Merge tags 'objtool-urgent-2021-06-28' and 'objtool-core-2021-06-28' of git://git...
[linux-2.6-microblaze.git] / drivers / scsi / sg.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  History:
4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
5  *           to allow user process control of SCSI devices.
6  *  Development Sponsored by Killy Corp. NY NY
7  *
8  * Original driver (sg.c):
9  *        Copyright (C) 1992 Lawrence Foard
10  * Version 2 and 3 extensions to driver:
11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
12  */
13
14 static int sg_version_num = 30536;      /* 2 digits for each component */
15 #define SG_VERSION_STR "3.5.36"
16
17 /*
18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
21  *        (otherwise the macros compile to empty statements).
22  *
23  */
24 #include <linux/module.h>
25
26 #include <linux/fs.h>
27 #include <linux/kernel.h>
28 #include <linux/sched.h>
29 #include <linux/string.h>
30 #include <linux/mm.h>
31 #include <linux/errno.h>
32 #include <linux/mtio.h>
33 #include <linux/ioctl.h>
34 #include <linux/slab.h>
35 #include <linux/fcntl.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/moduleparam.h>
39 #include <linux/cdev.h>
40 #include <linux/idr.h>
41 #include <linux/seq_file.h>
42 #include <linux/blkdev.h>
43 #include <linux/delay.h>
44 #include <linux/blktrace_api.h>
45 #include <linux/mutex.h>
46 #include <linux/atomic.h>
47 #include <linux/ratelimit.h>
48 #include <linux/uio.h>
49 #include <linux/cred.h> /* for sg_check_file_access() */
50
51 #include "scsi.h"
52 #include <scsi/scsi_dbg.h>
53 #include <scsi/scsi_host.h>
54 #include <scsi/scsi_driver.h>
55 #include <scsi/scsi_ioctl.h>
56 #include <scsi/sg.h>
57
58 #include "scsi_logging.h"
59
60 #ifdef CONFIG_SCSI_PROC_FS
61 #include <linux/proc_fs.h>
62 static char *sg_version_date = "20140603";
63
64 static int sg_proc_init(void);
65 #endif
66
67 #define SG_ALLOW_DIO_DEF 0
68
69 #define SG_MAX_DEVS 32768
70
71 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
72  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
73  * than 16 bytes are "variable length" whose length is a multiple of 4
74  */
75 #define SG_MAX_CDB_SIZE 252
76
77 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
78
79 int sg_big_buff = SG_DEF_RESERVED_SIZE;
80 /* N.B. This variable is readable and writeable via
81    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
82    of this size (or less if there is not enough memory) will be reserved
83    for use by this file descriptor. [Deprecated usage: this variable is also
84    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
85    the kernel (i.e. it is not a module).] */
86 static int def_reserved_size = -1;      /* picks up init parameter */
87 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
88
89 static int scatter_elem_sz = SG_SCATTER_SZ;
90 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
91
92 #define SG_SECTOR_SZ 512
93
94 static int sg_add_device(struct device *, struct class_interface *);
95 static void sg_remove_device(struct device *, struct class_interface *);
96
97 static DEFINE_IDR(sg_index_idr);
98 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
99                                                            file descriptor list for device */
100
101 static struct class_interface sg_interface = {
102         .add_dev        = sg_add_device,
103         .remove_dev     = sg_remove_device,
104 };
105
106 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
107         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
108         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
109         unsigned bufflen;       /* Size of (aggregate) data buffer */
110         struct page **pages;
111         int page_order;
112         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
113         unsigned char cmd_opcode; /* first byte of command */
114 } Sg_scatter_hold;
115
116 struct sg_device;               /* forward declarations */
117 struct sg_fd;
118
119 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
120         struct list_head entry; /* list entry */
121         struct sg_fd *parentfp; /* NULL -> not in use */
122         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
123         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
124         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
125         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
126         char orphan;            /* 1 -> drop on sight, 0 -> normal */
127         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
128         /* done protected by rq_list_lock */
129         char done;              /* 0->before bh, 1->before read, 2->read */
130         struct request *rq;
131         struct bio *bio;
132         struct execute_work ew;
133 } Sg_request;
134
135 typedef struct sg_fd {          /* holds the state of a file descriptor */
136         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
137         struct sg_device *parentdp;     /* owning device */
138         wait_queue_head_t read_wait;    /* queue read until command done */
139         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
140         struct mutex f_mutex;   /* protect against changes in this fd */
141         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
142         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
143         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
144         struct list_head rq_list; /* head of request list */
145         struct fasync_struct *async_qp; /* used by asynchronous notification */
146         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
147         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
148         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
149         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
150         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
151         char mmap_called;       /* 0 -> mmap() never called on this fd */
152         char res_in_use;        /* 1 -> 'reserve' array in use */
153         struct kref f_ref;
154         struct execute_work ew;
155 } Sg_fd;
156
157 typedef struct sg_device { /* holds the state of each scsi generic device */
158         struct scsi_device *device;
159         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
160         struct mutex open_rel_lock;     /* held when in open() or release() */
161         int sg_tablesize;       /* adapter's max scatter-gather table size */
162         u32 index;              /* device index number */
163         struct list_head sfds;
164         rwlock_t sfd_lock;      /* protect access to sfd list */
165         atomic_t detaching;     /* 0->device usable, 1->device detaching */
166         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
167         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
168         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
169         struct gendisk *disk;
170         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
171         struct kref d_ref;
172 } Sg_device;
173
174 /* tasklet or soft irq callback */
175 static void sg_rq_end_io(struct request *rq, blk_status_t status);
176 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
177 static int sg_finish_rem_req(Sg_request * srp);
178 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
179 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
180                            Sg_request * srp);
181 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
182                         const char __user *buf, size_t count, int blocking,
183                         int read_only, int sg_io_owned, Sg_request **o_srp);
184 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
185                            unsigned char *cmnd, int timeout, int blocking);
186 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
187 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
188 static void sg_build_reserve(Sg_fd * sfp, int req_size);
189 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
190 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
191 static Sg_fd *sg_add_sfp(Sg_device * sdp);
192 static void sg_remove_sfp(struct kref *);
193 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
194 static Sg_request *sg_add_request(Sg_fd * sfp);
195 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
196 static Sg_device *sg_get_dev(int dev);
197 static void sg_device_destroy(struct kref *kref);
198
199 #define SZ_SG_HEADER sizeof(struct sg_header)
200 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
201 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
202 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
203
204 #define sg_printk(prefix, sdp, fmt, a...) \
205         sdev_prefix_printk(prefix, (sdp)->device,               \
206                            (sdp)->disk->disk_name, fmt, ##a)
207
208 /*
209  * The SCSI interfaces that use read() and write() as an asynchronous variant of
210  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
211  * to trigger read() and write() calls from various contexts with elevated
212  * privileges. This can lead to kernel memory corruption (e.g. if these
213  * interfaces are called through splice()) and privilege escalation inside
214  * userspace (e.g. if a process with access to such a device passes a file
215  * descriptor to a SUID binary as stdin/stdout/stderr).
216  *
217  * This function provides protection for the legacy API by restricting the
218  * calling context.
219  */
220 static int sg_check_file_access(struct file *filp, const char *caller)
221 {
222         if (filp->f_cred != current_real_cred()) {
223                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
224                         caller, task_tgid_vnr(current), current->comm);
225                 return -EPERM;
226         }
227         if (uaccess_kernel()) {
228                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
229                         caller, task_tgid_vnr(current), current->comm);
230                 return -EACCES;
231         }
232         return 0;
233 }
234
235 static int sg_allow_access(struct file *filp, unsigned char *cmd)
236 {
237         struct sg_fd *sfp = filp->private_data;
238
239         if (sfp->parentdp->device->type == TYPE_SCANNER)
240                 return 0;
241
242         return blk_verify_command(cmd, filp->f_mode);
243 }
244
245 static int
246 open_wait(Sg_device *sdp, int flags)
247 {
248         int retval = 0;
249
250         if (flags & O_EXCL) {
251                 while (sdp->open_cnt > 0) {
252                         mutex_unlock(&sdp->open_rel_lock);
253                         retval = wait_event_interruptible(sdp->open_wait,
254                                         (atomic_read(&sdp->detaching) ||
255                                          !sdp->open_cnt));
256                         mutex_lock(&sdp->open_rel_lock);
257
258                         if (retval) /* -ERESTARTSYS */
259                                 return retval;
260                         if (atomic_read(&sdp->detaching))
261                                 return -ENODEV;
262                 }
263         } else {
264                 while (sdp->exclude) {
265                         mutex_unlock(&sdp->open_rel_lock);
266                         retval = wait_event_interruptible(sdp->open_wait,
267                                         (atomic_read(&sdp->detaching) ||
268                                          !sdp->exclude));
269                         mutex_lock(&sdp->open_rel_lock);
270
271                         if (retval) /* -ERESTARTSYS */
272                                 return retval;
273                         if (atomic_read(&sdp->detaching))
274                                 return -ENODEV;
275                 }
276         }
277
278         return retval;
279 }
280
281 /* Returns 0 on success, else a negated errno value */
282 static int
283 sg_open(struct inode *inode, struct file *filp)
284 {
285         int dev = iminor(inode);
286         int flags = filp->f_flags;
287         struct request_queue *q;
288         Sg_device *sdp;
289         Sg_fd *sfp;
290         int retval;
291
292         nonseekable_open(inode, filp);
293         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
294                 return -EPERM; /* Can't lock it with read only access */
295         sdp = sg_get_dev(dev);
296         if (IS_ERR(sdp))
297                 return PTR_ERR(sdp);
298
299         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
300                                       "sg_open: flags=0x%x\n", flags));
301
302         /* This driver's module count bumped by fops_get in <linux/fs.h> */
303         /* Prevent the device driver from vanishing while we sleep */
304         retval = scsi_device_get(sdp->device);
305         if (retval)
306                 goto sg_put;
307
308         retval = scsi_autopm_get_device(sdp->device);
309         if (retval)
310                 goto sdp_put;
311
312         /* scsi_block_when_processing_errors() may block so bypass
313          * check if O_NONBLOCK. Permits SCSI commands to be issued
314          * during error recovery. Tread carefully. */
315         if (!((flags & O_NONBLOCK) ||
316               scsi_block_when_processing_errors(sdp->device))) {
317                 retval = -ENXIO;
318                 /* we are in error recovery for this device */
319                 goto error_out;
320         }
321
322         mutex_lock(&sdp->open_rel_lock);
323         if (flags & O_NONBLOCK) {
324                 if (flags & O_EXCL) {
325                         if (sdp->open_cnt > 0) {
326                                 retval = -EBUSY;
327                                 goto error_mutex_locked;
328                         }
329                 } else {
330                         if (sdp->exclude) {
331                                 retval = -EBUSY;
332                                 goto error_mutex_locked;
333                         }
334                 }
335         } else {
336                 retval = open_wait(sdp, flags);
337                 if (retval) /* -ERESTARTSYS or -ENODEV */
338                         goto error_mutex_locked;
339         }
340
341         /* N.B. at this point we are holding the open_rel_lock */
342         if (flags & O_EXCL)
343                 sdp->exclude = true;
344
345         if (sdp->open_cnt < 1) {  /* no existing opens */
346                 sdp->sgdebug = 0;
347                 q = sdp->device->request_queue;
348                 sdp->sg_tablesize = queue_max_segments(q);
349         }
350         sfp = sg_add_sfp(sdp);
351         if (IS_ERR(sfp)) {
352                 retval = PTR_ERR(sfp);
353                 goto out_undo;
354         }
355
356         filp->private_data = sfp;
357         sdp->open_cnt++;
358         mutex_unlock(&sdp->open_rel_lock);
359
360         retval = 0;
361 sg_put:
362         kref_put(&sdp->d_ref, sg_device_destroy);
363         return retval;
364
365 out_undo:
366         if (flags & O_EXCL) {
367                 sdp->exclude = false;   /* undo if error */
368                 wake_up_interruptible(&sdp->open_wait);
369         }
370 error_mutex_locked:
371         mutex_unlock(&sdp->open_rel_lock);
372 error_out:
373         scsi_autopm_put_device(sdp->device);
374 sdp_put:
375         scsi_device_put(sdp->device);
376         goto sg_put;
377 }
378
379 /* Release resources associated with a successful sg_open()
380  * Returns 0 on success, else a negated errno value */
381 static int
382 sg_release(struct inode *inode, struct file *filp)
383 {
384         Sg_device *sdp;
385         Sg_fd *sfp;
386
387         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
388                 return -ENXIO;
389         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
390
391         mutex_lock(&sdp->open_rel_lock);
392         scsi_autopm_put_device(sdp->device);
393         kref_put(&sfp->f_ref, sg_remove_sfp);
394         sdp->open_cnt--;
395
396         /* possibly many open()s waiting on exlude clearing, start many;
397          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
398         if (sdp->exclude) {
399                 sdp->exclude = false;
400                 wake_up_interruptible_all(&sdp->open_wait);
401         } else if (0 == sdp->open_cnt) {
402                 wake_up_interruptible(&sdp->open_wait);
403         }
404         mutex_unlock(&sdp->open_rel_lock);
405         return 0;
406 }
407
408 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
409 {
410         struct sg_header __user *old_hdr = buf;
411         int reply_len;
412
413         if (count >= SZ_SG_HEADER) {
414                 /* negative reply_len means v3 format, otherwise v1/v2 */
415                 if (get_user(reply_len, &old_hdr->reply_len))
416                         return -EFAULT;
417
418                 if (reply_len >= 0)
419                         return get_user(*pack_id, &old_hdr->pack_id);
420
421                 if (in_compat_syscall() &&
422                     count >= sizeof(struct compat_sg_io_hdr)) {
423                         struct compat_sg_io_hdr __user *hp = buf;
424
425                         return get_user(*pack_id, &hp->pack_id);
426                 }
427
428                 if (count >= sizeof(struct sg_io_hdr)) {
429                         struct sg_io_hdr __user *hp = buf;
430
431                         return get_user(*pack_id, &hp->pack_id);
432                 }
433         }
434
435         /* no valid header was passed, so ignore the pack_id */
436         *pack_id = -1;
437         return 0;
438 }
439
440 static ssize_t
441 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
442 {
443         Sg_device *sdp;
444         Sg_fd *sfp;
445         Sg_request *srp;
446         int req_pack_id = -1;
447         sg_io_hdr_t *hp;
448         struct sg_header *old_hdr;
449         int retval;
450
451         /*
452          * This could cause a response to be stranded. Close the associated
453          * file descriptor to free up any resources being held.
454          */
455         retval = sg_check_file_access(filp, __func__);
456         if (retval)
457                 return retval;
458
459         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
460                 return -ENXIO;
461         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
462                                       "sg_read: count=%d\n", (int) count));
463
464         if (sfp->force_packid)
465                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
466         if (retval)
467                 return retval;
468
469         srp = sg_get_rq_mark(sfp, req_pack_id);
470         if (!srp) {             /* now wait on packet to arrive */
471                 if (atomic_read(&sdp->detaching))
472                         return -ENODEV;
473                 if (filp->f_flags & O_NONBLOCK)
474                         return -EAGAIN;
475                 retval = wait_event_interruptible(sfp->read_wait,
476                         (atomic_read(&sdp->detaching) ||
477                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
478                 if (atomic_read(&sdp->detaching))
479                         return -ENODEV;
480                 if (retval)
481                         /* -ERESTARTSYS as signal hit process */
482                         return retval;
483         }
484         if (srp->header.interface_id != '\0')
485                 return sg_new_read(sfp, buf, count, srp);
486
487         hp = &srp->header;
488         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
489         if (!old_hdr)
490                 return -ENOMEM;
491
492         old_hdr->reply_len = (int) hp->timeout;
493         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
494         old_hdr->pack_id = hp->pack_id;
495         old_hdr->twelve_byte =
496             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
497         old_hdr->target_status = hp->masked_status;
498         old_hdr->host_status = hp->host_status;
499         old_hdr->driver_status = hp->driver_status;
500         if ((CHECK_CONDITION & hp->masked_status) ||
501             (DRIVER_SENSE & hp->driver_status))
502                 memcpy(old_hdr->sense_buffer, srp->sense_b,
503                        sizeof (old_hdr->sense_buffer));
504         switch (hp->host_status) {
505         /* This setup of 'result' is for backward compatibility and is best
506            ignored by the user who should use target, host + driver status */
507         case DID_OK:
508         case DID_PASSTHROUGH:
509         case DID_SOFT_ERROR:
510                 old_hdr->result = 0;
511                 break;
512         case DID_NO_CONNECT:
513         case DID_BUS_BUSY:
514         case DID_TIME_OUT:
515                 old_hdr->result = EBUSY;
516                 break;
517         case DID_BAD_TARGET:
518         case DID_ABORT:
519         case DID_PARITY:
520         case DID_RESET:
521         case DID_BAD_INTR:
522                 old_hdr->result = EIO;
523                 break;
524         case DID_ERROR:
525                 old_hdr->result = (srp->sense_b[0] == 0 && 
526                                   hp->masked_status == GOOD) ? 0 : EIO;
527                 break;
528         default:
529                 old_hdr->result = EIO;
530                 break;
531         }
532
533         /* Now copy the result back to the user buffer.  */
534         if (count >= SZ_SG_HEADER) {
535                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
536                         retval = -EFAULT;
537                         goto free_old_hdr;
538                 }
539                 buf += SZ_SG_HEADER;
540                 if (count > old_hdr->reply_len)
541                         count = old_hdr->reply_len;
542                 if (count > SZ_SG_HEADER) {
543                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
544                                 retval = -EFAULT;
545                                 goto free_old_hdr;
546                         }
547                 }
548         } else
549                 count = (old_hdr->result == 0) ? 0 : -EIO;
550         sg_finish_rem_req(srp);
551         sg_remove_request(sfp, srp);
552         retval = count;
553 free_old_hdr:
554         kfree(old_hdr);
555         return retval;
556 }
557
558 static ssize_t
559 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
560 {
561         sg_io_hdr_t *hp = &srp->header;
562         int err = 0, err2;
563         int len;
564
565         if (in_compat_syscall()) {
566                 if (count < sizeof(struct compat_sg_io_hdr)) {
567                         err = -EINVAL;
568                         goto err_out;
569                 }
570         } else if (count < SZ_SG_IO_HDR) {
571                 err = -EINVAL;
572                 goto err_out;
573         }
574         hp->sb_len_wr = 0;
575         if ((hp->mx_sb_len > 0) && hp->sbp) {
576                 if ((CHECK_CONDITION & hp->masked_status) ||
577                     (DRIVER_SENSE & hp->driver_status)) {
578                         int sb_len = SCSI_SENSE_BUFFERSIZE;
579                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
580                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
581                         len = (len > sb_len) ? sb_len : len;
582                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
583                                 err = -EFAULT;
584                                 goto err_out;
585                         }
586                         hp->sb_len_wr = len;
587                 }
588         }
589         if (hp->masked_status || hp->host_status || hp->driver_status)
590                 hp->info |= SG_INFO_CHECK;
591         err = put_sg_io_hdr(hp, buf);
592 err_out:
593         err2 = sg_finish_rem_req(srp);
594         sg_remove_request(sfp, srp);
595         return err ? : err2 ? : count;
596 }
597
598 static ssize_t
599 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
600 {
601         int mxsize, cmd_size, k;
602         int input_size, blocking;
603         unsigned char opcode;
604         Sg_device *sdp;
605         Sg_fd *sfp;
606         Sg_request *srp;
607         struct sg_header old_hdr;
608         sg_io_hdr_t *hp;
609         unsigned char cmnd[SG_MAX_CDB_SIZE];
610         int retval;
611
612         retval = sg_check_file_access(filp, __func__);
613         if (retval)
614                 return retval;
615
616         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
617                 return -ENXIO;
618         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
619                                       "sg_write: count=%d\n", (int) count));
620         if (atomic_read(&sdp->detaching))
621                 return -ENODEV;
622         if (!((filp->f_flags & O_NONBLOCK) ||
623               scsi_block_when_processing_errors(sdp->device)))
624                 return -ENXIO;
625
626         if (count < SZ_SG_HEADER)
627                 return -EIO;
628         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
629                 return -EFAULT;
630         blocking = !(filp->f_flags & O_NONBLOCK);
631         if (old_hdr.reply_len < 0)
632                 return sg_new_write(sfp, filp, buf, count,
633                                     blocking, 0, 0, NULL);
634         if (count < (SZ_SG_HEADER + 6))
635                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
636
637         buf += SZ_SG_HEADER;
638         if (get_user(opcode, buf))
639                 return -EFAULT;
640
641         if (!(srp = sg_add_request(sfp))) {
642                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
643                                               "sg_write: queue full\n"));
644                 return -EDOM;
645         }
646         mutex_lock(&sfp->f_mutex);
647         if (sfp->next_cmd_len > 0) {
648                 cmd_size = sfp->next_cmd_len;
649                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
650         } else {
651                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
652                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
653                         cmd_size = 12;
654         }
655         mutex_unlock(&sfp->f_mutex);
656         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
657                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
658 /* Determine buffer size.  */
659         input_size = count - cmd_size;
660         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
661         mxsize -= SZ_SG_HEADER;
662         input_size -= SZ_SG_HEADER;
663         if (input_size < 0) {
664                 sg_remove_request(sfp, srp);
665                 return -EIO;    /* User did not pass enough bytes for this command. */
666         }
667         hp = &srp->header;
668         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
669         hp->cmd_len = (unsigned char) cmd_size;
670         hp->iovec_count = 0;
671         hp->mx_sb_len = 0;
672         if (input_size > 0)
673                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
674                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
675         else
676                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
677         hp->dxfer_len = mxsize;
678         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
679             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
680                 hp->dxferp = (char __user *)buf + cmd_size;
681         else
682                 hp->dxferp = NULL;
683         hp->sbp = NULL;
684         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
685         hp->flags = input_size; /* structure abuse ... */
686         hp->pack_id = old_hdr.pack_id;
687         hp->usr_ptr = NULL;
688         if (copy_from_user(cmnd, buf, cmd_size)) {
689                 sg_remove_request(sfp, srp);
690                 return -EFAULT;
691         }
692         /*
693          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
694          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
695          * is a non-zero input_size, so emit a warning.
696          */
697         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
698                 printk_ratelimited(KERN_WARNING
699                                    "sg_write: data in/out %d/%d bytes "
700                                    "for SCSI command 0x%x-- guessing "
701                                    "data in;\n   program %s not setting "
702                                    "count and/or reply_len properly\n",
703                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
704                                    input_size, (unsigned int) cmnd[0],
705                                    current->comm);
706         }
707         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
708         return (k < 0) ? k : count;
709 }
710
711 static ssize_t
712 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
713                  size_t count, int blocking, int read_only, int sg_io_owned,
714                  Sg_request **o_srp)
715 {
716         int k;
717         Sg_request *srp;
718         sg_io_hdr_t *hp;
719         unsigned char cmnd[SG_MAX_CDB_SIZE];
720         int timeout;
721         unsigned long ul_timeout;
722
723         if (count < SZ_SG_IO_HDR)
724                 return -EINVAL;
725
726         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
727         if (!(srp = sg_add_request(sfp))) {
728                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
729                                               "sg_new_write: queue full\n"));
730                 return -EDOM;
731         }
732         srp->sg_io_owned = sg_io_owned;
733         hp = &srp->header;
734         if (get_sg_io_hdr(hp, buf)) {
735                 sg_remove_request(sfp, srp);
736                 return -EFAULT;
737         }
738         if (hp->interface_id != 'S') {
739                 sg_remove_request(sfp, srp);
740                 return -ENOSYS;
741         }
742         if (hp->flags & SG_FLAG_MMAP_IO) {
743                 if (hp->dxfer_len > sfp->reserve.bufflen) {
744                         sg_remove_request(sfp, srp);
745                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
746                 }
747                 if (hp->flags & SG_FLAG_DIRECT_IO) {
748                         sg_remove_request(sfp, srp);
749                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
750                 }
751                 if (sfp->res_in_use) {
752                         sg_remove_request(sfp, srp);
753                         return -EBUSY;  /* reserve buffer already being used */
754                 }
755         }
756         ul_timeout = msecs_to_jiffies(srp->header.timeout);
757         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
758         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
759                 sg_remove_request(sfp, srp);
760                 return -EMSGSIZE;
761         }
762         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
763                 sg_remove_request(sfp, srp);
764                 return -EFAULT;
765         }
766         if (read_only && sg_allow_access(file, cmnd)) {
767                 sg_remove_request(sfp, srp);
768                 return -EPERM;
769         }
770         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
771         if (k < 0)
772                 return k;
773         if (o_srp)
774                 *o_srp = srp;
775         return count;
776 }
777
778 static int
779 sg_common_write(Sg_fd * sfp, Sg_request * srp,
780                 unsigned char *cmnd, int timeout, int blocking)
781 {
782         int k, at_head;
783         Sg_device *sdp = sfp->parentdp;
784         sg_io_hdr_t *hp = &srp->header;
785
786         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
787         hp->status = 0;
788         hp->masked_status = 0;
789         hp->msg_status = 0;
790         hp->info = 0;
791         hp->host_status = 0;
792         hp->driver_status = 0;
793         hp->resid = 0;
794         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
795                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
796                         (int) cmnd[0], (int) hp->cmd_len));
797
798         if (hp->dxfer_len >= SZ_256M) {
799                 sg_remove_request(sfp, srp);
800                 return -EINVAL;
801         }
802
803         k = sg_start_req(srp, cmnd);
804         if (k) {
805                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
806                         "sg_common_write: start_req err=%d\n", k));
807                 sg_finish_rem_req(srp);
808                 sg_remove_request(sfp, srp);
809                 return k;       /* probably out of space --> ENOMEM */
810         }
811         if (atomic_read(&sdp->detaching)) {
812                 if (srp->bio) {
813                         scsi_req_free_cmd(scsi_req(srp->rq));
814                         blk_put_request(srp->rq);
815                         srp->rq = NULL;
816                 }
817
818                 sg_finish_rem_req(srp);
819                 sg_remove_request(sfp, srp);
820                 return -ENODEV;
821         }
822
823         hp->duration = jiffies_to_msecs(jiffies);
824         if (hp->interface_id != '\0' && /* v3 (or later) interface */
825             (SG_FLAG_Q_AT_TAIL & hp->flags))
826                 at_head = 0;
827         else
828                 at_head = 1;
829
830         srp->rq->timeout = timeout;
831         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
832         blk_execute_rq_nowait(sdp->disk, srp->rq, at_head, sg_rq_end_io);
833         return 0;
834 }
835
836 static int srp_done(Sg_fd *sfp, Sg_request *srp)
837 {
838         unsigned long flags;
839         int ret;
840
841         read_lock_irqsave(&sfp->rq_list_lock, flags);
842         ret = srp->done;
843         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
844         return ret;
845 }
846
847 static int max_sectors_bytes(struct request_queue *q)
848 {
849         unsigned int max_sectors = queue_max_sectors(q);
850
851         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
852
853         return max_sectors << 9;
854 }
855
856 static void
857 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
858 {
859         Sg_request *srp;
860         int val;
861         unsigned int ms;
862
863         val = 0;
864         list_for_each_entry(srp, &sfp->rq_list, entry) {
865                 if (val >= SG_MAX_QUEUE)
866                         break;
867                 rinfo[val].req_state = srp->done + 1;
868                 rinfo[val].problem =
869                         srp->header.masked_status &
870                         srp->header.host_status &
871                         srp->header.driver_status;
872                 if (srp->done)
873                         rinfo[val].duration =
874                                 srp->header.duration;
875                 else {
876                         ms = jiffies_to_msecs(jiffies);
877                         rinfo[val].duration =
878                                 (ms > srp->header.duration) ?
879                                 (ms - srp->header.duration) : 0;
880                 }
881                 rinfo[val].orphan = srp->orphan;
882                 rinfo[val].sg_io_owned = srp->sg_io_owned;
883                 rinfo[val].pack_id = srp->header.pack_id;
884                 rinfo[val].usr_ptr = srp->header.usr_ptr;
885                 val++;
886         }
887 }
888
889 #ifdef CONFIG_COMPAT
890 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
891         char req_state;
892         char orphan;
893         char sg_io_owned;
894         char problem;
895         int pack_id;
896         compat_uptr_t usr_ptr;
897         unsigned int duration;
898         int unused;
899 };
900
901 static int put_compat_request_table(struct compat_sg_req_info __user *o,
902                                     struct sg_req_info *rinfo)
903 {
904         int i;
905         for (i = 0; i < SG_MAX_QUEUE; i++) {
906                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
907                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
908                     put_user(rinfo[i].duration, &o[i].duration) ||
909                     put_user(rinfo[i].unused, &o[i].unused))
910                         return -EFAULT;
911         }
912         return 0;
913 }
914 #endif
915
916 static long
917 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
918                 unsigned int cmd_in, void __user *p)
919 {
920         int __user *ip = p;
921         int result, val, read_only;
922         Sg_request *srp;
923         unsigned long iflags;
924
925         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
926                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
927         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
928
929         switch (cmd_in) {
930         case SG_IO:
931                 if (atomic_read(&sdp->detaching))
932                         return -ENODEV;
933                 if (!scsi_block_when_processing_errors(sdp->device))
934                         return -ENXIO;
935                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
936                                  1, read_only, 1, &srp);
937                 if (result < 0)
938                         return result;
939                 result = wait_event_interruptible(sfp->read_wait,
940                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
941                 if (atomic_read(&sdp->detaching))
942                         return -ENODEV;
943                 write_lock_irq(&sfp->rq_list_lock);
944                 if (srp->done) {
945                         srp->done = 2;
946                         write_unlock_irq(&sfp->rq_list_lock);
947                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
948                         return (result < 0) ? result : 0;
949                 }
950                 srp->orphan = 1;
951                 write_unlock_irq(&sfp->rq_list_lock);
952                 return result;  /* -ERESTARTSYS because signal hit process */
953         case SG_SET_TIMEOUT:
954                 result = get_user(val, ip);
955                 if (result)
956                         return result;
957                 if (val < 0)
958                         return -EIO;
959                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
960                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
961                                     INT_MAX);
962                 sfp->timeout_user = val;
963                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
964
965                 return 0;
966         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
967                                 /* strange ..., for backward compatibility */
968                 return sfp->timeout_user;
969         case SG_SET_FORCE_LOW_DMA:
970                 /*
971                  * N.B. This ioctl never worked properly, but failed to
972                  * return an error value. So returning '0' to keep compability
973                  * with legacy applications.
974                  */
975                 return 0;
976         case SG_GET_LOW_DMA:
977                 return put_user(0, ip);
978         case SG_GET_SCSI_ID:
979                 {
980                         sg_scsi_id_t v;
981
982                         if (atomic_read(&sdp->detaching))
983                                 return -ENODEV;
984                         memset(&v, 0, sizeof(v));
985                         v.host_no = sdp->device->host->host_no;
986                         v.channel = sdp->device->channel;
987                         v.scsi_id = sdp->device->id;
988                         v.lun = sdp->device->lun;
989                         v.scsi_type = sdp->device->type;
990                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
991                         v.d_queue_depth = sdp->device->queue_depth;
992                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
993                                 return -EFAULT;
994                         return 0;
995                 }
996         case SG_SET_FORCE_PACK_ID:
997                 result = get_user(val, ip);
998                 if (result)
999                         return result;
1000                 sfp->force_packid = val ? 1 : 0;
1001                 return 0;
1002         case SG_GET_PACK_ID:
1003                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1004                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1005                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1006                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1007                                                        iflags);
1008                                 return put_user(srp->header.pack_id, ip);
1009                         }
1010                 }
1011                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1012                 return put_user(-1, ip);
1013         case SG_GET_NUM_WAITING:
1014                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1015                 val = 0;
1016                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1017                         if ((1 == srp->done) && (!srp->sg_io_owned))
1018                                 ++val;
1019                 }
1020                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1021                 return put_user(val, ip);
1022         case SG_GET_SG_TABLESIZE:
1023                 return put_user(sdp->sg_tablesize, ip);
1024         case SG_SET_RESERVED_SIZE:
1025                 result = get_user(val, ip);
1026                 if (result)
1027                         return result;
1028                 if (val < 0)
1029                         return -EINVAL;
1030                 val = min_t(int, val,
1031                             max_sectors_bytes(sdp->device->request_queue));
1032                 mutex_lock(&sfp->f_mutex);
1033                 if (val != sfp->reserve.bufflen) {
1034                         if (sfp->mmap_called ||
1035                             sfp->res_in_use) {
1036                                 mutex_unlock(&sfp->f_mutex);
1037                                 return -EBUSY;
1038                         }
1039
1040                         sg_remove_scat(sfp, &sfp->reserve);
1041                         sg_build_reserve(sfp, val);
1042                 }
1043                 mutex_unlock(&sfp->f_mutex);
1044                 return 0;
1045         case SG_GET_RESERVED_SIZE:
1046                 val = min_t(int, sfp->reserve.bufflen,
1047                             max_sectors_bytes(sdp->device->request_queue));
1048                 return put_user(val, ip);
1049         case SG_SET_COMMAND_Q:
1050                 result = get_user(val, ip);
1051                 if (result)
1052                         return result;
1053                 sfp->cmd_q = val ? 1 : 0;
1054                 return 0;
1055         case SG_GET_COMMAND_Q:
1056                 return put_user((int) sfp->cmd_q, ip);
1057         case SG_SET_KEEP_ORPHAN:
1058                 result = get_user(val, ip);
1059                 if (result)
1060                         return result;
1061                 sfp->keep_orphan = val;
1062                 return 0;
1063         case SG_GET_KEEP_ORPHAN:
1064                 return put_user((int) sfp->keep_orphan, ip);
1065         case SG_NEXT_CMD_LEN:
1066                 result = get_user(val, ip);
1067                 if (result)
1068                         return result;
1069                 if (val > SG_MAX_CDB_SIZE)
1070                         return -ENOMEM;
1071                 sfp->next_cmd_len = (val > 0) ? val : 0;
1072                 return 0;
1073         case SG_GET_VERSION_NUM:
1074                 return put_user(sg_version_num, ip);
1075         case SG_GET_ACCESS_COUNT:
1076                 /* faked - we don't have a real access count anymore */
1077                 val = (sdp->device ? 1 : 0);
1078                 return put_user(val, ip);
1079         case SG_GET_REQUEST_TABLE:
1080                 {
1081                         sg_req_info_t *rinfo;
1082
1083                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1084                                         GFP_KERNEL);
1085                         if (!rinfo)
1086                                 return -ENOMEM;
1087                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1088                         sg_fill_request_table(sfp, rinfo);
1089                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1090         #ifdef CONFIG_COMPAT
1091                         if (in_compat_syscall())
1092                                 result = put_compat_request_table(p, rinfo);
1093                         else
1094         #endif
1095                                 result = copy_to_user(p, rinfo,
1096                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1097                         result = result ? -EFAULT : 0;
1098                         kfree(rinfo);
1099                         return result;
1100                 }
1101         case SG_EMULATED_HOST:
1102                 if (atomic_read(&sdp->detaching))
1103                         return -ENODEV;
1104                 return put_user(sdp->device->host->hostt->emulated, ip);
1105         case SCSI_IOCTL_SEND_COMMAND:
1106                 if (atomic_read(&sdp->detaching))
1107                         return -ENODEV;
1108                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1109         case SG_SET_DEBUG:
1110                 result = get_user(val, ip);
1111                 if (result)
1112                         return result;
1113                 sdp->sgdebug = (char) val;
1114                 return 0;
1115         case BLKSECTGET:
1116                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1117                                 ip);
1118         case BLKTRACESETUP:
1119                 return blk_trace_setup(sdp->device->request_queue,
1120                                        sdp->disk->disk_name,
1121                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1122                                        NULL, p);
1123         case BLKTRACESTART:
1124                 return blk_trace_startstop(sdp->device->request_queue, 1);
1125         case BLKTRACESTOP:
1126                 return blk_trace_startstop(sdp->device->request_queue, 0);
1127         case BLKTRACETEARDOWN:
1128                 return blk_trace_remove(sdp->device->request_queue);
1129         case SCSI_IOCTL_GET_IDLUN:
1130         case SCSI_IOCTL_GET_BUS_NUMBER:
1131         case SCSI_IOCTL_PROBE_HOST:
1132         case SG_GET_TRANSFORM:
1133         case SG_SCSI_RESET:
1134                 if (atomic_read(&sdp->detaching))
1135                         return -ENODEV;
1136                 break;
1137         default:
1138                 if (read_only)
1139                         return -EPERM;  /* don't know so take safe approach */
1140                 break;
1141         }
1142
1143         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1144                         cmd_in, filp->f_flags & O_NDELAY);
1145         if (result)
1146                 return result;
1147
1148         return -ENOIOCTLCMD;
1149 }
1150
1151 static long
1152 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1153 {
1154         void __user *p = (void __user *)arg;
1155         Sg_device *sdp;
1156         Sg_fd *sfp;
1157         int ret;
1158
1159         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1160                 return -ENXIO;
1161
1162         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1163         if (ret != -ENOIOCTLCMD)
1164                 return ret;
1165
1166         return scsi_ioctl(sdp->device, cmd_in, p);
1167 }
1168
1169 #ifdef CONFIG_COMPAT
1170 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1171 {
1172         void __user *p = compat_ptr(arg);
1173         Sg_device *sdp;
1174         Sg_fd *sfp;
1175         int ret;
1176
1177         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1178                 return -ENXIO;
1179
1180         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1181         if (ret != -ENOIOCTLCMD)
1182                 return ret;
1183
1184         return scsi_compat_ioctl(sdp->device, cmd_in, p);
1185 }
1186 #endif
1187
1188 static __poll_t
1189 sg_poll(struct file *filp, poll_table * wait)
1190 {
1191         __poll_t res = 0;
1192         Sg_device *sdp;
1193         Sg_fd *sfp;
1194         Sg_request *srp;
1195         int count = 0;
1196         unsigned long iflags;
1197
1198         sfp = filp->private_data;
1199         if (!sfp)
1200                 return EPOLLERR;
1201         sdp = sfp->parentdp;
1202         if (!sdp)
1203                 return EPOLLERR;
1204         poll_wait(filp, &sfp->read_wait, wait);
1205         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1206         list_for_each_entry(srp, &sfp->rq_list, entry) {
1207                 /* if any read waiting, flag it */
1208                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1209                         res = EPOLLIN | EPOLLRDNORM;
1210                 ++count;
1211         }
1212         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1213
1214         if (atomic_read(&sdp->detaching))
1215                 res |= EPOLLHUP;
1216         else if (!sfp->cmd_q) {
1217                 if (0 == count)
1218                         res |= EPOLLOUT | EPOLLWRNORM;
1219         } else if (count < SG_MAX_QUEUE)
1220                 res |= EPOLLOUT | EPOLLWRNORM;
1221         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1222                                       "sg_poll: res=0x%x\n", (__force u32) res));
1223         return res;
1224 }
1225
1226 static int
1227 sg_fasync(int fd, struct file *filp, int mode)
1228 {
1229         Sg_device *sdp;
1230         Sg_fd *sfp;
1231
1232         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1233                 return -ENXIO;
1234         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1235                                       "sg_fasync: mode=%d\n", mode));
1236
1237         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1238 }
1239
1240 static vm_fault_t
1241 sg_vma_fault(struct vm_fault *vmf)
1242 {
1243         struct vm_area_struct *vma = vmf->vma;
1244         Sg_fd *sfp;
1245         unsigned long offset, len, sa;
1246         Sg_scatter_hold *rsv_schp;
1247         int k, length;
1248
1249         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1250                 return VM_FAULT_SIGBUS;
1251         rsv_schp = &sfp->reserve;
1252         offset = vmf->pgoff << PAGE_SHIFT;
1253         if (offset >= rsv_schp->bufflen)
1254                 return VM_FAULT_SIGBUS;
1255         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1256                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1257                                       offset, rsv_schp->k_use_sg));
1258         sa = vma->vm_start;
1259         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1260         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1261                 len = vma->vm_end - sa;
1262                 len = (len < length) ? len : length;
1263                 if (offset < len) {
1264                         struct page *page = nth_page(rsv_schp->pages[k],
1265                                                      offset >> PAGE_SHIFT);
1266                         get_page(page); /* increment page count */
1267                         vmf->page = page;
1268                         return 0; /* success */
1269                 }
1270                 sa += len;
1271                 offset -= len;
1272         }
1273
1274         return VM_FAULT_SIGBUS;
1275 }
1276
1277 static const struct vm_operations_struct sg_mmap_vm_ops = {
1278         .fault = sg_vma_fault,
1279 };
1280
1281 static int
1282 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1283 {
1284         Sg_fd *sfp;
1285         unsigned long req_sz, len, sa;
1286         Sg_scatter_hold *rsv_schp;
1287         int k, length;
1288         int ret = 0;
1289
1290         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1291                 return -ENXIO;
1292         req_sz = vma->vm_end - vma->vm_start;
1293         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1294                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1295                                       (void *) vma->vm_start, (int) req_sz));
1296         if (vma->vm_pgoff)
1297                 return -EINVAL; /* want no offset */
1298         rsv_schp = &sfp->reserve;
1299         mutex_lock(&sfp->f_mutex);
1300         if (req_sz > rsv_schp->bufflen) {
1301                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1302                 goto out;
1303         }
1304
1305         sa = vma->vm_start;
1306         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1307         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1308                 len = vma->vm_end - sa;
1309                 len = (len < length) ? len : length;
1310                 sa += len;
1311         }
1312
1313         sfp->mmap_called = 1;
1314         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1315         vma->vm_private_data = sfp;
1316         vma->vm_ops = &sg_mmap_vm_ops;
1317 out:
1318         mutex_unlock(&sfp->f_mutex);
1319         return ret;
1320 }
1321
1322 static void
1323 sg_rq_end_io_usercontext(struct work_struct *work)
1324 {
1325         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1326         struct sg_fd *sfp = srp->parentfp;
1327
1328         sg_finish_rem_req(srp);
1329         sg_remove_request(sfp, srp);
1330         kref_put(&sfp->f_ref, sg_remove_sfp);
1331 }
1332
1333 /*
1334  * This function is a "bottom half" handler that is called by the mid
1335  * level when a command is completed (or has failed).
1336  */
1337 static void
1338 sg_rq_end_io(struct request *rq, blk_status_t status)
1339 {
1340         struct sg_request *srp = rq->end_io_data;
1341         struct scsi_request *req = scsi_req(rq);
1342         Sg_device *sdp;
1343         Sg_fd *sfp;
1344         unsigned long iflags;
1345         unsigned int ms;
1346         char *sense;
1347         int result, resid, done = 1;
1348
1349         if (WARN_ON(srp->done != 0))
1350                 return;
1351
1352         sfp = srp->parentfp;
1353         if (WARN_ON(sfp == NULL))
1354                 return;
1355
1356         sdp = sfp->parentdp;
1357         if (unlikely(atomic_read(&sdp->detaching)))
1358                 pr_info("%s: device detaching\n", __func__);
1359
1360         sense = req->sense;
1361         result = req->result;
1362         resid = req->resid_len;
1363
1364         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1365                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1366                                       srp->header.pack_id, result));
1367         srp->header.resid = resid;
1368         ms = jiffies_to_msecs(jiffies);
1369         srp->header.duration = (ms > srp->header.duration) ?
1370                                 (ms - srp->header.duration) : 0;
1371         if (0 != result) {
1372                 struct scsi_sense_hdr sshdr;
1373
1374                 srp->header.status = 0xff & result;
1375                 srp->header.masked_status = status_byte(result);
1376                 srp->header.msg_status = msg_byte(result);
1377                 srp->header.host_status = host_byte(result);
1378                 srp->header.driver_status = driver_byte(result);
1379                 if ((sdp->sgdebug > 0) &&
1380                     ((CHECK_CONDITION == srp->header.masked_status) ||
1381                      (COMMAND_TERMINATED == srp->header.masked_status)))
1382                         __scsi_print_sense(sdp->device, __func__, sense,
1383                                            SCSI_SENSE_BUFFERSIZE);
1384
1385                 /* Following if statement is a patch supplied by Eric Youngdale */
1386                 if (driver_byte(result) != 0
1387                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1388                     && !scsi_sense_is_deferred(&sshdr)
1389                     && sshdr.sense_key == UNIT_ATTENTION
1390                     && sdp->device->removable) {
1391                         /* Detected possible disc change. Set the bit - this */
1392                         /* may be used if there are filesystems using this device */
1393                         sdp->device->changed = 1;
1394                 }
1395         }
1396
1397         if (req->sense_len)
1398                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1399
1400         /* Rely on write phase to clean out srp status values, so no "else" */
1401
1402         /*
1403          * Free the request as soon as it is complete so that its resources
1404          * can be reused without waiting for userspace to read() the
1405          * result.  But keep the associated bio (if any) around until
1406          * blk_rq_unmap_user() can be called from user context.
1407          */
1408         srp->rq = NULL;
1409         scsi_req_free_cmd(scsi_req(rq));
1410         blk_put_request(rq);
1411
1412         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1413         if (unlikely(srp->orphan)) {
1414                 if (sfp->keep_orphan)
1415                         srp->sg_io_owned = 0;
1416                 else
1417                         done = 0;
1418         }
1419         srp->done = done;
1420         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1421
1422         if (likely(done)) {
1423                 /* Now wake up any sg_read() that is waiting for this
1424                  * packet.
1425                  */
1426                 wake_up_interruptible(&sfp->read_wait);
1427                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1428                 kref_put(&sfp->f_ref, sg_remove_sfp);
1429         } else {
1430                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1431                 schedule_work(&srp->ew.work);
1432         }
1433 }
1434
1435 static const struct file_operations sg_fops = {
1436         .owner = THIS_MODULE,
1437         .read = sg_read,
1438         .write = sg_write,
1439         .poll = sg_poll,
1440         .unlocked_ioctl = sg_ioctl,
1441 #ifdef CONFIG_COMPAT
1442         .compat_ioctl = sg_compat_ioctl,
1443 #endif
1444         .open = sg_open,
1445         .mmap = sg_mmap,
1446         .release = sg_release,
1447         .fasync = sg_fasync,
1448         .llseek = no_llseek,
1449 };
1450
1451 static struct class *sg_sysfs_class;
1452
1453 static int sg_sysfs_valid = 0;
1454
1455 static Sg_device *
1456 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1457 {
1458         struct request_queue *q = scsidp->request_queue;
1459         Sg_device *sdp;
1460         unsigned long iflags;
1461         int error;
1462         u32 k;
1463
1464         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1465         if (!sdp) {
1466                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1467                             "failure\n", __func__);
1468                 return ERR_PTR(-ENOMEM);
1469         }
1470
1471         idr_preload(GFP_KERNEL);
1472         write_lock_irqsave(&sg_index_lock, iflags);
1473
1474         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1475         if (error < 0) {
1476                 if (error == -ENOSPC) {
1477                         sdev_printk(KERN_WARNING, scsidp,
1478                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1479                                     scsidp->type, SG_MAX_DEVS - 1);
1480                         error = -ENODEV;
1481                 } else {
1482                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1483                                     "allocation Sg_device failure: %d\n",
1484                                     __func__, error);
1485                 }
1486                 goto out_unlock;
1487         }
1488         k = error;
1489
1490         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1491                                         "sg_alloc: dev=%d \n", k));
1492         sprintf(disk->disk_name, "sg%d", k);
1493         disk->first_minor = k;
1494         sdp->disk = disk;
1495         sdp->device = scsidp;
1496         mutex_init(&sdp->open_rel_lock);
1497         INIT_LIST_HEAD(&sdp->sfds);
1498         init_waitqueue_head(&sdp->open_wait);
1499         atomic_set(&sdp->detaching, 0);
1500         rwlock_init(&sdp->sfd_lock);
1501         sdp->sg_tablesize = queue_max_segments(q);
1502         sdp->index = k;
1503         kref_init(&sdp->d_ref);
1504         error = 0;
1505
1506 out_unlock:
1507         write_unlock_irqrestore(&sg_index_lock, iflags);
1508         idr_preload_end();
1509
1510         if (error) {
1511                 kfree(sdp);
1512                 return ERR_PTR(error);
1513         }
1514         return sdp;
1515 }
1516
1517 static int
1518 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1519 {
1520         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1521         struct gendisk *disk;
1522         Sg_device *sdp = NULL;
1523         struct cdev * cdev = NULL;
1524         int error;
1525         unsigned long iflags;
1526
1527         disk = alloc_disk(1);
1528         if (!disk) {
1529                 pr_warn("%s: alloc_disk failed\n", __func__);
1530                 return -ENOMEM;
1531         }
1532         disk->major = SCSI_GENERIC_MAJOR;
1533
1534         error = -ENOMEM;
1535         cdev = cdev_alloc();
1536         if (!cdev) {
1537                 pr_warn("%s: cdev_alloc failed\n", __func__);
1538                 goto out;
1539         }
1540         cdev->owner = THIS_MODULE;
1541         cdev->ops = &sg_fops;
1542
1543         sdp = sg_alloc(disk, scsidp);
1544         if (IS_ERR(sdp)) {
1545                 pr_warn("%s: sg_alloc failed\n", __func__);
1546                 error = PTR_ERR(sdp);
1547                 goto out;
1548         }
1549
1550         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1551         if (error)
1552                 goto cdev_add_err;
1553
1554         sdp->cdev = cdev;
1555         if (sg_sysfs_valid) {
1556                 struct device *sg_class_member;
1557
1558                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1559                                                 MKDEV(SCSI_GENERIC_MAJOR,
1560                                                       sdp->index),
1561                                                 sdp, "%s", disk->disk_name);
1562                 if (IS_ERR(sg_class_member)) {
1563                         pr_err("%s: device_create failed\n", __func__);
1564                         error = PTR_ERR(sg_class_member);
1565                         goto cdev_add_err;
1566                 }
1567                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1568                                           &sg_class_member->kobj, "generic");
1569                 if (error)
1570                         pr_err("%s: unable to make symlink 'generic' back "
1571                                "to sg%d\n", __func__, sdp->index);
1572         } else
1573                 pr_warn("%s: sg_sys Invalid\n", __func__);
1574
1575         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1576                     "type %d\n", sdp->index, scsidp->type);
1577
1578         dev_set_drvdata(cl_dev, sdp);
1579
1580         return 0;
1581
1582 cdev_add_err:
1583         write_lock_irqsave(&sg_index_lock, iflags);
1584         idr_remove(&sg_index_idr, sdp->index);
1585         write_unlock_irqrestore(&sg_index_lock, iflags);
1586         kfree(sdp);
1587
1588 out:
1589         put_disk(disk);
1590         if (cdev)
1591                 cdev_del(cdev);
1592         return error;
1593 }
1594
1595 static void
1596 sg_device_destroy(struct kref *kref)
1597 {
1598         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1599         unsigned long flags;
1600
1601         /* CAUTION!  Note that the device can still be found via idr_find()
1602          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1603          * any other cleanup.
1604          */
1605
1606         write_lock_irqsave(&sg_index_lock, flags);
1607         idr_remove(&sg_index_idr, sdp->index);
1608         write_unlock_irqrestore(&sg_index_lock, flags);
1609
1610         SCSI_LOG_TIMEOUT(3,
1611                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1612
1613         put_disk(sdp->disk);
1614         kfree(sdp);
1615 }
1616
1617 static void
1618 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1619 {
1620         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1621         Sg_device *sdp = dev_get_drvdata(cl_dev);
1622         unsigned long iflags;
1623         Sg_fd *sfp;
1624         int val;
1625
1626         if (!sdp)
1627                 return;
1628         /* want sdp->detaching non-zero as soon as possible */
1629         val = atomic_inc_return(&sdp->detaching);
1630         if (val > 1)
1631                 return; /* only want to do following once per device */
1632
1633         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1634                                       "%s\n", __func__));
1635
1636         read_lock_irqsave(&sdp->sfd_lock, iflags);
1637         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1638                 wake_up_interruptible_all(&sfp->read_wait);
1639                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1640         }
1641         wake_up_interruptible_all(&sdp->open_wait);
1642         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1643
1644         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1645         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1646         cdev_del(sdp->cdev);
1647         sdp->cdev = NULL;
1648
1649         kref_put(&sdp->d_ref, sg_device_destroy);
1650 }
1651
1652 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1653 module_param_named(def_reserved_size, def_reserved_size, int,
1654                    S_IRUGO | S_IWUSR);
1655 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1656
1657 MODULE_AUTHOR("Douglas Gilbert");
1658 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1659 MODULE_LICENSE("GPL");
1660 MODULE_VERSION(SG_VERSION_STR);
1661 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1662
1663 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1664                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1665 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1666 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1667
1668 static int __init
1669 init_sg(void)
1670 {
1671         int rc;
1672
1673         if (scatter_elem_sz < PAGE_SIZE) {
1674                 scatter_elem_sz = PAGE_SIZE;
1675                 scatter_elem_sz_prev = scatter_elem_sz;
1676         }
1677         if (def_reserved_size >= 0)
1678                 sg_big_buff = def_reserved_size;
1679         else
1680                 def_reserved_size = sg_big_buff;
1681
1682         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), 
1683                                     SG_MAX_DEVS, "sg");
1684         if (rc)
1685                 return rc;
1686         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1687         if ( IS_ERR(sg_sysfs_class) ) {
1688                 rc = PTR_ERR(sg_sysfs_class);
1689                 goto err_out;
1690         }
1691         sg_sysfs_valid = 1;
1692         rc = scsi_register_interface(&sg_interface);
1693         if (0 == rc) {
1694 #ifdef CONFIG_SCSI_PROC_FS
1695                 sg_proc_init();
1696 #endif                          /* CONFIG_SCSI_PROC_FS */
1697                 return 0;
1698         }
1699         class_destroy(sg_sysfs_class);
1700 err_out:
1701         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1702         return rc;
1703 }
1704
1705 static void __exit
1706 exit_sg(void)
1707 {
1708 #ifdef CONFIG_SCSI_PROC_FS
1709         remove_proc_subtree("scsi/sg", NULL);
1710 #endif                          /* CONFIG_SCSI_PROC_FS */
1711         scsi_unregister_interface(&sg_interface);
1712         class_destroy(sg_sysfs_class);
1713         sg_sysfs_valid = 0;
1714         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1715                                  SG_MAX_DEVS);
1716         idr_destroy(&sg_index_idr);
1717 }
1718
1719 static int
1720 sg_start_req(Sg_request *srp, unsigned char *cmd)
1721 {
1722         int res;
1723         struct request *rq;
1724         struct scsi_request *req;
1725         Sg_fd *sfp = srp->parentfp;
1726         sg_io_hdr_t *hp = &srp->header;
1727         int dxfer_len = (int) hp->dxfer_len;
1728         int dxfer_dir = hp->dxfer_direction;
1729         unsigned int iov_count = hp->iovec_count;
1730         Sg_scatter_hold *req_schp = &srp->data;
1731         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1732         struct request_queue *q = sfp->parentdp->device->request_queue;
1733         struct rq_map_data *md, map_data;
1734         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1735         unsigned char *long_cmdp = NULL;
1736
1737         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1738                                       "sg_start_req: dxfer_len=%d\n",
1739                                       dxfer_len));
1740
1741         if (hp->cmd_len > BLK_MAX_CDB) {
1742                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1743                 if (!long_cmdp)
1744                         return -ENOMEM;
1745         }
1746
1747         /*
1748          * NOTE
1749          *
1750          * With scsi-mq enabled, there are a fixed number of preallocated
1751          * requests equal in number to shost->can_queue.  If all of the
1752          * preallocated requests are already in use, then blk_get_request()
1753          * will sleep until an active command completes, freeing up a request.
1754          * Although waiting in an asynchronous interface is less than ideal, we
1755          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1756          * not expect an EWOULDBLOCK from this condition.
1757          */
1758         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1759                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
1760         if (IS_ERR(rq)) {
1761                 kfree(long_cmdp);
1762                 return PTR_ERR(rq);
1763         }
1764         req = scsi_req(rq);
1765
1766         if (hp->cmd_len > BLK_MAX_CDB)
1767                 req->cmd = long_cmdp;
1768         memcpy(req->cmd, cmd, hp->cmd_len);
1769         req->cmd_len = hp->cmd_len;
1770
1771         srp->rq = rq;
1772         rq->end_io_data = srp;
1773         req->retries = SG_DEFAULT_RETRIES;
1774
1775         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1776                 return 0;
1777
1778         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1779             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1780             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1781                 md = NULL;
1782         else
1783                 md = &map_data;
1784
1785         if (md) {
1786                 mutex_lock(&sfp->f_mutex);
1787                 if (dxfer_len <= rsv_schp->bufflen &&
1788                     !sfp->res_in_use) {
1789                         sfp->res_in_use = 1;
1790                         sg_link_reserve(sfp, srp, dxfer_len);
1791                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1792                         res = -EBUSY; /* sfp->res_in_use == 1 */
1793                         if (dxfer_len > rsv_schp->bufflen)
1794                                 res = -ENOMEM;
1795                         mutex_unlock(&sfp->f_mutex);
1796                         return res;
1797                 } else {
1798                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1799                         if (res) {
1800                                 mutex_unlock(&sfp->f_mutex);
1801                                 return res;
1802                         }
1803                 }
1804                 mutex_unlock(&sfp->f_mutex);
1805
1806                 md->pages = req_schp->pages;
1807                 md->page_order = req_schp->page_order;
1808                 md->nr_entries = req_schp->k_use_sg;
1809                 md->offset = 0;
1810                 md->null_mapped = hp->dxferp ? 0 : 1;
1811                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1812                         md->from_user = 1;
1813                 else
1814                         md->from_user = 0;
1815         }
1816
1817         if (iov_count) {
1818                 struct iovec *iov = NULL;
1819                 struct iov_iter i;
1820
1821                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1822                 if (res < 0)
1823                         return res;
1824
1825                 iov_iter_truncate(&i, hp->dxfer_len);
1826                 if (!iov_iter_count(&i)) {
1827                         kfree(iov);
1828                         return -EINVAL;
1829                 }
1830
1831                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1832                 kfree(iov);
1833         } else
1834                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1835                                       hp->dxfer_len, GFP_ATOMIC);
1836
1837         if (!res) {
1838                 srp->bio = rq->bio;
1839
1840                 if (!md) {
1841                         req_schp->dio_in_use = 1;
1842                         hp->info |= SG_INFO_DIRECT_IO;
1843                 }
1844         }
1845         return res;
1846 }
1847
1848 static int
1849 sg_finish_rem_req(Sg_request *srp)
1850 {
1851         int ret = 0;
1852
1853         Sg_fd *sfp = srp->parentfp;
1854         Sg_scatter_hold *req_schp = &srp->data;
1855
1856         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1857                                       "sg_finish_rem_req: res_used=%d\n",
1858                                       (int) srp->res_used));
1859         if (srp->bio)
1860                 ret = blk_rq_unmap_user(srp->bio);
1861
1862         if (srp->rq) {
1863                 scsi_req_free_cmd(scsi_req(srp->rq));
1864                 blk_put_request(srp->rq);
1865         }
1866
1867         if (srp->res_used)
1868                 sg_unlink_reserve(sfp, srp);
1869         else
1870                 sg_remove_scat(sfp, req_schp);
1871
1872         return ret;
1873 }
1874
1875 static int
1876 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1877 {
1878         int sg_bufflen = tablesize * sizeof(struct page *);
1879         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1880
1881         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1882         if (!schp->pages)
1883                 return -ENOMEM;
1884         schp->sglist_len = sg_bufflen;
1885         return tablesize;       /* number of scat_gath elements allocated */
1886 }
1887
1888 static int
1889 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1890 {
1891         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1892         int sg_tablesize = sfp->parentdp->sg_tablesize;
1893         int blk_size = buff_size, order;
1894         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1895
1896         if (blk_size < 0)
1897                 return -EFAULT;
1898         if (0 == blk_size)
1899                 ++blk_size;     /* don't know why */
1900         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1901         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1902         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1903                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1904                 buff_size, blk_size));
1905
1906         /* N.B. ret_sz carried into this block ... */
1907         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1908         if (mx_sc_elems < 0)
1909                 return mx_sc_elems;     /* most likely -ENOMEM */
1910
1911         num = scatter_elem_sz;
1912         if (unlikely(num != scatter_elem_sz_prev)) {
1913                 if (num < PAGE_SIZE) {
1914                         scatter_elem_sz = PAGE_SIZE;
1915                         scatter_elem_sz_prev = PAGE_SIZE;
1916                 } else
1917                         scatter_elem_sz_prev = num;
1918         }
1919
1920         order = get_order(num);
1921 retry:
1922         ret_sz = 1 << (PAGE_SHIFT + order);
1923
1924         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1925              k++, rem_sz -= ret_sz) {
1926
1927                 num = (rem_sz > scatter_elem_sz_prev) ?
1928                         scatter_elem_sz_prev : rem_sz;
1929
1930                 schp->pages[k] = alloc_pages(gfp_mask, order);
1931                 if (!schp->pages[k])
1932                         goto out;
1933
1934                 if (num == scatter_elem_sz_prev) {
1935                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1936                                 scatter_elem_sz = ret_sz;
1937                                 scatter_elem_sz_prev = ret_sz;
1938                         }
1939                 }
1940
1941                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1942                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1943                                  k, num, ret_sz));
1944         }               /* end of for loop */
1945
1946         schp->page_order = order;
1947         schp->k_use_sg = k;
1948         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1949                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1950                          k, rem_sz));
1951
1952         schp->bufflen = blk_size;
1953         if (rem_sz > 0) /* must have failed */
1954                 return -ENOMEM;
1955         return 0;
1956 out:
1957         for (i = 0; i < k; i++)
1958                 __free_pages(schp->pages[i], order);
1959
1960         if (--order >= 0)
1961                 goto retry;
1962
1963         return -ENOMEM;
1964 }
1965
1966 static void
1967 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1968 {
1969         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1970                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1971         if (schp->pages && schp->sglist_len > 0) {
1972                 if (!schp->dio_in_use) {
1973                         int k;
1974
1975                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1976                                 SCSI_LOG_TIMEOUT(5,
1977                                         sg_printk(KERN_INFO, sfp->parentdp,
1978                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1979                                         k, schp->pages[k]));
1980                                 __free_pages(schp->pages[k], schp->page_order);
1981                         }
1982
1983                         kfree(schp->pages);
1984                 }
1985         }
1986         memset(schp, 0, sizeof (*schp));
1987 }
1988
1989 static int
1990 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1991 {
1992         Sg_scatter_hold *schp = &srp->data;
1993         int k, num;
1994
1995         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1996                          "sg_read_oxfer: num_read_xfer=%d\n",
1997                          num_read_xfer));
1998         if ((!outp) || (num_read_xfer <= 0))
1999                 return 0;
2000
2001         num = 1 << (PAGE_SHIFT + schp->page_order);
2002         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2003                 if (num > num_read_xfer) {
2004                         if (copy_to_user(outp, page_address(schp->pages[k]),
2005                                            num_read_xfer))
2006                                 return -EFAULT;
2007                         break;
2008                 } else {
2009                         if (copy_to_user(outp, page_address(schp->pages[k]),
2010                                            num))
2011                                 return -EFAULT;
2012                         num_read_xfer -= num;
2013                         if (num_read_xfer <= 0)
2014                                 break;
2015                         outp += num;
2016                 }
2017         }
2018
2019         return 0;
2020 }
2021
2022 static void
2023 sg_build_reserve(Sg_fd * sfp, int req_size)
2024 {
2025         Sg_scatter_hold *schp = &sfp->reserve;
2026
2027         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2028                          "sg_build_reserve: req_size=%d\n", req_size));
2029         do {
2030                 if (req_size < PAGE_SIZE)
2031                         req_size = PAGE_SIZE;
2032                 if (0 == sg_build_indirect(schp, sfp, req_size))
2033                         return;
2034                 else
2035                         sg_remove_scat(sfp, schp);
2036                 req_size >>= 1; /* divide by 2 */
2037         } while (req_size > (PAGE_SIZE / 2));
2038 }
2039
2040 static void
2041 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2042 {
2043         Sg_scatter_hold *req_schp = &srp->data;
2044         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2045         int k, num, rem;
2046
2047         srp->res_used = 1;
2048         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2049                          "sg_link_reserve: size=%d\n", size));
2050         rem = size;
2051
2052         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2053         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2054                 if (rem <= num) {
2055                         req_schp->k_use_sg = k + 1;
2056                         req_schp->sglist_len = rsv_schp->sglist_len;
2057                         req_schp->pages = rsv_schp->pages;
2058
2059                         req_schp->bufflen = size;
2060                         req_schp->page_order = rsv_schp->page_order;
2061                         break;
2062                 } else
2063                         rem -= num;
2064         }
2065
2066         if (k >= rsv_schp->k_use_sg)
2067                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2068                                  "sg_link_reserve: BAD size\n"));
2069 }
2070
2071 static void
2072 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2073 {
2074         Sg_scatter_hold *req_schp = &srp->data;
2075
2076         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2077                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2078                                       (int) req_schp->k_use_sg));
2079         req_schp->k_use_sg = 0;
2080         req_schp->bufflen = 0;
2081         req_schp->pages = NULL;
2082         req_schp->page_order = 0;
2083         req_schp->sglist_len = 0;
2084         srp->res_used = 0;
2085         /* Called without mutex lock to avoid deadlock */
2086         sfp->res_in_use = 0;
2087 }
2088
2089 static Sg_request *
2090 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2091 {
2092         Sg_request *resp;
2093         unsigned long iflags;
2094
2095         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2096         list_for_each_entry(resp, &sfp->rq_list, entry) {
2097                 /* look for requests that are ready + not SG_IO owned */
2098                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2099                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2100                         resp->done = 2; /* guard against other readers */
2101                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2102                         return resp;
2103                 }
2104         }
2105         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2106         return NULL;
2107 }
2108
2109 /* always adds to end of list */
2110 static Sg_request *
2111 sg_add_request(Sg_fd * sfp)
2112 {
2113         int k;
2114         unsigned long iflags;
2115         Sg_request *rp = sfp->req_arr;
2116
2117         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2118         if (!list_empty(&sfp->rq_list)) {
2119                 if (!sfp->cmd_q)
2120                         goto out_unlock;
2121
2122                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2123                         if (!rp->parentfp)
2124                                 break;
2125                 }
2126                 if (k >= SG_MAX_QUEUE)
2127                         goto out_unlock;
2128         }
2129         memset(rp, 0, sizeof (Sg_request));
2130         rp->parentfp = sfp;
2131         rp->header.duration = jiffies_to_msecs(jiffies);
2132         list_add_tail(&rp->entry, &sfp->rq_list);
2133         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2134         return rp;
2135 out_unlock:
2136         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2137         return NULL;
2138 }
2139
2140 /* Return of 1 for found; 0 for not found */
2141 static int
2142 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2143 {
2144         unsigned long iflags;
2145         int res = 0;
2146
2147         if (!sfp || !srp || list_empty(&sfp->rq_list))
2148                 return res;
2149         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2150         if (!list_empty(&srp->entry)) {
2151                 list_del(&srp->entry);
2152                 srp->parentfp = NULL;
2153                 res = 1;
2154         }
2155         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2156         return res;
2157 }
2158
2159 static Sg_fd *
2160 sg_add_sfp(Sg_device * sdp)
2161 {
2162         Sg_fd *sfp;
2163         unsigned long iflags;
2164         int bufflen;
2165
2166         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2167         if (!sfp)
2168                 return ERR_PTR(-ENOMEM);
2169
2170         init_waitqueue_head(&sfp->read_wait);
2171         rwlock_init(&sfp->rq_list_lock);
2172         INIT_LIST_HEAD(&sfp->rq_list);
2173         kref_init(&sfp->f_ref);
2174         mutex_init(&sfp->f_mutex);
2175         sfp->timeout = SG_DEFAULT_TIMEOUT;
2176         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2177         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2178         sfp->cmd_q = SG_DEF_COMMAND_Q;
2179         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2180         sfp->parentdp = sdp;
2181         write_lock_irqsave(&sdp->sfd_lock, iflags);
2182         if (atomic_read(&sdp->detaching)) {
2183                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2184                 kfree(sfp);
2185                 return ERR_PTR(-ENODEV);
2186         }
2187         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2188         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2189         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2190                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2191         if (unlikely(sg_big_buff != def_reserved_size))
2192                 sg_big_buff = def_reserved_size;
2193
2194         bufflen = min_t(int, sg_big_buff,
2195                         max_sectors_bytes(sdp->device->request_queue));
2196         sg_build_reserve(sfp, bufflen);
2197         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2198                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2199                                       sfp->reserve.bufflen,
2200                                       sfp->reserve.k_use_sg));
2201
2202         kref_get(&sdp->d_ref);
2203         __module_get(THIS_MODULE);
2204         return sfp;
2205 }
2206
2207 static void
2208 sg_remove_sfp_usercontext(struct work_struct *work)
2209 {
2210         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2211         struct sg_device *sdp = sfp->parentdp;
2212         Sg_request *srp;
2213         unsigned long iflags;
2214
2215         /* Cleanup any responses which were never read(). */
2216         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2217         while (!list_empty(&sfp->rq_list)) {
2218                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2219                 sg_finish_rem_req(srp);
2220                 list_del(&srp->entry);
2221                 srp->parentfp = NULL;
2222         }
2223         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2224
2225         if (sfp->reserve.bufflen > 0) {
2226                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2227                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2228                                 (int) sfp->reserve.bufflen,
2229                                 (int) sfp->reserve.k_use_sg));
2230                 sg_remove_scat(sfp, &sfp->reserve);
2231         }
2232
2233         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2234                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2235         kfree(sfp);
2236
2237         scsi_device_put(sdp->device);
2238         kref_put(&sdp->d_ref, sg_device_destroy);
2239         module_put(THIS_MODULE);
2240 }
2241
2242 static void
2243 sg_remove_sfp(struct kref *kref)
2244 {
2245         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2246         struct sg_device *sdp = sfp->parentdp;
2247         unsigned long iflags;
2248
2249         write_lock_irqsave(&sdp->sfd_lock, iflags);
2250         list_del(&sfp->sfd_siblings);
2251         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2252
2253         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2254         schedule_work(&sfp->ew.work);
2255 }
2256
2257 #ifdef CONFIG_SCSI_PROC_FS
2258 static int
2259 sg_idr_max_id(int id, void *p, void *data)
2260 {
2261         int *k = data;
2262
2263         if (*k < id)
2264                 *k = id;
2265
2266         return 0;
2267 }
2268
2269 static int
2270 sg_last_dev(void)
2271 {
2272         int k = -1;
2273         unsigned long iflags;
2274
2275         read_lock_irqsave(&sg_index_lock, iflags);
2276         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2277         read_unlock_irqrestore(&sg_index_lock, iflags);
2278         return k + 1;           /* origin 1 */
2279 }
2280 #endif
2281
2282 /* must be called with sg_index_lock held */
2283 static Sg_device *sg_lookup_dev(int dev)
2284 {
2285         return idr_find(&sg_index_idr, dev);
2286 }
2287
2288 static Sg_device *
2289 sg_get_dev(int dev)
2290 {
2291         struct sg_device *sdp;
2292         unsigned long flags;
2293
2294         read_lock_irqsave(&sg_index_lock, flags);
2295         sdp = sg_lookup_dev(dev);
2296         if (!sdp)
2297                 sdp = ERR_PTR(-ENXIO);
2298         else if (atomic_read(&sdp->detaching)) {
2299                 /* If sdp->detaching, then the refcount may already be 0, in
2300                  * which case it would be a bug to do kref_get().
2301                  */
2302                 sdp = ERR_PTR(-ENODEV);
2303         } else
2304                 kref_get(&sdp->d_ref);
2305         read_unlock_irqrestore(&sg_index_lock, flags);
2306
2307         return sdp;
2308 }
2309
2310 #ifdef CONFIG_SCSI_PROC_FS
2311 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2312
2313 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2314 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2315                                   size_t count, loff_t *off);
2316 static const struct proc_ops adio_proc_ops = {
2317         .proc_open      = sg_proc_single_open_adio,
2318         .proc_read      = seq_read,
2319         .proc_lseek     = seq_lseek,
2320         .proc_write     = sg_proc_write_adio,
2321         .proc_release   = single_release,
2322 };
2323
2324 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2325 static ssize_t sg_proc_write_dressz(struct file *filp, 
2326                 const char __user *buffer, size_t count, loff_t *off);
2327 static const struct proc_ops dressz_proc_ops = {
2328         .proc_open      = sg_proc_single_open_dressz,
2329         .proc_read      = seq_read,
2330         .proc_lseek     = seq_lseek,
2331         .proc_write     = sg_proc_write_dressz,
2332         .proc_release   = single_release,
2333 };
2334
2335 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2336 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2337 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2338 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2339 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2340 static void dev_seq_stop(struct seq_file *s, void *v);
2341 static const struct seq_operations dev_seq_ops = {
2342         .start = dev_seq_start,
2343         .next  = dev_seq_next,
2344         .stop  = dev_seq_stop,
2345         .show  = sg_proc_seq_show_dev,
2346 };
2347
2348 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2349 static const struct seq_operations devstrs_seq_ops = {
2350         .start = dev_seq_start,
2351         .next  = dev_seq_next,
2352         .stop  = dev_seq_stop,
2353         .show  = sg_proc_seq_show_devstrs,
2354 };
2355
2356 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2357 static const struct seq_operations debug_seq_ops = {
2358         .start = dev_seq_start,
2359         .next  = dev_seq_next,
2360         .stop  = dev_seq_stop,
2361         .show  = sg_proc_seq_show_debug,
2362 };
2363
2364 static int
2365 sg_proc_init(void)
2366 {
2367         struct proc_dir_entry *p;
2368
2369         p = proc_mkdir("scsi/sg", NULL);
2370         if (!p)
2371                 return 1;
2372
2373         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2374         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2375         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2376         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2377         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2378         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2379         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2380         return 0;
2381 }
2382
2383
2384 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2385 {
2386         seq_printf(s, "%d\n", *((int *)s->private));
2387         return 0;
2388 }
2389
2390 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2391 {
2392         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2393 }
2394
2395 static ssize_t 
2396 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2397                    size_t count, loff_t *off)
2398 {
2399         int err;
2400         unsigned long num;
2401
2402         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2403                 return -EACCES;
2404         err = kstrtoul_from_user(buffer, count, 0, &num);
2405         if (err)
2406                 return err;
2407         sg_allow_dio = num ? 1 : 0;
2408         return count;
2409 }
2410
2411 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2412 {
2413         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2414 }
2415
2416 static ssize_t 
2417 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2418                      size_t count, loff_t *off)
2419 {
2420         int err;
2421         unsigned long k = ULONG_MAX;
2422
2423         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2424                 return -EACCES;
2425
2426         err = kstrtoul_from_user(buffer, count, 0, &k);
2427         if (err)
2428                 return err;
2429         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2430                 sg_big_buff = k;
2431                 return count;
2432         }
2433         return -ERANGE;
2434 }
2435
2436 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2437 {
2438         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2439                    sg_version_date);
2440         return 0;
2441 }
2442
2443 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2444 {
2445         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2446         return 0;
2447 }
2448
2449 struct sg_proc_deviter {
2450         loff_t  index;
2451         size_t  max;
2452 };
2453
2454 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2455 {
2456         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2457
2458         s->private = it;
2459         if (! it)
2460                 return NULL;
2461
2462         it->index = *pos;
2463         it->max = sg_last_dev();
2464         if (it->index >= it->max)
2465                 return NULL;
2466         return it;
2467 }
2468
2469 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2470 {
2471         struct sg_proc_deviter * it = s->private;
2472
2473         *pos = ++it->index;
2474         return (it->index < it->max) ? it : NULL;
2475 }
2476
2477 static void dev_seq_stop(struct seq_file *s, void *v)
2478 {
2479         kfree(s->private);
2480 }
2481
2482 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2483 {
2484         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2485         Sg_device *sdp;
2486         struct scsi_device *scsidp;
2487         unsigned long iflags;
2488
2489         read_lock_irqsave(&sg_index_lock, iflags);
2490         sdp = it ? sg_lookup_dev(it->index) : NULL;
2491         if ((NULL == sdp) || (NULL == sdp->device) ||
2492             (atomic_read(&sdp->detaching)))
2493                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2494         else {
2495                 scsidp = sdp->device;
2496                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2497                               scsidp->host->host_no, scsidp->channel,
2498                               scsidp->id, scsidp->lun, (int) scsidp->type,
2499                               1,
2500                               (int) scsidp->queue_depth,
2501                               (int) scsi_device_busy(scsidp),
2502                               (int) scsi_device_online(scsidp));
2503         }
2504         read_unlock_irqrestore(&sg_index_lock, iflags);
2505         return 0;
2506 }
2507
2508 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2509 {
2510         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2511         Sg_device *sdp;
2512         struct scsi_device *scsidp;
2513         unsigned long iflags;
2514
2515         read_lock_irqsave(&sg_index_lock, iflags);
2516         sdp = it ? sg_lookup_dev(it->index) : NULL;
2517         scsidp = sdp ? sdp->device : NULL;
2518         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2519                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2520                            scsidp->vendor, scsidp->model, scsidp->rev);
2521         else
2522                 seq_puts(s, "<no active device>\n");
2523         read_unlock_irqrestore(&sg_index_lock, iflags);
2524         return 0;
2525 }
2526
2527 /* must be called while holding sg_index_lock */
2528 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2529 {
2530         int k, new_interface, blen, usg;
2531         Sg_request *srp;
2532         Sg_fd *fp;
2533         const sg_io_hdr_t *hp;
2534         const char * cp;
2535         unsigned int ms;
2536
2537         k = 0;
2538         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2539                 k++;
2540                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2541                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2542                            "(res)sgat=%d low_dma=%d\n", k,
2543                            jiffies_to_msecs(fp->timeout),
2544                            fp->reserve.bufflen,
2545                            (int) fp->reserve.k_use_sg, 0);
2546                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2547                            (int) fp->cmd_q, (int) fp->force_packid,
2548                            (int) fp->keep_orphan);
2549                 list_for_each_entry(srp, &fp->rq_list, entry) {
2550                         hp = &srp->header;
2551                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2552                         if (srp->res_used) {
2553                                 if (new_interface &&
2554                                     (SG_FLAG_MMAP_IO & hp->flags))
2555                                         cp = "     mmap>> ";
2556                                 else
2557                                         cp = "     rb>> ";
2558                         } else {
2559                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2560                                         cp = "     dio>> ";
2561                                 else
2562                                         cp = "     ";
2563                         }
2564                         seq_puts(s, cp);
2565                         blen = srp->data.bufflen;
2566                         usg = srp->data.k_use_sg;
2567                         seq_puts(s, srp->done ?
2568                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2569                                   : "act:");
2570                         seq_printf(s, " id=%d blen=%d",
2571                                    srp->header.pack_id, blen);
2572                         if (srp->done)
2573                                 seq_printf(s, " dur=%d", hp->duration);
2574                         else {
2575                                 ms = jiffies_to_msecs(jiffies);
2576                                 seq_printf(s, " t_o/elap=%d/%d",
2577                                         (new_interface ? hp->timeout :
2578                                                   jiffies_to_msecs(fp->timeout)),
2579                                         (ms > hp->duration ? ms - hp->duration : 0));
2580                         }
2581                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2582                                    (int) srp->data.cmd_opcode);
2583                 }
2584                 if (list_empty(&fp->rq_list))
2585                         seq_puts(s, "     No requests active\n");
2586                 read_unlock(&fp->rq_list_lock);
2587         }
2588 }
2589
2590 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2591 {
2592         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2593         Sg_device *sdp;
2594         unsigned long iflags;
2595
2596         if (it && (0 == it->index))
2597                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2598                            (int)it->max, sg_big_buff);
2599
2600         read_lock_irqsave(&sg_index_lock, iflags);
2601         sdp = it ? sg_lookup_dev(it->index) : NULL;
2602         if (NULL == sdp)
2603                 goto skip;
2604         read_lock(&sdp->sfd_lock);
2605         if (!list_empty(&sdp->sfds)) {
2606                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2607                 if (atomic_read(&sdp->detaching))
2608                         seq_puts(s, "detaching pending close ");
2609                 else if (sdp->device) {
2610                         struct scsi_device *scsidp = sdp->device;
2611
2612                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2613                                    scsidp->host->host_no,
2614                                    scsidp->channel, scsidp->id,
2615                                    scsidp->lun,
2616                                    scsidp->host->hostt->emulated);
2617                 }
2618                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2619                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2620                 sg_proc_debug_helper(s, sdp);
2621         }
2622         read_unlock(&sdp->sfd_lock);
2623 skip:
2624         read_unlock_irqrestore(&sg_index_lock, iflags);
2625         return 0;
2626 }
2627
2628 #endif                          /* CONFIG_SCSI_PROC_FS */
2629
2630 module_init(init_sg);
2631 module_exit(exit_sg);