xref: /openbmc/linux/drivers/scsi/sg.c (revision 19c233b7)
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 /*
83  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
84  * Then when using 32 bit integers x * m may overflow during the calculation.
85  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
86  * calculates the same, but prevents the overflow when both m and d
87  * are "small" numbers (like HZ and USER_HZ).
88  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
89  * in 32 bits.
90  */
91 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
92 
93 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
94 
95 int sg_big_buff = SG_DEF_RESERVED_SIZE;
96 /* N.B. This variable is readable and writeable via
97    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
98    of this size (or less if there is not enough memory) will be reserved
99    for use by this file descriptor. [Deprecated usage: this variable is also
100    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
101    the kernel (i.e. it is not a module).] */
102 static int def_reserved_size = -1;	/* picks up init parameter */
103 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
104 
105 static int scatter_elem_sz = SG_SCATTER_SZ;
106 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
107 
108 #define SG_SECTOR_SZ 512
109 
110 static int sg_add_device(struct device *, struct class_interface *);
111 static void sg_remove_device(struct device *, struct class_interface *);
112 
113 static DEFINE_IDR(sg_index_idr);
114 static DEFINE_RWLOCK(sg_index_lock);	/* Also used to lock
115 							   file descriptor list for device */
116 
117 static struct class_interface sg_interface = {
118 	.add_dev        = sg_add_device,
119 	.remove_dev     = sg_remove_device,
120 };
121 
122 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
123 	unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
124 	unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
125 	unsigned bufflen;	/* Size of (aggregate) data buffer */
126 	struct page **pages;
127 	int page_order;
128 	char dio_in_use;	/* 0->indirect IO (or mmap), 1->dio */
129 	unsigned char cmd_opcode; /* first byte of command */
130 } Sg_scatter_hold;
131 
132 struct sg_device;		/* forward declarations */
133 struct sg_fd;
134 
135 typedef struct sg_request {	/* SG_MAX_QUEUE requests outstanding per file */
136 	struct sg_request *nextrp;	/* NULL -> tail request (slist) */
137 	struct sg_fd *parentfp;	/* NULL -> not in use */
138 	Sg_scatter_hold data;	/* hold buffer, perhaps scatter list */
139 	sg_io_hdr_t header;	/* scsi command+info, see <scsi/sg.h> */
140 	unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
141 	char res_used;		/* 1 -> using reserve buffer, 0 -> not ... */
142 	char orphan;		/* 1 -> drop on sight, 0 -> normal */
143 	char sg_io_owned;	/* 1 -> packet belongs to SG_IO */
144 	/* done protected by rq_list_lock */
145 	char done;		/* 0->before bh, 1->before read, 2->read */
146 	struct request *rq;
147 	struct bio *bio;
148 	struct execute_work ew;
149 } Sg_request;
150 
151 typedef struct sg_fd {		/* holds the state of a file descriptor */
152 	struct list_head sfd_siblings;  /* protected by device's sfd_lock */
153 	struct sg_device *parentdp;	/* owning device */
154 	wait_queue_head_t read_wait;	/* queue read until command done */
155 	rwlock_t rq_list_lock;	/* protect access to list in req_arr */
156 	int timeout;		/* defaults to SG_DEFAULT_TIMEOUT      */
157 	int timeout_user;	/* defaults to SG_DEFAULT_TIMEOUT_USER */
158 	Sg_scatter_hold reserve;	/* buffer held for this file descriptor */
159 	unsigned save_scat_len;	/* original length of trunc. scat. element */
160 	Sg_request *headrp;	/* head of request slist, NULL->empty */
161 	struct fasync_struct *async_qp;	/* used by asynchronous notification */
162 	Sg_request req_arr[SG_MAX_QUEUE];	/* used as singly-linked list */
163 	char low_dma;		/* as in parent but possibly overridden to 1 */
164 	char force_packid;	/* 1 -> pack_id input to read(), 0 -> ignored */
165 	char cmd_q;		/* 1 -> allow command queuing, 0 -> don't */
166 	unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
167 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
168 	char mmap_called;	/* 0 -> mmap() never called on this fd */
169 	struct kref f_ref;
170 	struct execute_work ew;
171 } Sg_fd;
172 
173 typedef struct sg_device { /* holds the state of each scsi generic device */
174 	struct scsi_device *device;
175 	wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
176 	struct mutex open_rel_lock;     /* held when in open() or release() */
177 	int sg_tablesize;	/* adapter's max scatter-gather table size */
178 	u32 index;		/* device index number */
179 	struct list_head sfds;
180 	rwlock_t sfd_lock;      /* protect access to sfd list */
181 	atomic_t detaching;     /* 0->device usable, 1->device detaching */
182 	bool exclude;		/* 1->open(O_EXCL) succeeded and is active */
183 	int open_cnt;		/* count of opens (perhaps < num(sfds) ) */
184 	char sgdebug;		/* 0->off, 1->sense, 9->dump dev, 10-> all devs */
185 	struct gendisk *disk;
186 	struct cdev * cdev;	/* char_dev [sysfs: /sys/cdev/major/sg<n>] */
187 	struct kref d_ref;
188 } Sg_device;
189 
190 /* tasklet or soft irq callback */
191 static void sg_rq_end_io(struct request *rq, int uptodate);
192 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
193 static int sg_finish_rem_req(Sg_request * srp);
194 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
195 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
196 			   Sg_request * srp);
197 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
198 			const char __user *buf, size_t count, int blocking,
199 			int read_only, int sg_io_owned, Sg_request **o_srp);
200 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
201 			   unsigned char *cmnd, int timeout, int blocking);
202 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
203 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
204 static void sg_build_reserve(Sg_fd * sfp, int req_size);
205 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
206 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
207 static Sg_fd *sg_add_sfp(Sg_device * sdp);
208 static void sg_remove_sfp(struct kref *);
209 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
210 static Sg_request *sg_add_request(Sg_fd * sfp);
211 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
212 static int sg_res_in_use(Sg_fd * sfp);
213 static Sg_device *sg_get_dev(int dev);
214 static void sg_device_destroy(struct kref *kref);
215 
216 #define SZ_SG_HEADER sizeof(struct sg_header)
217 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
218 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
219 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
220 
221 #define sg_printk(prefix, sdp, fmt, a...) \
222 	sdev_prefix_printk(prefix, (sdp)->device,		\
223 			   (sdp)->disk->disk_name, fmt, ##a)
224 
225 static int sg_allow_access(struct file *filp, unsigned char *cmd)
226 {
227 	struct sg_fd *sfp = filp->private_data;
228 
229 	if (sfp->parentdp->device->type == TYPE_SCANNER)
230 		return 0;
231 
232 	return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
233 }
234 
235 static int
236 open_wait(Sg_device *sdp, int flags)
237 {
238 	int retval = 0;
239 
240 	if (flags & O_EXCL) {
241 		while (sdp->open_cnt > 0) {
242 			mutex_unlock(&sdp->open_rel_lock);
243 			retval = wait_event_interruptible(sdp->open_wait,
244 					(atomic_read(&sdp->detaching) ||
245 					 !sdp->open_cnt));
246 			mutex_lock(&sdp->open_rel_lock);
247 
248 			if (retval) /* -ERESTARTSYS */
249 				return retval;
250 			if (atomic_read(&sdp->detaching))
251 				return -ENODEV;
252 		}
253 	} else {
254 		while (sdp->exclude) {
255 			mutex_unlock(&sdp->open_rel_lock);
256 			retval = wait_event_interruptible(sdp->open_wait,
257 					(atomic_read(&sdp->detaching) ||
258 					 !sdp->exclude));
259 			mutex_lock(&sdp->open_rel_lock);
260 
261 			if (retval) /* -ERESTARTSYS */
262 				return retval;
263 			if (atomic_read(&sdp->detaching))
264 				return -ENODEV;
265 		}
266 	}
267 
268 	return retval;
269 }
270 
271 /* Returns 0 on success, else a negated errno value */
272 static int
273 sg_open(struct inode *inode, struct file *filp)
274 {
275 	int dev = iminor(inode);
276 	int flags = filp->f_flags;
277 	struct request_queue *q;
278 	Sg_device *sdp;
279 	Sg_fd *sfp;
280 	int retval;
281 
282 	nonseekable_open(inode, filp);
283 	if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
284 		return -EPERM; /* Can't lock it with read only access */
285 	sdp = sg_get_dev(dev);
286 	if (IS_ERR(sdp))
287 		return PTR_ERR(sdp);
288 
289 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
290 				      "sg_open: flags=0x%x\n", flags));
291 
292 	/* This driver's module count bumped by fops_get in <linux/fs.h> */
293 	/* Prevent the device driver from vanishing while we sleep */
294 	retval = scsi_device_get(sdp->device);
295 	if (retval)
296 		goto sg_put;
297 
298 	retval = scsi_autopm_get_device(sdp->device);
299 	if (retval)
300 		goto sdp_put;
301 
302 	/* scsi_block_when_processing_errors() may block so bypass
303 	 * check if O_NONBLOCK. Permits SCSI commands to be issued
304 	 * during error recovery. Tread carefully. */
305 	if (!((flags & O_NONBLOCK) ||
306 	      scsi_block_when_processing_errors(sdp->device))) {
307 		retval = -ENXIO;
308 		/* we are in error recovery for this device */
309 		goto error_out;
310 	}
311 
312 	mutex_lock(&sdp->open_rel_lock);
313 	if (flags & O_NONBLOCK) {
314 		if (flags & O_EXCL) {
315 			if (sdp->open_cnt > 0) {
316 				retval = -EBUSY;
317 				goto error_mutex_locked;
318 			}
319 		} else {
320 			if (sdp->exclude) {
321 				retval = -EBUSY;
322 				goto error_mutex_locked;
323 			}
324 		}
325 	} else {
326 		retval = open_wait(sdp, flags);
327 		if (retval) /* -ERESTARTSYS or -ENODEV */
328 			goto error_mutex_locked;
329 	}
330 
331 	/* N.B. at this point we are holding the open_rel_lock */
332 	if (flags & O_EXCL)
333 		sdp->exclude = true;
334 
335 	if (sdp->open_cnt < 1) {  /* no existing opens */
336 		sdp->sgdebug = 0;
337 		q = sdp->device->request_queue;
338 		sdp->sg_tablesize = queue_max_segments(q);
339 	}
340 	sfp = sg_add_sfp(sdp);
341 	if (IS_ERR(sfp)) {
342 		retval = PTR_ERR(sfp);
343 		goto out_undo;
344 	}
345 
346 	filp->private_data = sfp;
347 	sdp->open_cnt++;
348 	mutex_unlock(&sdp->open_rel_lock);
349 
350 	retval = 0;
351 sg_put:
352 	kref_put(&sdp->d_ref, sg_device_destroy);
353 	return retval;
354 
355 out_undo:
356 	if (flags & O_EXCL) {
357 		sdp->exclude = false;   /* undo if error */
358 		wake_up_interruptible(&sdp->open_wait);
359 	}
360 error_mutex_locked:
361 	mutex_unlock(&sdp->open_rel_lock);
362 error_out:
363 	scsi_autopm_put_device(sdp->device);
364 sdp_put:
365 	scsi_device_put(sdp->device);
366 	goto sg_put;
367 }
368 
369 /* Release resources associated with a successful sg_open()
370  * Returns 0 on success, else a negated errno value */
371 static int
372 sg_release(struct inode *inode, struct file *filp)
373 {
374 	Sg_device *sdp;
375 	Sg_fd *sfp;
376 
377 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
378 		return -ENXIO;
379 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
380 
381 	mutex_lock(&sdp->open_rel_lock);
382 	scsi_autopm_put_device(sdp->device);
383 	kref_put(&sfp->f_ref, sg_remove_sfp);
384 	sdp->open_cnt--;
385 
386 	/* possibly many open()s waiting on exlude clearing, start many;
387 	 * only open(O_EXCL)s wait on 0==open_cnt so only start one */
388 	if (sdp->exclude) {
389 		sdp->exclude = false;
390 		wake_up_interruptible_all(&sdp->open_wait);
391 	} else if (0 == sdp->open_cnt) {
392 		wake_up_interruptible(&sdp->open_wait);
393 	}
394 	mutex_unlock(&sdp->open_rel_lock);
395 	return 0;
396 }
397 
398 static ssize_t
399 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
400 {
401 	Sg_device *sdp;
402 	Sg_fd *sfp;
403 	Sg_request *srp;
404 	int req_pack_id = -1;
405 	sg_io_hdr_t *hp;
406 	struct sg_header *old_hdr = NULL;
407 	int retval = 0;
408 
409 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
410 		return -ENXIO;
411 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
412 				      "sg_read: count=%d\n", (int) count));
413 
414 	if (!access_ok(VERIFY_WRITE, buf, count))
415 		return -EFAULT;
416 	if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
417 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
418 		if (!old_hdr)
419 			return -ENOMEM;
420 		if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
421 			retval = -EFAULT;
422 			goto free_old_hdr;
423 		}
424 		if (old_hdr->reply_len < 0) {
425 			if (count >= SZ_SG_IO_HDR) {
426 				sg_io_hdr_t *new_hdr;
427 				new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
428 				if (!new_hdr) {
429 					retval = -ENOMEM;
430 					goto free_old_hdr;
431 				}
432 				retval =__copy_from_user
433 				    (new_hdr, buf, SZ_SG_IO_HDR);
434 				req_pack_id = new_hdr->pack_id;
435 				kfree(new_hdr);
436 				if (retval) {
437 					retval = -EFAULT;
438 					goto free_old_hdr;
439 				}
440 			}
441 		} else
442 			req_pack_id = old_hdr->pack_id;
443 	}
444 	srp = sg_get_rq_mark(sfp, req_pack_id);
445 	if (!srp) {		/* now wait on packet to arrive */
446 		if (atomic_read(&sdp->detaching)) {
447 			retval = -ENODEV;
448 			goto free_old_hdr;
449 		}
450 		if (filp->f_flags & O_NONBLOCK) {
451 			retval = -EAGAIN;
452 			goto free_old_hdr;
453 		}
454 		retval = wait_event_interruptible(sfp->read_wait,
455 			(atomic_read(&sdp->detaching) ||
456 			(srp = sg_get_rq_mark(sfp, req_pack_id))));
457 		if (atomic_read(&sdp->detaching)) {
458 			retval = -ENODEV;
459 			goto free_old_hdr;
460 		}
461 		if (retval) {
462 			/* -ERESTARTSYS as signal hit process */
463 			goto free_old_hdr;
464 		}
465 	}
466 	if (srp->header.interface_id != '\0') {
467 		retval = sg_new_read(sfp, buf, count, srp);
468 		goto free_old_hdr;
469 	}
470 
471 	hp = &srp->header;
472 	if (old_hdr == NULL) {
473 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
474 		if (! old_hdr) {
475 			retval = -ENOMEM;
476 			goto free_old_hdr;
477 		}
478 	}
479 	memset(old_hdr, 0, SZ_SG_HEADER);
480 	old_hdr->reply_len = (int) hp->timeout;
481 	old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
482 	old_hdr->pack_id = hp->pack_id;
483 	old_hdr->twelve_byte =
484 	    ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
485 	old_hdr->target_status = hp->masked_status;
486 	old_hdr->host_status = hp->host_status;
487 	old_hdr->driver_status = hp->driver_status;
488 	if ((CHECK_CONDITION & hp->masked_status) ||
489 	    (DRIVER_SENSE & hp->driver_status))
490 		memcpy(old_hdr->sense_buffer, srp->sense_b,
491 		       sizeof (old_hdr->sense_buffer));
492 	switch (hp->host_status) {
493 	/* This setup of 'result' is for backward compatibility and is best
494 	   ignored by the user who should use target, host + driver status */
495 	case DID_OK:
496 	case DID_PASSTHROUGH:
497 	case DID_SOFT_ERROR:
498 		old_hdr->result = 0;
499 		break;
500 	case DID_NO_CONNECT:
501 	case DID_BUS_BUSY:
502 	case DID_TIME_OUT:
503 		old_hdr->result = EBUSY;
504 		break;
505 	case DID_BAD_TARGET:
506 	case DID_ABORT:
507 	case DID_PARITY:
508 	case DID_RESET:
509 	case DID_BAD_INTR:
510 		old_hdr->result = EIO;
511 		break;
512 	case DID_ERROR:
513 		old_hdr->result = (srp->sense_b[0] == 0 &&
514 				  hp->masked_status == GOOD) ? 0 : EIO;
515 		break;
516 	default:
517 		old_hdr->result = EIO;
518 		break;
519 	}
520 
521 	/* Now copy the result back to the user buffer.  */
522 	if (count >= SZ_SG_HEADER) {
523 		if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
524 			retval = -EFAULT;
525 			goto free_old_hdr;
526 		}
527 		buf += SZ_SG_HEADER;
528 		if (count > old_hdr->reply_len)
529 			count = old_hdr->reply_len;
530 		if (count > SZ_SG_HEADER) {
531 			if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
532 				retval = -EFAULT;
533 				goto free_old_hdr;
534 			}
535 		}
536 	} else
537 		count = (old_hdr->result == 0) ? 0 : -EIO;
538 	sg_finish_rem_req(srp);
539 	retval = count;
540 free_old_hdr:
541 	kfree(old_hdr);
542 	return retval;
543 }
544 
545 static ssize_t
546 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
547 {
548 	sg_io_hdr_t *hp = &srp->header;
549 	int err = 0, err2;
550 	int len;
551 
552 	if (count < SZ_SG_IO_HDR) {
553 		err = -EINVAL;
554 		goto err_out;
555 	}
556 	hp->sb_len_wr = 0;
557 	if ((hp->mx_sb_len > 0) && hp->sbp) {
558 		if ((CHECK_CONDITION & hp->masked_status) ||
559 		    (DRIVER_SENSE & hp->driver_status)) {
560 			int sb_len = SCSI_SENSE_BUFFERSIZE;
561 			sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
562 			len = 8 + (int) srp->sense_b[7];	/* Additional sense length field */
563 			len = (len > sb_len) ? sb_len : len;
564 			if (copy_to_user(hp->sbp, srp->sense_b, len)) {
565 				err = -EFAULT;
566 				goto err_out;
567 			}
568 			hp->sb_len_wr = len;
569 		}
570 	}
571 	if (hp->masked_status || hp->host_status || hp->driver_status)
572 		hp->info |= SG_INFO_CHECK;
573 	if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
574 		err = -EFAULT;
575 		goto err_out;
576 	}
577 err_out:
578 	err2 = sg_finish_rem_req(srp);
579 	return err ? : err2 ? : count;
580 }
581 
582 static ssize_t
583 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
584 {
585 	int mxsize, cmd_size, k;
586 	int input_size, blocking;
587 	unsigned char opcode;
588 	Sg_device *sdp;
589 	Sg_fd *sfp;
590 	Sg_request *srp;
591 	struct sg_header old_hdr;
592 	sg_io_hdr_t *hp;
593 	unsigned char cmnd[SG_MAX_CDB_SIZE];
594 
595 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
596 		return -ENXIO;
597 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
598 				      "sg_write: count=%d\n", (int) count));
599 	if (atomic_read(&sdp->detaching))
600 		return -ENODEV;
601 	if (!((filp->f_flags & O_NONBLOCK) ||
602 	      scsi_block_when_processing_errors(sdp->device)))
603 		return -ENXIO;
604 
605 	if (!access_ok(VERIFY_READ, buf, count))
606 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
607 	if (count < SZ_SG_HEADER)
608 		return -EIO;
609 	if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
610 		return -EFAULT;
611 	blocking = !(filp->f_flags & O_NONBLOCK);
612 	if (old_hdr.reply_len < 0)
613 		return sg_new_write(sfp, filp, buf, count,
614 				    blocking, 0, 0, NULL);
615 	if (count < (SZ_SG_HEADER + 6))
616 		return -EIO;	/* The minimum scsi command length is 6 bytes. */
617 
618 	if (!(srp = sg_add_request(sfp))) {
619 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
620 					      "sg_write: queue full\n"));
621 		return -EDOM;
622 	}
623 	buf += SZ_SG_HEADER;
624 	__get_user(opcode, buf);
625 	if (sfp->next_cmd_len > 0) {
626 		cmd_size = sfp->next_cmd_len;
627 		sfp->next_cmd_len = 0;	/* reset so only this write() effected */
628 	} else {
629 		cmd_size = COMMAND_SIZE(opcode);	/* based on SCSI command group */
630 		if ((opcode >= 0xc0) && old_hdr.twelve_byte)
631 			cmd_size = 12;
632 	}
633 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
634 		"sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
635 /* Determine buffer size.  */
636 	input_size = count - cmd_size;
637 	mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
638 	mxsize -= SZ_SG_HEADER;
639 	input_size -= SZ_SG_HEADER;
640 	if (input_size < 0) {
641 		sg_remove_request(sfp, srp);
642 		return -EIO;	/* User did not pass enough bytes for this command. */
643 	}
644 	hp = &srp->header;
645 	hp->interface_id = '\0';	/* indicator of old interface tunnelled */
646 	hp->cmd_len = (unsigned char) cmd_size;
647 	hp->iovec_count = 0;
648 	hp->mx_sb_len = 0;
649 	if (input_size > 0)
650 		hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
651 		    SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
652 	else
653 		hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
654 	hp->dxfer_len = mxsize;
655 	if (hp->dxfer_direction == SG_DXFER_TO_DEV)
656 		hp->dxferp = (char __user *)buf + cmd_size;
657 	else
658 		hp->dxferp = NULL;
659 	hp->sbp = NULL;
660 	hp->timeout = old_hdr.reply_len;	/* structure abuse ... */
661 	hp->flags = input_size;	/* structure abuse ... */
662 	hp->pack_id = old_hdr.pack_id;
663 	hp->usr_ptr = NULL;
664 	if (__copy_from_user(cmnd, buf, cmd_size))
665 		return -EFAULT;
666 	/*
667 	 * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
668 	 * but is is possible that the app intended SG_DXFER_TO_DEV, because there
669 	 * is a non-zero input_size, so emit a warning.
670 	 */
671 	if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
672 		static char cmd[TASK_COMM_LEN];
673 		if (strcmp(current->comm, cmd)) {
674 			printk_ratelimited(KERN_WARNING
675 					   "sg_write: data in/out %d/%d bytes "
676 					   "for SCSI command 0x%x-- guessing "
677 					   "data in;\n   program %s not setting "
678 					   "count and/or reply_len properly\n",
679 					   old_hdr.reply_len - (int)SZ_SG_HEADER,
680 					   input_size, (unsigned int) cmnd[0],
681 					   current->comm);
682 			strcpy(cmd, current->comm);
683 		}
684 	}
685 	k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
686 	return (k < 0) ? k : count;
687 }
688 
689 static ssize_t
690 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
691 		 size_t count, int blocking, int read_only, int sg_io_owned,
692 		 Sg_request **o_srp)
693 {
694 	int k;
695 	Sg_request *srp;
696 	sg_io_hdr_t *hp;
697 	unsigned char cmnd[SG_MAX_CDB_SIZE];
698 	int timeout;
699 	unsigned long ul_timeout;
700 
701 	if (count < SZ_SG_IO_HDR)
702 		return -EINVAL;
703 	if (!access_ok(VERIFY_READ, buf, count))
704 		return -EFAULT; /* protects following copy_from_user()s + get_user()s */
705 
706 	sfp->cmd_q = 1;	/* when sg_io_hdr seen, set command queuing on */
707 	if (!(srp = sg_add_request(sfp))) {
708 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
709 					      "sg_new_write: queue full\n"));
710 		return -EDOM;
711 	}
712 	srp->sg_io_owned = sg_io_owned;
713 	hp = &srp->header;
714 	if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
715 		sg_remove_request(sfp, srp);
716 		return -EFAULT;
717 	}
718 	if (hp->interface_id != 'S') {
719 		sg_remove_request(sfp, srp);
720 		return -ENOSYS;
721 	}
722 	if (hp->flags & SG_FLAG_MMAP_IO) {
723 		if (hp->dxfer_len > sfp->reserve.bufflen) {
724 			sg_remove_request(sfp, srp);
725 			return -ENOMEM;	/* MMAP_IO size must fit in reserve buffer */
726 		}
727 		if (hp->flags & SG_FLAG_DIRECT_IO) {
728 			sg_remove_request(sfp, srp);
729 			return -EINVAL;	/* either MMAP_IO or DIRECT_IO (not both) */
730 		}
731 		if (sg_res_in_use(sfp)) {
732 			sg_remove_request(sfp, srp);
733 			return -EBUSY;	/* reserve buffer already being used */
734 		}
735 	}
736 	ul_timeout = msecs_to_jiffies(srp->header.timeout);
737 	timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
738 	if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
739 		sg_remove_request(sfp, srp);
740 		return -EMSGSIZE;
741 	}
742 	if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
743 		sg_remove_request(sfp, srp);
744 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
745 	}
746 	if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
747 		sg_remove_request(sfp, srp);
748 		return -EFAULT;
749 	}
750 	if (read_only && sg_allow_access(file, cmnd)) {
751 		sg_remove_request(sfp, srp);
752 		return -EPERM;
753 	}
754 	k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
755 	if (k < 0)
756 		return k;
757 	if (o_srp)
758 		*o_srp = srp;
759 	return count;
760 }
761 
762 static int
763 sg_common_write(Sg_fd * sfp, Sg_request * srp,
764 		unsigned char *cmnd, int timeout, int blocking)
765 {
766 	int k, at_head;
767 	Sg_device *sdp = sfp->parentdp;
768 	sg_io_hdr_t *hp = &srp->header;
769 
770 	srp->data.cmd_opcode = cmnd[0];	/* hold opcode of command */
771 	hp->status = 0;
772 	hp->masked_status = 0;
773 	hp->msg_status = 0;
774 	hp->info = 0;
775 	hp->host_status = 0;
776 	hp->driver_status = 0;
777 	hp->resid = 0;
778 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
779 			"sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
780 			(int) cmnd[0], (int) hp->cmd_len));
781 
782 	k = sg_start_req(srp, cmnd);
783 	if (k) {
784 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
785 			"sg_common_write: start_req err=%d\n", k));
786 		sg_finish_rem_req(srp);
787 		return k;	/* probably out of space --> ENOMEM */
788 	}
789 	if (atomic_read(&sdp->detaching)) {
790 		if (srp->bio)
791 			blk_end_request_all(srp->rq, -EIO);
792 		sg_finish_rem_req(srp);
793 		return -ENODEV;
794 	}
795 
796 	hp->duration = jiffies_to_msecs(jiffies);
797 	if (hp->interface_id != '\0' &&	/* v3 (or later) interface */
798 	    (SG_FLAG_Q_AT_TAIL & hp->flags))
799 		at_head = 0;
800 	else
801 		at_head = 1;
802 
803 	srp->rq->timeout = timeout;
804 	kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
805 	blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
806 			      srp->rq, at_head, sg_rq_end_io);
807 	return 0;
808 }
809 
810 static int srp_done(Sg_fd *sfp, Sg_request *srp)
811 {
812 	unsigned long flags;
813 	int ret;
814 
815 	read_lock_irqsave(&sfp->rq_list_lock, flags);
816 	ret = srp->done;
817 	read_unlock_irqrestore(&sfp->rq_list_lock, flags);
818 	return ret;
819 }
820 
821 static int max_sectors_bytes(struct request_queue *q)
822 {
823 	unsigned int max_sectors = queue_max_sectors(q);
824 
825 	max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
826 
827 	return max_sectors << 9;
828 }
829 
830 static long
831 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
832 {
833 	void __user *p = (void __user *)arg;
834 	int __user *ip = p;
835 	int result, val, read_only;
836 	Sg_device *sdp;
837 	Sg_fd *sfp;
838 	Sg_request *srp;
839 	unsigned long iflags;
840 
841 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
842 		return -ENXIO;
843 
844 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
845 				   "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
846 	read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
847 
848 	switch (cmd_in) {
849 	case SG_IO:
850 		if (atomic_read(&sdp->detaching))
851 			return -ENODEV;
852 		if (!scsi_block_when_processing_errors(sdp->device))
853 			return -ENXIO;
854 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
855 			return -EFAULT;
856 		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
857 				 1, read_only, 1, &srp);
858 		if (result < 0)
859 			return result;
860 		result = wait_event_interruptible(sfp->read_wait,
861 			(srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
862 		if (atomic_read(&sdp->detaching))
863 			return -ENODEV;
864 		write_lock_irq(&sfp->rq_list_lock);
865 		if (srp->done) {
866 			srp->done = 2;
867 			write_unlock_irq(&sfp->rq_list_lock);
868 			result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
869 			return (result < 0) ? result : 0;
870 		}
871 		srp->orphan = 1;
872 		write_unlock_irq(&sfp->rq_list_lock);
873 		return result;	/* -ERESTARTSYS because signal hit process */
874 	case SG_SET_TIMEOUT:
875 		result = get_user(val, ip);
876 		if (result)
877 			return result;
878 		if (val < 0)
879 			return -EIO;
880 		if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
881 		    val = MULDIV (INT_MAX, USER_HZ, HZ);
882 		sfp->timeout_user = val;
883 		sfp->timeout = MULDIV (val, HZ, USER_HZ);
884 
885 		return 0;
886 	case SG_GET_TIMEOUT:	/* N.B. User receives timeout as return value */
887 				/* strange ..., for backward compatibility */
888 		return sfp->timeout_user;
889 	case SG_SET_FORCE_LOW_DMA:
890 		result = get_user(val, ip);
891 		if (result)
892 			return result;
893 		if (val) {
894 			sfp->low_dma = 1;
895 			if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
896 				val = (int) sfp->reserve.bufflen;
897 				sg_remove_scat(sfp, &sfp->reserve);
898 				sg_build_reserve(sfp, val);
899 			}
900 		} else {
901 			if (atomic_read(&sdp->detaching))
902 				return -ENODEV;
903 			sfp->low_dma = sdp->device->host->unchecked_isa_dma;
904 		}
905 		return 0;
906 	case SG_GET_LOW_DMA:
907 		return put_user((int) sfp->low_dma, ip);
908 	case SG_GET_SCSI_ID:
909 		if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
910 			return -EFAULT;
911 		else {
912 			sg_scsi_id_t __user *sg_idp = p;
913 
914 			if (atomic_read(&sdp->detaching))
915 				return -ENODEV;
916 			__put_user((int) sdp->device->host->host_no,
917 				   &sg_idp->host_no);
918 			__put_user((int) sdp->device->channel,
919 				   &sg_idp->channel);
920 			__put_user((int) sdp->device->id, &sg_idp->scsi_id);
921 			__put_user((int) sdp->device->lun, &sg_idp->lun);
922 			__put_user((int) sdp->device->type, &sg_idp->scsi_type);
923 			__put_user((short) sdp->device->host->cmd_per_lun,
924 				   &sg_idp->h_cmd_per_lun);
925 			__put_user((short) sdp->device->queue_depth,
926 				   &sg_idp->d_queue_depth);
927 			__put_user(0, &sg_idp->unused[0]);
928 			__put_user(0, &sg_idp->unused[1]);
929 			return 0;
930 		}
931 	case SG_SET_FORCE_PACK_ID:
932 		result = get_user(val, ip);
933 		if (result)
934 			return result;
935 		sfp->force_packid = val ? 1 : 0;
936 		return 0;
937 	case SG_GET_PACK_ID:
938 		if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
939 			return -EFAULT;
940 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
941 		for (srp = sfp->headrp; srp; srp = srp->nextrp) {
942 			if ((1 == srp->done) && (!srp->sg_io_owned)) {
943 				read_unlock_irqrestore(&sfp->rq_list_lock,
944 						       iflags);
945 				__put_user(srp->header.pack_id, ip);
946 				return 0;
947 			}
948 		}
949 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
950 		__put_user(-1, ip);
951 		return 0;
952 	case SG_GET_NUM_WAITING:
953 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
954 		for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
955 			if ((1 == srp->done) && (!srp->sg_io_owned))
956 				++val;
957 		}
958 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
959 		return put_user(val, ip);
960 	case SG_GET_SG_TABLESIZE:
961 		return put_user(sdp->sg_tablesize, ip);
962 	case SG_SET_RESERVED_SIZE:
963 		result = get_user(val, ip);
964 		if (result)
965 			return result;
966                 if (val < 0)
967                         return -EINVAL;
968 		val = min_t(int, val,
969 			    max_sectors_bytes(sdp->device->request_queue));
970 		if (val != sfp->reserve.bufflen) {
971 			if (sg_res_in_use(sfp) || sfp->mmap_called)
972 				return -EBUSY;
973 			sg_remove_scat(sfp, &sfp->reserve);
974 			sg_build_reserve(sfp, val);
975 		}
976 		return 0;
977 	case SG_GET_RESERVED_SIZE:
978 		val = min_t(int, sfp->reserve.bufflen,
979 			    max_sectors_bytes(sdp->device->request_queue));
980 		return put_user(val, ip);
981 	case SG_SET_COMMAND_Q:
982 		result = get_user(val, ip);
983 		if (result)
984 			return result;
985 		sfp->cmd_q = val ? 1 : 0;
986 		return 0;
987 	case SG_GET_COMMAND_Q:
988 		return put_user((int) sfp->cmd_q, ip);
989 	case SG_SET_KEEP_ORPHAN:
990 		result = get_user(val, ip);
991 		if (result)
992 			return result;
993 		sfp->keep_orphan = val;
994 		return 0;
995 	case SG_GET_KEEP_ORPHAN:
996 		return put_user((int) sfp->keep_orphan, ip);
997 	case SG_NEXT_CMD_LEN:
998 		result = get_user(val, ip);
999 		if (result)
1000 			return result;
1001 		sfp->next_cmd_len = (val > 0) ? val : 0;
1002 		return 0;
1003 	case SG_GET_VERSION_NUM:
1004 		return put_user(sg_version_num, ip);
1005 	case SG_GET_ACCESS_COUNT:
1006 		/* faked - we don't have a real access count anymore */
1007 		val = (sdp->device ? 1 : 0);
1008 		return put_user(val, ip);
1009 	case SG_GET_REQUEST_TABLE:
1010 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1011 			return -EFAULT;
1012 		else {
1013 			sg_req_info_t *rinfo;
1014 			unsigned int ms;
1015 
1016 			rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1017 								GFP_KERNEL);
1018 			if (!rinfo)
1019 				return -ENOMEM;
1020 			read_lock_irqsave(&sfp->rq_list_lock, iflags);
1021 			for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
1022 			     ++val, srp = srp ? srp->nextrp : srp) {
1023 				memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
1024 				if (srp) {
1025 					rinfo[val].req_state = srp->done + 1;
1026 					rinfo[val].problem =
1027 					    srp->header.masked_status &
1028 					    srp->header.host_status &
1029 					    srp->header.driver_status;
1030 					if (srp->done)
1031 						rinfo[val].duration =
1032 							srp->header.duration;
1033 					else {
1034 						ms = jiffies_to_msecs(jiffies);
1035 						rinfo[val].duration =
1036 						    (ms > srp->header.duration) ?
1037 						    (ms - srp->header.duration) : 0;
1038 					}
1039 					rinfo[val].orphan = srp->orphan;
1040 					rinfo[val].sg_io_owned =
1041 							srp->sg_io_owned;
1042 					rinfo[val].pack_id =
1043 							srp->header.pack_id;
1044 					rinfo[val].usr_ptr =
1045 							srp->header.usr_ptr;
1046 				}
1047 			}
1048 			read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1049 			result = __copy_to_user(p, rinfo,
1050 						SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1051 			result = result ? -EFAULT : 0;
1052 			kfree(rinfo);
1053 			return result;
1054 		}
1055 	case SG_EMULATED_HOST:
1056 		if (atomic_read(&sdp->detaching))
1057 			return -ENODEV;
1058 		return put_user(sdp->device->host->hostt->emulated, ip);
1059 	case SCSI_IOCTL_SEND_COMMAND:
1060 		if (atomic_read(&sdp->detaching))
1061 			return -ENODEV;
1062 		if (read_only) {
1063 			unsigned char opcode = WRITE_6;
1064 			Scsi_Ioctl_Command __user *siocp = p;
1065 
1066 			if (copy_from_user(&opcode, siocp->data, 1))
1067 				return -EFAULT;
1068 			if (sg_allow_access(filp, &opcode))
1069 				return -EPERM;
1070 		}
1071 		return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1072 	case SG_SET_DEBUG:
1073 		result = get_user(val, ip);
1074 		if (result)
1075 			return result;
1076 		sdp->sgdebug = (char) val;
1077 		return 0;
1078 	case BLKSECTGET:
1079 		return put_user(max_sectors_bytes(sdp->device->request_queue),
1080 				ip);
1081 	case BLKTRACESETUP:
1082 		return blk_trace_setup(sdp->device->request_queue,
1083 				       sdp->disk->disk_name,
1084 				       MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1085 				       NULL,
1086 				       (char *)arg);
1087 	case BLKTRACESTART:
1088 		return blk_trace_startstop(sdp->device->request_queue, 1);
1089 	case BLKTRACESTOP:
1090 		return blk_trace_startstop(sdp->device->request_queue, 0);
1091 	case BLKTRACETEARDOWN:
1092 		return blk_trace_remove(sdp->device->request_queue);
1093 	case SCSI_IOCTL_GET_IDLUN:
1094 	case SCSI_IOCTL_GET_BUS_NUMBER:
1095 	case SCSI_IOCTL_PROBE_HOST:
1096 	case SG_GET_TRANSFORM:
1097 	case SG_SCSI_RESET:
1098 		if (atomic_read(&sdp->detaching))
1099 			return -ENODEV;
1100 		break;
1101 	default:
1102 		if (read_only)
1103 			return -EPERM;	/* don't know so take safe approach */
1104 		break;
1105 	}
1106 
1107 	result = scsi_ioctl_block_when_processing_errors(sdp->device,
1108 			cmd_in, filp->f_flags & O_NDELAY);
1109 	if (result)
1110 		return result;
1111 	return scsi_ioctl(sdp->device, cmd_in, p);
1112 }
1113 
1114 #ifdef CONFIG_COMPAT
1115 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1116 {
1117 	Sg_device *sdp;
1118 	Sg_fd *sfp;
1119 	struct scsi_device *sdev;
1120 
1121 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1122 		return -ENXIO;
1123 
1124 	sdev = sdp->device;
1125 	if (sdev->host->hostt->compat_ioctl) {
1126 		int ret;
1127 
1128 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1129 
1130 		return ret;
1131 	}
1132 
1133 	return -ENOIOCTLCMD;
1134 }
1135 #endif
1136 
1137 static unsigned int
1138 sg_poll(struct file *filp, poll_table * wait)
1139 {
1140 	unsigned int res = 0;
1141 	Sg_device *sdp;
1142 	Sg_fd *sfp;
1143 	Sg_request *srp;
1144 	int count = 0;
1145 	unsigned long iflags;
1146 
1147 	sfp = filp->private_data;
1148 	if (!sfp)
1149 		return POLLERR;
1150 	sdp = sfp->parentdp;
1151 	if (!sdp)
1152 		return POLLERR;
1153 	poll_wait(filp, &sfp->read_wait, wait);
1154 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
1155 	for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1156 		/* if any read waiting, flag it */
1157 		if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1158 			res = POLLIN | POLLRDNORM;
1159 		++count;
1160 	}
1161 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1162 
1163 	if (atomic_read(&sdp->detaching))
1164 		res |= POLLHUP;
1165 	else if (!sfp->cmd_q) {
1166 		if (0 == count)
1167 			res |= POLLOUT | POLLWRNORM;
1168 	} else if (count < SG_MAX_QUEUE)
1169 		res |= POLLOUT | POLLWRNORM;
1170 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1171 				      "sg_poll: res=0x%x\n", (int) res));
1172 	return res;
1173 }
1174 
1175 static int
1176 sg_fasync(int fd, struct file *filp, int mode)
1177 {
1178 	Sg_device *sdp;
1179 	Sg_fd *sfp;
1180 
1181 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1182 		return -ENXIO;
1183 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1184 				      "sg_fasync: mode=%d\n", mode));
1185 
1186 	return fasync_helper(fd, filp, mode, &sfp->async_qp);
1187 }
1188 
1189 static int
1190 sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1191 {
1192 	Sg_fd *sfp;
1193 	unsigned long offset, len, sa;
1194 	Sg_scatter_hold *rsv_schp;
1195 	int k, length;
1196 
1197 	if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1198 		return VM_FAULT_SIGBUS;
1199 	rsv_schp = &sfp->reserve;
1200 	offset = vmf->pgoff << PAGE_SHIFT;
1201 	if (offset >= rsv_schp->bufflen)
1202 		return VM_FAULT_SIGBUS;
1203 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1204 				      "sg_vma_fault: offset=%lu, scatg=%d\n",
1205 				      offset, rsv_schp->k_use_sg));
1206 	sa = vma->vm_start;
1207 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1208 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1209 		len = vma->vm_end - sa;
1210 		len = (len < length) ? len : length;
1211 		if (offset < len) {
1212 			struct page *page = nth_page(rsv_schp->pages[k],
1213 						     offset >> PAGE_SHIFT);
1214 			get_page(page);	/* increment page count */
1215 			vmf->page = page;
1216 			return 0; /* success */
1217 		}
1218 		sa += len;
1219 		offset -= len;
1220 	}
1221 
1222 	return VM_FAULT_SIGBUS;
1223 }
1224 
1225 static const struct vm_operations_struct sg_mmap_vm_ops = {
1226 	.fault = sg_vma_fault,
1227 };
1228 
1229 static int
1230 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1231 {
1232 	Sg_fd *sfp;
1233 	unsigned long req_sz, len, sa;
1234 	Sg_scatter_hold *rsv_schp;
1235 	int k, length;
1236 
1237 	if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1238 		return -ENXIO;
1239 	req_sz = vma->vm_end - vma->vm_start;
1240 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1241 				      "sg_mmap starting, vm_start=%p, len=%d\n",
1242 				      (void *) vma->vm_start, (int) req_sz));
1243 	if (vma->vm_pgoff)
1244 		return -EINVAL;	/* want no offset */
1245 	rsv_schp = &sfp->reserve;
1246 	if (req_sz > rsv_schp->bufflen)
1247 		return -ENOMEM;	/* cannot map more than reserved buffer */
1248 
1249 	sa = vma->vm_start;
1250 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1251 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1252 		len = vma->vm_end - sa;
1253 		len = (len < length) ? len : length;
1254 		sa += len;
1255 	}
1256 
1257 	sfp->mmap_called = 1;
1258 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1259 	vma->vm_private_data = sfp;
1260 	vma->vm_ops = &sg_mmap_vm_ops;
1261 	return 0;
1262 }
1263 
1264 static void
1265 sg_rq_end_io_usercontext(struct work_struct *work)
1266 {
1267 	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1268 	struct sg_fd *sfp = srp->parentfp;
1269 
1270 	sg_finish_rem_req(srp);
1271 	kref_put(&sfp->f_ref, sg_remove_sfp);
1272 }
1273 
1274 /*
1275  * This function is a "bottom half" handler that is called by the mid
1276  * level when a command is completed (or has failed).
1277  */
1278 static void
1279 sg_rq_end_io(struct request *rq, int uptodate)
1280 {
1281 	struct sg_request *srp = rq->end_io_data;
1282 	Sg_device *sdp;
1283 	Sg_fd *sfp;
1284 	unsigned long iflags;
1285 	unsigned int ms;
1286 	char *sense;
1287 	int result, resid, done = 1;
1288 
1289 	if (WARN_ON(srp->done != 0))
1290 		return;
1291 
1292 	sfp = srp->parentfp;
1293 	if (WARN_ON(sfp == NULL))
1294 		return;
1295 
1296 	sdp = sfp->parentdp;
1297 	if (unlikely(atomic_read(&sdp->detaching)))
1298 		pr_info("%s: device detaching\n", __func__);
1299 
1300 	sense = rq->sense;
1301 	result = rq->errors;
1302 	resid = rq->resid_len;
1303 
1304 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1305 				      "sg_cmd_done: pack_id=%d, res=0x%x\n",
1306 				      srp->header.pack_id, result));
1307 	srp->header.resid = resid;
1308 	ms = jiffies_to_msecs(jiffies);
1309 	srp->header.duration = (ms > srp->header.duration) ?
1310 				(ms - srp->header.duration) : 0;
1311 	if (0 != result) {
1312 		struct scsi_sense_hdr sshdr;
1313 
1314 		srp->header.status = 0xff & result;
1315 		srp->header.masked_status = status_byte(result);
1316 		srp->header.msg_status = msg_byte(result);
1317 		srp->header.host_status = host_byte(result);
1318 		srp->header.driver_status = driver_byte(result);
1319 		if ((sdp->sgdebug > 0) &&
1320 		    ((CHECK_CONDITION == srp->header.masked_status) ||
1321 		     (COMMAND_TERMINATED == srp->header.masked_status)))
1322 			__scsi_print_sense(sdp->device, __func__, sense,
1323 					   SCSI_SENSE_BUFFERSIZE);
1324 
1325 		/* Following if statement is a patch supplied by Eric Youngdale */
1326 		if (driver_byte(result) != 0
1327 		    && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1328 		    && !scsi_sense_is_deferred(&sshdr)
1329 		    && sshdr.sense_key == UNIT_ATTENTION
1330 		    && sdp->device->removable) {
1331 			/* Detected possible disc change. Set the bit - this */
1332 			/* may be used if there are filesystems using this device */
1333 			sdp->device->changed = 1;
1334 		}
1335 	}
1336 	/* Rely on write phase to clean out srp status values, so no "else" */
1337 
1338 	/*
1339 	 * Free the request as soon as it is complete so that its resources
1340 	 * can be reused without waiting for userspace to read() the
1341 	 * result.  But keep the associated bio (if any) around until
1342 	 * blk_rq_unmap_user() can be called from user context.
1343 	 */
1344 	srp->rq = NULL;
1345 	if (rq->cmd != rq->__cmd)
1346 		kfree(rq->cmd);
1347 	__blk_put_request(rq->q, rq);
1348 
1349 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1350 	if (unlikely(srp->orphan)) {
1351 		if (sfp->keep_orphan)
1352 			srp->sg_io_owned = 0;
1353 		else
1354 			done = 0;
1355 	}
1356 	srp->done = done;
1357 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1358 
1359 	if (likely(done)) {
1360 		/* Now wake up any sg_read() that is waiting for this
1361 		 * packet.
1362 		 */
1363 		wake_up_interruptible(&sfp->read_wait);
1364 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1365 		kref_put(&sfp->f_ref, sg_remove_sfp);
1366 	} else {
1367 		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1368 		schedule_work(&srp->ew.work);
1369 	}
1370 }
1371 
1372 static const struct file_operations sg_fops = {
1373 	.owner = THIS_MODULE,
1374 	.read = sg_read,
1375 	.write = sg_write,
1376 	.poll = sg_poll,
1377 	.unlocked_ioctl = sg_ioctl,
1378 #ifdef CONFIG_COMPAT
1379 	.compat_ioctl = sg_compat_ioctl,
1380 #endif
1381 	.open = sg_open,
1382 	.mmap = sg_mmap,
1383 	.release = sg_release,
1384 	.fasync = sg_fasync,
1385 	.llseek = no_llseek,
1386 };
1387 
1388 static struct class *sg_sysfs_class;
1389 
1390 static int sg_sysfs_valid = 0;
1391 
1392 static Sg_device *
1393 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1394 {
1395 	struct request_queue *q = scsidp->request_queue;
1396 	Sg_device *sdp;
1397 	unsigned long iflags;
1398 	int error;
1399 	u32 k;
1400 
1401 	sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1402 	if (!sdp) {
1403 		sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1404 			    "failure\n", __func__);
1405 		return ERR_PTR(-ENOMEM);
1406 	}
1407 
1408 	idr_preload(GFP_KERNEL);
1409 	write_lock_irqsave(&sg_index_lock, iflags);
1410 
1411 	error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1412 	if (error < 0) {
1413 		if (error == -ENOSPC) {
1414 			sdev_printk(KERN_WARNING, scsidp,
1415 				    "Unable to attach sg device type=%d, minor number exceeds %d\n",
1416 				    scsidp->type, SG_MAX_DEVS - 1);
1417 			error = -ENODEV;
1418 		} else {
1419 			sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1420 				    "allocation Sg_device failure: %d\n",
1421 				    __func__, error);
1422 		}
1423 		goto out_unlock;
1424 	}
1425 	k = error;
1426 
1427 	SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1428 					"sg_alloc: dev=%d \n", k));
1429 	sprintf(disk->disk_name, "sg%d", k);
1430 	disk->first_minor = k;
1431 	sdp->disk = disk;
1432 	sdp->device = scsidp;
1433 	mutex_init(&sdp->open_rel_lock);
1434 	INIT_LIST_HEAD(&sdp->sfds);
1435 	init_waitqueue_head(&sdp->open_wait);
1436 	atomic_set(&sdp->detaching, 0);
1437 	rwlock_init(&sdp->sfd_lock);
1438 	sdp->sg_tablesize = queue_max_segments(q);
1439 	sdp->index = k;
1440 	kref_init(&sdp->d_ref);
1441 	error = 0;
1442 
1443 out_unlock:
1444 	write_unlock_irqrestore(&sg_index_lock, iflags);
1445 	idr_preload_end();
1446 
1447 	if (error) {
1448 		kfree(sdp);
1449 		return ERR_PTR(error);
1450 	}
1451 	return sdp;
1452 }
1453 
1454 static int
1455 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1456 {
1457 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1458 	struct gendisk *disk;
1459 	Sg_device *sdp = NULL;
1460 	struct cdev * cdev = NULL;
1461 	int error;
1462 	unsigned long iflags;
1463 
1464 	disk = alloc_disk(1);
1465 	if (!disk) {
1466 		pr_warn("%s: alloc_disk failed\n", __func__);
1467 		return -ENOMEM;
1468 	}
1469 	disk->major = SCSI_GENERIC_MAJOR;
1470 
1471 	error = -ENOMEM;
1472 	cdev = cdev_alloc();
1473 	if (!cdev) {
1474 		pr_warn("%s: cdev_alloc failed\n", __func__);
1475 		goto out;
1476 	}
1477 	cdev->owner = THIS_MODULE;
1478 	cdev->ops = &sg_fops;
1479 
1480 	sdp = sg_alloc(disk, scsidp);
1481 	if (IS_ERR(sdp)) {
1482 		pr_warn("%s: sg_alloc failed\n", __func__);
1483 		error = PTR_ERR(sdp);
1484 		goto out;
1485 	}
1486 
1487 	error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1488 	if (error)
1489 		goto cdev_add_err;
1490 
1491 	sdp->cdev = cdev;
1492 	if (sg_sysfs_valid) {
1493 		struct device *sg_class_member;
1494 
1495 		sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1496 						MKDEV(SCSI_GENERIC_MAJOR,
1497 						      sdp->index),
1498 						sdp, "%s", disk->disk_name);
1499 		if (IS_ERR(sg_class_member)) {
1500 			pr_err("%s: device_create failed\n", __func__);
1501 			error = PTR_ERR(sg_class_member);
1502 			goto cdev_add_err;
1503 		}
1504 		error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1505 					  &sg_class_member->kobj, "generic");
1506 		if (error)
1507 			pr_err("%s: unable to make symlink 'generic' back "
1508 			       "to sg%d\n", __func__, sdp->index);
1509 	} else
1510 		pr_warn("%s: sg_sys Invalid\n", __func__);
1511 
1512 	sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1513 		    "type %d\n", sdp->index, scsidp->type);
1514 
1515 	dev_set_drvdata(cl_dev, sdp);
1516 
1517 	return 0;
1518 
1519 cdev_add_err:
1520 	write_lock_irqsave(&sg_index_lock, iflags);
1521 	idr_remove(&sg_index_idr, sdp->index);
1522 	write_unlock_irqrestore(&sg_index_lock, iflags);
1523 	kfree(sdp);
1524 
1525 out:
1526 	put_disk(disk);
1527 	if (cdev)
1528 		cdev_del(cdev);
1529 	return error;
1530 }
1531 
1532 static void
1533 sg_device_destroy(struct kref *kref)
1534 {
1535 	struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1536 	unsigned long flags;
1537 
1538 	/* CAUTION!  Note that the device can still be found via idr_find()
1539 	 * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1540 	 * any other cleanup.
1541 	 */
1542 
1543 	write_lock_irqsave(&sg_index_lock, flags);
1544 	idr_remove(&sg_index_idr, sdp->index);
1545 	write_unlock_irqrestore(&sg_index_lock, flags);
1546 
1547 	SCSI_LOG_TIMEOUT(3,
1548 		sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1549 
1550 	put_disk(sdp->disk);
1551 	kfree(sdp);
1552 }
1553 
1554 static void
1555 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1556 {
1557 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1558 	Sg_device *sdp = dev_get_drvdata(cl_dev);
1559 	unsigned long iflags;
1560 	Sg_fd *sfp;
1561 	int val;
1562 
1563 	if (!sdp)
1564 		return;
1565 	/* want sdp->detaching non-zero as soon as possible */
1566 	val = atomic_inc_return(&sdp->detaching);
1567 	if (val > 1)
1568 		return; /* only want to do following once per device */
1569 
1570 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1571 				      "%s\n", __func__));
1572 
1573 	read_lock_irqsave(&sdp->sfd_lock, iflags);
1574 	list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1575 		wake_up_interruptible_all(&sfp->read_wait);
1576 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1577 	}
1578 	wake_up_interruptible_all(&sdp->open_wait);
1579 	read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1580 
1581 	sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1582 	device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1583 	cdev_del(sdp->cdev);
1584 	sdp->cdev = NULL;
1585 
1586 	kref_put(&sdp->d_ref, sg_device_destroy);
1587 }
1588 
1589 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1590 module_param_named(def_reserved_size, def_reserved_size, int,
1591 		   S_IRUGO | S_IWUSR);
1592 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1593 
1594 MODULE_AUTHOR("Douglas Gilbert");
1595 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1596 MODULE_LICENSE("GPL");
1597 MODULE_VERSION(SG_VERSION_STR);
1598 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1599 
1600 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1601                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1602 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1603 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1604 
1605 static int __init
1606 init_sg(void)
1607 {
1608 	int rc;
1609 
1610 	if (scatter_elem_sz < PAGE_SIZE) {
1611 		scatter_elem_sz = PAGE_SIZE;
1612 		scatter_elem_sz_prev = scatter_elem_sz;
1613 	}
1614 	if (def_reserved_size >= 0)
1615 		sg_big_buff = def_reserved_size;
1616 	else
1617 		def_reserved_size = sg_big_buff;
1618 
1619 	rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1620 				    SG_MAX_DEVS, "sg");
1621 	if (rc)
1622 		return rc;
1623         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1624         if ( IS_ERR(sg_sysfs_class) ) {
1625 		rc = PTR_ERR(sg_sysfs_class);
1626 		goto err_out;
1627         }
1628 	sg_sysfs_valid = 1;
1629 	rc = scsi_register_interface(&sg_interface);
1630 	if (0 == rc) {
1631 #ifdef CONFIG_SCSI_PROC_FS
1632 		sg_proc_init();
1633 #endif				/* CONFIG_SCSI_PROC_FS */
1634 		return 0;
1635 	}
1636 	class_destroy(sg_sysfs_class);
1637 err_out:
1638 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1639 	return rc;
1640 }
1641 
1642 static void __exit
1643 exit_sg(void)
1644 {
1645 #ifdef CONFIG_SCSI_PROC_FS
1646 	sg_proc_cleanup();
1647 #endif				/* CONFIG_SCSI_PROC_FS */
1648 	scsi_unregister_interface(&sg_interface);
1649 	class_destroy(sg_sysfs_class);
1650 	sg_sysfs_valid = 0;
1651 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1652 				 SG_MAX_DEVS);
1653 	idr_destroy(&sg_index_idr);
1654 }
1655 
1656 static int
1657 sg_start_req(Sg_request *srp, unsigned char *cmd)
1658 {
1659 	int res;
1660 	struct request *rq;
1661 	Sg_fd *sfp = srp->parentfp;
1662 	sg_io_hdr_t *hp = &srp->header;
1663 	int dxfer_len = (int) hp->dxfer_len;
1664 	int dxfer_dir = hp->dxfer_direction;
1665 	unsigned int iov_count = hp->iovec_count;
1666 	Sg_scatter_hold *req_schp = &srp->data;
1667 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1668 	struct request_queue *q = sfp->parentdp->device->request_queue;
1669 	struct rq_map_data *md, map_data;
1670 	int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1671 	unsigned char *long_cmdp = NULL;
1672 
1673 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1674 				      "sg_start_req: dxfer_len=%d\n",
1675 				      dxfer_len));
1676 
1677 	if (hp->cmd_len > BLK_MAX_CDB) {
1678 		long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1679 		if (!long_cmdp)
1680 			return -ENOMEM;
1681 	}
1682 
1683 	/*
1684 	 * NOTE
1685 	 *
1686 	 * With scsi-mq enabled, there are a fixed number of preallocated
1687 	 * requests equal in number to shost->can_queue.  If all of the
1688 	 * preallocated requests are already in use, then using GFP_ATOMIC with
1689 	 * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1690 	 * will cause blk_get_request() to sleep until an active command
1691 	 * completes, freeing up a request.  Neither option is ideal, but
1692 	 * GFP_KERNEL is the better choice to prevent userspace from getting an
1693 	 * unexpected EWOULDBLOCK.
1694 	 *
1695 	 * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1696 	 * does not sleep except under memory pressure.
1697 	 */
1698 	rq = blk_get_request(q, rw, GFP_KERNEL);
1699 	if (IS_ERR(rq)) {
1700 		kfree(long_cmdp);
1701 		return PTR_ERR(rq);
1702 	}
1703 
1704 	blk_rq_set_block_pc(rq);
1705 
1706 	if (hp->cmd_len > BLK_MAX_CDB)
1707 		rq->cmd = long_cmdp;
1708 	memcpy(rq->cmd, cmd, hp->cmd_len);
1709 	rq->cmd_len = hp->cmd_len;
1710 
1711 	srp->rq = rq;
1712 	rq->end_io_data = srp;
1713 	rq->sense = srp->sense_b;
1714 	rq->retries = SG_DEFAULT_RETRIES;
1715 
1716 	if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1717 		return 0;
1718 
1719 	if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1720 	    dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1721 	    !sfp->parentdp->device->host->unchecked_isa_dma &&
1722 	    blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1723 		md = NULL;
1724 	else
1725 		md = &map_data;
1726 
1727 	if (md) {
1728 		if (!sg_res_in_use(sfp) && dxfer_len <= rsv_schp->bufflen)
1729 			sg_link_reserve(sfp, srp, dxfer_len);
1730 		else {
1731 			res = sg_build_indirect(req_schp, sfp, dxfer_len);
1732 			if (res)
1733 				return res;
1734 		}
1735 
1736 		md->pages = req_schp->pages;
1737 		md->page_order = req_schp->page_order;
1738 		md->nr_entries = req_schp->k_use_sg;
1739 		md->offset = 0;
1740 		md->null_mapped = hp->dxferp ? 0 : 1;
1741 		if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1742 			md->from_user = 1;
1743 		else
1744 			md->from_user = 0;
1745 	}
1746 
1747 	if (iov_count) {
1748 		struct iovec *iov = NULL;
1749 		struct iov_iter i;
1750 
1751 		res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1752 		if (res < 0)
1753 			return res;
1754 
1755 		iov_iter_truncate(&i, hp->dxfer_len);
1756 
1757 		res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1758 		kfree(iov);
1759 	} else
1760 		res = blk_rq_map_user(q, rq, md, hp->dxferp,
1761 				      hp->dxfer_len, GFP_ATOMIC);
1762 
1763 	if (!res) {
1764 		srp->bio = rq->bio;
1765 
1766 		if (!md) {
1767 			req_schp->dio_in_use = 1;
1768 			hp->info |= SG_INFO_DIRECT_IO;
1769 		}
1770 	}
1771 	return res;
1772 }
1773 
1774 static int
1775 sg_finish_rem_req(Sg_request *srp)
1776 {
1777 	int ret = 0;
1778 
1779 	Sg_fd *sfp = srp->parentfp;
1780 	Sg_scatter_hold *req_schp = &srp->data;
1781 
1782 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1783 				      "sg_finish_rem_req: res_used=%d\n",
1784 				      (int) srp->res_used));
1785 	if (srp->bio)
1786 		ret = blk_rq_unmap_user(srp->bio);
1787 
1788 	if (srp->rq) {
1789 		if (srp->rq->cmd != srp->rq->__cmd)
1790 			kfree(srp->rq->cmd);
1791 		blk_put_request(srp->rq);
1792 	}
1793 
1794 	if (srp->res_used)
1795 		sg_unlink_reserve(sfp, srp);
1796 	else
1797 		sg_remove_scat(sfp, req_schp);
1798 
1799 	sg_remove_request(sfp, srp);
1800 
1801 	return ret;
1802 }
1803 
1804 static int
1805 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1806 {
1807 	int sg_bufflen = tablesize * sizeof(struct page *);
1808 	gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1809 
1810 	schp->pages = kzalloc(sg_bufflen, gfp_flags);
1811 	if (!schp->pages)
1812 		return -ENOMEM;
1813 	schp->sglist_len = sg_bufflen;
1814 	return tablesize;	/* number of scat_gath elements allocated */
1815 }
1816 
1817 static int
1818 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1819 {
1820 	int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1821 	int sg_tablesize = sfp->parentdp->sg_tablesize;
1822 	int blk_size = buff_size, order;
1823 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1824 
1825 	if (blk_size < 0)
1826 		return -EFAULT;
1827 	if (0 == blk_size)
1828 		++blk_size;	/* don't know why */
1829 	/* round request up to next highest SG_SECTOR_SZ byte boundary */
1830 	blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1831 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1832 		"sg_build_indirect: buff_size=%d, blk_size=%d\n",
1833 		buff_size, blk_size));
1834 
1835 	/* N.B. ret_sz carried into this block ... */
1836 	mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1837 	if (mx_sc_elems < 0)
1838 		return mx_sc_elems;	/* most likely -ENOMEM */
1839 
1840 	num = scatter_elem_sz;
1841 	if (unlikely(num != scatter_elem_sz_prev)) {
1842 		if (num < PAGE_SIZE) {
1843 			scatter_elem_sz = PAGE_SIZE;
1844 			scatter_elem_sz_prev = PAGE_SIZE;
1845 		} else
1846 			scatter_elem_sz_prev = num;
1847 	}
1848 
1849 	if (sfp->low_dma)
1850 		gfp_mask |= GFP_DMA;
1851 
1852 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1853 		gfp_mask |= __GFP_ZERO;
1854 
1855 	order = get_order(num);
1856 retry:
1857 	ret_sz = 1 << (PAGE_SHIFT + order);
1858 
1859 	for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1860 	     k++, rem_sz -= ret_sz) {
1861 
1862 		num = (rem_sz > scatter_elem_sz_prev) ?
1863 			scatter_elem_sz_prev : rem_sz;
1864 
1865 		schp->pages[k] = alloc_pages(gfp_mask, order);
1866 		if (!schp->pages[k])
1867 			goto out;
1868 
1869 		if (num == scatter_elem_sz_prev) {
1870 			if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1871 				scatter_elem_sz = ret_sz;
1872 				scatter_elem_sz_prev = ret_sz;
1873 			}
1874 		}
1875 
1876 		SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1877 				 "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1878 				 k, num, ret_sz));
1879 	}		/* end of for loop */
1880 
1881 	schp->page_order = order;
1882 	schp->k_use_sg = k;
1883 	SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1884 			 "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1885 			 k, rem_sz));
1886 
1887 	schp->bufflen = blk_size;
1888 	if (rem_sz > 0)	/* must have failed */
1889 		return -ENOMEM;
1890 	return 0;
1891 out:
1892 	for (i = 0; i < k; i++)
1893 		__free_pages(schp->pages[i], order);
1894 
1895 	if (--order >= 0)
1896 		goto retry;
1897 
1898 	return -ENOMEM;
1899 }
1900 
1901 static void
1902 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1903 {
1904 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1905 			 "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1906 	if (schp->pages && schp->sglist_len > 0) {
1907 		if (!schp->dio_in_use) {
1908 			int k;
1909 
1910 			for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1911 				SCSI_LOG_TIMEOUT(5,
1912 					sg_printk(KERN_INFO, sfp->parentdp,
1913 					"sg_remove_scat: k=%d, pg=0x%p\n",
1914 					k, schp->pages[k]));
1915 				__free_pages(schp->pages[k], schp->page_order);
1916 			}
1917 
1918 			kfree(schp->pages);
1919 		}
1920 	}
1921 	memset(schp, 0, sizeof (*schp));
1922 }
1923 
1924 static int
1925 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1926 {
1927 	Sg_scatter_hold *schp = &srp->data;
1928 	int k, num;
1929 
1930 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1931 			 "sg_read_oxfer: num_read_xfer=%d\n",
1932 			 num_read_xfer));
1933 	if ((!outp) || (num_read_xfer <= 0))
1934 		return 0;
1935 
1936 	num = 1 << (PAGE_SHIFT + schp->page_order);
1937 	for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1938 		if (num > num_read_xfer) {
1939 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1940 					   num_read_xfer))
1941 				return -EFAULT;
1942 			break;
1943 		} else {
1944 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1945 					   num))
1946 				return -EFAULT;
1947 			num_read_xfer -= num;
1948 			if (num_read_xfer <= 0)
1949 				break;
1950 			outp += num;
1951 		}
1952 	}
1953 
1954 	return 0;
1955 }
1956 
1957 static void
1958 sg_build_reserve(Sg_fd * sfp, int req_size)
1959 {
1960 	Sg_scatter_hold *schp = &sfp->reserve;
1961 
1962 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1963 			 "sg_build_reserve: req_size=%d\n", req_size));
1964 	do {
1965 		if (req_size < PAGE_SIZE)
1966 			req_size = PAGE_SIZE;
1967 		if (0 == sg_build_indirect(schp, sfp, req_size))
1968 			return;
1969 		else
1970 			sg_remove_scat(sfp, schp);
1971 		req_size >>= 1;	/* divide by 2 */
1972 	} while (req_size > (PAGE_SIZE / 2));
1973 }
1974 
1975 static void
1976 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
1977 {
1978 	Sg_scatter_hold *req_schp = &srp->data;
1979 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1980 	int k, num, rem;
1981 
1982 	srp->res_used = 1;
1983 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1984 			 "sg_link_reserve: size=%d\n", size));
1985 	rem = size;
1986 
1987 	num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1988 	for (k = 0; k < rsv_schp->k_use_sg; k++) {
1989 		if (rem <= num) {
1990 			req_schp->k_use_sg = k + 1;
1991 			req_schp->sglist_len = rsv_schp->sglist_len;
1992 			req_schp->pages = rsv_schp->pages;
1993 
1994 			req_schp->bufflen = size;
1995 			req_schp->page_order = rsv_schp->page_order;
1996 			break;
1997 		} else
1998 			rem -= num;
1999 	}
2000 
2001 	if (k >= rsv_schp->k_use_sg)
2002 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2003 				 "sg_link_reserve: BAD size\n"));
2004 }
2005 
2006 static void
2007 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2008 {
2009 	Sg_scatter_hold *req_schp = &srp->data;
2010 
2011 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2012 				      "sg_unlink_reserve: req->k_use_sg=%d\n",
2013 				      (int) req_schp->k_use_sg));
2014 	req_schp->k_use_sg = 0;
2015 	req_schp->bufflen = 0;
2016 	req_schp->pages = NULL;
2017 	req_schp->page_order = 0;
2018 	req_schp->sglist_len = 0;
2019 	sfp->save_scat_len = 0;
2020 	srp->res_used = 0;
2021 }
2022 
2023 static Sg_request *
2024 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2025 {
2026 	Sg_request *resp;
2027 	unsigned long iflags;
2028 
2029 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2030 	for (resp = sfp->headrp; resp; resp = resp->nextrp) {
2031 		/* look for requests that are ready + not SG_IO owned */
2032 		if ((1 == resp->done) && (!resp->sg_io_owned) &&
2033 		    ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2034 			resp->done = 2;	/* guard against other readers */
2035 			break;
2036 		}
2037 	}
2038 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2039 	return resp;
2040 }
2041 
2042 /* always adds to end of list */
2043 static Sg_request *
2044 sg_add_request(Sg_fd * sfp)
2045 {
2046 	int k;
2047 	unsigned long iflags;
2048 	Sg_request *resp;
2049 	Sg_request *rp = sfp->req_arr;
2050 
2051 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2052 	resp = sfp->headrp;
2053 	if (!resp) {
2054 		memset(rp, 0, sizeof (Sg_request));
2055 		rp->parentfp = sfp;
2056 		resp = rp;
2057 		sfp->headrp = resp;
2058 	} else {
2059 		if (0 == sfp->cmd_q)
2060 			resp = NULL;	/* command queuing disallowed */
2061 		else {
2062 			for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2063 				if (!rp->parentfp)
2064 					break;
2065 			}
2066 			if (k < SG_MAX_QUEUE) {
2067 				memset(rp, 0, sizeof (Sg_request));
2068 				rp->parentfp = sfp;
2069 				while (resp->nextrp)
2070 					resp = resp->nextrp;
2071 				resp->nextrp = rp;
2072 				resp = rp;
2073 			} else
2074 				resp = NULL;
2075 		}
2076 	}
2077 	if (resp) {
2078 		resp->nextrp = NULL;
2079 		resp->header.duration = jiffies_to_msecs(jiffies);
2080 	}
2081 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2082 	return resp;
2083 }
2084 
2085 /* Return of 1 for found; 0 for not found */
2086 static int
2087 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2088 {
2089 	Sg_request *prev_rp;
2090 	Sg_request *rp;
2091 	unsigned long iflags;
2092 	int res = 0;
2093 
2094 	if ((!sfp) || (!srp) || (!sfp->headrp))
2095 		return res;
2096 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2097 	prev_rp = sfp->headrp;
2098 	if (srp == prev_rp) {
2099 		sfp->headrp = prev_rp->nextrp;
2100 		prev_rp->parentfp = NULL;
2101 		res = 1;
2102 	} else {
2103 		while ((rp = prev_rp->nextrp)) {
2104 			if (srp == rp) {
2105 				prev_rp->nextrp = rp->nextrp;
2106 				rp->parentfp = NULL;
2107 				res = 1;
2108 				break;
2109 			}
2110 			prev_rp = rp;
2111 		}
2112 	}
2113 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2114 	return res;
2115 }
2116 
2117 static Sg_fd *
2118 sg_add_sfp(Sg_device * sdp)
2119 {
2120 	Sg_fd *sfp;
2121 	unsigned long iflags;
2122 	int bufflen;
2123 
2124 	sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2125 	if (!sfp)
2126 		return ERR_PTR(-ENOMEM);
2127 
2128 	init_waitqueue_head(&sfp->read_wait);
2129 	rwlock_init(&sfp->rq_list_lock);
2130 
2131 	kref_init(&sfp->f_ref);
2132 	sfp->timeout = SG_DEFAULT_TIMEOUT;
2133 	sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2134 	sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2135 	sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2136 	    sdp->device->host->unchecked_isa_dma : 1;
2137 	sfp->cmd_q = SG_DEF_COMMAND_Q;
2138 	sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2139 	sfp->parentdp = sdp;
2140 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2141 	if (atomic_read(&sdp->detaching)) {
2142 		write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2143 		return ERR_PTR(-ENODEV);
2144 	}
2145 	list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2146 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2147 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2148 				      "sg_add_sfp: sfp=0x%p\n", sfp));
2149 	if (unlikely(sg_big_buff != def_reserved_size))
2150 		sg_big_buff = def_reserved_size;
2151 
2152 	bufflen = min_t(int, sg_big_buff,
2153 			max_sectors_bytes(sdp->device->request_queue));
2154 	sg_build_reserve(sfp, bufflen);
2155 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2156 				      "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2157 				      sfp->reserve.bufflen,
2158 				      sfp->reserve.k_use_sg));
2159 
2160 	kref_get(&sdp->d_ref);
2161 	__module_get(THIS_MODULE);
2162 	return sfp;
2163 }
2164 
2165 static void
2166 sg_remove_sfp_usercontext(struct work_struct *work)
2167 {
2168 	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2169 	struct sg_device *sdp = sfp->parentdp;
2170 
2171 	/* Cleanup any responses which were never read(). */
2172 	while (sfp->headrp)
2173 		sg_finish_rem_req(sfp->headrp);
2174 
2175 	if (sfp->reserve.bufflen > 0) {
2176 		SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2177 				"sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2178 				(int) sfp->reserve.bufflen,
2179 				(int) sfp->reserve.k_use_sg));
2180 		sg_remove_scat(sfp, &sfp->reserve);
2181 	}
2182 
2183 	SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2184 			"sg_remove_sfp: sfp=0x%p\n", sfp));
2185 	kfree(sfp);
2186 
2187 	scsi_device_put(sdp->device);
2188 	kref_put(&sdp->d_ref, sg_device_destroy);
2189 	module_put(THIS_MODULE);
2190 }
2191 
2192 static void
2193 sg_remove_sfp(struct kref *kref)
2194 {
2195 	struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2196 	struct sg_device *sdp = sfp->parentdp;
2197 	unsigned long iflags;
2198 
2199 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2200 	list_del(&sfp->sfd_siblings);
2201 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2202 
2203 	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2204 	schedule_work(&sfp->ew.work);
2205 }
2206 
2207 static int
2208 sg_res_in_use(Sg_fd * sfp)
2209 {
2210 	const Sg_request *srp;
2211 	unsigned long iflags;
2212 
2213 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
2214 	for (srp = sfp->headrp; srp; srp = srp->nextrp)
2215 		if (srp->res_used)
2216 			break;
2217 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2218 	return srp ? 1 : 0;
2219 }
2220 
2221 #ifdef CONFIG_SCSI_PROC_FS
2222 static int
2223 sg_idr_max_id(int id, void *p, void *data)
2224 {
2225 	int *k = data;
2226 
2227 	if (*k < id)
2228 		*k = id;
2229 
2230 	return 0;
2231 }
2232 
2233 static int
2234 sg_last_dev(void)
2235 {
2236 	int k = -1;
2237 	unsigned long iflags;
2238 
2239 	read_lock_irqsave(&sg_index_lock, iflags);
2240 	idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2241 	read_unlock_irqrestore(&sg_index_lock, iflags);
2242 	return k + 1;		/* origin 1 */
2243 }
2244 #endif
2245 
2246 /* must be called with sg_index_lock held */
2247 static Sg_device *sg_lookup_dev(int dev)
2248 {
2249 	return idr_find(&sg_index_idr, dev);
2250 }
2251 
2252 static Sg_device *
2253 sg_get_dev(int dev)
2254 {
2255 	struct sg_device *sdp;
2256 	unsigned long flags;
2257 
2258 	read_lock_irqsave(&sg_index_lock, flags);
2259 	sdp = sg_lookup_dev(dev);
2260 	if (!sdp)
2261 		sdp = ERR_PTR(-ENXIO);
2262 	else if (atomic_read(&sdp->detaching)) {
2263 		/* If sdp->detaching, then the refcount may already be 0, in
2264 		 * which case it would be a bug to do kref_get().
2265 		 */
2266 		sdp = ERR_PTR(-ENODEV);
2267 	} else
2268 		kref_get(&sdp->d_ref);
2269 	read_unlock_irqrestore(&sg_index_lock, flags);
2270 
2271 	return sdp;
2272 }
2273 
2274 #ifdef CONFIG_SCSI_PROC_FS
2275 
2276 static struct proc_dir_entry *sg_proc_sgp = NULL;
2277 
2278 static char sg_proc_sg_dirname[] = "scsi/sg";
2279 
2280 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2281 
2282 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2283 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2284 			          size_t count, loff_t *off);
2285 static const struct file_operations adio_fops = {
2286 	.owner = THIS_MODULE,
2287 	.open = sg_proc_single_open_adio,
2288 	.read = seq_read,
2289 	.llseek = seq_lseek,
2290 	.write = sg_proc_write_adio,
2291 	.release = single_release,
2292 };
2293 
2294 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2295 static ssize_t sg_proc_write_dressz(struct file *filp,
2296 		const char __user *buffer, size_t count, loff_t *off);
2297 static const struct file_operations dressz_fops = {
2298 	.owner = THIS_MODULE,
2299 	.open = sg_proc_single_open_dressz,
2300 	.read = seq_read,
2301 	.llseek = seq_lseek,
2302 	.write = sg_proc_write_dressz,
2303 	.release = single_release,
2304 };
2305 
2306 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2307 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2308 static const struct file_operations version_fops = {
2309 	.owner = THIS_MODULE,
2310 	.open = sg_proc_single_open_version,
2311 	.read = seq_read,
2312 	.llseek = seq_lseek,
2313 	.release = single_release,
2314 };
2315 
2316 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2317 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2318 static const struct file_operations devhdr_fops = {
2319 	.owner = THIS_MODULE,
2320 	.open = sg_proc_single_open_devhdr,
2321 	.read = seq_read,
2322 	.llseek = seq_lseek,
2323 	.release = single_release,
2324 };
2325 
2326 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2327 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2328 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2329 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2330 static void dev_seq_stop(struct seq_file *s, void *v);
2331 static const struct file_operations dev_fops = {
2332 	.owner = THIS_MODULE,
2333 	.open = sg_proc_open_dev,
2334 	.read = seq_read,
2335 	.llseek = seq_lseek,
2336 	.release = seq_release,
2337 };
2338 static const struct seq_operations dev_seq_ops = {
2339 	.start = dev_seq_start,
2340 	.next  = dev_seq_next,
2341 	.stop  = dev_seq_stop,
2342 	.show  = sg_proc_seq_show_dev,
2343 };
2344 
2345 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2346 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2347 static const struct file_operations devstrs_fops = {
2348 	.owner = THIS_MODULE,
2349 	.open = sg_proc_open_devstrs,
2350 	.read = seq_read,
2351 	.llseek = seq_lseek,
2352 	.release = seq_release,
2353 };
2354 static const struct seq_operations devstrs_seq_ops = {
2355 	.start = dev_seq_start,
2356 	.next  = dev_seq_next,
2357 	.stop  = dev_seq_stop,
2358 	.show  = sg_proc_seq_show_devstrs,
2359 };
2360 
2361 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2362 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2363 static const struct file_operations debug_fops = {
2364 	.owner = THIS_MODULE,
2365 	.open = sg_proc_open_debug,
2366 	.read = seq_read,
2367 	.llseek = seq_lseek,
2368 	.release = seq_release,
2369 };
2370 static const struct seq_operations debug_seq_ops = {
2371 	.start = dev_seq_start,
2372 	.next  = dev_seq_next,
2373 	.stop  = dev_seq_stop,
2374 	.show  = sg_proc_seq_show_debug,
2375 };
2376 
2377 
2378 struct sg_proc_leaf {
2379 	const char * name;
2380 	const struct file_operations * fops;
2381 };
2382 
2383 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2384 	{"allow_dio", &adio_fops},
2385 	{"debug", &debug_fops},
2386 	{"def_reserved_size", &dressz_fops},
2387 	{"device_hdr", &devhdr_fops},
2388 	{"devices", &dev_fops},
2389 	{"device_strs", &devstrs_fops},
2390 	{"version", &version_fops}
2391 };
2392 
2393 static int
2394 sg_proc_init(void)
2395 {
2396 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2397 	int k;
2398 
2399 	sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2400 	if (!sg_proc_sgp)
2401 		return 1;
2402 	for (k = 0; k < num_leaves; ++k) {
2403 		const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2404 		umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2405 		proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2406 	}
2407 	return 0;
2408 }
2409 
2410 static void
2411 sg_proc_cleanup(void)
2412 {
2413 	int k;
2414 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2415 
2416 	if (!sg_proc_sgp)
2417 		return;
2418 	for (k = 0; k < num_leaves; ++k)
2419 		remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2420 	remove_proc_entry(sg_proc_sg_dirname, NULL);
2421 }
2422 
2423 
2424 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2425 {
2426 	seq_printf(s, "%d\n", *((int *)s->private));
2427 	return 0;
2428 }
2429 
2430 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2431 {
2432 	return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2433 }
2434 
2435 static ssize_t
2436 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2437 		   size_t count, loff_t *off)
2438 {
2439 	int err;
2440 	unsigned long num;
2441 
2442 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2443 		return -EACCES;
2444 	err = kstrtoul_from_user(buffer, count, 0, &num);
2445 	if (err)
2446 		return err;
2447 	sg_allow_dio = num ? 1 : 0;
2448 	return count;
2449 }
2450 
2451 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2452 {
2453 	return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2454 }
2455 
2456 static ssize_t
2457 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2458 		     size_t count, loff_t *off)
2459 {
2460 	int err;
2461 	unsigned long k = ULONG_MAX;
2462 
2463 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2464 		return -EACCES;
2465 
2466 	err = kstrtoul_from_user(buffer, count, 0, &k);
2467 	if (err)
2468 		return err;
2469 	if (k <= 1048576) {	/* limit "big buff" to 1 MB */
2470 		sg_big_buff = k;
2471 		return count;
2472 	}
2473 	return -ERANGE;
2474 }
2475 
2476 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2477 {
2478 	seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2479 		   sg_version_date);
2480 	return 0;
2481 }
2482 
2483 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2484 {
2485 	return single_open(file, sg_proc_seq_show_version, NULL);
2486 }
2487 
2488 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2489 {
2490 	seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2491 	return 0;
2492 }
2493 
2494 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2495 {
2496 	return single_open(file, sg_proc_seq_show_devhdr, NULL);
2497 }
2498 
2499 struct sg_proc_deviter {
2500 	loff_t	index;
2501 	size_t	max;
2502 };
2503 
2504 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2505 {
2506 	struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2507 
2508 	s->private = it;
2509 	if (! it)
2510 		return NULL;
2511 
2512 	it->index = *pos;
2513 	it->max = sg_last_dev();
2514 	if (it->index >= it->max)
2515 		return NULL;
2516 	return it;
2517 }
2518 
2519 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2520 {
2521 	struct sg_proc_deviter * it = s->private;
2522 
2523 	*pos = ++it->index;
2524 	return (it->index < it->max) ? it : NULL;
2525 }
2526 
2527 static void dev_seq_stop(struct seq_file *s, void *v)
2528 {
2529 	kfree(s->private);
2530 }
2531 
2532 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2533 {
2534         return seq_open(file, &dev_seq_ops);
2535 }
2536 
2537 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2538 {
2539 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2540 	Sg_device *sdp;
2541 	struct scsi_device *scsidp;
2542 	unsigned long iflags;
2543 
2544 	read_lock_irqsave(&sg_index_lock, iflags);
2545 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2546 	if ((NULL == sdp) || (NULL == sdp->device) ||
2547 	    (atomic_read(&sdp->detaching)))
2548 		seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2549 	else {
2550 		scsidp = sdp->device;
2551 		seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2552 			      scsidp->host->host_no, scsidp->channel,
2553 			      scsidp->id, scsidp->lun, (int) scsidp->type,
2554 			      1,
2555 			      (int) scsidp->queue_depth,
2556 			      (int) atomic_read(&scsidp->device_busy),
2557 			      (int) scsi_device_online(scsidp));
2558 	}
2559 	read_unlock_irqrestore(&sg_index_lock, iflags);
2560 	return 0;
2561 }
2562 
2563 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2564 {
2565         return seq_open(file, &devstrs_seq_ops);
2566 }
2567 
2568 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2569 {
2570 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2571 	Sg_device *sdp;
2572 	struct scsi_device *scsidp;
2573 	unsigned long iflags;
2574 
2575 	read_lock_irqsave(&sg_index_lock, iflags);
2576 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2577 	scsidp = sdp ? sdp->device : NULL;
2578 	if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2579 		seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2580 			   scsidp->vendor, scsidp->model, scsidp->rev);
2581 	else
2582 		seq_puts(s, "<no active device>\n");
2583 	read_unlock_irqrestore(&sg_index_lock, iflags);
2584 	return 0;
2585 }
2586 
2587 /* must be called while holding sg_index_lock */
2588 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2589 {
2590 	int k, m, new_interface, blen, usg;
2591 	Sg_request *srp;
2592 	Sg_fd *fp;
2593 	const sg_io_hdr_t *hp;
2594 	const char * cp;
2595 	unsigned int ms;
2596 
2597 	k = 0;
2598 	list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2599 		k++;
2600 		read_lock(&fp->rq_list_lock); /* irqs already disabled */
2601 		seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2602 			   "(res)sgat=%d low_dma=%d\n", k,
2603 			   jiffies_to_msecs(fp->timeout),
2604 			   fp->reserve.bufflen,
2605 			   (int) fp->reserve.k_use_sg,
2606 			   (int) fp->low_dma);
2607 		seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2608 			   (int) fp->cmd_q, (int) fp->force_packid,
2609 			   (int) fp->keep_orphan);
2610 		for (m = 0, srp = fp->headrp;
2611 				srp != NULL;
2612 				++m, srp = srp->nextrp) {
2613 			hp = &srp->header;
2614 			new_interface = (hp->interface_id == '\0') ? 0 : 1;
2615 			if (srp->res_used) {
2616 				if (new_interface &&
2617 				    (SG_FLAG_MMAP_IO & hp->flags))
2618 					cp = "     mmap>> ";
2619 				else
2620 					cp = "     rb>> ";
2621 			} else {
2622 				if (SG_INFO_DIRECT_IO_MASK & hp->info)
2623 					cp = "     dio>> ";
2624 				else
2625 					cp = "     ";
2626 			}
2627 			seq_puts(s, cp);
2628 			blen = srp->data.bufflen;
2629 			usg = srp->data.k_use_sg;
2630 			seq_puts(s, srp->done ?
2631 				 ((1 == srp->done) ?  "rcv:" : "fin:")
2632 				  : "act:");
2633 			seq_printf(s, " id=%d blen=%d",
2634 				   srp->header.pack_id, blen);
2635 			if (srp->done)
2636 				seq_printf(s, " dur=%d", hp->duration);
2637 			else {
2638 				ms = jiffies_to_msecs(jiffies);
2639 				seq_printf(s, " t_o/elap=%d/%d",
2640 					(new_interface ? hp->timeout :
2641 						  jiffies_to_msecs(fp->timeout)),
2642 					(ms > hp->duration ? ms - hp->duration : 0));
2643 			}
2644 			seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2645 				   (int) srp->data.cmd_opcode);
2646 		}
2647 		if (0 == m)
2648 			seq_puts(s, "     No requests active\n");
2649 		read_unlock(&fp->rq_list_lock);
2650 	}
2651 }
2652 
2653 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2654 {
2655         return seq_open(file, &debug_seq_ops);
2656 }
2657 
2658 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2659 {
2660 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2661 	Sg_device *sdp;
2662 	unsigned long iflags;
2663 
2664 	if (it && (0 == it->index))
2665 		seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2666 			   (int)it->max, sg_big_buff);
2667 
2668 	read_lock_irqsave(&sg_index_lock, iflags);
2669 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2670 	if (NULL == sdp)
2671 		goto skip;
2672 	read_lock(&sdp->sfd_lock);
2673 	if (!list_empty(&sdp->sfds)) {
2674 		seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2675 		if (atomic_read(&sdp->detaching))
2676 			seq_puts(s, "detaching pending close ");
2677 		else if (sdp->device) {
2678 			struct scsi_device *scsidp = sdp->device;
2679 
2680 			seq_printf(s, "%d:%d:%d:%llu   em=%d",
2681 				   scsidp->host->host_no,
2682 				   scsidp->channel, scsidp->id,
2683 				   scsidp->lun,
2684 				   scsidp->host->hostt->emulated);
2685 		}
2686 		seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2687 			   sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2688 		sg_proc_debug_helper(s, sdp);
2689 	}
2690 	read_unlock(&sdp->sfd_lock);
2691 skip:
2692 	read_unlock_irqrestore(&sg_index_lock, iflags);
2693 	return 0;
2694 }
2695 
2696 #endif				/* CONFIG_SCSI_PROC_FS */
2697 
2698 module_init(init_sg);
2699 module_exit(exit_sg);
2700