kernel/drivers/scsi/sg.c

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