		/* 0->off, 1->sense, 9->dump dev, 10-> all devs */
174 	struct gendisk *disk;
175 	struct cdev * cdev;	/* char_dev [sysfs: /sys/cdev/major/sg<n>] */
176 	struct kref d_ref;
177 } Sg_device;
178 
179 /* tasklet or soft irq callback */
180 static void sg_rq_end_io(struct request *rq, int uptodate);
181 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
182 static int sg_finish_rem_req(Sg_request * srp);
183 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
184 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
185 			   Sg_request * srp);
186 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
187 			const char __user *buf, size_t count, int blocking,
188 			int read_only, int sg_io_owned, Sg_request **o_srp);
189 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
190 			   unsigned char *cmnd, int timeout, int blocking);
191 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
192 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
193 static void sg_build_reserve(Sg_fd * sfp, int req_size);
194 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
195 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
196 static Sg_fd *sg_add_sfp(Sg_device * sdp);
197 static void sg_remove_sfp(struct kref *);
198 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
199 static Sg_request *sg_add_request(Sg_fd * sfp);
200 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
201 static Sg_device *sg_get_dev(int dev);
202 static void sg_device_destroy(struct kref *kref);
203 
204 #define SZ_SG_HEADER sizeof(struct sg_header)
205 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
206 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
207 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
208 
209 #define sg_printk(prefix, sdp, fmt, a...) \
210 	sdev_prefix_printk(prefix, (sdp)->device,		\
211 			   (sdp)->disk->disk_name, fmt, ##a)
212 
213 static int sg_allow_access(struct file *filp, unsigned char *cmd)
214 {
215 	struct sg_fd *sfp = filp->private_data;
216 
217 	if (sfp->parentdp->device->type == TYPE_SCANNER)
218 		return 0;
219 
220 	return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
221 }
222 
223 static int
224 open_wait(Sg_device *sdp, int flags)
225 {
226 	int retval = 0;
227 
228 	if (flags & O_EXCL) {
229 		while (sdp->open_cnt > 0) {
230 			mutex_unlock(&sdp->open_rel_lock);
231 			retval = wait_event_interruptible(sdp->open_wait,
232 					(atomic_read(&sdp->detaching) ||
233 					 !sdp->open_cnt));
234 			mutex_lock(&sdp->open_rel_lock);
235 
236 			if (retval) /* -ERESTARTSYS */
237 				return retval;
238 			if (atomic_read(&sdp->detaching))
239 				return -ENODEV;
240 		}
241 	} else {
242 		while (sdp->exclude) {
243 			mutex_unlock(&sdp->open_rel_lock);
244 			retval = wait_event_interruptible(sdp->open_wait,
245 					(atomic_read(&sdp->detaching) ||
246 					 !sdp->exclude));
247 			mutex_lock(&sdp->open_rel_lock);
248 
249 			if (retval) /* -ERESTARTSYS */
250 				return retval;
251 			if (atomic_read(&sdp->detaching))
252 				return -ENODEV;
253 		}
254 	}
255 
256 	return retval;
257 }
258 
259 /* Returns 0 on success, else a negated errno value */
260 static int
261 sg_open(struct inode *inode, struct file *filp)
262 {
263 	int dev = iminor(inode);
264 	int flags = filp->f_flags;
265 	struct request_queue *q;
266 	Sg_device *sdp;
267 	Sg_fd *sfp;
268 	int retval;
269 
270 	nonseekable_open(inode, filp);
271 	if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
272 		return -EPERM; /* Can't lock it with read only access */
273 	sdp = sg_get_dev(dev);
274 	if (IS_ERR(sdp))
275 		return PTR_ERR(sdp);
276 
277 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
278 				      "sg_open: flags=0x%x\n", flags));
279 
280 	/* This driver's module count bumped by fops_get in <linux/fs.h> */
281 	/* Prevent the device driver from vanishing while we sleep */
282 	retval = scsi_device_get(sdp->device);
283 	if (retval)
284 		goto sg_put;
285 
286 	retval = scsi_autopm_get_device(sdp->device);
287 	if (retval)
288 		goto sdp_put;
289 
290 	/* scsi_block_when_processing_errors() may block so bypass
291 	 * check if O_NONBLOCK. Permits SCSI commands to be issued
292 	 * during error recovery. Tread carefully. */
293 	if (!((flags & O_NONBLOCK) ||
294 	      scsi_block_when_processing_errors(sdp->device))) {
295 		retval = -ENXIO;
296 		/* we are in error recovery for this device */
297 		goto error_out;
298 	}
299 
300 	mutex_lock(&sdp->open_rel_lock);
301 	if (flags & O_NONBLOCK) {
302 		if (flags & O_EXCL) {
303 			if (sdp->open_cnt > 0) {
304 				retval = -EBUSY;
305 				goto error_mutex_locked;
306 			}
307 		} else {
308 			if (sdp->exclude) {
309 				retval = -EBUSY;
310 				goto error_mutex_locked;
311 			}
312 		}
313 	} else {
314 		retval = open_wait(sdp, flags);
315 		if (retval) /* -ERESTARTSYS or -ENODEV */
316 			goto error_mutex_locked;
317 	}
318 
319 	/* N.B. at this point we are holding the open_rel_lock */
320 	if (flags & O_EXCL)
321 		sdp->exclude = true;
322 
323 	if (sdp->open_cnt < 1) {  /* no existing opens */
324 		sdp->sgdebug = 0;
325 		q = sdp->device->request_queue;
326 		sdp->sg_tablesize = queue_max_segments(q);
327 	}
328 	sfp = sg_add_sfp(sdp);
329 	if (IS_ERR(sfp)) {
330 		retval = PTR_ERR(sfp);
331 		goto out_undo;
332 	}
333 
334 	filp->private_data = sfp;
335 	sdp->open_cnt++;
336 	mutex_unlock(&sdp->open_rel_lock);
337 
338 	retval = 0;
339 sg_put:
340 	kref_put(&sdp->d_ref, sg_device_destroy);
341 	return retval;
342 
343 out_undo:
344 	if (flags & O_EXCL) {
345 		sdp->exclude = false;   /* undo if error */
346 		wake_up_interruptible(&sdp->open_wait);
347 	}
348 error_mutex_locked:
349 	mutex_unlock(&sdp->open_rel_lock);
350 error_out:
351 	scsi_autopm_put_device(sdp->device);
352 sdp_put:
353 	scsi_device_put(sdp->device);
354 	goto sg_put;
355 }
356 
357 /* Release resources associated with a successful sg_open()
358  * Returns 0 on success, else a negated errno value */
359 static int
360 sg_release(struct inode *inode, struct file *filp)
361 {
362 	Sg_device *sdp;
363 	Sg_fd *sfp;
364 
365 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
366 		return -ENXIO;
367 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
368 
369 	mutex_lock(&sdp->open_rel_lock);
370 	scsi_autopm_put_device(sdp->device);
371 	kref_put(&sfp->f_ref, sg_remove_sfp);
372 	sdp->open_cnt--;
373 
374 	/* possibly many open()s waiting on exlude clearing, start many;
375 	 * only open(O_EXCL)s wait on 0==open_cnt so only start one */
376 	if (sdp->exclude) {
377 		sdp->exclude = false;
378 		wake_up_interruptible_all(&sdp->open_wait);
379 	} else if (0 == sdp->open_cnt) {
380 		wake_up_interruptible(&sdp->open_wait);
381 	}
382 	mutex_unlock(&sdp->open_rel_lock);
383 	return 0;
384 }
385 
386 static ssize_t
387 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
388 {
389 	Sg_device *sdp;
390 	Sg_fd *sfp;
391 	Sg_request *srp;
392 	int req_pack_id = -1;
393 	sg_io_hdr_t *hp;
394 	struct sg_header *old_hdr = NULL;
395 	int retval = 0;
396 
397 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
398 		return -ENXIO;
399 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
400 				      "sg_read: count=%d\n", (int) count));
401 
402 	if (!access_ok(VERIFY_WRITE, buf, count))
403 		return -EFAULT;
404 	if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
405 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
406 		if (!old_hdr)
407 			return -ENOMEM;
408 		if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
409 			retval = -EFAULT;
410 			goto free_old_hdr;
411 		}
412 		if (old_hdr->reply_len < 0) {
413 			if (count >= SZ_SG_IO_HDR) {
414 				sg_io_hdr_t *new_hdr;
415 				new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
416 				if (!new_hdr) {
417 					retval = -ENOMEM;
418 					goto free_old_hdr;
419 				}
420 				retval =__copy_from_user
421 				    (new_hdr, buf, SZ_SG_IO_HDR);
422 				req_pack_id = new_hdr->pack_id;
423 				kfree(new_hdr);
424 				if (retval) {
425 					retval = -EFAULT;
426 					goto free_old_hdr;
427 				}
428 			}
429 		} else
430 			req_pack_id = old_hdr->pack_id;
431 	}
432 	srp = sg_get_rq_mark(sfp, req_pack_id);
433 	if (!srp) {		/* now wait on packet to arrive */
434 		if (atomic_read(&sdp->detaching)) {
435 			retval = -ENODEV;
436 			goto free_old_hdr;
437 		}
438 		if (filp->f_flags & O_NONBLOCK) {
439 			retval = -EAGAIN;
440 			goto free_old_hdr;
441 		}
442 		retval = wait_event_interruptible(sfp->read_wait,
443 			(atomic_read(&sdp->detaching) ||
444 			(srp = sg_get_rq_mark(sfp, req_pack_id))));
445 		if (atomic_read(&sdp->detaching)) {
446 			retval = -ENODEV;
447 			goto free_old_hdr;
448 		}
449 		if (retval) {
450 			/* -ERESTARTSYS as signal hit process */
451 			goto free_old_hdr;
452 		}
453 	}
454 	if (srp->header.interface_id != '\0') {
455 		retval = sg_new_read(sfp, buf, count, srp);
456 		goto free_old_hdr;
457 	}
458 
459 	hp = &srp->header;
460 	if (old_hdr == NULL) {
461 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
462 		if (! old_hdr) {
463 			retval = -ENOMEM;
464 			goto free_old_hdr;
465 		}
466 	}
467 	memset(old_hdr, 0, SZ_SG_HEADER);
468 	old_hdr->reply_len = (int) hp->timeout;
469 	old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
470 	old_hdr->pack_id = hp->pack_id;
471 	old_hdr->twelve_byte =
472 	    ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
473 	old_hdr->target_status = hp->masked_status;
474 	old_hdr->host_status = hp->host_status;
475 	old_hdr->driver_status = hp->driver_status;
476 	if ((CHECK_CONDITION & hp->masked_status) ||
477 	    (DRIVER_SENSE & hp->driver_status))
478 		memcpy(old_hdr->sense_buffer, srp->sense_b,
479 		       sizeof (old_hdr->sense_buffer));
480 	switch (hp->host_status) {
481 	/* This setup of 'result' is for backward compatibility and is best
482 	   ignored by the user who should use target, host + driver status */
483 	case DID_OK:
484 	case DID_PASSTHROUGH:
485 	case DID_SOFT_ERROR:
486 		old_hdr->result = 0;
487 		break;
488 	case DID_NO_CONNECT:
489 	case DID_BUS_BUSY:
490 	case DID_TIME_OUT:
491 		old_hdr->result = EBUSY;
492 		break;
493 	case DID_BAD_TARGET:
494 	case DID_ABORT:
495 	case DID_PARITY:
496 	case DID_RESET:
497 	case DID_BAD_INTR:
498 		old_hdr->result = EIO;
499 		break;
500 	case DID_ERROR:
501 		old_hdr->result = (srp->sense_b[0] == 0 &&
502 				  hp->masked_status == GOOD) ? 0 : EIO;
503 		break;
504 	default:
505 		old_hdr->result = EIO;
506 		break;
507 	}
508 
509 	/* Now copy the result back to the user buffer.  */
510 	if (count >= SZ_SG_HEADER) {
511 		if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
512 			retval = -EFAULT;
513 			goto free_old_hdr;
514 		}
515 		buf += SZ_SG_HEADER;
516 		if (count > old_hdr->reply_len)
517 			count = old_hdr->reply_len;
518 		if (count > SZ_SG_HEADER) {
519 			if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
520 				retval = -EFAULT;
521 				goto free_old_hdr;
522 			}
523 		}
524 	} else
525 		count = (old_hdr->result == 0) ? 0 : -EIO;
526 	sg_finish_rem_req(srp);
527 	sg_remove_request(sfp, srp);
528 	retval = count;
529 free_old_hdr:
530 	kfree(old_hdr);
531 	return retval;
532 }
533 
534 static ssize_t
535 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
536 {
537 	sg_io_hdr_t *hp = &srp->header;
538 	int err = 0, err2;
539 	int len;
540 
541 	if (count < SZ_SG_IO_HDR) {
542 		err = -EINVAL;
543 		goto err_out;
544 	}
545 	hp->sb_len_wr = 0;
546 	if ((hp->mx_sb_len > 0) && hp->sbp) {
547 		if ((CHECK_CONDITION & hp->masked_status) ||
548 		    (DRIVER_SENSE & hp->driver_status)) {
549 			int sb_len = SCSI_SENSE_BUFFERSIZE;
550 			sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
551 			len = 8 + (int) srp->sense_b[7];	/* Additional sense length field */
552 			len = (len > sb_len) ? sb_len : len;
553 			if (copy_to_user(hp->sbp, srp->sense_b, len)) {
554 				err = -EFAULT;
555 				goto err_out;
556 			}
557 			hp->sb_len_wr = len;
558 		}
559 	}
560 	if (hp->masked_status || hp->host_status || hp->driver_status)
561 		hp->info |= SG_INFO_CHECK;
562 	if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
563 		err = -EFAULT;
564 		goto err_out;
565 	}
566 err_out:
567 	err2 = sg_finish_rem_req(srp);
568 	sg_remove_request(sfp, srp);
569 	return err ? : err2 ? : count;
570 }
571 
572 static ssize_t
573 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
574 {
575 	int mxsize, cmd_size, k;
576 	int input_size, blocking;
577 	unsigned char opcode;
578 	Sg_device *sdp;
579 	Sg_fd *sfp;
580 	Sg_request *srp;
581 	struct sg_header old_hdr;
582 	sg_io_hdr_t *hp;
583 	unsigned char cmnd[SG_MAX_CDB_SIZE];
584 
585 	if (unlikely(uaccess_kernel()))
586 		return -EINVAL;
587 
588 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
589 		return -ENXIO;
590 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
591 				      "sg_write: count=%d\n", (int) count));
592 	if (atomic_read(&sdp->detaching))
593 		return -ENODEV;
594 	if (!((filp->f_flags & O_NONBLOCK) ||
595 	      scsi_block_when_processing_errors(sdp->device)))
596 		return -ENXIO;
597 
598 	if (!access_ok(VERIFY_READ, buf, count))
599 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
600 	if (count < SZ_SG_HEADER)
601 		return -EIO;
602 	if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
603 		return -EFAULT;
604 	blocking = !(filp->f_flags & O_NONBLOCK);
605 	if (old_hdr.reply_len < 0)
606 		return sg_new_write(sfp, filp, buf, count,
607 				    blocking, 0, 0, NULL);
608 	if (count < (SZ_SG_HEADER + 6))
609 		return -EIO;	/* The minimum scsi command length is 6 bytes. */
610 
611 	if (!(srp = sg_add_request(sfp))) {
612 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
613 					      "sg_write: queue full\n"));
614 		return -EDOM;
615 	}
616 	buf += SZ_SG_HEADER;
617 	__get_user(opcode, buf);
618 	mutex_lock(&sfp->f_mutex);
619 	if (sfp->next_cmd_len > 0) {
620 		cmd_size = sfp->next_cmd_len;
621 		sfp->next_cmd_len = 0;	/* reset so only this write() effected */
622 	} else {
623 		cmd_size = COMMAND_SIZE(opcode);	/* based on SCSI command group */
624 		if ((opcode >= 0xc0) && old_hdr.twelve_byte)
625 			cmd_size = 12;
626 	}
627 	mutex_unlock(&sfp->f_mutex);
628 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
629 		"sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
630 /* Determine buffer size.  */
631 	input_size = count - cmd_size;
632 	mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
633 	mxsize -= SZ_SG_HEADER;
634 	input_size -= SZ_SG_HEADER;
635 	if (input_size < 0) {
636 		sg_remove_request(sfp, srp);
637 		return -EIO;	/* User did not pass enough bytes for this command. */
638 	}
639 	hp = &srp->header;
640 	hp->interface_id = '\0';	/* indicator of old interface tunnelled */
641 	hp->cmd_len = (unsigned char) cmd_size;
642 	hp->iovec_count = 0;
643 	hp->mx_sb_len = 0;
644 	if (input_size > 0)
645 		hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
646 		    SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
647 	else
648 		hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
649 	hp->dxfer_len = mxsize;
650 	if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
651 	    (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
652 		hp->dxferp = (char __user *)buf + cmd_size;
653 	else
654 		hp->dxferp = NULL;
655 	hp->sbp = NULL;
656 	hp->timeout = old_hdr.reply_len;	/* structure abuse ... */
657 	hp->flags = input_size;	/* structure abuse ... */
658 	hp->pack_id = old_hdr.pack_id;
659 	hp->usr_ptr = NULL;
660 	if (__copy_from_user(cmnd, buf, cmd_size))
661 		return -EFAULT;
662 	/*
663 	 * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
664 	 * but is is possible that the app intended SG_DXFER_TO_DEV, because there
665 	 * is a non-zero input_size, so emit a warning.
666 	 */
667 	if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
668 		printk_ratelimited(KERN_WARNING
669 				   "sg_write: data in/out %d/%d bytes "
670 				   "for SCSI command 0x%x-- guessing "
671 				   "data in;\n   program %s not setting "
672 				   "count and/or reply_len properly\n",
673 				   old_hdr.reply_len - (int)SZ_SG_HEADER,
674 				   input_size, (unsigned int) cmnd[0],
675 				   current->comm);
676 	}
677 	k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
678 	return (k < 0) ? k : count;
679 }
680 
681 static ssize_t
682 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
683 		 size_t count, int blocking, int read_only, int sg_io_owned,
684 		 Sg_request **o_srp)
685 {
686 	int k;
687 	Sg_request *srp;
688 	sg_io_hdr_t *hp;
689 	unsigned char cmnd[SG_MAX_CDB_SIZE];
690 	int timeout;
691 	unsigned long ul_timeout;
692 
693 	if (count < SZ_SG_IO_HDR)
694 		return -EINVAL;
695 	if (!access_ok(VERIFY_READ, buf, count))
696 		return -EFAULT; /* protects following copy_from_user()s + get_user()s */
697 
698 	sfp->cmd_q = 1;	/* when sg_io_hdr seen, set command queuing on */
699 	if (!(srp = sg_add_request(sfp))) {
700 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
701 					      "sg_new_write: queue full\n"));
702 		return -EDOM;
703 	}
704 	srp->sg_io_owned = sg_io_owned;
705 	hp = &srp->header;
706 	if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
707 		sg_remove_request(sfp, srp);
708 		return -EFAULT;
709 	}
710 	if (hp->interface_id != 'S') {
711 		sg_remove_request(sfp, srp);
712 		return -ENOSYS;
713 	}
714 	if (hp->flags & SG_FLAG_MMAP_IO) {
715 		if (hp->dxfer_len > sfp->reserve.bufflen) {
716 			sg_remove_request(sfp, srp);
717 			return -ENOMEM;	/* MMAP_IO size must fit in reserve buffer */
718 		}
719 		if (hp->flags & SG_FLAG_DIRECT_IO) {
720 			sg_remove_request(sfp, srp);
721 			return -EINVAL;	/* either MMAP_IO or DIRECT_IO (not both) */
722 		}
723 		if (sfp->res_in_use) {
724 			sg_remove_request(sfp, srp);
725 			return -EBUSY;	/* reserve buffer already being used */
726 		}
727 	}
728 	ul_timeout = msecs_to_jiffies(srp->header.timeout);
729 	timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
730 	if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
731 		sg_remove_request(sfp, srp);
732 		return -EMSGSIZE;
733 	}
734 	if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
735 		sg_remove_request(sfp, srp);
736 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
737 	}
738 	if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
739 		sg_remove_request(sfp, srp);
740 		return -EFAULT;
741 	}
742 	if (read_only && sg_allow_access(file, cmnd)) {
743 		sg_remove_request(sfp, srp);
744 		return -EPERM;
745 	}
746 	k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
747 	if (k < 0)
748 		return k;
749 	if (o_srp)
750 		*o_srp = srp;
751 	return count;
752 }
753 
754 static bool sg_is_valid_dxfer(sg_io_hdr_t *hp)
755 {
756 	switch (hp->dxfer_direction) {
757 	case SG_DXFER_NONE:
758 		if (hp->dxferp || hp->dxfer_len > 0)
759 			return false;
760 		return true;
761 	case SG_DXFER_TO_DEV:
762 	case SG_DXFER_FROM_DEV:
763 	case SG_DXFER_TO_FROM_DEV:
764 		if (!hp->dxferp || hp->dxfer_len == 0)
765 			return false;
766 		return true;
767 	case SG_DXFER_UNKNOWN:
768 		if ((!hp->dxferp && hp->dxfer_len) ||
769 		    (hp->dxferp && hp->dxfer_len == 0))
770 			return false;
771 		return true;
772 	default:
773 		return false;
774 	}
775 }
776 
777 static int
778 sg_common_write(Sg_fd * sfp, Sg_request * srp,
779 		unsigned char *cmnd, int timeout, int blocking)
780 {
781 	int k, at_head;
782 	Sg_device *sdp = sfp->parentdp;
783 	sg_io_hdr_t *hp = &srp->header;
784 
785 	srp->data.cmd_opcode = cmnd[0];	/* hold opcode of command */
786 	hp->status = 0;
787 	hp->masked_status = 0;
788 	hp->msg_status = 0;
789 	hp->info = 0;
790 	hp->host_status = 0;
791 	hp->driver_status = 0;
792 	hp->resid = 0;
793 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
794 			"sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
795 			(int) cmnd[0], (int) hp->cmd_len));
796 
797 	if (!sg_is_valid_dxfer(hp))
798 		return -EINVAL;
799 
800 	k = sg_start_req(srp, cmnd);
801 	if (k) {
802 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
803 			"sg_common_write: start_req err=%d\n", k));
804 		sg_finish_rem_req(srp);
805 		sg_remove_request(sfp, srp);
806 		return k;	/* probably out of space --> ENOMEM */
807 	}
808 	if (atomic_read(&sdp->detaching)) {
809 		if (srp->bio) {
810 			scsi_req_free_cmd(scsi_req(srp->rq));
811 			blk_end_request_all(srp->rq, -EIO);
812 			srp->rq = NULL;
813 		}
814 
815 		sg_finish_rem_req(srp);
816 		sg_remove_request(sfp, srp);
817 		return -ENODEV;
818 	}
819 
820 	hp->duration = jiffies_to_msecs(jiffies);
821 	if (hp->interface_id != '\0' &&	/* v3 (or later) interface */
822 	    (SG_FLAG_Q_AT_TAIL & hp->flags))
823 		at_head = 0;
824 	else
825 		at_head = 1;
826 
827 	srp->rq->timeout = timeout;
828 	kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
829 	blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
830 			      srp->rq, at_head, sg_rq_end_io);
831 	return 0;
832 }
833 
834 static int srp_done(Sg_fd *sfp, Sg_request *srp)
835 {
836 	unsigned long flags;
837 	int ret;
838 
839 	read_lock_irqsave(&sfp->rq_list_lock, flags);
840 	ret = srp->done;
841 	read_unlock_irqrestore(&sfp->rq_list_lock, flags);
842 	return ret;
843 }
844 
845 static int max_sectors_bytes(struct request_queue *q)
846 {
847 	unsigned int max_sectors = queue_max_sectors(q);
848 
849 	max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
850 
851 	return max_sectors << 9;
852 }
853 
854 static long
855 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
856 {
857 	void __user *p = (void __user *)arg;
858 	int __user *ip = p;
859 	int result, val, read_only;
860 	Sg_device *sdp;
861 	Sg_fd *sfp;
862 	Sg_request *srp;
863 	unsigned long iflags;
864 
865 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
866 		return -ENXIO;
867 
868 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
869 				   "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
870 	read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
871 
872 	switch (cmd_in) {
873 	case SG_IO:
874 		if (atomic_read(&sdp->detaching))
875 			return -ENODEV;
876 		if (!scsi_block_when_processing_errors(sdp->device))
877 			return -ENXIO;
878 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
879 			return -EFAULT;
880 		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
881 				 1, read_only, 1, &srp);
882 		if (result < 0)
883 			return result;
884 		result = wait_event_interruptible(sfp->read_wait,
885 			(srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
886 		if (atomic_read(&sdp->detaching))
887 			return -ENODEV;
888 		write_lock_irq(&sfp->rq_list_lock);
889 		if (srp->done) {
890 			srp->done = 2;
891 			write_unlock_irq(&sfp->rq_list_lock);
892 			result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
893 			return (result < 0) ? result : 0;
894 		}
895 		srp->orphan = 1;
896 		write_unlock_irq(&sfp->rq_list_lock);
897 		return result;	/* -ERESTARTSYS because signal hit process */
898 	case SG_SET_TIMEOUT:
899 		result = get_user(val, ip);
900 		if (result)
901 			return result;
902 		if (val < 0)
903 			return -EIO;
904 		if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
905 			val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
906 				    INT_MAX);
907 		sfp->timeout_user = val;
908 		sfp->timeout = mult_frac(val, HZ, USER_HZ);
909 
910 		return 0;
911 	case SG_GET_TIMEOUT:	/* N.B. User receives timeout as return value */
912 				/* strange ..., for backward compatibility */
913 		return sfp->timeout_user;
914 	case SG_SET_FORCE_LOW_DMA:
915 		/*
916 		 * N.B. This ioctl never worked properly, but failed to
917 		 * return an error value. So returning '0' to keep compability
918 		 * with legacy applications.
919 		 */
920 		return 0;
921 	case SG_GET_LOW_DMA:
922 		return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
923 	case SG_GET_SCSI_ID:
924 		if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
925 			return -EFAULT;
926 		else {
927 			sg_scsi_id_t __user *sg_idp = p;
928 
929 			if (atomic_read(&sdp->detaching))
930 				return -ENODEV;
931 			__put_user((int) sdp->device->host->host_no,
932 				   &sg_idp->host_no);
933 			__put_user((int) sdp->device->channel,
934 				   &sg_idp->channel);
935 			__put_user((int) sdp->device->id, &sg_idp->scsi_id);
936 			__put_user((int) sdp->device->lun, &sg_idp->lun);
937 			__put_user((int) sdp->device->type, &sg_idp->scsi_type);
938 			__put_user((short) sdp->device->host->cmd_per_lun,
939 				   &sg_idp->h_cmd_per_lun);
940 			__put_user((short) sdp->device->queue_depth,
941 				   &sg_idp->d_queue_depth);
942 			__put_user(0, &sg_idp->unused[0]);
943 			__put_user(0, &sg_idp->unused[1]);
944 			return 0;
945 		}
946 	case SG_SET_FORCE_PACK_ID:
947 		result = get_user(val, ip);
948 		if (result)
949 			return result;
950 		sfp->force_packid = val ? 1 : 0;
951 		return 0;
952 	case SG_GET_PACK_ID:
953 		if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
954 			return -EFAULT;
955 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
956 		list_for_each_entry(srp, &sfp->rq_list, entry) {
957 			if ((1 == srp->done) && (!srp->sg_io_owned)) {
958 				read_unlock_irqrestore(&sfp->rq_list_lock,
959 						       iflags);
960 				__put_user(srp->header.pack_id, ip);
961 				return 0;
962 			}
963 		}
964 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
965 		__put_user(-1, ip);
966 		return 0;
967 	case SG_GET_NUM_WAITING:
968 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
969 		val = 0;
970 		list_for_each_entry(srp, &sfp->rq_list, entry) {
971 			if ((1 == srp->done) && (!srp->sg_io_owned))
972 				++val;
973 		}
974 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
975 		return put_user(val, ip);
976 	case SG_GET_SG_TABLESIZE:
977 		return put_user(sdp->sg_tablesize, ip);
978 	case SG_SET_RESERVED_SIZE:
979 		result = get_user(val, ip);
980 		if (result)
981 			return result;
982                 if (val < 0)
983                         return -EINVAL;
984 		val = min_t(int, val,
985 			    max_sectors_bytes(sdp->device->request_queue));
986 		mutex_lock(&sfp->f_mutex);
987 		if (val != sfp->reserve.bufflen) {
988 			if (sfp->mmap_called ||
989 			    sfp->res_in_use) {
990 				mutex_unlock(&sfp->f_mutex);
991 				return -EBUSY;
992 			}
993 
994 			sg_remove_scat(sfp, &sfp->reserve);
995 			sg_build_reserve(sfp, val);
996 		}
997 		mutex_unlock(&sfp->f_mutex);
998 		return 0;
999 	case SG_GET_RESERVED_SIZE:
1000 		val = min_t(int, sfp->reserve.bufflen,
1001 			    max_sectors_bytes(sdp->device->request_queue));
1002 		return put_user(val, ip);
1003 	case SG_SET_COMMAND_Q:
1004 		result = get_user(val, ip);
1005 		if (result)
1006 			return result;
1007 		sfp->cmd_q = val ? 1 : 0;
1008 		return 0;
1009 	case SG_GET_COMMAND_Q:
1010 		return put_user((int) sfp->cmd_q, ip);
1011 	case SG_SET_KEEP_ORPHAN:
1012 		result = get_user(val, ip);
1013 		if (result)
1014 			return result;
1015 		sfp->keep_orphan = val;
1016 		return 0;
1017 	case SG_GET_KEEP_ORPHAN:
1018 		return put_user((int) sfp->keep_orphan, ip);
1019 	case SG_NEXT_CMD_LEN:
1020 		result = get_user(val, ip);
1021 		if (result)
1022 			return result;
1023 		if (val > SG_MAX_CDB_SIZE)
1024 			return -ENOMEM;
1025 		sfp->next_cmd_len = (val > 0) ? val : 0;
1026 		return 0;
1027 	case SG_GET_VERSION_NUM:
1028 		return put_user(sg_version_num, ip);
1029 	case SG_GET_ACCESS_COUNT:
1030 		/* faked - we don't have a real access count anymore */
1031 		val = (sdp->device ? 1 : 0);
1032 		return put_user(val, ip);
1033 	case SG_GET_REQUEST_TABLE:
1034 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1035 			return -EFAULT;
1036 		else {
1037 			sg_req_info_t *rinfo;
1038 			unsigned int ms;
1039 
1040 			rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1041 								GFP_KERNEL);
1042 			if (!rinfo)
1043 				return -ENOMEM;
1044 			read_lock_irqsave(&sfp->rq_list_lock, iflags);
1045 			val = 0;
1046 			list_for_each_entry(srp, &sfp->rq_list, entry) {
1047 				if (val > SG_MAX_QUEUE)
1048 					break;
1049 				memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
1050 				rinfo[val].req_state = srp->done + 1;
1051 				rinfo[val].problem =
1052 					srp->header.masked_status &
1053 					srp->header.host_status &
1054 					srp->header.driver_status;
1055 				if (srp->done)
1056 					rinfo[val].duration =
1057 						srp->header.duration;
1058 				else {
1059 					ms = jiffies_to_msecs(jiffies);
1060 					rinfo[val].duration =
1061 						(ms > srp->header.duration) ?
1062 						(ms - srp->header.duration) : 0;
1063 				}
1064 				rinfo[val].orphan = srp->orphan;
1065 				rinfo[val].sg_io_owned = srp->sg_io_owned;
1066 				rinfo[val].pack_id = srp->header.pack_id;
1067 				rinfo[val].usr_ptr = srp->header.usr_ptr;
1068 				val++;
1069 			}
1070 			read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1071 			result = __copy_to_user(p, rinfo,
1072 						SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1073 			result = result ? -EFAULT : 0;
1074 			kfree(rinfo);
1075 			return result;
1076 		}
1077 	case SG_EMULATED_HOST:
1078 		if (atomic_read(&sdp->detaching))
1079 			return -ENODEV;
1080 		return put_user(sdp->device->host->hostt->emulated, ip);
1081 	case SCSI_IOCTL_SEND_COMMAND:
1082 		if (atomic_read(&sdp->detaching))
1083 			return -ENODEV;
1084 		if (read_only) {
1085 			unsigned char opcode = WRITE_6;
1086 			Scsi_Ioctl_Command __user *siocp = p;
1087 
1088 			if (copy_from_user(&opcode, siocp->data, 1))
1089 				return -EFAULT;
1090 			if (sg_allow_access(filp, &opcode))
1091 				return -EPERM;
1092 		}
1093 		return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1094 	case SG_SET_DEBUG:
1095 		result = get_user(val, ip);
1096 		if (result)
1097 			return result;
1098 		sdp->sgdebug = (char) val;
1099 		return 0;
1100 	case BLKSECTGET:
1101 		return put_user(max_sectors_bytes(sdp->device->request_queue),
1102 				ip);
1103 	case BLKTRACESETUP:
1104 		return blk_trace_setup(sdp->device->request_queue,
1105 				       sdp->disk->disk_name,
1106 				       MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1107 				       NULL,
1108 				       (char *)arg);
1109 	case BLKTRACESTART:
1110 		return blk_trace_startstop(sdp->device->request_queue, 1);
1111 	case BLKTRACESTOP:
1112 		return blk_trace_startstop(sdp->device->request_queue, 0);
1113 	case BLKTRACETEARDOWN:
1114 		return blk_trace_remove(sdp->device->request_queue);
1115 	case SCSI_IOCTL_GET_IDLUN:
1116 	case SCSI_IOCTL_GET_BUS_NUMBER:
1117 	case SCSI_IOCTL_PROBE_HOST:
1118 	case SG_GET_TRANSFORM:
1119 	case SG_SCSI_RESET:
1120 		if (atomic_read(&sdp->detaching))
1121 			return -ENODEV;
1122 		break;
1123 	default:
1124 		if (read_only)
1125 			return -EPERM;	/* don't know so take safe approach */
1126 		break;
1127 	}
1128 
1129 	result = scsi_ioctl_block_when_processing_errors(sdp->device,
1130 			cmd_in, filp->f_flags & O_NDELAY);
1131 	if (result)
1132 		return result;
1133 	return scsi_ioctl(sdp->device, cmd_in, p);
1134 }
1135 
1136 #ifdef CONFIG_COMPAT
1137 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1138 {
1139 	Sg_device *sdp;
1140 	Sg_fd *sfp;
1141 	struct scsi_device *sdev;
1142 
1143 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1144 		return -ENXIO;
1145 
1146 	sdev = sdp->device;
1147 	if (sdev->host->hostt->compat_ioctl) {
1148 		int ret;
1149 
1150 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1151 
1152 		return ret;
1153 	}
1154 
1155 	return -ENOIOCTLCMD;
1156 }
1157 #endif
1158 
1159 static unsigned int
1160 sg_poll(struct file *filp, poll_table * wait)
1161 {
1162 	unsigned int res = 0;
1163 	Sg_device *sdp;
1164 	Sg_fd *sfp;
1165 	Sg_request *srp;
1166 	int count = 0;
1167 	unsigned long iflags;
1168 
1169 	sfp = filp->private_data;
1170 	if (!sfp)
1171 		return POLLERR;
1172 	sdp = sfp->parentdp;
1173 	if (!sdp)
1174 		return POLLERR;
1175 	poll_wait(filp, &sfp->read_wait, wait);
1176 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
1177 	list_for_each_entry(srp, &sfp->rq_list, entry) {
1178 		/* if any read waiting, flag it */
1179 		if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1180 			res = POLLIN | POLLRDNORM;
1181 		++count;
1182 	}
1183 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1184 
1185 	if (atomic_read(&sdp->detaching))
1186 		res |= POLLHUP;
1187 	else if (!sfp->cmd_q) {
1188 		if (0 == count)
1189 			res |= POLLOUT | POLLWRNORM;
1190 	} else if (count < SG_MAX_QUEUE)
1191 		res |= POLLOUT | POLLWRNORM;
1192 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1193 				      "sg_poll: res=0x%x\n", (int) res));
1194 	return res;
1195 }
1196 
1197 static int
1198 sg_fasync(int fd, struct file *filp, int mode)
1199 {
1200 	Sg_device *sdp;
1201 	Sg_fd *sfp;
1202 
1203 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1204 		return -ENXIO;
1205 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1206 				      "sg_fasync: mode=%d\n", mode));
1207 
1208 	return fasync_helper(fd, filp, mode, &sfp->async_qp);
1209 }
1210 
1211 static int
1212 sg_vma_fault(struct vm_fault *vmf)
1213 {
1214 	struct vm_area_struct *vma = vmf->vma;
1215 	Sg_fd *sfp;
1216 	unsigned long offset, len, sa;
1217 	Sg_scatter_hold *rsv_schp;
1218 	int k, length;
1219 
1220 	if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1221 		return VM_FAULT_SIGBUS;
1222 	rsv_schp = &sfp->reserve;
1223 	offset = vmf->pgoff << PAGE_SHIFT;
1224 	if (offset >= rsv_schp->bufflen)
1225 		return VM_FAULT_SIGBUS;
1226 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1227 				      "sg_vma_fault: offset=%lu, scatg=%d\n",
1228 				      offset, rsv_schp->k_use_sg));
1229 	sa = vma->vm_start;
1230 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1231 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1232 		len = vma->vm_end - sa;
1233 		len = (len < length) ? len : length;
1234 		if (offset < len) {
1235 			struct page *page = nth_page(rsv_schp->pages[k],
1236 						     offset >> PAGE_SHIFT);
1237 			get_page(page);	/* increment page count */
1238 			vmf->page = page;
1239 			return 0; /* success */
1240 		}
1241 		sa += len;
1242 		offset -= len;
1243 	}
1244 
1245 	return VM_FAULT_SIGBUS;
1246 }
1247 
1248 static const struct vm_operations_struct sg_mmap_vm_ops = {
1249 	.fault = sg_vma_fault,
1250 };
1251 
1252 static int
1253 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1254 {
1255 	Sg_fd *sfp;
1256 	unsigned long req_sz, len, sa;
1257 	Sg_scatter_hold *rsv_schp;
1258 	int k, length;
1259 
1260 	if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1261 		return -ENXIO;
1262 	req_sz = vma->vm_end - vma->vm_start;
1263 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1264 				      "sg_mmap starting, vm_start=%p, len=%d\n",
1265 				      (void *) vma->vm_start, (int) req_sz));
1266 	if (vma->vm_pgoff)
1267 		return -EINVAL;	/* want no offset */
1268 	rsv_schp = &sfp->reserve;
1269 	if (req_sz > rsv_schp->bufflen)
1270 		return -ENOMEM;	/* cannot map more than reserved buffer */
1271 
1272 	sa = vma->vm_start;
1273 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1274 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1275 		len = vma->vm_end - sa;
1276 		len = (len < length) ? len : length;
1277 		sa += len;
1278 	}
1279 
1280 	sfp->mmap_called = 1;
1281 	vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1282 	vma->vm_private_data = sfp;
1283 	vma->vm_ops = &sg_mmap_vm_ops;
1284 	return 0;
1285 }
1286 
1287 static void
1288 sg_rq_end_io_usercontext(struct work_struct *work)
1289 {
1290 	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1291 	struct sg_fd *sfp = srp->parentfp;
1292 
1293 	sg_finish_rem_req(srp);
1294 	sg_remove_request(sfp, srp);
1295 	kref_put(&sfp->f_ref, sg_remove_sfp);
1296 }
1297 
1298 /*
1299  * This function is a "bottom half" handler that is called by the mid
1300  * level when a command is completed (or has failed).
1301  */
1302 static void
1303 sg_rq_end_io(struct request *rq, int uptodate)
1304 {
1305 	struct sg_request *srp = rq->end_io_data;
1306 	struct scsi_request *req = scsi_req(rq);
1307 	Sg_device *sdp;
1308 	Sg_fd *sfp;
1309 	unsigned long iflags;
1310 	unsigned int ms;
1311 	char *sense;
1312 	int result, resid, done = 1;
1313 
1314 	if (WARN_ON(srp->done != 0))
1315 		return;
1316 
1317 	sfp = srp->parentfp;
1318 	if (WARN_ON(sfp == NULL))
1319 		return;
1320 
1321 	sdp = sfp->parentdp;
1322 	if (unlikely(atomic_read(&sdp->detaching)))
1323 		pr_info("%s: device detaching\n", __func__);
1324 
1325 	sense = req->sense;
1326 	result = req->result;
1327 	resid = req->resid_len;
1328 
1329 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1330 				      "sg_cmd_done: pack_id=%d, res=0x%x\n",
1331 				      srp->header.pack_id, result));
1332 	srp->header.resid = resid;
1333 	ms = jiffies_to_msecs(jiffies);
1334 	srp->header.duration = (ms > srp->header.duration) ?
1335 				(ms - srp->header.duration) : 0;
1336 	if (0 != result) {
1337 		struct scsi_sense_hdr sshdr;
1338 
1339 		srp->header.status = 0xff & result;
1340 		srp->header.masked_status = status_byte(result);
1341 		srp->header.msg_status = msg_byte(result);
1342 		srp->header.host_status = host_byte(result);
1343 		srp->header.driver_status = driver_byte(result);
1344 		if ((sdp->sgdebug > 0) &&
1345 		    ((CHECK_CONDITION == srp->header.masked_status) ||
1346 		     (COMMAND_TERMINATED == srp->header.masked_status)))
1347 			__scsi_print_sense(sdp->device, __func__, sense,
1348 					   SCSI_SENSE_BUFFERSIZE);
1349 
1350 		/* Following if statement is a patch supplied by Eric Youngdale */
1351 		if (driver_byte(result) != 0
1352 		    && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1353 		    && !scsi_sense_is_deferred(&sshdr)
1354 		    && sshdr.sense_key == UNIT_ATTENTION
1355 		    && sdp->device->removable) {
1356 			/* Detected possible disc change. Set the bit - this */
1357 			/* may be used if there are filesystems using this device */
1358 			sdp->device->changed = 1;
1359 		}
1360 	}
1361 
1362 	if (req->sense_len)
1363 		memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1364 
1365 	/* Rely on write phase to clean out srp status values, so no "else" */
1366 
1367 	/*
1368 	 * Free the request as soon as it is complete so that its resources
1369 	 * can be reused without waiting for userspace to read() the
1370 	 * result.  But keep the associated bio (if any) around until
1371 	 * blk_rq_unmap_user() can be called from user context.
1372 	 */
1373 	srp->rq = NULL;
1374 	scsi_req_free_cmd(scsi_req(rq));
1375 	__blk_put_request(rq->q, rq);
1376 
1377 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1378 	if (unlikely(srp->orphan)) {
1379 		if (sfp->keep_orphan)
1380 			srp->sg_io_owned = 0;
1381 		else
1382 			done = 0;
1383 	}
1384 	srp->done = done;
1385 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1386 
1387 	if (likely(done)) {
1388 		/* Now wake up any sg_read() that is waiting for this
1389 		 * packet.
1390 		 */
1391 		wake_up_interruptible(&sfp->read_wait);
1392 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1393 		kref_put(&sfp->f_ref, sg_remove_sfp);
1394 	} else {
1395 		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1396 		schedule_work(&srp->ew.work);
1397 	}
1398 }
1399 
1400 static const struct file_operations sg_fops = {
1401 	.owner = THIS_MODULE,
1402 	.read = sg_read,
1403 	.write = sg_write,
1404 	.poll = sg_poll,
1405 	.unlocked_ioctl = sg_ioctl,
1406 #ifdef CONFIG_COMPAT
1407 	.compat_ioctl = sg_compat_ioctl,
1408 #endif
1409 	.open = sg_open,
1410 	.mmap = sg_mmap,
1411 	.release = sg_release,
1412 	.fasync = sg_fasync,
1413 	.llseek = no_llseek,
1414 };
1415 
1416 static struct class *sg_sysfs_class;
1417 
1418 static int sg_sysfs_valid = 0;
1419 
1420 static Sg_device *
1421 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1422 {
1423 	struct request_queue *q = scsidp->request_queue;
1424 	Sg_device *sdp;
1425 	unsigned long iflags;
1426 	int error;
1427 	u32 k;
1428 
1429 	sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1430 	if (!sdp) {
1431 		sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1432 			    "failure\n", __func__);
1433 		return ERR_PTR(-ENOMEM);
1434 	}
1435 
1436 	idr_preload(GFP_KERNEL);
1437 	write_lock_irqsave(&sg_index_lock, iflags);
1438 
1439 	error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1440 	if (error < 0) {
1441 		if (error == -ENOSPC) {
1442 			sdev_printk(KERN_WARNING, scsidp,
1443 				    "Unable to attach sg device type=%d, minor number exceeds %d\n",
1444 				    scsidp->type, SG_MAX_DEVS - 1);
1445 			error = -ENODEV;
1446 		} else {
1447 			sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1448 				    "allocation Sg_device failure: %d\n",
1449 				    __func__, error);
1450 		}
1451 		goto out_unlock;
1452 	}
1453 	k = error;
1454 
1455 	SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1456 					"sg_alloc: dev=%d \n", k));
1457 	sprintf(disk->disk_name, "sg%d", k);
1458 	disk->first_minor = k;
1459 	sdp->disk = disk;
1460 	sdp->device = scsidp;
1461 	mutex_init(&sdp->open_rel_lock);
1462 	INIT_LIST_HEAD(&sdp->sfds);
1463 	init_waitqueue_head(&sdp->open_wait);
1464 	atomic_set(&sdp->detaching, 0);
1465 	rwlock_init(&sdp->sfd_lock);
1466 	sdp->sg_tablesize = queue_max_segments(q);
1467 	sdp->index = k;
1468 	kref_init(&sdp->d_ref);
1469 	error = 0;
1470 
1471 out_unlock:
1472 	write_unlock_irqrestore(&sg_index_lock, iflags);
1473 	idr_preload_end();
1474 
1475 	if (error) {
1476 		kfree(sdp);
1477 		return ERR_PTR(error);
1478 	}
1479 	return sdp;
1480 }
1481 
1482 static int
1483 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1484 {
1485 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1486 	struct gendisk *disk;
1487 	Sg_device *sdp = NULL;
1488 	struct cdev * cdev = NULL;
1489 	int error;
1490 	unsigned long iflags;
1491 
1492 	disk = alloc_disk(1);
1493 	if (!disk) {
1494 		pr_warn("%s: alloc_disk failed\n", __func__);
1495 		return -ENOMEM;
1496 	}
1497 	disk->major = SCSI_GENERIC_MAJOR;
1498 
1499 	error = -ENOMEM;
1500 	cdev = cdev_alloc();
1501 	if (!cdev) {
1502 		pr_warn("%s: cdev_alloc failed\n", __func__);
1503 		goto out;
1504 	}
1505 	cdev->owner = THIS_MODULE;
1506 	cdev->ops = &sg_fops;
1507 
1508 	sdp = sg_alloc(disk, scsidp);
1509 	if (IS_ERR(sdp)) {
1510 		pr_warn("%s: sg_alloc failed\n", __func__);
1511 		error = PTR_ERR(sdp);
1512 		goto out;
1513 	}
1514 
1515 	error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1516 	if (error)
1517 		goto cdev_add_err;
1518 
1519 	sdp->cdev = cdev;
1520 	if (sg_sysfs_valid) {
1521 		struct device *sg_class_member;
1522 
1523 		sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1524 						MKDEV(SCSI_GENERIC_MAJOR,
1525 						      sdp->index),
1526 						sdp, "%s", disk->disk_name);
1527 		if (IS_ERR(sg_class_member)) {
1528 			pr_err("%s: device_create failed\n", __func__);
1529 			error = PTR_ERR(sg_class_member);
1530 			goto cdev_add_err;
1531 		}
1532 		error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1533 					  &sg_class_member->kobj, "generic");
1534 		if (error)
1535 			pr_err("%s: unable to make symlink 'generic' back "
1536 			       "to sg%d\n", __func__, sdp->index);
1537 	} else
1538 		pr_warn("%s: sg_sys Invalid\n", __func__);
1539 
1540 	sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1541 		    "type %d\n", sdp->index, scsidp->type);
1542 
1543 	dev_set_drvdata(cl_dev, sdp);
1544 
1545 	return 0;
1546 
1547 cdev_add_err:
1548 	write_lock_irqsave(&sg_index_lock, iflags);
1549 	idr_remove(&sg_index_idr, sdp->index);
1550 	write_unlock_irqrestore(&sg_index_lock, iflags);
1551 	kfree(sdp);
1552 
1553 out:
1554 	put_disk(disk);
1555 	if (cdev)
1556 		cdev_del(cdev);
1557 	return error;
1558 }
1559 
1560 static void
1561 sg_device_destroy(struct kref *kref)
1562 {
1563 	struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1564 	unsigned long flags;
1565 
1566 	/* CAUTION!  Note that the device can still be found via idr_find()
1567 	 * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1568 	 * any other cleanup.
1569 	 */
1570 
1571 	write_lock_irqsave(&sg_index_lock, flags);
1572 	idr_remove(&sg_index_idr, sdp->index);
1573 	write_unlock_irqrestore(&sg_index_lock, flags);
1574 
1575 	SCSI_LOG_TIMEOUT(3,
1576 		sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1577 
1578 	put_disk(sdp->disk);
1579 	kfree(sdp);
1580 }
1581 
1582 static void
1583 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1584 {
1585 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1586 	Sg_device *sdp = dev_get_drvdata(cl_dev);
1587 	unsigned long iflags;
1588 	Sg_fd *sfp;
1589 	int val;
1590 
1591 	if (!sdp)
1592 		return;
1593 	/* want sdp->detaching non-zero as soon as possible */
1594 	val = atomic_inc_return(&sdp->detaching);
1595 	if (val > 1)
1596 		return; /* only want to do following once per device */
1597 
1598 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1599 				      "%s\n", __func__));
1600 
1601 	read_lock_irqsave(&sdp->sfd_lock, iflags);
1602 	list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1603 		wake_up_interruptible_all(&sfp->read_wait);
1604 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1605 	}
1606 	wake_up_interruptible_all(&sdp->open_wait);
1607 	read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1608 
1609 	sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1610 	device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1611 	cdev_del(sdp->cdev);
1612 	sdp->cdev = NULL;
1613 
1614 	kref_put(&sdp->d_ref, sg_device_destroy);
1615 }
1616 
1617 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1618 module_param_named(def_reserved_size, def_reserved_size, int,
1619 		   S_IRUGO | S_IWUSR);
1620 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1621 
1622 MODULE_AUTHOR("Douglas Gilbert");
1623 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1624 MODULE_LICENSE("GPL");
1625 MODULE_VERSION(SG_VERSION_STR);
1626 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1627 
1628 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1629                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1630 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1631 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1632 
1633 static int __init
1634 init_sg(void)
1635 {
1636 	int rc;
1637 
1638 	if (scatter_elem_sz < PAGE_SIZE) {
1639 		scatter_elem_sz = PAGE_SIZE;
1640 		scatter_elem_sz_prev = scatter_elem_sz;
1641 	}
1642 	if (def_reserved_size >= 0)
1643 		sg_big_buff = def_reserved_size;
1644 	else
1645 		def_reserved_size = sg_big_buff;
1646 
1647 	rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1648 				    SG_MAX_DEVS, "sg");
1649 	if (rc)
1650 		return rc;
1651         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1652         if ( IS_ERR(sg_sysfs_class) ) {
1653 		rc = PTR_ERR(sg_sysfs_class);
1654 		goto err_out;
1655         }
1656 	sg_sysfs_valid = 1;
1657 	rc = scsi_register_interface(&sg_interface);
1658 	if (0 == rc) {
1659 #ifdef CONFIG_SCSI_PROC_FS
1660 		sg_proc_init();
1661 #endif				/* CONFIG_SCSI_PROC_FS */
1662 		return 0;
1663 	}
1664 	class_destroy(sg_sysfs_class);
1665 err_out:
1666 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1667 	return rc;
1668 }
1669 
1670 static void __exit
1671 exit_sg(void)
1672 {
1673 #ifdef CONFIG_SCSI_PROC_FS
1674 	sg_proc_cleanup();
1675 #endif				/* CONFIG_SCSI_PROC_FS */
1676 	scsi_unregister_interface(&sg_interface);
1677 	class_destroy(sg_sysfs_class);
1678 	sg_sysfs_valid = 0;
1679 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1680 				 SG_MAX_DEVS);
1681 	idr_destroy(&sg_index_idr);
1682 }
1683 
1684 static int
1685 sg_start_req(Sg_request *srp, unsigned char *cmd)
1686 {
1687 	int res;
1688 	struct request *rq;
1689 	struct scsi_request *req;
1690 	Sg_fd *sfp = srp->parentfp;
1691 	sg_io_hdr_t *hp = &srp->header;
1692 	int dxfer_len = (int) hp->dxfer_len;
1693 	int dxfer_dir = hp->dxfer_direction;
1694 	unsigned int iov_count = hp->iovec_count;
1695 	Sg_scatter_hold *req_schp = &srp->data;
1696 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1697 	struct request_queue *q = sfp->parentdp->device->request_queue;
1698 	struct rq_map_data *md, map_data;
1699 	int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1700 	unsigned char *long_cmdp = NULL;
1701 
1702 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1703 				      "sg_start_req: dxfer_len=%d\n",
1704 				      dxfer_len));
1705 
1706 	if (hp->cmd_len > BLK_MAX_CDB) {
1707 		long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1708 		if (!long_cmdp)
1709 			return -ENOMEM;
1710 	}
1711 
1712 	/*
1713 	 * NOTE
1714 	 *
1715 	 * With scsi-mq enabled, there are a fixed number of preallocated
1716 	 * requests equal in number to shost->can_queue.  If all of the
1717 	 * preallocated requests are already in use, then using GFP_ATOMIC with
1718 	 * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1719 	 * will cause blk_get_request() to sleep until an active command
1720 	 * completes, freeing up a request.  Neither option is ideal, but
1721 	 * GFP_KERNEL is the better choice to prevent userspace from getting an
1722 	 * unexpected EWOULDBLOCK.
1723 	 *
1724 	 * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1725 	 * does not sleep except under memory pressure.
1726 	 */
1727 	rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1728 			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, GFP_KERNEL);
1729 	if (IS_ERR(rq)) {
1730 		kfree(long_cmdp);
1731 		return PTR_ERR(rq);
1732 	}
1733 	req = scsi_req(rq);
1734 
1735 	scsi_req_init(rq);
1736 
1737 	if (hp->cmd_len > BLK_MAX_CDB)
1738 		req->cmd = long_cmdp;
1739 	memcpy(req->cmd, cmd, hp->cmd_len);
1740 	req->cmd_len = hp->cmd_len;
1741 
1742 	srp->rq = rq;
1743 	rq->end_io_data = srp;
1744 	req->retries = SG_DEFAULT_RETRIES;
1745 
1746 	if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1747 		return 0;
1748 
1749 	if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1750 	    dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1751 	    !sfp->parentdp->device->host->unchecked_isa_dma &&
1752 	    blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1753 		md = NULL;
1754 	else
1755 		md = &map_data;
1756 
1757 	if (md) {
1758 		mutex_lock(&sfp->f_mutex);
1759 		if (dxfer_len <= rsv_schp->bufflen &&
1760 		    !sfp->res_in_use) {
1761 			sfp->res_in_use = 1;
1762 			sg_link_reserve(sfp, srp, dxfer_len);
1763 		} else if ((hp->flags & SG_FLAG_MMAP_IO) && sfp->res_in_use) {
1764 			mutex_unlock(&sfp->f_mutex);
1765 			return -EBUSY;
1766 		} else {
1767 			res = sg_build_indirect(req_schp, sfp, dxfer_len);
1768 			if (res) {
1769 				mutex_unlock(&sfp->f_mutex);
1770 				return res;
1771 			}
1772 		}
1773 		mutex_unlock(&sfp->f_mutex);
1774 
1775 		md->pages = req_schp->pages;
1776 		md->page_order = req_schp->page_order;
1777 		md->nr_entries = req_schp->k_use_sg;
1778 		md->offset = 0;
1779 		md->null_mapped = hp->dxferp ? 0 : 1;
1780 		if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1781 			md->from_user = 1;
1782 		else
1783 			md->from_user = 0;
1784 	}
1785 
1786 	if (iov_count) {
1787 		struct iovec *iov = NULL;
1788 		struct iov_iter i;
1789 
1790 		res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1791 		if (res < 0)
1792 			return res;
1793 
1794 		iov_iter_truncate(&i, hp->dxfer_len);
1795 		if (!iov_iter_count(&i)) {
1796 			kfree(iov);
1797 			return -EINVAL;
1798 		}
1799 
1800 		res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1801 		kfree(iov);
1802 	} else
1803 		res = blk_rq_map_user(q, rq, md, hp->dxferp,
1804 				      hp->dxfer_len, GFP_ATOMIC);
1805 
1806 	if (!res) {
1807 		srp->bio = rq->bio;
1808 
1809 		if (!md) {
1810 			req_schp->dio_in_use = 1;
1811 			hp->info |= SG_INFO_DIRECT_IO;
1812 		}
1813 	}
1814 	return res;
1815 }
1816 
1817 static int
1818 sg_finish_rem_req(Sg_request *srp)
1819 {
1820 	int ret = 0;
1821 
1822 	Sg_fd *sfp = srp->parentfp;
1823 	Sg_scatter_hold *req_schp = &srp->data;
1824 
1825 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1826 				      "sg_finish_rem_req: res_used=%d\n",
1827 				      (int) srp->res_used));
1828 	if (srp->bio)
1829 		ret = blk_rq_unmap_user(srp->bio);
1830 
1831 	if (srp->rq) {
1832 		scsi_req_free_cmd(scsi_req(srp->rq));
1833 		blk_put_request(srp->rq);
1834 	}
1835 
1836 	if (srp->res_used)
1837 		sg_unlink_reserve(sfp, srp);
1838 	else
1839 		sg_remove_scat(sfp, req_schp);
1840 
1841 	return ret;
1842 }
1843 
1844 static int
1845 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1846 {
1847 	int sg_bufflen = tablesize * sizeof(struct page *);
1848 	gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1849 
1850 	schp->pages = kzalloc(sg_bufflen, gfp_flags);
1851 	if (!schp->pages)
1852 		return -ENOMEM;
1853 	schp->sglist_len = sg_bufflen;
1854 	return tablesize;	/* number of scat_gath elements allocated */
1855 }
1856 
1857 static int
1858 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1859 {
1860 	int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1861 	int sg_tablesize = sfp->parentdp->sg_tablesize;
1862 	int blk_size = buff_size, order;
1863 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1864 	struct sg_device *sdp = sfp->parentdp;
1865 
1866 	if (blk_size < 0)
1867 		return -EFAULT;
1868 	if (0 == blk_size)
1869 		++blk_size;	/* don't know why */
1870 	/* round request up to next highest SG_SECTOR_SZ byte boundary */
1871 	blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1872 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1873 		"sg_build_indirect: buff_size=%d, blk_size=%d\n",
1874 		buff_size, blk_size));
1875 
1876 	/* N.B. ret_sz carried into this block ... */
1877 	mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1878 	if (mx_sc_elems < 0)
1879 		return mx_sc_elems;	/* most likely -ENOMEM */
1880 
1881 	num = scatter_elem_sz;
1882 	if (unlikely(num != scatter_elem_sz_prev)) {
1883 		if (num < PAGE_SIZE) {
1884 			scatter_elem_sz = PAGE_SIZE;
1885 			scatter_elem_sz_prev = PAGE_SIZE;
1886 		} else
1887 			scatter_elem_sz_prev = num;
1888 	}
1889 
1890 	if (sdp->device->host->unchecked_isa_dma)
1891 		gfp_mask |= GFP_DMA;
1892 
1893 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1894 		gfp_mask |= __GFP_ZERO;
1895 
1896 	order = get_order(num);
1897 retry:
1898 	ret_sz = 1 << (PAGE_SHIFT + order);
1899 
1900 	for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1901 	     k++, rem_sz -= ret_sz) {
1902 
1903 		num = (rem_sz > scatter_elem_sz_prev) ?
1904 			scatter_elem_sz_prev : rem_sz;
1905 
1906 		schp->pages[k] = alloc_pages(gfp_mask, order);
1907 		if (!schp->pages[k])
1908 			goto out;
1909 
1910 		if (num == scatter_elem_sz_prev) {
1911 			if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1912 				scatter_elem_sz = ret_sz;
1913 				scatter_elem_sz_prev = ret_sz;
1914 			}
1915 		}
1916 
1917 		SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1918 				 "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1919 				 k, num, ret_sz));
1920 	}		/* end of for loop */
1921 
1922 	schp->page_order = order;
1923 	schp->k_use_sg = k;
1924 	SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1925 			 "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1926 			 k, rem_sz));
1927 
1928 	schp->bufflen = blk_size;
1929 	if (rem_sz > 0)	/* must have failed */
1930 		return -ENOMEM;
1931 	return 0;
1932 out:
1933 	for (i = 0; i < k; i++)
1934 		__free_pages(schp->pages[i], order);
1935 
1936 	if (--order >= 0)
1937 		goto retry;
1938 
1939 	return -ENOMEM;
1940 }
1941 
1942 static void
1943 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1944 {
1945 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1946 			 "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1947 	if (schp->pages && schp->sglist_len > 0) {
1948 		if (!schp->dio_in_use) {
1949 			int k;
1950 
1951 			for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1952 				SCSI_LOG_TIMEOUT(5,
1953 					sg_printk(KERN_INFO, sfp->parentdp,
1954 					"sg_remove_scat: k=%d, pg=0x%p\n",
1955 					k, schp->pages[k]));
1956 				__free_pages(schp->pages[k], schp->page_order);
1957 			}
1958 
1959 			kfree(schp->pages);
1960 		}
1961 	}
1962 	memset(schp, 0, sizeof (*schp));
1963 }
1964 
1965 static int
1966 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1967 {
1968 	Sg_scatter_hold *schp = &srp->data;
1969 	int k, num;
1970 
1971 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1972 			 "sg_read_oxfer: num_read_xfer=%d\n",
1973 			 num_read_xfer));
1974 	if ((!outp) || (num_read_xfer <= 0))
1975 		return 0;
1976 
1977 	num = 1 << (PAGE_SHIFT + schp->page_order);
1978 	for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1979 		if (num > num_read_xfer) {
1980 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1981 					   num_read_xfer))
1982 				return -EFAULT;
1983 			break;
1984 		} else {
1985 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1986 					   num))
1987 				return -EFAULT;
1988 			num_read_xfer -= num;
1989 			if (num_read_xfer <= 0)
1990 				break;
1991 			outp += num;
1992 		}
1993 	}
1994 
1995 	return 0;
1996 }
1997 
1998 static void
1999 sg_build_reserve(Sg_fd * sfp, int req_size)
2000 {
2001 	Sg_scatter_hold *schp = &sfp->reserve;
2002 
2003 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2004 			 "sg_build_reserve: req_size=%d\n", req_size));
2005 	do {
2006 		if (req_size < PAGE_SIZE)
2007 			req_size = PAGE_SIZE;
2008 		if (0 == sg_build_indirect(schp, sfp, req_size))
2009 			return;
2010 		else
2011 			sg_remove_scat(sfp, schp);
2012 		req_size >>= 1;	/* divide by 2 */
2013 	} while (req_size > (PAGE_SIZE / 2));
2014 }
2015 
2016 static void
2017 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2018 {
2019 	Sg_scatter_hold *req_schp = &srp->data;
2020 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
2021 	int k, num, rem;
2022 
2023 	srp->res_used = 1;
2024 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2025 			 "sg_link_reserve: size=%d\n", size));
2026 	rem = size;
2027 
2028 	num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2029 	for (k = 0; k < rsv_schp->k_use_sg; k++) {
2030 		if (rem <= num) {
2031 			req_schp->k_use_sg = k + 1;
2032 			req_schp->sglist_len = rsv_schp->sglist_len;
2033 			req_schp->pages = rsv_schp->pages;
2034 
2035 			req_schp->bufflen = size;
2036 			req_schp->page_order = rsv_schp->page_order;
2037 			break;
2038 		} else
2039 			rem -= num;
2040 	}
2041 
2042 	if (k >= rsv_schp->k_use_sg)
2043 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2044 				 "sg_link_reserve: BAD size\n"));
2045 }
2046 
2047 static void
2048 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2049 {
2050 	Sg_scatter_hold *req_schp = &srp->data;
2051 
2052 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2053 				      "sg_unlink_reserve: req->k_use_sg=%d\n",
2054 				      (int) req_schp->k_use_sg));
2055 	req_schp->k_use_sg = 0;
2056 	req_schp->bufflen = 0;
2057 	req_schp->pages = NULL;
2058 	req_schp->page_order = 0;
2059 	req_schp->sglist_len = 0;
2060 	srp->res_used = 0;
2061 	/* Called without mutex lock to avoid deadlock */
2062 	sfp->res_in_use = 0;
2063 }
2064 
2065 static Sg_request *
2066 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2067 {
2068 	Sg_request *resp;
2069 	unsigned long iflags;
2070 
2071 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2072 	list_for_each_entry(resp, &sfp->rq_list, entry) {
2073 		/* look for requests that are ready + not SG_IO owned */
2074 		if ((1 == resp->done) && (!resp->sg_io_owned) &&
2075 		    ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2076 			resp->done = 2;	/* guard against other readers */
2077 			write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2078 			return resp;
2079 		}
2080 	}
2081 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2082 	return NULL;
2083 }
2084 
2085 /* always adds to end of list */
2086 static Sg_request *
2087 sg_add_request(Sg_fd * sfp)
2088 {
2089 	int k;
2090 	unsigned long iflags;
2091 	Sg_request *rp = sfp->req_arr;
2092 
2093 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2094 	if (!list_empty(&sfp->rq_list)) {
2095 		if (!sfp->cmd_q)
2096 			goto out_unlock;
2097 
2098 		for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2099 			if (!rp->parentfp)
2100 				break;
2101 		}
2102 		if (k >= SG_MAX_QUEUE)
2103 			goto out_unlock;
2104 	}
2105 	memset(rp, 0, sizeof (Sg_request));
2106 	rp->parentfp = sfp;
2107 	rp->header.duration = jiffies_to_msecs(jiffies);
2108 	list_add_tail(&rp->entry, &sfp->rq_list);
2109 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2110 	return rp;
2111 out_unlock:
2112 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2113 	return NULL;
2114 }
2115 
2116 /* Return of 1 for found; 0 for not found */
2117 static int
2118 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2119 {
2120 	unsigned long iflags;
2121 	int res = 0;
2122 
2123 	if (!sfp || !srp || list_empty(&sfp->rq_list))
2124 		return res;
2125 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2126 	if (!list_empty(&srp->entry)) {
2127 		list_del(&srp->entry);
2128 		srp->parentfp = NULL;
2129 		res = 1;
2130 	}
2131 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2132 	return res;
2133 }
2134 
2135 static Sg_fd *
2136 sg_add_sfp(Sg_device * sdp)
2137 {
2138 	Sg_fd *sfp;
2139 	unsigned long iflags;
2140 	int bufflen;
2141 
2142 	sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2143 	if (!sfp)
2144 		return ERR_PTR(-ENOMEM);
2145 
2146 	init_waitqueue_head(&sfp->read_wait);
2147 	rwlock_init(&sfp->rq_list_lock);
2148 	INIT_LIST_HEAD(&sfp->rq_list);
2149 	kref_init(&sfp->f_ref);
2150 	mutex_init(&sfp->f_mutex);
2151 	sfp->timeout = SG_DEFAULT_TIMEOUT;
2152 	sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2153 	sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2154 	sfp->cmd_q = SG_DEF_COMMAND_Q;
2155 	sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2156 	sfp->parentdp = sdp;
2157 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2158 	if (atomic_read(&sdp->detaching)) {
2159 		write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2160 		return ERR_PTR(-ENODEV);
2161 	}
2162 	list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2163 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2164 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2165 				      "sg_add_sfp: sfp=0x%p\n", sfp));
2166 	if (unlikely(sg_big_buff != def_reserved_size))
2167 		sg_big_buff = def_reserved_size;
2168 
2169 	bufflen = min_t(int, sg_big_buff,
2170 			max_sectors_bytes(sdp->device->request_queue));
2171 	sg_build_reserve(sfp, bufflen);
2172 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2173 				      "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2174 				      sfp->reserve.bufflen,
2175 				      sfp->reserve.k_use_sg));
2176 
2177 	kref_get(&sdp->d_ref);
2178 	__module_get(THIS_MODULE);
2179 	return sfp;
2180 }
2181 
2182 static void
2183 sg_remove_sfp_usercontext(struct work_struct *work)
2184 {
2185 	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2186 	struct sg_device *sdp = sfp->parentdp;
2187 	Sg_request *srp;
2188 	unsigned long iflags;
2189 
2190 	/* Cleanup any responses which were never read(). */
2191 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2192 	while (!list_empty(&sfp->rq_list)) {
2193 		srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2194 		sg_finish_rem_req(srp);
2195 		list_del(&srp->entry);
2196 		srp->parentfp = NULL;
2197 	}
2198 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2199 
2200 	if (sfp->reserve.bufflen > 0) {
2201 		SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2202 				"sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2203 				(int) sfp->reserve.bufflen,
2204 				(int) sfp->reserve.k_use_sg));
2205 		sg_remove_scat(sfp, &sfp->reserve);
2206 	}
2207 
2208 	SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2209 			"sg_remove_sfp: sfp=0x%p\n", sfp));
2210 	kfree(sfp);
2211 
2212 	scsi_device_put(sdp->device);
2213 	kref_put(&sdp->d_ref, sg_device_destroy);
2214 	module_put(THIS_MODULE);
2215 }
2216 
2217 static void
2218 sg_remove_sfp(struct kref *kref)
2219 {
2220 	struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2221 	struct sg_device *sdp = sfp->parentdp;
2222 	unsigned long iflags;
2223 
2224 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2225 	list_del(&sfp->sfd_siblings);
2226 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2227 
2228 	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2229 	schedule_work(&sfp->ew.work);
2230 }
2231 
2232 #ifdef CONFIG_SCSI_PROC_FS
2233 static int
2234 sg_idr_max_id(int id, void *p, void *data)
2235 {
2236 	int *k = data;
2237 
2238 	if (*k < id)
2239 		*k = id;
2240 
2241 	return 0;
2242 }
2243 
2244 static int
2245 sg_last_dev(void)
2246 {
2247 	int k = -1;
2248 	unsigned long iflags;
2249 
2250 	read_lock_irqsave(&sg_index_lock, iflags);
2251 	idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2252 	read_unlock_irqrestore(&sg_index_lock, iflags);
2253 	return k + 1;		/* origin 1 */
2254 }
2255 #endif
2256 
2257 /* must be called with sg_index_lock held */
2258 static Sg_device *sg_lookup_dev(int dev)
2259 {
2260 	return idr_find(&sg_index_idr, dev);
2261 }
2262 
2263 static Sg_device *
2264 sg_get_dev(int dev)
2265 {
2266 	struct sg_device *sdp;
2267 	unsigned long flags;
2268 
2269 	read_lock_irqsave(&sg_index_lock, flags);
2270 	sdp = sg_lookup_dev(dev);
2271 	if (!sdp)
2272 		sdp = ERR_PTR(-ENXIO);
2273 	else if (atomic_read(&sdp->detaching)) {
2274 		/* If sdp->detaching, then the refcount may already be 0, in
2275 		 * which case it would be a bug to do kref_get().
2276 		 */
2277 		sdp = ERR_PTR(-ENODEV);
2278 	} else
2279 		kref_get(&sdp->d_ref);
2280 	read_unlock_irqrestore(&sg_index_lock, flags);
2281 
2282 	return sdp;
2283 }
2284 
2285 #ifdef CONFIG_SCSI_PROC_FS
2286 
2287 static struct proc_dir_entry *sg_proc_sgp = NULL;
2288 
2289 static char sg_proc_sg_dirname[] = "scsi/sg";
2290 
2291 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2292 
2293 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2294 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2295 			          size_t count, loff_t *off);
2296 static const struct file_operations adio_fops = {
2297 	.owner = THIS_MODULE,
2298 	.open = sg_proc_single_open_adio,
2299 	.read = seq_read,
2300 	.llseek = seq_lseek,
2301 	.write = sg_proc_write_adio,
2302 	.release = single_release,
2303 };
2304 
2305 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2306 static ssize_t sg_proc_write_dressz(struct file *filp,
2307 		const char __user *buffer, size_t count, loff_t *off);
2308 static const struct file_operations dressz_fops = {
2309 	.owner = THIS_MODULE,
2310 	.open = sg_proc_single_open_dressz,
2311 	.read = seq_read,
2312 	.llseek = seq_lseek,
2313 	.write = sg_proc_write_dressz,
2314 	.release = single_release,
2315 };
2316 
2317 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2318 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2319 static const struct file_operations version_fops = {
2320 	.owner = THIS_MODULE,
2321 	.open = sg_proc_single_open_version,
2322 	.read = seq_read,
2323 	.llseek = seq_lseek,
2324 	.release = single_release,
2325 };
2326 
2327 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2328 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2329 static const struct file_operations devhdr_fops = {
2330 	.owner = THIS_MODULE,
2331 	.open = sg_proc_single_open_devhdr,
2332 	.read = seq_read,
2333 	.llseek = seq_lseek,
2334 	.release = single_release,
2335 };
2336 
2337 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2338 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2339 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2340 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2341 static void dev_seq_stop(struct seq_file *s, void *v);
2342 static const struct file_operations dev_fops = {
2343 	.owner = THIS_MODULE,
2344 	.open = sg_proc_open_dev,
2345 	.read = seq_read,
2346 	.llseek = seq_lseek,
2347 	.release = seq_release,
2348 };
2349 static const struct seq_operations dev_seq_ops = {
2350 	.start = dev_seq_start,
2351 	.next  = dev_seq_next,
2352 	.stop  = dev_seq_stop,
2353 	.show  = sg_proc_seq_show_dev,
2354 };
2355 
2356 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2357 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2358 static const struct file_operations devstrs_fops = {
2359 	.owner = THIS_MODULE,
2360 	.open = sg_proc_open_devstrs,
2361 	.read = seq_read,
2362 	.llseek = seq_lseek,
2363 	.release = seq_release,
2364 };
2365 static const struct seq_operations devstrs_seq_ops = {
2366 	.start = dev_seq_start,
2367 	.next  = dev_seq_next,
2368 	.stop  = dev_seq_stop,
2369 	.show  = sg_proc_seq_show_devstrs,
2370 };
2371 
2372 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2373 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2374 static const struct file_operations debug_fops = {
2375 	.owner = THIS_MODULE,
2376 	.open = sg_proc_open_debug,
2377 	.read = seq_read,
2378 	.llseek = seq_lseek,
2379 	.release = seq_release,
2380 };
2381 static const struct seq_operations debug_seq_ops = {
2382 	.start = dev_seq_start,
2383 	.next  = dev_seq_next,
2384 	.stop  = dev_seq_stop,
2385 	.show  = sg_proc_seq_show_debug,
2386 };
2387 
2388 
2389 struct sg_proc_leaf {
2390 	const char * name;
2391 	const struct file_operations * fops;
2392 };
2393 
2394 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2395 	{"allow_dio", &adio_fops},
2396 	{"debug", &debug_fops},
2397 	{"def_reserved_size", &dressz_fops},
2398 	{"device_hdr", &devhdr_fops},
2399 	{"devices", &dev_fops},
2400 	{"device_strs", &devstrs_fops},
2401 	{"version", &version_fops}
2402 };
2403 
2404 static int
2405 sg_proc_init(void)
2406 {
2407 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2408 	int k;
2409 
2410 	sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2411 	if (!sg_proc_sgp)
2412 		return 1;
2413 	for (k = 0; k < num_leaves; ++k) {
2414 		const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2415 		umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2416 		proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2417 	}
2418 	return 0;
2419 }
2420 
2421 static void
2422 sg_proc_cleanup(void)
2423 {
2424 	int k;
2425 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2426 
2427 	if (!sg_proc_sgp)
2428 		return;
2429 	for (k = 0; k < num_leaves; ++k)
2430 		remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2431 	remove_proc_entry(sg_proc_sg_dirname, NULL);
2432 }
2433 
2434 
2435 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2436 {
2437 	seq_printf(s, "%d\n", *((int *)s->private));
2438 	return 0;
2439 }
2440 
2441 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2442 {
2443 	return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2444 }
2445 
2446 static ssize_t
2447 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2448 		   size_t count, loff_t *off)
2449 {
2450 	int err;
2451 	unsigned long num;
2452 
2453 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2454 		return -EACCES;
2455 	err = kstrtoul_from_user(buffer, count, 0, &num);
2456 	if (err)
2457 		return err;
2458 	sg_allow_dio = num ? 1 : 0;
2459 	return count;
2460 }
2461 
2462 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2463 {
2464 	return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2465 }
2466 
2467 static ssize_t
2468 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2469 		     size_t count, loff_t *off)
2470 {
2471 	int err;
2472 	unsigned long k = ULONG_MAX;
2473 
2474 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2475 		return -EACCES;
2476 
2477 	err = kstrtoul_from_user(buffer, count, 0, &k);
2478 	if (err)
2479 		return err;
2480 	if (k <= 1048576) {	/* limit "big buff" to 1 MB */
2481 		sg_big_buff = k;
2482 		return count;
2483 	}
2484 	return -ERANGE;
2485 }
2486 
2487 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2488 {
2489 	seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2490 		   sg_version_date);
2491 	return 0;
2492 }
2493 
2494 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2495 {
2496 	return single_open(file, sg_proc_seq_show_version, NULL);
2497 }
2498 
2499 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2500 {
2501 	seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2502 	return 0;
2503 }
2504 
2505 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2506 {
2507 	return single_open(file, sg_proc_seq_show_devhdr, NULL);
2508 }
2509 
2510 struct sg_proc_deviter {
2511 	loff_t	index;
2512 	size_t	max;
2513 };
2514 
2515 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2516 {
2517 	struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2518 
2519 	s->private = it;
2520 	if (! it)
2521 		return NULL;
2522 
2523 	it->index = *pos;
2524 	it->max = sg_last_dev();
2525 	if (it->index >= it->max)
2526 		return NULL;
2527 	return it;
2528 }
2529 
2530 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2531 {
2532 	struct sg_proc_deviter * it = s->private;
2533 
2534 	*pos = ++it->index;
2535 	return (it->index < it->max) ? it : NULL;
2536 }
2537 
2538 static void dev_seq_stop(struct seq_file *s, void *v)
2539 {
2540 	kfree(s->private);
2541 }
2542 
2543 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2544 {
2545         return seq_open(file, &dev_seq_ops);
2546 }
2547 
2548 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2549 {
2550 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2551 	Sg_device *sdp;
2552 	struct scsi_device *scsidp;
2553 	unsigned long iflags;
2554 
2555 	read_lock_irqsave(&sg_index_lock, iflags);
2556 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2557 	if ((NULL == sdp) || (NULL == sdp->device) ||
2558 	    (atomic_read(&sdp->detaching)))
2559 		seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2560 	else {
2561 		scsidp = sdp->device;
2562 		seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2563 			      scsidp->host->host_no, scsidp->channel,
2564 			      scsidp->id, scsidp->lun, (int) scsidp->type,
2565 			      1,
2566 			      (int) scsidp->queue_depth,
2567 			      (int) atomic_read(&scsidp->device_busy),
2568 			      (int) scsi_device_online(scsidp));
2569 	}
2570 	read_unlock_irqrestore(&sg_index_lock, iflags);
2571 	return 0;
2572 }
2573 
2574 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2575 {
2576         return seq_open(file, &devstrs_seq_ops);
2577 }
2578 
2579 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2580 {
2581 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2582 	Sg_device *sdp;
2583 	struct scsi_device *scsidp;
2584 	unsigned long iflags;
2585 
2586 	read_lock_irqsave(&sg_index_lock, iflags);
2587 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2588 	scsidp = sdp ? sdp->device : NULL;
2589 	if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2590 		seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2591 			   scsidp->vendor, scsidp->model, scsidp->rev);
2592 	else
2593 		seq_puts(s, "<no active device>\n");
2594 	read_unlock_irqrestore(&sg_index_lock, iflags);
2595 	return 0;
2596 }
2597 
2598 /* must be called while holding sg_index_lock */
2599 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2600 {
2601 	int k, new_interface, blen, usg;
2602 	Sg_request *srp;
2603 	Sg_fd *fp;
2604 	const sg_io_hdr_t *hp;
2605 	const char * cp;
2606 	unsigned int ms;
2607 
2608 	k = 0;
2609 	list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2610 		k++;
2611 		read_lock(&fp->rq_list_lock); /* irqs already disabled */
2612 		seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2613 			   "(res)sgat=%d low_dma=%d\n", k,
2614 			   jiffies_to_msecs(fp->timeout),
2615 			   fp->reserve.bufflen,
2616 			   (int) fp->reserve.k_use_sg,
2617 			   (int) sdp->device->host->unchecked_isa_dma);
2618 		seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2619 			   (int) fp->cmd_q, (int) fp->force_packid,
2620 			   (int) fp->keep_orphan);
2621 		list_for_each_entry(srp, &fp->rq_list, entry) {
2622 			hp = &srp->header;
2623 			new_interface = (hp->interface_id == '\0') ? 0 : 1;
2624 			if (srp->res_used) {
2625 				if (new_interface &&
2626 				    (SG_FLAG_MMAP_IO & hp->flags))
2627 					cp = "     mmap>> ";
2628 				else
2629 					cp = "     rb>> ";
2630 			} else {
2631 				if (SG_INFO_DIRECT_IO_MASK & hp->info)
2632 					cp = "     dio>> ";
2633 				else
2634 					cp = "     ";
2635 			}
2636 			seq_puts(s, cp);
2637 			blen = srp->data.bufflen;
2638 			usg = srp->data.k_use_sg;
2639 			seq_puts(s, srp->done ?
2640 				 ((1 == srp->done) ?  "rcv:" : "fin:")
2641 				  : "act:");
2642 			seq_printf(s, " id=%d blen=%d",
2643 				   srp->header.pack_id, blen);
2644 			if (srp->done)
2645 				seq_printf(s, " dur=%d", hp->duration);
2646 			else {
2647 				ms = jiffies_to_msecs(jiffies);
2648 				seq_printf(s, " t_o/elap=%d/%d",
2649 					(new_interface ? hp->timeout :
2650 						  jiffies_to_msecs(fp->timeout)),
2651 					(ms > hp->duration ? ms - hp->duration : 0));
2652 			}
2653 			seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2654 				   (int) srp->data.cmd_opcode);
2655 		}
2656 		if (list_empty(&fp->rq_list))
2657 			seq_puts(s, "     No requests active\n");
2658 		read_unlock(&fp->rq_list_lock);
2659 	}
2660 }
2661 
2662 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2663 {
2664         return seq_open(file, &debug_seq_ops);
2665 }
2666 
2667 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2668 {
2669 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2670 	Sg_device *sdp;
2671 	unsigned long iflags;
2672 
2673 	if (it && (0 == it->index))
2674 		seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2675 			   (int)it->max, sg_big_buff);
2676 
2677 	read_lock_irqsave(&sg_index_lock, iflags);
2678 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2679 	if (NULL == sdp)
2680 		goto skip;
2681 	read_lock(&sdp->sfd_lock);
2682 	if (!list_empty(&sdp->sfds)) {
2683 		seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2684 		if (atomic_read(&sdp->detaching))
2685 			seq_puts(s, "detaching pending close ");
2686 		else if (sdp->device) {
2687 			struct scsi_device *scsidp = sdp->device;
2688 
2689 			seq_printf(s, "%d:%d:%d:%llu   em=%d",
2690 				   scsidp->host->host_no,
2691 				   scsidp->channel, scsidp->id,
2692 				   scsidp->lun,
2693 				   scsidp->host->hostt->emulated);
2694 		}
2695 		seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2696 			   sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2697 		sg_proc_debug_helper(s, sdp);
2698 	}
2699 	read_unlock(&sdp->sfd_lock);
2700 skip:
2701 	read_unlock_irqrestore(&sg_index_lock, iflags);
2702 	return 0;
2703 }
2704 
2705 #endif				/* CONFIG_SCSI_PROC_FS */
2706 
2707 module_init(init_sg);
2708 module_exit(exit_sg);
2709