drivers/scsi/sg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  History:
   4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   5  *           to allow user process control of SCSI devices.
   6  *  Development Sponsored by Killy Corp. NY NY
   7  *
   8  * Original driver (sg.c):
   9  *        Copyright (C) 1992 Lawrence Foard
  10  * Version 2 and 3 extensions to driver:
  11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  12  */
  13
  14 static int sg_version_num = 30536;      /* 2 digits for each component */
  15 #define SG_VERSION_STR "3.5.36"
  16
  17 /*
  18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  21  *        (otherwise the macros compile to empty statements).
  22  *
  23  */
  24 #include <linux/module.h>
  25
  26 #include <linux/fs.h>
  27 #include <linux/kernel.h>
  28 #include <linux/sched.h>
  29 #include <linux/string.h>
  30 #include <linux/mm.h>
  31 #include <linux/errno.h>
  32 #include <linux/mtio.h>
  33 #include <linux/ioctl.h>
  34 #include <linux/slab.h>
  35 #include <linux/fcntl.h>
  36 #include <linux/init.h>
  37 #include <linux/poll.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/cdev.h>
  40 #include <linux/idr.h>
  41 #include <linux/seq_file.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/delay.h>
  44 #include <linux/blktrace_api.h>
  45 #include <linux/mutex.h>
  46 #include <linux/atomic.h>
  47 #include <linux/ratelimit.h>
  48 #include <linux/uio.h>
  49 #include <linux/cred.h> /* for sg_check_file_access() */
  50
  51 #include "scsi.h"
  52 #include <scsi/scsi_dbg.h>
  53 #include <scsi/scsi_host.h>
  54 #include <scsi/scsi_driver.h>
  55 #include <scsi/scsi_ioctl.h>
  56 #include <scsi/sg.h>
  57
  58 #include "scsi_logging.h"
  59
  60 #ifdef CONFIG_SCSI_PROC_FS
  61 #include <linux/proc_fs.h>
  62 static char *sg_version_date = "20140603";
  63
  64 static int sg_proc_init(void);
  65 #endif
  66
  67 #define SG_ALLOW_DIO_DEF 0
  68
  69 #define SG_MAX_DEVS 32768
  70
  71 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  72  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  73  * than 16 bytes are "variable length" whose length is a multiple of 4
  74  */
  75 #define SG_MAX_CDB_SIZE 252
  76
  77 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  78
  79 int sg_big_buff = SG_DEF_RESERVED_SIZE;
  80 /* N.B. This variable is readable and writeable via
  81    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  82    of this size (or less if there is not enough memory) will be reserved
  83    for use by this file descriptor. [Deprecated usage: this variable is also
  84    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  85    the kernel (i.e. it is not a module).] */
  86 static int def_reserved_size = -1;      /* picks up init parameter */
  87 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  88
  89 static int scatter_elem_sz = SG_SCATTER_SZ;
  90 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  91
  92 #define SG_SECTOR_SZ 512
  93
  94 static int sg_add_device(struct device *, struct class_interface *);
  95 static void sg_remove_device(struct device *, struct class_interface *);
  96
  97 static DEFINE_IDR(sg_index_idr);
  98 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
  99                                                            file descriptor list for device */
 100
 101 static struct class_interface sg_interface = {
 102         .add_dev        = sg_add_device,
 103         .remove_dev     = sg_remove_device,
 104 };
 105
 106 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 107         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 108         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 109         unsigned bufflen;       /* Size of (aggregate) data buffer */
 110         struct page **pages;
 111         int page_order;
 112         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 113         unsigned char cmd_opcode; /* first byte of command */
 114 } Sg_scatter_hold;
 115
 116 struct sg_device;               /* forward declarations */
 117 struct sg_fd;
 118
 119 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 120         struct list_head entry; /* list entry */
 121         struct sg_fd *parentfp; /* NULL -> not in use */
 122         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 123         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 124         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 125         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 126         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 127         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 128         /* done protected by rq_list_lock */
 129         char done;              /* 0->before bh, 1->before read, 2->read */
 130         struct request *rq;
 131         struct bio *bio;
 132         struct execute_work ew;
 133 } Sg_request;
 134
 135 typedef struct sg_fd {          /* holds the state of a file descriptor */
 136         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 137         struct sg_device *parentdp;     /* owning device */
 138         wait_queue_head_t read_wait;    /* queue read until command done */
 139         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 140         struct mutex f_mutex;   /* protect against changes in this fd */
 141         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 142         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 143         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 144         struct list_head rq_list; /* head of request list */
 145         struct fasync_struct *async_qp; /* used by asynchronous notification */
 146         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 147         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 148         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 149         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 150         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 151         char mmap_called;       /* 0 -> mmap() never called on this fd */
 152         char res_in_use;        /* 1 -> 'reserve' array in use */
 153         struct kref f_ref;
 154         struct execute_work ew;
 155 } Sg_fd;
 156
 157 typedef struct sg_device { /* holds the state of each scsi generic device */
 158         struct scsi_device *device;
 159         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 160         struct mutex open_rel_lock;     /* held when in open() or release() */
 161         int sg_tablesize;       /* adapter's max scatter-gather table size */
 162         u32 index;              /* device index number */
 163         struct list_head sfds;
 164         rwlock_t sfd_lock;      /* protect access to sfd list */
 165         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 166         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 167         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 168         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 169         struct gendisk *disk;
 170         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 171         struct kref d_ref;
 172 } Sg_device;
 173
 174 /* tasklet or soft irq callback */
 175 static void sg_rq_end_io(struct request *rq, blk_status_t status);
 176 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 177 static int sg_finish_rem_req(Sg_request * srp);
 178 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 179 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 180                            Sg_request * srp);
 181 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 182                         const char __user *buf, size_t count, int blocking,
 183                         int read_only, int sg_io_owned, Sg_request **o_srp);
 184 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 185                            unsigned char *cmnd, int timeout, int blocking);
 186 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 187 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 188 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 189 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 190 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 191 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 192 static void sg_remove_sfp(struct kref *);
 193 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 194 static Sg_request *sg_add_request(Sg_fd * sfp);
 195 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 196 static Sg_device *sg_get_dev(int dev);
 197 static void sg_device_destroy(struct kref *kref);
 198
 199 #define SZ_SG_HEADER sizeof(struct sg_header)
 200 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 201 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 202 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 203
 204 #define sg_printk(prefix, sdp, fmt, a...) \
 205         sdev_prefix_printk(prefix, (sdp)->device,               \
 206                            (sdp)->disk->disk_name, fmt, ##a)
 207
 208 /*
 209  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 210  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 211  * to trigger read() and write() calls from various contexts with elevated
 212  * privileges. This can lead to kernel memory corruption (e.g. if these
 213  * interfaces are called through splice()) and privilege escalation inside
 214  * userspace (e.g. if a process with access to such a device passes a file
 215  * descriptor to a SUID binary as stdin/stdout/stderr).
 216  *
 217  * This function provides protection for the legacy API by restricting the
 218  * calling context.
 219  */
 220 static int sg_check_file_access(struct file *filp, const char *caller)
 221 {
 222         if (filp->f_cred != current_real_cred()) {
 223                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 224                         caller, task_tgid_vnr(current), current->comm);
 225                 return -EPERM;
 226         }
 227         if (uaccess_kernel()) {
 228                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
 229                         caller, task_tgid_vnr(current), current->comm);
 230                 return -EACCES;
 231         }
 232         return 0;
 233 }
 234
 235 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 236 {
 237         struct sg_fd *sfp = filp->private_data;
 238
 239         if (sfp->parentdp->device->type == TYPE_SCANNER)
 240                 return 0;
 241
 242         return blk_verify_command(cmd, filp->f_mode);
 243 }
 244
 245 static int
 246 open_wait(Sg_device *sdp, int flags)
 247 {
 248         int retval = 0;
 249
 250         if (flags & O_EXCL) {
 251                 while (sdp->open_cnt > 0) {
 252                         mutex_unlock(&sdp->open_rel_lock);
 253                         retval = wait_event_interruptible(sdp->open_wait,
 254                                         (atomic_read(&sdp->detaching) ||
 255                                          !sdp->open_cnt));
 256                         mutex_lock(&sdp->open_rel_lock);
 257
 258                         if (retval) /* -ERESTARTSYS */
 259                                 return retval;
 260                         if (atomic_read(&sdp->detaching))
 261                                 return -ENODEV;
 262                 }
 263         } else {
 264                 while (sdp->exclude) {
 265                         mutex_unlock(&sdp->open_rel_lock);
 266                         retval = wait_event_interruptible(sdp->open_wait,
 267                                         (atomic_read(&sdp->detaching) ||
 268                                          !sdp->exclude));
 269                         mutex_lock(&sdp->open_rel_lock);
 270
 271                         if (retval) /* -ERESTARTSYS */
 272                                 return retval;
 273                         if (atomic_read(&sdp->detaching))
 274                                 return -ENODEV;
 275                 }
 276         }
 277
 278         return retval;
 279 }
 280
 281 /* Returns 0 on success, else a negated errno value */
 282 static int
 283 sg_open(struct inode *inode, struct file *filp)
 284 {
 285         int dev = iminor(inode);
 286         int flags = filp->f_flags;
 287         struct request_queue *q;
 288         Sg_device *sdp;
 289         Sg_fd *sfp;
 290         int retval;
 291
 292         nonseekable_open(inode, filp);
 293         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 294                 return -EPERM; /* Can't lock it with read only access */
 295         sdp = sg_get_dev(dev);
 296         if (IS_ERR(sdp))
 297                 return PTR_ERR(sdp);
 298
 299         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 300                                       "sg_open: flags=0x%x\n", flags));
 301
 302         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 303         /* Prevent the device driver from vanishing while we sleep */
 304         retval = scsi_device_get(sdp->device);
 305         if (retval)
 306                 goto sg_put;
 307
 308         retval = scsi_autopm_get_device(sdp->device);
 309         if (retval)
 310                 goto sdp_put;
 311
 312         /* scsi_block_when_processing_errors() may block so bypass
 313          * check if O_NONBLOCK. Permits SCSI commands to be issued
 314          * during error recovery. Tread carefully. */
 315         if (!((flags & O_NONBLOCK) ||
 316               scsi_block_when_processing_errors(sdp->device))) {
 317                 retval = -ENXIO;
 318                 /* we are in error recovery for this device */
 319                 goto error_out;
 320         }
 321
 322         mutex_lock(&sdp->open_rel_lock);
 323         if (flags & O_NONBLOCK) {
 324                 if (flags & O_EXCL) {
 325                         if (sdp->open_cnt > 0) {
 326                                 retval = -EBUSY;
 327                                 goto error_mutex_locked;
 328                         }
 329                 } else {
 330                         if (sdp->exclude) {
 331                                 retval = -EBUSY;
 332                                 goto error_mutex_locked;
 333                         }
 334                 }
 335         } else {
 336                 retval = open_wait(sdp, flags);
 337                 if (retval) /* -ERESTARTSYS or -ENODEV */
 338                         goto error_mutex_locked;
 339         }
 340
 341         /* N.B. at this point we are holding the open_rel_lock */
 342         if (flags & O_EXCL)
 343                 sdp->exclude = true;
 344
 345         if (sdp->open_cnt < 1) {  /* no existing opens */
 346                 sdp->sgdebug = 0;
 347                 q = sdp->device->request_queue;
 348                 sdp->sg_tablesize = queue_max_segments(q);
 349         }
 350         sfp = sg_add_sfp(sdp);
 351         if (IS_ERR(sfp)) {
 352                 retval = PTR_ERR(sfp);
 353                 goto out_undo;
 354         }
 355
 356         filp->private_data = sfp;
 357         sdp->open_cnt++;
 358         mutex_unlock(&sdp->open_rel_lock);
 359
 360         retval = 0;
 361 sg_put:
 362         kref_put(&sdp->d_ref, sg_device_destroy);
 363         return retval;
 364
 365 out_undo:
 366         if (flags & O_EXCL) {
 367                 sdp->exclude = false;   /* undo if error */
 368                 wake_up_interruptible(&sdp->open_wait);
 369         }
 370 error_mutex_locked:
 371         mutex_unlock(&sdp->open_rel_lock);
 372 error_out:
 373         scsi_autopm_put_device(sdp->device);
 374 sdp_put:
 375         scsi_device_put(sdp->device);
 376         goto sg_put;
 377 }
 378
 379 /* Release resources associated with a successful sg_open()
 380  * Returns 0 on success, else a negated errno value */
 381 static int
 382 sg_release(struct inode *inode, struct file *filp)
 383 {
 384         Sg_device *sdp;
 385         Sg_fd *sfp;
 386
 387         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 388                 return -ENXIO;
 389         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 390
 391         mutex_lock(&sdp->open_rel_lock);
 392         scsi_autopm_put_device(sdp->device);
 393         kref_put(&sfp->f_ref, sg_remove_sfp);
 394         sdp->open_cnt--;
 395
 396         /* possibly many open()s waiting on exlude clearing, start many;
 397          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 398         if (sdp->exclude) {
 399                 sdp->exclude = false;
 400                 wake_up_interruptible_all(&sdp->open_wait);
 401         } else if (0 == sdp->open_cnt) {
 402                 wake_up_interruptible(&sdp->open_wait);
 403         }
 404         mutex_unlock(&sdp->open_rel_lock);
 405         return 0;
 406 }
 407
 408 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 409 {
 410         struct sg_header __user *old_hdr = buf;
 411         int reply_len;
 412
 413         if (count >= SZ_SG_HEADER) {
 414                 /* negative reply_len means v3 format, otherwise v1/v2 */
 415                 if (get_user(reply_len, &old_hdr->reply_len))
 416                         return -EFAULT;
 417
 418                 if (reply_len >= 0)
 419                         return get_user(*pack_id, &old_hdr->pack_id);
 420
 421                 if (in_compat_syscall() &&
 422                     count >= sizeof(struct compat_sg_io_hdr)) {
 423                         struct compat_sg_io_hdr __user *hp = buf;
 424
 425                         return get_user(*pack_id, &hp->pack_id);
 426                 }
 427
 428                 if (count >= sizeof(struct sg_io_hdr)) {
 429                         struct sg_io_hdr __user *hp = buf;
 430
 431                         return get_user(*pack_id, &hp->pack_id);
 432                 }
 433         }
 434
 435         /* no valid header was passed, so ignore the pack_id */
 436         *pack_id = -1;
 437         return 0;
 438 }
 439
 440 static ssize_t
 441 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 442 {
 443         Sg_device *sdp;
 444         Sg_fd *sfp;
 445         Sg_request *srp;
 446         int req_pack_id = -1;
 447         sg_io_hdr_t *hp;
 448         struct sg_header *old_hdr;
 449         int retval;
 450
 451         /*
 452          * This could cause a response to be stranded. Close the associated
 453          * file descriptor to free up any resources being held.
 454          */
 455         retval = sg_check_file_access(filp, __func__);
 456         if (retval)
 457                 return retval;
 458
 459         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 460                 return -ENXIO;
 461         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 462                                       "sg_read: count=%d\n", (int) count));
 463
 464         if (sfp->force_packid)
 465                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 466         if (retval)
 467                 return retval;
 468
 469         srp = sg_get_rq_mark(sfp, req_pack_id);
 470         if (!srp) {             /* now wait on packet to arrive */
 471                 if (atomic_read(&sdp->detaching))
 472                         return -ENODEV;
 473                 if (filp->f_flags & O_NONBLOCK)
 474                         return -EAGAIN;
 475                 retval = wait_event_interruptible(sfp->read_wait,
 476                         (atomic_read(&sdp->detaching) ||
 477                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 478                 if (atomic_read(&sdp->detaching))
 479                         return -ENODEV;
 480                 if (retval)
 481                         /* -ERESTARTSYS as signal hit process */
 482                         return retval;
 483         }
 484         if (srp->header.interface_id != '\0')
 485                 return sg_new_read(sfp, buf, count, srp);
 486
 487         hp = &srp->header;
 488         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 489         if (!old_hdr)
 490                 return -ENOMEM;
 491
 492         old_hdr->reply_len = (int) hp->timeout;
 493         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 494         old_hdr->pack_id = hp->pack_id;
 495         old_hdr->twelve_byte =
 496             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 497         old_hdr->target_status = hp->masked_status;
 498         old_hdr->host_status = hp->host_status;
 499         old_hdr->driver_status = hp->driver_status;
 500         if ((CHECK_CONDITION & hp->masked_status) ||
 501             (DRIVER_SENSE & hp->driver_status))
 502                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 503                        sizeof (old_hdr->sense_buffer));
 504         switch (hp->host_status) {
 505         /* This setup of 'result' is for backward compatibility and is best
 506            ignored by the user who should use target, host + driver status */
 507         case DID_OK:
 508         case DID_PASSTHROUGH:
 509         case DID_SOFT_ERROR:
 510                 old_hdr->result = 0;
 511                 break;
 512         case DID_NO_CONNECT:
 513         case DID_BUS_BUSY:
 514         case DID_TIME_OUT:
 515                 old_hdr->result = EBUSY;
 516                 break;
 517         case DID_BAD_TARGET:
 518         case DID_ABORT:
 519         case DID_PARITY:
 520         case DID_RESET:
 521         case DID_BAD_INTR:
 522                 old_hdr->result = EIO;
 523                 break;
 524         case DID_ERROR:
 525                 old_hdr->result = (srp->sense_b[0] == 0 &&
 526                                   hp->masked_status == GOOD) ? 0 : EIO;
 527                 break;
 528         default:
 529                 old_hdr->result = EIO;
 530                 break;
 531         }
 532
 533         /* Now copy the result back to the user buffer.  */
 534         if (count >= SZ_SG_HEADER) {
 535                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 536                         retval = -EFAULT;
 537                         goto free_old_hdr;
 538                 }
 539                 buf += SZ_SG_HEADER;
 540                 if (count > old_hdr->reply_len)
 541                         count = old_hdr->reply_len;
 542                 if (count > SZ_SG_HEADER) {
 543                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 544                                 retval = -EFAULT;
 545                                 goto free_old_hdr;
 546                         }
 547                 }
 548         } else
 549                 count = (old_hdr->result == 0) ? 0 : -EIO;
 550         sg_finish_rem_req(srp);
 551         sg_remove_request(sfp, srp);
 552         retval = count;
 553 free_old_hdr:
 554         kfree(old_hdr);
 555         return retval;
 556 }
 557
 558 static ssize_t
 559 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 560 {
 561         sg_io_hdr_t *hp = &srp->header;
 562         int err = 0, err2;
 563         int len;
 564
 565         if (in_compat_syscall()) {
 566                 if (count < sizeof(struct compat_sg_io_hdr)) {
 567                         err = -EINVAL;
 568                         goto err_out;
 569                 }
 570         } else if (count < SZ_SG_IO_HDR) {
 571                 err = -EINVAL;
 572                 goto err_out;
 573         }
 574         hp->sb_len_wr = 0;
 575         if ((hp->mx_sb_len > 0) && hp->sbp) {
 576                 if ((CHECK_CONDITION & hp->masked_status) ||
 577                     (DRIVER_SENSE & hp->driver_status)) {
 578                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 579                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 580                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 581                         len = (len > sb_len) ? sb_len : len;
 582                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 583                                 err = -EFAULT;
 584                                 goto err_out;
 585                         }
 586                         hp->sb_len_wr = len;
 587                 }
 588         }
 589         if (hp->masked_status || hp->host_status || hp->driver_status)
 590                 hp->info |= SG_INFO_CHECK;
 591         err = put_sg_io_hdr(hp, buf);
 592 err_out:
 593         err2 = sg_finish_rem_req(srp);
 594         sg_remove_request(sfp, srp);
 595         return err ? : err2 ? : count;
 596 }
 597
 598 static ssize_t
 599 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 600 {
 601         int mxsize, cmd_size, k;
 602         int input_size, blocking;
 603         unsigned char opcode;
 604         Sg_device *sdp;
 605         Sg_fd *sfp;
 606         Sg_request *srp;
 607         struct sg_header old_hdr;
 608         sg_io_hdr_t *hp;
 609         unsigned char cmnd[SG_MAX_CDB_SIZE];
 610         int retval;
 611
 612         retval = sg_check_file_access(filp, __func__);
 613         if (retval)
 614                 return retval;
 615
 616         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 617                 return -ENXIO;
 618         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 619                                       "sg_write: count=%d\n", (int) count));
 620         if (atomic_read(&sdp->detaching))
 621                 return -ENODEV;
 622         if (!((filp->f_flags & O_NONBLOCK) ||
 623               scsi_block_when_processing_errors(sdp->device)))
 624                 return -ENXIO;
 625
 626         if (count < SZ_SG_HEADER)
 627                 return -EIO;
 628         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 629                 return -EFAULT;
 630         blocking = !(filp->f_flags & O_NONBLOCK);
 631         if (old_hdr.reply_len < 0)
 632                 return sg_new_write(sfp, filp, buf, count,
 633                                     blocking, 0, 0, NULL);
 634         if (count < (SZ_SG_HEADER + 6))
 635                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 636
 637         buf += SZ_SG_HEADER;
 638         if (get_user(opcode, buf))
 639                 return -EFAULT;
 640
 641         if (!(srp = sg_add_request(sfp))) {
 642                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 643                                               "sg_write: queue full\n"));
 644                 return -EDOM;
 645         }
 646         mutex_lock(&sfp->f_mutex);
 647         if (sfp->next_cmd_len > 0) {
 648                 cmd_size = sfp->next_cmd_len;
 649                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 650         } else {
 651                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 652                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 653                         cmd_size = 12;
 654         }
 655         mutex_unlock(&sfp->f_mutex);
 656         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 657                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 658 /* Determine buffer size.  */
 659         input_size = count - cmd_size;
 660         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 661         mxsize -= SZ_SG_HEADER;
 662         input_size -= SZ_SG_HEADER;
 663         if (input_size < 0) {
 664                 sg_remove_request(sfp, srp);
 665                 return -EIO;    /* User did not pass enough bytes for this command. */
 666         }
 667         hp = &srp->header;
 668         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 669         hp->cmd_len = (unsigned char) cmd_size;
 670         hp->iovec_count = 0;
 671         hp->mx_sb_len = 0;
 672         if (input_size > 0)
 673                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 674                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 675         else
 676                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 677         hp->dxfer_len = mxsize;
 678         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 679             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 680                 hp->dxferp = (char __user *)buf + cmd_size;
 681         else
 682                 hp->dxferp = NULL;
 683         hp->sbp = NULL;
 684         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 685         hp->flags = input_size; /* structure abuse ... */
 686         hp->pack_id = old_hdr.pack_id;
 687         hp->usr_ptr = NULL;
 688         if (copy_from_user(cmnd, buf, cmd_size))
 689                 return -EFAULT;
 690         /*
 691          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 692          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 693          * is a non-zero input_size, so emit a warning.
 694          */
 695         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 696                 printk_ratelimited(KERN_WARNING
 697                                    "sg_write: data in/out %d/%d bytes "
 698                                    "for SCSI command 0x%x-- guessing "
 699                                    "data in;\n   program %s not setting "
 700                                    "count and/or reply_len properly\n",
 701                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 702                                    input_size, (unsigned int) cmnd[0],
 703                                    current->comm);
 704         }
 705         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 706         return (k < 0) ? k : count;
 707 }
 708
 709 static ssize_t
 710 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 711                  size_t count, int blocking, int read_only, int sg_io_owned,
 712                  Sg_request **o_srp)
 713 {
 714         int k;
 715         Sg_request *srp;
 716         sg_io_hdr_t *hp;
 717         unsigned char cmnd[SG_MAX_CDB_SIZE];
 718         int timeout;
 719         unsigned long ul_timeout;
 720
 721         if (count < SZ_SG_IO_HDR)
 722                 return -EINVAL;
 723
 724         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 725         if (!(srp = sg_add_request(sfp))) {
 726                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 727                                               "sg_new_write: queue full\n"));
 728                 return -EDOM;
 729         }
 730         srp->sg_io_owned = sg_io_owned;
 731         hp = &srp->header;
 732         if (get_sg_io_hdr(hp, buf)) {
 733                 sg_remove_request(sfp, srp);
 734                 return -EFAULT;
 735         }
 736         if (hp->interface_id != 'S') {
 737                 sg_remove_request(sfp, srp);
 738                 return -ENOSYS;
 739         }
 740         if (hp->flags & SG_FLAG_MMAP_IO) {
 741                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 742                         sg_remove_request(sfp, srp);
 743                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 744                 }
 745                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 746                         sg_remove_request(sfp, srp);
 747                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 748                 }
 749                 if (sfp->res_in_use) {
 750                         sg_remove_request(sfp, srp);
 751                         return -EBUSY;  /* reserve buffer already being used */
 752                 }
 753         }
 754         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 755         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 756         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 757                 sg_remove_request(sfp, srp);
 758                 return -EMSGSIZE;
 759         }
 760         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 761                 sg_remove_request(sfp, srp);
 762                 return -EFAULT;
 763         }
 764         if (read_only && sg_allow_access(file, cmnd)) {
 765                 sg_remove_request(sfp, srp);
 766                 return -EPERM;
 767         }
 768         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 769         if (k < 0)
 770                 return k;
 771         if (o_srp)
 772                 *o_srp = srp;
 773         return count;
 774 }
 775
 776 static int
 777 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 778                 unsigned char *cmnd, int timeout, int blocking)
 779 {
 780         int k, at_head;
 781         Sg_device *sdp = sfp->parentdp;
 782         sg_io_hdr_t *hp = &srp->header;
 783
 784         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 785         hp->status = 0;
 786         hp->masked_status = 0;
 787         hp->msg_status = 0;
 788         hp->info = 0;
 789         hp->host_status = 0;
 790         hp->driver_status = 0;
 791         hp->resid = 0;
 792         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 793                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 794                         (int) cmnd[0], (int) hp->cmd_len));
 795
 796         if (hp->dxfer_len >= SZ_256M)
 797                 return -EINVAL;
 798
 799         k = sg_start_req(srp, cmnd);
 800         if (k) {
 801                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 802                         "sg_common_write: start_req err=%d\n", k));
 803                 sg_finish_rem_req(srp);
 804                 sg_remove_request(sfp, srp);
 805                 return k;       /* probably out of space --> ENOMEM */
 806         }
 807         if (atomic_read(&sdp->detaching)) {
 808                 if (srp->bio) {
 809                         scsi_req_free_cmd(scsi_req(srp->rq));
 810                         blk_put_request(srp->rq);
 811                         srp->rq = NULL;
 812                 }
 813
 814                 sg_finish_rem_req(srp);
 815                 sg_remove_request(sfp, srp);
 816                 return -ENODEV;
 817         }
 818
 819         hp->duration = jiffies_to_msecs(jiffies);
 820         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 821             (SG_FLAG_Q_AT_TAIL & hp->flags))
 822                 at_head = 0;
 823         else
 824                 at_head = 1;
 825
 826         srp->rq->timeout = timeout;
 827         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 828         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
 829                               srp->rq, at_head, sg_rq_end_io);
 830         return 0;
 831 }
 832
 833 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 834 {
 835         unsigned long flags;
 836         int ret;
 837
 838         read_lock_irqsave(&sfp->rq_list_lock, flags);
 839         ret = srp->done;
 840         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 841         return ret;
 842 }
 843
 844 static int max_sectors_bytes(struct request_queue *q)
 845 {
 846         unsigned int max_sectors = queue_max_sectors(q);
 847
 848         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 849
 850         return max_sectors << 9;
 851 }
 852
 853 static void
 854 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 855 {
 856         Sg_request *srp;
 857         int val;
 858         unsigned int ms;
 859
 860         val = 0;
 861         list_for_each_entry(srp, &sfp->rq_list, entry) {
 862                 if (val >= SG_MAX_QUEUE)
 863                         break;
 864                 rinfo[val].req_state = srp->done + 1;
 865                 rinfo[val].problem =
 866                         srp->header.masked_status &
 867                         srp->header.host_status &
 868                         srp->header.driver_status;
 869                 if (srp->done)
 870                         rinfo[val].duration =
 871                                 srp->header.duration;
 872                 else {
 873                         ms = jiffies_to_msecs(jiffies);
 874                         rinfo[val].duration =
 875                                 (ms > srp->header.duration) ?
 876                                 (ms - srp->header.duration) : 0;
 877                 }
 878                 rinfo[val].orphan = srp->orphan;
 879                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 880                 rinfo[val].pack_id = srp->header.pack_id;
 881                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 882                 val++;
 883         }
 884 }
 885
 886 #ifdef CONFIG_COMPAT
 887 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 888         char req_state;
 889         char orphan;
 890         char sg_io_owned;
 891         char problem;
 892         int pack_id;
 893         compat_uptr_t usr_ptr;
 894         unsigned int duration;
 895         int unused;
 896 };
 897
 898 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 899                                     struct sg_req_info *rinfo)
 900 {
 901         int i;
 902         for (i = 0; i < SG_MAX_QUEUE; i++) {
 903                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 904                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 905                     put_user(rinfo[i].duration, &o[i].duration) ||
 906                     put_user(rinfo[i].unused, &o[i].unused))
 907                         return -EFAULT;
 908         }
 909         return 0;
 910 }
 911 #endif
 912
 913 static long
 914 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 915                 unsigned int cmd_in, void __user *p)
 916 {
 917         int __user *ip = p;
 918         int result, val, read_only;
 919         Sg_request *srp;
 920         unsigned long iflags;
 921
 922         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 923                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 924         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 925
 926         switch (cmd_in) {
 927         case SG_IO:
 928                 if (atomic_read(&sdp->detaching))
 929                         return -ENODEV;
 930                 if (!scsi_block_when_processing_errors(sdp->device))
 931                         return -ENXIO;
 932                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 933                                  1, read_only, 1, &srp);
 934                 if (result < 0)
 935                         return result;
 936                 result = wait_event_interruptible(sfp->read_wait,
 937                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 938                 if (atomic_read(&sdp->detaching))
 939                         return -ENODEV;
 940                 write_lock_irq(&sfp->rq_list_lock);
 941                 if (srp->done) {
 942                         srp->done = 2;
 943                         write_unlock_irq(&sfp->rq_list_lock);
 944                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 945                         return (result < 0) ? result : 0;
 946                 }
 947                 srp->orphan = 1;
 948                 write_unlock_irq(&sfp->rq_list_lock);
 949                 return result;  /* -ERESTARTSYS because signal hit process */
 950         case SG_SET_TIMEOUT:
 951                 result = get_user(val, ip);
 952                 if (result)
 953                         return result;
 954                 if (val < 0)
 955                         return -EIO;
 956                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 957                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 958                                     INT_MAX);
 959                 sfp->timeout_user = val;
 960                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 961
 962                 return 0;
 963         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 964                                 /* strange ..., for backward compatibility */
 965                 return sfp->timeout_user;
 966         case SG_SET_FORCE_LOW_DMA:
 967                 /*
 968                  * N.B. This ioctl never worked properly, but failed to
 969                  * return an error value. So returning '0' to keep compability
 970                  * with legacy applications.
 971                  */
 972                 return 0;
 973         case SG_GET_LOW_DMA:
 974                 return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
 975         case SG_GET_SCSI_ID:
 976                 {
 977                         sg_scsi_id_t v;
 978
 979                         if (atomic_read(&sdp->detaching))
 980                                 return -ENODEV;
 981                         memset(&v, 0, sizeof(v));
 982                         v.host_no = sdp->device->host->host_no;
 983                         v.channel = sdp->device->channel;
 984                         v.scsi_id = sdp->device->id;
 985                         v.lun = sdp->device->lun;
 986                         v.scsi_type = sdp->device->type;
 987                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 988                         v.d_queue_depth = sdp->device->queue_depth;
 989                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 990                                 return -EFAULT;
 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                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1001                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1002                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1003                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1004                                                        iflags);
1005                                 return put_user(srp->header.pack_id, ip);
1006                         }
1007                 }
1008                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1009                 return put_user(-1, ip);
1010         case SG_GET_NUM_WAITING:
1011                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1012                 val = 0;
1013                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1014                         if ((1 == srp->done) && (!srp->sg_io_owned))
1015                                 ++val;
1016                 }
1017                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1018                 return put_user(val, ip);
1019         case SG_GET_SG_TABLESIZE:
1020                 return put_user(sdp->sg_tablesize, ip);
1021         case SG_SET_RESERVED_SIZE:
1022                 result = get_user(val, ip);
1023                 if (result)
1024                         return result;
1025                 if (val < 0)
1026                         return -EINVAL;
1027                 val = min_t(int, val,
1028                             max_sectors_bytes(sdp->device->request_queue));
1029                 mutex_lock(&sfp->f_mutex);
1030                 if (val != sfp->reserve.bufflen) {
1031                         if (sfp->mmap_called ||
1032                             sfp->res_in_use) {
1033                                 mutex_unlock(&sfp->f_mutex);
1034                                 return -EBUSY;
1035                         }
1036
1037                         sg_remove_scat(sfp, &sfp->reserve);
1038                         sg_build_reserve(sfp, val);
1039                 }
1040                 mutex_unlock(&sfp->f_mutex);
1041                 return 0;
1042         case SG_GET_RESERVED_SIZE:
1043                 val = min_t(int, sfp->reserve.bufflen,
1044                             max_sectors_bytes(sdp->device->request_queue));
1045                 return put_user(val, ip);
1046         case SG_SET_COMMAND_Q:
1047                 result = get_user(val, ip);
1048                 if (result)
1049                         return result;
1050                 sfp->cmd_q = val ? 1 : 0;
1051                 return 0;
1052         case SG_GET_COMMAND_Q:
1053                 return put_user((int) sfp->cmd_q, ip);
1054         case SG_SET_KEEP_ORPHAN:
1055                 result = get_user(val, ip);
1056                 if (result)
1057                         return result;
1058                 sfp->keep_orphan = val;
1059                 return 0;
1060         case SG_GET_KEEP_ORPHAN:
1061                 return put_user((int) sfp->keep_orphan, ip);
1062         case SG_NEXT_CMD_LEN:
1063                 result = get_user(val, ip);
1064                 if (result)
1065                         return result;
1066                 if (val > SG_MAX_CDB_SIZE)
1067                         return -ENOMEM;
1068                 sfp->next_cmd_len = (val > 0) ? val : 0;
1069                 return 0;
1070         case SG_GET_VERSION_NUM:
1071                 return put_user(sg_version_num, ip);
1072         case SG_GET_ACCESS_COUNT:
1073                 /* faked - we don't have a real access count anymore */
1074                 val = (sdp->device ? 1 : 0);
1075                 return put_user(val, ip);
1076         case SG_GET_REQUEST_TABLE:
1077                 {
1078                         sg_req_info_t *rinfo;
1079
1080                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1081                                         GFP_KERNEL);
1082                         if (!rinfo)
1083                                 return -ENOMEM;
1084                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1085                         sg_fill_request_table(sfp, rinfo);
1086                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1087         #ifdef CONFIG_COMPAT
1088                         if (in_compat_syscall())
1089                                 result = put_compat_request_table(p, rinfo);
1090                         else
1091         #endif
1092                                 result = copy_to_user(p, rinfo,
1093                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1094                         result = result ? -EFAULT : 0;
1095                         kfree(rinfo);
1096                         return result;
1097                 }
1098         case SG_EMULATED_HOST:
1099                 if (atomic_read(&sdp->detaching))
1100                         return -ENODEV;
1101                 return put_user(sdp->device->host->hostt->emulated, ip);
1102         case SCSI_IOCTL_SEND_COMMAND:
1103                 if (atomic_read(&sdp->detaching))
1104                         return -ENODEV;
1105                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1106         case SG_SET_DEBUG:
1107                 result = get_user(val, ip);
1108                 if (result)
1109                         return result;
1110                 sdp->sgdebug = (char) val;
1111                 return 0;
1112         case BLKSECTGET:
1113                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1114                                 ip);
1115         case BLKTRACESETUP:
1116                 return blk_trace_setup(sdp->device->request_queue,
1117                                        sdp->disk->disk_name,
1118                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1119                                        NULL, p);
1120         case BLKTRACESTART:
1121                 return blk_trace_startstop(sdp->device->request_queue, 1);
1122         case BLKTRACESTOP:
1123                 return blk_trace_startstop(sdp->device->request_queue, 0);
1124         case BLKTRACETEARDOWN:
1125                 return blk_trace_remove(sdp->device->request_queue);
1126         case SCSI_IOCTL_GET_IDLUN:
1127         case SCSI_IOCTL_GET_BUS_NUMBER:
1128         case SCSI_IOCTL_PROBE_HOST:
1129         case SG_GET_TRANSFORM:
1130         case SG_SCSI_RESET:
1131                 if (atomic_read(&sdp->detaching))
1132                         return -ENODEV;
1133                 break;
1134         default:
1135                 if (read_only)
1136                         return -EPERM;  /* don't know so take safe approach */
1137                 break;
1138         }
1139
1140         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1141                         cmd_in, filp->f_flags & O_NDELAY);
1142         if (result)
1143                 return result;
1144
1145         return -ENOIOCTLCMD;
1146 }
1147
1148 static long
1149 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1150 {
1151         void __user *p = (void __user *)arg;
1152         Sg_device *sdp;
1153         Sg_fd *sfp;
1154         int ret;
1155
1156         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1157                 return -ENXIO;
1158
1159         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1160         if (ret != -ENOIOCTLCMD)
1161                 return ret;
1162
1163         return scsi_ioctl(sdp->device, cmd_in, p);
1164 }
1165
1166 #ifdef CONFIG_COMPAT
1167 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1168 {
1169         void __user *p = compat_ptr(arg);
1170         Sg_device *sdp;
1171         Sg_fd *sfp;
1172         int ret;
1173
1174         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1175                 return -ENXIO;
1176
1177         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1178         if (ret != -ENOIOCTLCMD)
1179                 return ret;
1180
1181         return scsi_compat_ioctl(sdp->device, cmd_in, p);
1182 }
1183 #endif
1184
1185 static __poll_t
1186 sg_poll(struct file *filp, poll_table * wait)
1187 {
1188         __poll_t res = 0;
1189         Sg_device *sdp;
1190         Sg_fd *sfp;
1191         Sg_request *srp;
1192         int count = 0;
1193         unsigned long iflags;
1194
1195         sfp = filp->private_data;
1196         if (!sfp)
1197                 return EPOLLERR;
1198         sdp = sfp->parentdp;
1199         if (!sdp)
1200                 return EPOLLERR;
1201         poll_wait(filp, &sfp->read_wait, wait);
1202         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1203         list_for_each_entry(srp, &sfp->rq_list, entry) {
1204                 /* if any read waiting, flag it */
1205                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1206                         res = EPOLLIN | EPOLLRDNORM;
1207                 ++count;
1208         }
1209         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1210
1211         if (atomic_read(&sdp->detaching))
1212                 res |= EPOLLHUP;
1213         else if (!sfp->cmd_q) {
1214                 if (0 == count)
1215                         res |= EPOLLOUT | EPOLLWRNORM;
1216         } else if (count < SG_MAX_QUEUE)
1217                 res |= EPOLLOUT | EPOLLWRNORM;
1218         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1219                                       "sg_poll: res=0x%x\n", (__force u32) res));
1220         return res;
1221 }
1222
1223 static int
1224 sg_fasync(int fd, struct file *filp, int mode)
1225 {
1226         Sg_device *sdp;
1227         Sg_fd *sfp;
1228
1229         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1230                 return -ENXIO;
1231         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1232                                       "sg_fasync: mode=%d\n", mode));
1233
1234         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1235 }
1236
1237 static vm_fault_t
1238 sg_vma_fault(struct vm_fault *vmf)
1239 {
1240         struct vm_area_struct *vma = vmf->vma;
1241         Sg_fd *sfp;
1242         unsigned long offset, len, sa;
1243         Sg_scatter_hold *rsv_schp;
1244         int k, length;
1245
1246         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1247                 return VM_FAULT_SIGBUS;
1248         rsv_schp = &sfp->reserve;
1249         offset = vmf->pgoff << PAGE_SHIFT;
1250         if (offset >= rsv_schp->bufflen)
1251                 return VM_FAULT_SIGBUS;
1252         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1253                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1254                                       offset, rsv_schp->k_use_sg));
1255         sa = vma->vm_start;
1256         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1257         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1258                 len = vma->vm_end - sa;
1259                 len = (len < length) ? len : length;
1260                 if (offset < len) {
1261                         struct page *page = nth_page(rsv_schp->pages[k],
1262                                                      offset >> PAGE_SHIFT);
1263                         get_page(page); /* increment page count */
1264                         vmf->page = page;
1265                         return 0; /* success */
1266                 }
1267                 sa += len;
1268                 offset -= len;
1269         }
1270
1271         return VM_FAULT_SIGBUS;
1272 }
1273
1274 static const struct vm_operations_struct sg_mmap_vm_ops = {
1275         .fault = sg_vma_fault,
1276 };
1277
1278 static int
1279 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1280 {
1281         Sg_fd *sfp;
1282         unsigned long req_sz, len, sa;
1283         Sg_scatter_hold *rsv_schp;
1284         int k, length;
1285         int ret = 0;
1286
1287         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1288                 return -ENXIO;
1289         req_sz = vma->vm_end - vma->vm_start;
1290         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1291                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1292                                       (void *) vma->vm_start, (int) req_sz));
1293         if (vma->vm_pgoff)
1294                 return -EINVAL; /* want no offset */
1295         rsv_schp = &sfp->reserve;
1296         mutex_lock(&sfp->f_mutex);
1297         if (req_sz > rsv_schp->bufflen) {
1298                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1299                 goto out;
1300         }
1301
1302         sa = vma->vm_start;
1303         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1304         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1305                 len = vma->vm_end - sa;
1306                 len = (len < length) ? len : length;
1307                 sa += len;
1308         }
1309
1310         sfp->mmap_called = 1;
1311         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1312         vma->vm_private_data = sfp;
1313         vma->vm_ops = &sg_mmap_vm_ops;
1314 out:
1315         mutex_unlock(&sfp->f_mutex);
1316         return ret;
1317 }
1318
1319 static void
1320 sg_rq_end_io_usercontext(struct work_struct *work)
1321 {
1322         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1323         struct sg_fd *sfp = srp->parentfp;
1324
1325         sg_finish_rem_req(srp);
1326         sg_remove_request(sfp, srp);
1327         kref_put(&sfp->f_ref, sg_remove_sfp);
1328 }
1329
1330 /*
1331  * This function is a "bottom half" handler that is called by the mid
1332  * level when a command is completed (or has failed).
1333  */
1334 static void
1335 sg_rq_end_io(struct request *rq, blk_status_t status)
1336 {
1337         struct sg_request *srp = rq->end_io_data;
1338         struct scsi_request *req = scsi_req(rq);
1339         Sg_device *sdp;
1340         Sg_fd *sfp;
1341         unsigned long iflags;
1342         unsigned int ms;
1343         char *sense;
1344         int result, resid, done = 1;
1345
1346         if (WARN_ON(srp->done != 0))
1347                 return;
1348
1349         sfp = srp->parentfp;
1350         if (WARN_ON(sfp == NULL))
1351                 return;
1352
1353         sdp = sfp->parentdp;
1354         if (unlikely(atomic_read(&sdp->detaching)))
1355                 pr_info("%s: device detaching\n", __func__);
1356
1357         sense = req->sense;
1358         result = req->result;
1359         resid = req->resid_len;
1360
1361         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1362                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1363                                       srp->header.pack_id, result));
1364         srp->header.resid = resid;
1365         ms = jiffies_to_msecs(jiffies);
1366         srp->header.duration = (ms > srp->header.duration) ?
1367                                 (ms - srp->header.duration) : 0;
1368         if (0 != result) {
1369                 struct scsi_sense_hdr sshdr;
1370
1371                 srp->header.status = 0xff & result;
1372                 srp->header.masked_status = status_byte(result);
1373                 srp->header.msg_status = msg_byte(result);
1374                 srp->header.host_status = host_byte(result);
1375                 srp->header.driver_status = driver_byte(result);
1376                 if ((sdp->sgdebug > 0) &&
1377                     ((CHECK_CONDITION == srp->header.masked_status) ||
1378                      (COMMAND_TERMINATED == srp->header.masked_status)))
1379                         __scsi_print_sense(sdp->device, __func__, sense,
1380                                            SCSI_SENSE_BUFFERSIZE);
1381
1382                 /* Following if statement is a patch supplied by Eric Youngdale */
1383                 if (driver_byte(result) != 0
1384                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1385                     && !scsi_sense_is_deferred(&sshdr)
1386                     && sshdr.sense_key == UNIT_ATTENTION
1387                     && sdp->device->removable) {
1388                         /* Detected possible disc change. Set the bit - this */
1389                         /* may be used if there are filesystems using this device */
1390                         sdp->device->changed = 1;
1391                 }
1392         }
1393
1394         if (req->sense_len)
1395                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1396
1397         /* Rely on write phase to clean out srp status values, so no "else" */
1398
1399         /*
1400          * Free the request as soon as it is complete so that its resources
1401          * can be reused without waiting for userspace to read() the
1402          * result.  But keep the associated bio (if any) around until
1403          * blk_rq_unmap_user() can be called from user context.
1404          */
1405         srp->rq = NULL;
1406         scsi_req_free_cmd(scsi_req(rq));
1407         blk_put_request(rq);
1408
1409         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1410         if (unlikely(srp->orphan)) {
1411                 if (sfp->keep_orphan)
1412                         srp->sg_io_owned = 0;
1413                 else
1414                         done = 0;
1415         }
1416         srp->done = done;
1417         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1418
1419         if (likely(done)) {
1420                 /* Now wake up any sg_read() that is waiting for this
1421                  * packet.
1422                  */
1423                 wake_up_interruptible(&sfp->read_wait);
1424                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1425                 kref_put(&sfp->f_ref, sg_remove_sfp);
1426         } else {
1427                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1428                 schedule_work(&srp->ew.work);
1429         }
1430 }
1431
1432 static const struct file_operations sg_fops = {
1433         .owner = THIS_MODULE,
1434         .read = sg_read,
1435         .write = sg_write,
1436         .poll = sg_poll,
1437         .unlocked_ioctl = sg_ioctl,
1438 #ifdef CONFIG_COMPAT
1439         .compat_ioctl = sg_compat_ioctl,
1440 #endif
1441         .open = sg_open,
1442         .mmap = sg_mmap,
1443         .release = sg_release,
1444         .fasync = sg_fasync,
1445         .llseek = no_llseek,
1446 };
1447
1448 static struct class *sg_sysfs_class;
1449
1450 static int sg_sysfs_valid = 0;
1451
1452 static Sg_device *
1453 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1454 {
1455         struct request_queue *q = scsidp->request_queue;
1456         Sg_device *sdp;
1457         unsigned long iflags;
1458         int error;
1459         u32 k;
1460
1461         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1462         if (!sdp) {
1463                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1464                             "failure\n", __func__);
1465                 return ERR_PTR(-ENOMEM);
1466         }
1467
1468         idr_preload(GFP_KERNEL);
1469         write_lock_irqsave(&sg_index_lock, iflags);
1470
1471         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1472         if (error < 0) {
1473                 if (error == -ENOSPC) {
1474                         sdev_printk(KERN_WARNING, scsidp,
1475                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1476                                     scsidp->type, SG_MAX_DEVS - 1);
1477                         error = -ENODEV;
1478                 } else {
1479                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1480                                     "allocation Sg_device failure: %d\n",
1481                                     __func__, error);
1482                 }
1483                 goto out_unlock;
1484         }
1485         k = error;
1486
1487         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1488                                         "sg_alloc: dev=%d \n", k));
1489         sprintf(disk->disk_name, "sg%d", k);
1490         disk->first_minor = k;
1491         sdp->disk = disk;
1492         sdp->device = scsidp;
1493         mutex_init(&sdp->open_rel_lock);
1494         INIT_LIST_HEAD(&sdp->sfds);
1495         init_waitqueue_head(&sdp->open_wait);
1496         atomic_set(&sdp->detaching, 0);
1497         rwlock_init(&sdp->sfd_lock);
1498         sdp->sg_tablesize = queue_max_segments(q);
1499         sdp->index = k;
1500         kref_init(&sdp->d_ref);
1501         error = 0;
1502
1503 out_unlock:
1504         write_unlock_irqrestore(&sg_index_lock, iflags);
1505         idr_preload_end();
1506
1507         if (error) {
1508                 kfree(sdp);
1509                 return ERR_PTR(error);
1510         }
1511         return sdp;
1512 }
1513
1514 static int
1515 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1516 {
1517         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1518         struct gendisk *disk;
1519         Sg_device *sdp = NULL;
1520         struct cdev * cdev = NULL;
1521         int error;
1522         unsigned long iflags;
1523
1524         disk = alloc_disk(1);
1525         if (!disk) {
1526                 pr_warn("%s: alloc_disk failed\n", __func__);
1527                 return -ENOMEM;
1528         }
1529         disk->major = SCSI_GENERIC_MAJOR;
1530
1531         error = -ENOMEM;
1532         cdev = cdev_alloc();
1533         if (!cdev) {
1534                 pr_warn("%s: cdev_alloc failed\n", __func__);
1535                 goto out;
1536         }
1537         cdev->owner = THIS_MODULE;
1538         cdev->ops = &sg_fops;
1539
1540         sdp = sg_alloc(disk, scsidp);
1541         if (IS_ERR(sdp)) {
1542                 pr_warn("%s: sg_alloc failed\n", __func__);
1543                 error = PTR_ERR(sdp);
1544                 goto out;
1545         }
1546
1547         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1548         if (error)
1549                 goto cdev_add_err;
1550
1551         sdp->cdev = cdev;
1552         if (sg_sysfs_valid) {
1553                 struct device *sg_class_member;
1554
1555                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1556                                                 MKDEV(SCSI_GENERIC_MAJOR,
1557                                                       sdp->index),
1558                                                 sdp, "%s", disk->disk_name);
1559                 if (IS_ERR(sg_class_member)) {
1560                         pr_err("%s: device_create failed\n", __func__);
1561                         error = PTR_ERR(sg_class_member);
1562                         goto cdev_add_err;
1563                 }
1564                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1565                                           &sg_class_member->kobj, "generic");
1566                 if (error)
1567                         pr_err("%s: unable to make symlink 'generic' back "
1568                                "to sg%d\n", __func__, sdp->index);
1569         } else
1570                 pr_warn("%s: sg_sys Invalid\n", __func__);
1571
1572         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1573                     "type %d\n", sdp->index, scsidp->type);
1574
1575         dev_set_drvdata(cl_dev, sdp);
1576
1577         return 0;
1578
1579 cdev_add_err:
1580         write_lock_irqsave(&sg_index_lock, iflags);
1581         idr_remove(&sg_index_idr, sdp->index);
1582         write_unlock_irqrestore(&sg_index_lock, iflags);
1583         kfree(sdp);
1584
1585 out:
1586         put_disk(disk);
1587         if (cdev)
1588                 cdev_del(cdev);
1589         return error;
1590 }
1591
1592 static void
1593 sg_device_destroy(struct kref *kref)
1594 {
1595         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1596         unsigned long flags;
1597
1598         /* CAUTION!  Note that the device can still be found via idr_find()
1599          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1600          * any other cleanup.
1601          */
1602
1603         write_lock_irqsave(&sg_index_lock, flags);
1604         idr_remove(&sg_index_idr, sdp->index);
1605         write_unlock_irqrestore(&sg_index_lock, flags);
1606
1607         SCSI_LOG_TIMEOUT(3,
1608                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1609
1610         put_disk(sdp->disk);
1611         kfree(sdp);
1612 }
1613
1614 static void
1615 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1616 {
1617         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1618         Sg_device *sdp = dev_get_drvdata(cl_dev);
1619         unsigned long iflags;
1620         Sg_fd *sfp;
1621         int val;
1622
1623         if (!sdp)
1624                 return;
1625         /* want sdp->detaching non-zero as soon as possible */
1626         val = atomic_inc_return(&sdp->detaching);
1627         if (val > 1)
1628                 return; /* only want to do following once per device */
1629
1630         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1631                                       "%s\n", __func__));
1632
1633         read_lock_irqsave(&sdp->sfd_lock, iflags);
1634         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1635                 wake_up_interruptible_all(&sfp->read_wait);
1636                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1637         }
1638         wake_up_interruptible_all(&sdp->open_wait);
1639         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1640
1641         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1642         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1643         cdev_del(sdp->cdev);
1644         sdp->cdev = NULL;
1645
1646         kref_put(&sdp->d_ref, sg_device_destroy);
1647 }
1648
1649 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1650 module_param_named(def_reserved_size, def_reserved_size, int,
1651                    S_IRUGO | S_IWUSR);
1652 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1653
1654 MODULE_AUTHOR("Douglas Gilbert");
1655 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1656 MODULE_LICENSE("GPL");
1657 MODULE_VERSION(SG_VERSION_STR);
1658 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1659
1660 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1661                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1662 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1663 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1664
1665 static int __init
1666 init_sg(void)
1667 {
1668         int rc;
1669
1670         if (scatter_elem_sz < PAGE_SIZE) {
1671                 scatter_elem_sz = PAGE_SIZE;
1672                 scatter_elem_sz_prev = scatter_elem_sz;
1673         }
1674         if (def_reserved_size >= 0)
1675                 sg_big_buff = def_reserved_size;
1676         else
1677                 def_reserved_size = sg_big_buff;
1678
1679         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1680                                     SG_MAX_DEVS, "sg");
1681         if (rc)
1682                 return rc;
1683         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1684         if ( IS_ERR(sg_sysfs_class) ) {
1685                 rc = PTR_ERR(sg_sysfs_class);
1686                 goto err_out;
1687         }
1688         sg_sysfs_valid = 1;
1689         rc = scsi_register_interface(&sg_interface);
1690         if (0 == rc) {
1691 #ifdef CONFIG_SCSI_PROC_FS
1692                 sg_proc_init();
1693 #endif                          /* CONFIG_SCSI_PROC_FS */
1694                 return 0;
1695         }
1696         class_destroy(sg_sysfs_class);
1697 err_out:
1698         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1699         return rc;
1700 }
1701
1702 static void __exit
1703 exit_sg(void)
1704 {
1705 #ifdef CONFIG_SCSI_PROC_FS
1706         remove_proc_subtree("scsi/sg", NULL);
1707 #endif                          /* CONFIG_SCSI_PROC_FS */
1708         scsi_unregister_interface(&sg_interface);
1709         class_destroy(sg_sysfs_class);
1710         sg_sysfs_valid = 0;
1711         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1712                                  SG_MAX_DEVS);
1713         idr_destroy(&sg_index_idr);
1714 }
1715
1716 static int
1717 sg_start_req(Sg_request *srp, unsigned char *cmd)
1718 {
1719         int res;
1720         struct request *rq;
1721         struct scsi_request *req;
1722         Sg_fd *sfp = srp->parentfp;
1723         sg_io_hdr_t *hp = &srp->header;
1724         int dxfer_len = (int) hp->dxfer_len;
1725         int dxfer_dir = hp->dxfer_direction;
1726         unsigned int iov_count = hp->iovec_count;
1727         Sg_scatter_hold *req_schp = &srp->data;
1728         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1729         struct request_queue *q = sfp->parentdp->device->request_queue;
1730         struct rq_map_data *md, map_data;
1731         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1732         unsigned char *long_cmdp = NULL;
1733
1734         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1735                                       "sg_start_req: dxfer_len=%d\n",
1736                                       dxfer_len));
1737
1738         if (hp->cmd_len > BLK_MAX_CDB) {
1739                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1740                 if (!long_cmdp)
1741                         return -ENOMEM;
1742         }
1743
1744         /*
1745          * NOTE
1746          *
1747          * With scsi-mq enabled, there are a fixed number of preallocated
1748          * requests equal in number to shost->can_queue.  If all of the
1749          * preallocated requests are already in use, then blk_get_request()
1750          * will sleep until an active command completes, freeing up a request.
1751          * Although waiting in an asynchronous interface is less than ideal, we
1752          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1753          * not expect an EWOULDBLOCK from this condition.
1754          */
1755         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1756                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
1757         if (IS_ERR(rq)) {
1758                 kfree(long_cmdp);
1759                 return PTR_ERR(rq);
1760         }
1761         req = scsi_req(rq);
1762
1763         if (hp->cmd_len > BLK_MAX_CDB)
1764                 req->cmd = long_cmdp;
1765         memcpy(req->cmd, cmd, hp->cmd_len);
1766         req->cmd_len = hp->cmd_len;
1767
1768         srp->rq = rq;
1769         rq->end_io_data = srp;
1770         req->retries = SG_DEFAULT_RETRIES;
1771
1772         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1773                 return 0;
1774
1775         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1776             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1777             !sfp->parentdp->device->host->unchecked_isa_dma &&
1778             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1779                 md = NULL;
1780         else
1781                 md = &map_data;
1782
1783         if (md) {
1784                 mutex_lock(&sfp->f_mutex);
1785                 if (dxfer_len <= rsv_schp->bufflen &&
1786                     !sfp->res_in_use) {
1787                         sfp->res_in_use = 1;
1788                         sg_link_reserve(sfp, srp, dxfer_len);
1789                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1790                         res = -EBUSY; /* sfp->res_in_use == 1 */
1791                         if (dxfer_len > rsv_schp->bufflen)
1792                                 res = -ENOMEM;
1793                         mutex_unlock(&sfp->f_mutex);
1794                         return res;
1795                 } else {
1796                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1797                         if (res) {
1798                                 mutex_unlock(&sfp->f_mutex);
1799                                 return res;
1800                         }
1801                 }
1802                 mutex_unlock(&sfp->f_mutex);
1803
1804                 md->pages = req_schp->pages;
1805                 md->page_order = req_schp->page_order;
1806                 md->nr_entries = req_schp->k_use_sg;
1807                 md->offset = 0;
1808                 md->null_mapped = hp->dxferp ? 0 : 1;
1809                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1810                         md->from_user = 1;
1811                 else
1812                         md->from_user = 0;
1813         }
1814
1815         if (iov_count) {
1816                 struct iovec *iov = NULL;
1817                 struct iov_iter i;
1818
1819 #ifdef CONFIG_COMPAT
1820                 if (in_compat_syscall())
1821                         res = compat_import_iovec(rw, hp->dxferp, iov_count,
1822                                                   0, &iov, &i);
1823                 else
1824 #endif
1825                         res = import_iovec(rw, hp->dxferp, iov_count,
1826                                            0, &iov, &i);
1827                 if (res < 0)
1828                         return res;
1829
1830                 iov_iter_truncate(&i, hp->dxfer_len);
1831                 if (!iov_iter_count(&i)) {
1832                         kfree(iov);
1833                         return -EINVAL;
1834                 }
1835
1836                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1837                 kfree(iov);
1838         } else
1839                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1840                                       hp->dxfer_len, GFP_ATOMIC);
1841
1842         if (!res) {
1843                 srp->bio = rq->bio;
1844
1845                 if (!md) {
1846                         req_schp->dio_in_use = 1;
1847                         hp->info |= SG_INFO_DIRECT_IO;
1848                 }
1849         }
1850         return res;
1851 }
1852
1853 static int
1854 sg_finish_rem_req(Sg_request *srp)
1855 {
1856         int ret = 0;
1857
1858         Sg_fd *sfp = srp->parentfp;
1859         Sg_scatter_hold *req_schp = &srp->data;
1860
1861         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1862                                       "sg_finish_rem_req: res_used=%d\n",
1863                                       (int) srp->res_used));
1864         if (srp->bio)
1865                 ret = blk_rq_unmap_user(srp->bio);
1866
1867         if (srp->rq) {
1868                 scsi_req_free_cmd(scsi_req(srp->rq));
1869                 blk_put_request(srp->rq);
1870         }
1871
1872         if (srp->res_used)
1873                 sg_unlink_reserve(sfp, srp);
1874         else
1875                 sg_remove_scat(sfp, req_schp);
1876
1877         return ret;
1878 }
1879
1880 static int
1881 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1882 {
1883         int sg_bufflen = tablesize * sizeof(struct page *);
1884         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1885
1886         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1887         if (!schp->pages)
1888                 return -ENOMEM;
1889         schp->sglist_len = sg_bufflen;
1890         return tablesize;       /* number of scat_gath elements allocated */
1891 }
1892
1893 static int
1894 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1895 {
1896         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1897         int sg_tablesize = sfp->parentdp->sg_tablesize;
1898         int blk_size = buff_size, order;
1899         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1900         struct sg_device *sdp = sfp->parentdp;
1901
1902         if (blk_size < 0)
1903                 return -EFAULT;
1904         if (0 == blk_size)
1905                 ++blk_size;     /* don't know why */
1906         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1907         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1908         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1909                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1910                 buff_size, blk_size));
1911
1912         /* N.B. ret_sz carried into this block ... */
1913         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1914         if (mx_sc_elems < 0)
1915                 return mx_sc_elems;     /* most likely -ENOMEM */
1916
1917         num = scatter_elem_sz;
1918         if (unlikely(num != scatter_elem_sz_prev)) {
1919                 if (num < PAGE_SIZE) {
1920                         scatter_elem_sz = PAGE_SIZE;
1921                         scatter_elem_sz_prev = PAGE_SIZE;
1922                 } else
1923                         scatter_elem_sz_prev = num;
1924         }
1925
1926         if (sdp->device->host->unchecked_isa_dma)
1927                 gfp_mask |= GFP_DMA;
1928
1929         order = get_order(num);
1930 retry:
1931         ret_sz = 1 << (PAGE_SHIFT + order);
1932
1933         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1934              k++, rem_sz -= ret_sz) {
1935
1936                 num = (rem_sz > scatter_elem_sz_prev) ?
1937                         scatter_elem_sz_prev : rem_sz;
1938
1939                 schp->pages[k] = alloc_pages(gfp_mask, order);
1940                 if (!schp->pages[k])
1941                         goto out;
1942
1943                 if (num == scatter_elem_sz_prev) {
1944                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1945                                 scatter_elem_sz = ret_sz;
1946                                 scatter_elem_sz_prev = ret_sz;
1947                         }
1948                 }
1949
1950                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1951                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1952                                  k, num, ret_sz));
1953         }               /* end of for loop */
1954
1955         schp->page_order = order;
1956         schp->k_use_sg = k;
1957         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1958                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1959                          k, rem_sz));
1960
1961         schp->bufflen = blk_size;
1962         if (rem_sz > 0) /* must have failed */
1963                 return -ENOMEM;
1964         return 0;
1965 out:
1966         for (i = 0; i < k; i++)
1967                 __free_pages(schp->pages[i], order);
1968
1969         if (--order >= 0)
1970                 goto retry;
1971
1972         return -ENOMEM;
1973 }
1974
1975 static void
1976 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1977 {
1978         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1979                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1980         if (schp->pages && schp->sglist_len > 0) {
1981                 if (!schp->dio_in_use) {
1982                         int k;
1983
1984                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1985                                 SCSI_LOG_TIMEOUT(5,
1986                                         sg_printk(KERN_INFO, sfp->parentdp,
1987                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1988                                         k, schp->pages[k]));
1989                                 __free_pages(schp->pages[k], schp->page_order);
1990                         }
1991
1992                         kfree(schp->pages);
1993                 }
1994         }
1995         memset(schp, 0, sizeof (*schp));
1996 }
1997
1998 static int
1999 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
2000 {
2001         Sg_scatter_hold *schp = &srp->data;
2002         int k, num;
2003
2004         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2005                          "sg_read_oxfer: num_read_xfer=%d\n",
2006                          num_read_xfer));
2007         if ((!outp) || (num_read_xfer <= 0))
2008                 return 0;
2009
2010         num = 1 << (PAGE_SHIFT + schp->page_order);
2011         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2012                 if (num > num_read_xfer) {
2013                         if (copy_to_user(outp, page_address(schp->pages[k]),
2014                                            num_read_xfer))
2015                                 return -EFAULT;
2016                         break;
2017                 } else {
2018                         if (copy_to_user(outp, page_address(schp->pages[k]),
2019                                            num))
2020                                 return -EFAULT;
2021                         num_read_xfer -= num;
2022                         if (num_read_xfer <= 0)
2023                                 break;
2024                         outp += num;
2025                 }
2026         }
2027
2028         return 0;
2029 }
2030
2031 static void
2032 sg_build_reserve(Sg_fd * sfp, int req_size)
2033 {
2034         Sg_scatter_hold *schp = &sfp->reserve;
2035
2036         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2037                          "sg_build_reserve: req_size=%d\n", req_size));
2038         do {
2039                 if (req_size < PAGE_SIZE)
2040                         req_size = PAGE_SIZE;
2041                 if (0 == sg_build_indirect(schp, sfp, req_size))
2042                         return;
2043                 else
2044                         sg_remove_scat(sfp, schp);
2045                 req_size >>= 1; /* divide by 2 */
2046         } while (req_size > (PAGE_SIZE / 2));
2047 }
2048
2049 static void
2050 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2051 {
2052         Sg_scatter_hold *req_schp = &srp->data;
2053         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2054         int k, num, rem;
2055
2056         srp->res_used = 1;
2057         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2058                          "sg_link_reserve: size=%d\n", size));
2059         rem = size;
2060
2061         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2062         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2063                 if (rem <= num) {
2064                         req_schp->k_use_sg = k + 1;
2065                         req_schp->sglist_len = rsv_schp->sglist_len;
2066                         req_schp->pages = rsv_schp->pages;
2067
2068                         req_schp->bufflen = size;
2069                         req_schp->page_order = rsv_schp->page_order;
2070                         break;
2071                 } else
2072                         rem -= num;
2073         }
2074
2075         if (k >= rsv_schp->k_use_sg)
2076                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2077                                  "sg_link_reserve: BAD size\n"));
2078 }
2079
2080 static void
2081 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2082 {
2083         Sg_scatter_hold *req_schp = &srp->data;
2084
2085         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2086                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2087                                       (int) req_schp->k_use_sg));
2088         req_schp->k_use_sg = 0;
2089         req_schp->bufflen = 0;
2090         req_schp->pages = NULL;
2091         req_schp->page_order = 0;
2092         req_schp->sglist_len = 0;
2093         srp->res_used = 0;
2094         /* Called without mutex lock to avoid deadlock */
2095         sfp->res_in_use = 0;
2096 }
2097
2098 static Sg_request *
2099 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2100 {
2101         Sg_request *resp;
2102         unsigned long iflags;
2103
2104         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2105         list_for_each_entry(resp, &sfp->rq_list, entry) {
2106                 /* look for requests that are ready + not SG_IO owned */
2107                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2108                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2109                         resp->done = 2; /* guard against other readers */
2110                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2111                         return resp;
2112                 }
2113         }
2114         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2115         return NULL;
2116 }
2117
2118 /* always adds to end of list */
2119 static Sg_request *
2120 sg_add_request(Sg_fd * sfp)
2121 {
2122         int k;
2123         unsigned long iflags;
2124         Sg_request *rp = sfp->req_arr;
2125
2126         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2127         if (!list_empty(&sfp->rq_list)) {
2128                 if (!sfp->cmd_q)
2129                         goto out_unlock;
2130
2131                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2132                         if (!rp->parentfp)
2133                                 break;
2134                 }
2135                 if (k >= SG_MAX_QUEUE)
2136                         goto out_unlock;
2137         }
2138         memset(rp, 0, sizeof (Sg_request));
2139         rp->parentfp = sfp;
2140         rp->header.duration = jiffies_to_msecs(jiffies);
2141         list_add_tail(&rp->entry, &sfp->rq_list);
2142         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2143         return rp;
2144 out_unlock:
2145         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2146         return NULL;
2147 }
2148
2149 /* Return of 1 for found; 0 for not found */
2150 static int
2151 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2152 {
2153         unsigned long iflags;
2154         int res = 0;
2155
2156         if (!sfp || !srp || list_empty(&sfp->rq_list))
2157                 return res;
2158         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2159         if (!list_empty(&srp->entry)) {
2160                 list_del(&srp->entry);
2161                 srp->parentfp = NULL;
2162                 res = 1;
2163         }
2164         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2165         return res;
2166 }
2167
2168 static Sg_fd *
2169 sg_add_sfp(Sg_device * sdp)
2170 {
2171         Sg_fd *sfp;
2172         unsigned long iflags;
2173         int bufflen;
2174
2175         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2176         if (!sfp)
2177                 return ERR_PTR(-ENOMEM);
2178
2179         init_waitqueue_head(&sfp->read_wait);
2180         rwlock_init(&sfp->rq_list_lock);
2181         INIT_LIST_HEAD(&sfp->rq_list);
2182         kref_init(&sfp->f_ref);
2183         mutex_init(&sfp->f_mutex);
2184         sfp->timeout = SG_DEFAULT_TIMEOUT;
2185         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2186         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2187         sfp->cmd_q = SG_DEF_COMMAND_Q;
2188         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2189         sfp->parentdp = sdp;
2190         write_lock_irqsave(&sdp->sfd_lock, iflags);
2191         if (atomic_read(&sdp->detaching)) {
2192                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2193                 kfree(sfp);
2194                 return ERR_PTR(-ENODEV);
2195         }
2196         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2197         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2198         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2199                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2200         if (unlikely(sg_big_buff != def_reserved_size))
2201                 sg_big_buff = def_reserved_size;
2202
2203         bufflen = min_t(int, sg_big_buff,
2204                         max_sectors_bytes(sdp->device->request_queue));
2205         sg_build_reserve(sfp, bufflen);
2206         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2207                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2208                                       sfp->reserve.bufflen,
2209                                       sfp->reserve.k_use_sg));
2210
2211         kref_get(&sdp->d_ref);
2212         __module_get(THIS_MODULE);
2213         return sfp;
2214 }
2215
2216 static void
2217 sg_remove_sfp_usercontext(struct work_struct *work)
2218 {
2219         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2220         struct sg_device *sdp = sfp->parentdp;
2221         Sg_request *srp;
2222         unsigned long iflags;
2223
2224         /* Cleanup any responses which were never read(). */
2225         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2226         while (!list_empty(&sfp->rq_list)) {
2227                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2228                 sg_finish_rem_req(srp);
2229                 list_del(&srp->entry);
2230                 srp->parentfp = NULL;
2231         }
2232         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2233
2234         if (sfp->reserve.bufflen > 0) {
2235                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2236                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2237                                 (int) sfp->reserve.bufflen,
2238                                 (int) sfp->reserve.k_use_sg));
2239                 sg_remove_scat(sfp, &sfp->reserve);
2240         }
2241
2242         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2243                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2244         kfree(sfp);
2245
2246         scsi_device_put(sdp->device);
2247         kref_put(&sdp->d_ref, sg_device_destroy);
2248         module_put(THIS_MODULE);
2249 }
2250
2251 static void
2252 sg_remove_sfp(struct kref *kref)
2253 {
2254         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2255         struct sg_device *sdp = sfp->parentdp;
2256         unsigned long iflags;
2257
2258         write_lock_irqsave(&sdp->sfd_lock, iflags);
2259         list_del(&sfp->sfd_siblings);
2260         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2261
2262         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2263         schedule_work(&sfp->ew.work);
2264 }
2265
2266 #ifdef CONFIG_SCSI_PROC_FS
2267 static int
2268 sg_idr_max_id(int id, void *p, void *data)
2269 {
2270         int *k = data;
2271
2272         if (*k < id)
2273                 *k = id;
2274
2275         return 0;
2276 }
2277
2278 static int
2279 sg_last_dev(void)
2280 {
2281         int k = -1;
2282         unsigned long iflags;
2283
2284         read_lock_irqsave(&sg_index_lock, iflags);
2285         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2286         read_unlock_irqrestore(&sg_index_lock, iflags);
2287         return k + 1;           /* origin 1 */
2288 }
2289 #endif
2290
2291 /* must be called with sg_index_lock held */
2292 static Sg_device *sg_lookup_dev(int dev)
2293 {
2294         return idr_find(&sg_index_idr, dev);
2295 }
2296
2297 static Sg_device *
2298 sg_get_dev(int dev)
2299 {
2300         struct sg_device *sdp;
2301         unsigned long flags;
2302
2303         read_lock_irqsave(&sg_index_lock, flags);
2304         sdp = sg_lookup_dev(dev);
2305         if (!sdp)
2306                 sdp = ERR_PTR(-ENXIO);
2307         else if (atomic_read(&sdp->detaching)) {
2308                 /* If sdp->detaching, then the refcount may already be 0, in
2309                  * which case it would be a bug to do kref_get().
2310                  */
2311                 sdp = ERR_PTR(-ENODEV);
2312         } else
2313                 kref_get(&sdp->d_ref);
2314         read_unlock_irqrestore(&sg_index_lock, flags);
2315
2316         return sdp;
2317 }
2318
2319 #ifdef CONFIG_SCSI_PROC_FS
2320 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2321
2322 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2323 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2324                                   size_t count, loff_t *off);
2325 static const struct file_operations adio_fops = {
2326         .owner = THIS_MODULE,
2327         .open = sg_proc_single_open_adio,
2328         .read = seq_read,
2329         .llseek = seq_lseek,
2330         .write = sg_proc_write_adio,
2331         .release = single_release,
2332 };
2333
2334 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2335 static ssize_t sg_proc_write_dressz(struct file *filp,
2336                 const char __user *buffer, size_t count, loff_t *off);
2337 static const struct file_operations dressz_fops = {
2338         .owner = THIS_MODULE,
2339         .open = sg_proc_single_open_dressz,
2340         .read = seq_read,
2341         .llseek = seq_lseek,
2342         .write = sg_proc_write_dressz,
2343         .release = single_release,
2344 };
2345
2346 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2347 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2348 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2349 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2350 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2351 static void dev_seq_stop(struct seq_file *s, void *v);
2352 static const struct seq_operations dev_seq_ops = {
2353         .start = dev_seq_start,
2354         .next  = dev_seq_next,
2355         .stop  = dev_seq_stop,
2356         .show  = sg_proc_seq_show_dev,
2357 };
2358
2359 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2360 static const struct seq_operations devstrs_seq_ops = {
2361         .start = dev_seq_start,
2362         .next  = dev_seq_next,
2363         .stop  = dev_seq_stop,
2364         .show  = sg_proc_seq_show_devstrs,
2365 };
2366
2367 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2368 static const struct seq_operations debug_seq_ops = {
2369         .start = dev_seq_start,
2370         .next  = dev_seq_next,
2371         .stop  = dev_seq_stop,
2372         .show  = sg_proc_seq_show_debug,
2373 };
2374
2375 static int
2376 sg_proc_init(void)
2377 {
2378         struct proc_dir_entry *p;
2379
2380         p = proc_mkdir("scsi/sg", NULL);
2381         if (!p)
2382                 return 1;
2383
2384         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_fops);
2385         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2386         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_fops);
2387         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2388         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2389         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2390         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2391         return 0;
2392 }
2393
2394
2395 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2396 {
2397         seq_printf(s, "%d\n", *((int *)s->private));
2398         return 0;
2399 }
2400
2401 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2402 {
2403         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2404 }
2405
2406 static ssize_t
2407 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2408                    size_t count, loff_t *off)
2409 {
2410         int err;
2411         unsigned long num;
2412
2413         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2414                 return -EACCES;
2415         err = kstrtoul_from_user(buffer, count, 0, &num);
2416         if (err)
2417                 return err;
2418         sg_allow_dio = num ? 1 : 0;
2419         return count;
2420 }
2421
2422 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2423 {
2424         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2425 }
2426
2427 static ssize_t
2428 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2429                      size_t count, loff_t *off)
2430 {
2431         int err;
2432         unsigned long k = ULONG_MAX;
2433
2434         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2435                 return -EACCES;
2436
2437         err = kstrtoul_from_user(buffer, count, 0, &k);
2438         if (err)
2439                 return err;
2440         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2441                 sg_big_buff = k;
2442                 return count;
2443         }
2444         return -ERANGE;
2445 }
2446
2447 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2448 {
2449         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2450                    sg_version_date);
2451         return 0;
2452 }
2453
2454 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2455 {
2456         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2457         return 0;
2458 }
2459
2460 struct sg_proc_deviter {
2461         loff_t  index;
2462         size_t  max;
2463 };
2464
2465 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2466 {
2467         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2468
2469         s->private = it;
2470         if (! it)
2471                 return NULL;
2472
2473         it->index = *pos;
2474         it->max = sg_last_dev();
2475         if (it->index >= it->max)
2476                 return NULL;
2477         return it;
2478 }
2479
2480 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2481 {
2482         struct sg_proc_deviter * it = s->private;
2483
2484         *pos = ++it->index;
2485         return (it->index < it->max) ? it : NULL;
2486 }
2487
2488 static void dev_seq_stop(struct seq_file *s, void *v)
2489 {
2490         kfree(s->private);
2491 }
2492
2493 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2494 {
2495         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2496         Sg_device *sdp;
2497         struct scsi_device *scsidp;
2498         unsigned long iflags;
2499
2500         read_lock_irqsave(&sg_index_lock, iflags);
2501         sdp = it ? sg_lookup_dev(it->index) : NULL;
2502         if ((NULL == sdp) || (NULL == sdp->device) ||
2503             (atomic_read(&sdp->detaching)))
2504                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2505         else {
2506                 scsidp = sdp->device;
2507                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2508                               scsidp->host->host_no, scsidp->channel,
2509                               scsidp->id, scsidp->lun, (int) scsidp->type,
2510                               1,
2511                               (int) scsidp->queue_depth,
2512                               (int) atomic_read(&scsidp->device_busy),
2513                               (int) scsi_device_online(scsidp));
2514         }
2515         read_unlock_irqrestore(&sg_index_lock, iflags);
2516         return 0;
2517 }
2518
2519 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2520 {
2521         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2522         Sg_device *sdp;
2523         struct scsi_device *scsidp;
2524         unsigned long iflags;
2525
2526         read_lock_irqsave(&sg_index_lock, iflags);
2527         sdp = it ? sg_lookup_dev(it->index) : NULL;
2528         scsidp = sdp ? sdp->device : NULL;
2529         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2530                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2531                            scsidp->vendor, scsidp->model, scsidp->rev);
2532         else
2533                 seq_puts(s, "<no active device>\n");
2534         read_unlock_irqrestore(&sg_index_lock, iflags);
2535         return 0;
2536 }
2537
2538 /* must be called while holding sg_index_lock */
2539 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2540 {
2541         int k, new_interface, blen, usg;
2542         Sg_request *srp;
2543         Sg_fd *fp;
2544         const sg_io_hdr_t *hp;
2545         const char * cp;
2546         unsigned int ms;
2547
2548         k = 0;
2549         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2550                 k++;
2551                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2552                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2553                            "(res)sgat=%d low_dma=%d\n", k,
2554                            jiffies_to_msecs(fp->timeout),
2555                            fp->reserve.bufflen,
2556                            (int) fp->reserve.k_use_sg,
2557                            (int) sdp->device->host->unchecked_isa_dma);
2558                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2559                            (int) fp->cmd_q, (int) fp->force_packid,
2560                            (int) fp->keep_orphan);
2561                 list_for_each_entry(srp, &fp->rq_list, entry) {
2562                         hp = &srp->header;
2563                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2564                         if (srp->res_used) {
2565                                 if (new_interface &&
2566                                     (SG_FLAG_MMAP_IO & hp->flags))
2567                                         cp = "     mmap>> ";
2568                                 else
2569                                         cp = "     rb>> ";
2570                         } else {
2571                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2572                                         cp = "     dio>> ";
2573                                 else
2574                                         cp = "     ";
2575                         }
2576                         seq_puts(s, cp);
2577                         blen = srp->data.bufflen;
2578                         usg = srp->data.k_use_sg;
2579                         seq_puts(s, srp->done ?
2580                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2581                                   : "act:");
2582                         seq_printf(s, " id=%d blen=%d",
2583                                    srp->header.pack_id, blen);
2584                         if (srp->done)
2585                                 seq_printf(s, " dur=%d", hp->duration);
2586                         else {
2587                                 ms = jiffies_to_msecs(jiffies);
2588                                 seq_printf(s, " t_o/elap=%d/%d",
2589                                         (new_interface ? hp->timeout :
2590                                                   jiffies_to_msecs(fp->timeout)),
2591                                         (ms > hp->duration ? ms - hp->duration : 0));
2592                         }
2593                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2594                                    (int) srp->data.cmd_opcode);
2595                 }
2596                 if (list_empty(&fp->rq_list))
2597                         seq_puts(s, "     No requests active\n");
2598                 read_unlock(&fp->rq_list_lock);
2599         }
2600 }
2601
2602 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2603 {
2604         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2605         Sg_device *sdp;
2606         unsigned long iflags;
2607
2608         if (it && (0 == it->index))
2609                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2610                            (int)it->max, sg_big_buff);
2611
2612         read_lock_irqsave(&sg_index_lock, iflags);
2613         sdp = it ? sg_lookup_dev(it->index) : NULL;
2614         if (NULL == sdp)
2615                 goto skip;
2616         read_lock(&sdp->sfd_lock);
2617         if (!list_empty(&sdp->sfds)) {
2618                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2619                 if (atomic_read(&sdp->detaching))
2620                         seq_puts(s, "detaching pending close ");
2621                 else if (sdp->device) {
2622                         struct scsi_device *scsidp = sdp->device;
2623
2624                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2625                                    scsidp->host->host_no,
2626                                    scsidp->channel, scsidp->id,
2627                                    scsidp->lun,
2628                                    scsidp->host->hostt->emulated);
2629                 }
2630                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2631                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2632                 sg_proc_debug_helper(s, sdp);
2633         }
2634         read_unlock(&sdp->sfd_lock);
2635 skip:
2636         read_unlock_irqrestore(&sg_index_lock, iflags);
2637         return 0;
2638 }
2639
2640 #endif                          /* CONFIG_SCSI_PROC_FS */
2641
2642 module_init(init_sg);
2643 module_exit(exit_sg);