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