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         char name[DISK_NAME_LEN];
 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, (sdp)->name, fmt, ##a)
 206
 207 /*
 208  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 209  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 210  * to trigger read() and write() calls from various contexts with elevated
 211  * privileges. This can lead to kernel memory corruption (e.g. if these
 212  * interfaces are called through splice()) and privilege escalation inside
 213  * userspace (e.g. if a process with access to such a device passes a file
 214  * descriptor to a SUID binary as stdin/stdout/stderr).
 215  *
 216  * This function provides protection for the legacy API by restricting the
 217  * calling context.
 218  */
 219 static int sg_check_file_access(struct file *filp, const char *caller)
 220 {
 221         if (filp->f_cred != current_real_cred()) {
 222                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 223                         caller, task_tgid_vnr(current), current->comm);
 224                 return -EPERM;
 225         }
 226         if (uaccess_kernel()) {
 227                 pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
 228                         caller, task_tgid_vnr(current), current->comm);
 229                 return -EACCES;
 230         }
 231         return 0;
 232 }
 233
 234 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 235 {
 236         struct sg_fd *sfp = filp->private_data;
 237
 238         if (sfp->parentdp->device->type == TYPE_SCANNER)
 239                 return 0;
 240
 241         return blk_verify_command(cmd, filp->f_mode);
 242 }
 243
 244 static int
 245 open_wait(Sg_device *sdp, int flags)
 246 {
 247         int retval = 0;
 248
 249         if (flags & O_EXCL) {
 250                 while (sdp->open_cnt > 0) {
 251                         mutex_unlock(&sdp->open_rel_lock);
 252                         retval = wait_event_interruptible(sdp->open_wait,
 253                                         (atomic_read(&sdp->detaching) ||
 254                                          !sdp->open_cnt));
 255                         mutex_lock(&sdp->open_rel_lock);
 256
 257                         if (retval) /* -ERESTARTSYS */
 258                                 return retval;
 259                         if (atomic_read(&sdp->detaching))
 260                                 return -ENODEV;
 261                 }
 262         } else {
 263                 while (sdp->exclude) {
 264                         mutex_unlock(&sdp->open_rel_lock);
 265                         retval = wait_event_interruptible(sdp->open_wait,
 266                                         (atomic_read(&sdp->detaching) ||
 267                                          !sdp->exclude));
 268                         mutex_lock(&sdp->open_rel_lock);
 269
 270                         if (retval) /* -ERESTARTSYS */
 271                                 return retval;
 272                         if (atomic_read(&sdp->detaching))
 273                                 return -ENODEV;
 274                 }
 275         }
 276
 277         return retval;
 278 }
 279
 280 /* Returns 0 on success, else a negated errno value */
 281 static int
 282 sg_open(struct inode *inode, struct file *filp)
 283 {
 284         int dev = iminor(inode);
 285         int flags = filp->f_flags;
 286         struct request_queue *q;
 287         Sg_device *sdp;
 288         Sg_fd *sfp;
 289         int retval;
 290
 291         nonseekable_open(inode, filp);
 292         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 293                 return -EPERM; /* Can't lock it with read only access */
 294         sdp = sg_get_dev(dev);
 295         if (IS_ERR(sdp))
 296                 return PTR_ERR(sdp);
 297
 298         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 299                                       "sg_open: flags=0x%x\n", flags));
 300
 301         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 302         /* Prevent the device driver from vanishing while we sleep */
 303         retval = scsi_device_get(sdp->device);
 304         if (retval)
 305                 goto sg_put;
 306
 307         retval = scsi_autopm_get_device(sdp->device);
 308         if (retval)
 309                 goto sdp_put;
 310
 311         /* scsi_block_when_processing_errors() may block so bypass
 312          * check if O_NONBLOCK. Permits SCSI commands to be issued
 313          * during error recovery. Tread carefully. */
 314         if (!((flags & O_NONBLOCK) ||
 315               scsi_block_when_processing_errors(sdp->device))) {
 316                 retval = -ENXIO;
 317                 /* we are in error recovery for this device */
 318                 goto error_out;
 319         }
 320
 321         mutex_lock(&sdp->open_rel_lock);
 322         if (flags & O_NONBLOCK) {
 323                 if (flags & O_EXCL) {
 324                         if (sdp->open_cnt > 0) {
 325                                 retval = -EBUSY;
 326                                 goto error_mutex_locked;
 327                         }
 328                 } else {
 329                         if (sdp->exclude) {
 330                                 retval = -EBUSY;
 331                                 goto error_mutex_locked;
 332                         }
 333                 }
 334         } else {
 335                 retval = open_wait(sdp, flags);
 336                 if (retval) /* -ERESTARTSYS or -ENODEV */
 337                         goto error_mutex_locked;
 338         }
 339
 340         /* N.B. at this point we are holding the open_rel_lock */
 341         if (flags & O_EXCL)
 342                 sdp->exclude = true;
 343
 344         if (sdp->open_cnt < 1) {  /* no existing opens */
 345                 sdp->sgdebug = 0;
 346                 q = sdp->device->request_queue;
 347                 sdp->sg_tablesize = queue_max_segments(q);
 348         }
 349         sfp = sg_add_sfp(sdp);
 350         if (IS_ERR(sfp)) {
 351                 retval = PTR_ERR(sfp);
 352                 goto out_undo;
 353         }
 354
 355         filp->private_data = sfp;
 356         sdp->open_cnt++;
 357         mutex_unlock(&sdp->open_rel_lock);
 358
 359         retval = 0;
 360 sg_put:
 361         kref_put(&sdp->d_ref, sg_device_destroy);
 362         return retval;
 363
 364 out_undo:
 365         if (flags & O_EXCL) {
 366                 sdp->exclude = false;   /* undo if error */
 367                 wake_up_interruptible(&sdp->open_wait);
 368         }
 369 error_mutex_locked:
 370         mutex_unlock(&sdp->open_rel_lock);
 371 error_out:
 372         scsi_autopm_put_device(sdp->device);
 373 sdp_put:
 374         scsi_device_put(sdp->device);
 375         goto sg_put;
 376 }
 377
 378 /* Release resources associated with a successful sg_open()
 379  * Returns 0 on success, else a negated errno value */
 380 static int
 381 sg_release(struct inode *inode, struct file *filp)
 382 {
 383         Sg_device *sdp;
 384         Sg_fd *sfp;
 385
 386         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 387                 return -ENXIO;
 388         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 389
 390         mutex_lock(&sdp->open_rel_lock);
 391         scsi_autopm_put_device(sdp->device);
 392         kref_put(&sfp->f_ref, sg_remove_sfp);
 393         sdp->open_cnt--;
 394
 395         /* possibly many open()s waiting on exlude clearing, start many;
 396          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 397         if (sdp->exclude) {
 398                 sdp->exclude = false;
 399                 wake_up_interruptible_all(&sdp->open_wait);
 400         } else if (0 == sdp->open_cnt) {
 401                 wake_up_interruptible(&sdp->open_wait);
 402         }
 403         mutex_unlock(&sdp->open_rel_lock);
 404         return 0;
 405 }
 406
 407 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 408 {
 409         struct sg_header __user *old_hdr = buf;
 410         int reply_len;
 411
 412         if (count >= SZ_SG_HEADER) {
 413                 /* negative reply_len means v3 format, otherwise v1/v2 */
 414                 if (get_user(reply_len, &old_hdr->reply_len))
 415                         return -EFAULT;
 416
 417                 if (reply_len >= 0)
 418                         return get_user(*pack_id, &old_hdr->pack_id);
 419
 420                 if (in_compat_syscall() &&
 421                     count >= sizeof(struct compat_sg_io_hdr)) {
 422                         struct compat_sg_io_hdr __user *hp = buf;
 423
 424                         return get_user(*pack_id, &hp->pack_id);
 425                 }
 426
 427                 if (count >= sizeof(struct sg_io_hdr)) {
 428                         struct sg_io_hdr __user *hp = buf;
 429
 430                         return get_user(*pack_id, &hp->pack_id);
 431                 }
 432         }
 433
 434         /* no valid header was passed, so ignore the pack_id */
 435         *pack_id = -1;
 436         return 0;
 437 }
 438
 439 static ssize_t
 440 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 441 {
 442         Sg_device *sdp;
 443         Sg_fd *sfp;
 444         Sg_request *srp;
 445         int req_pack_id = -1;
 446         sg_io_hdr_t *hp;
 447         struct sg_header *old_hdr;
 448         int retval;
 449
 450         /*
 451          * This could cause a response to be stranded. Close the associated
 452          * file descriptor to free up any resources being held.
 453          */
 454         retval = sg_check_file_access(filp, __func__);
 455         if (retval)
 456                 return retval;
 457
 458         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 459                 return -ENXIO;
 460         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 461                                       "sg_read: count=%d\n", (int) count));
 462
 463         if (sfp->force_packid)
 464                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 465         if (retval)
 466                 return retval;
 467
 468         srp = sg_get_rq_mark(sfp, req_pack_id);
 469         if (!srp) {             /* now wait on packet to arrive */
 470                 if (atomic_read(&sdp->detaching))
 471                         return -ENODEV;
 472                 if (filp->f_flags & O_NONBLOCK)
 473                         return -EAGAIN;
 474                 retval = wait_event_interruptible(sfp->read_wait,
 475                         (atomic_read(&sdp->detaching) ||
 476                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 477                 if (atomic_read(&sdp->detaching))
 478                         return -ENODEV;
 479                 if (retval)
 480                         /* -ERESTARTSYS as signal hit process */
 481                         return retval;
 482         }
 483         if (srp->header.interface_id != '\0')
 484                 return sg_new_read(sfp, buf, count, srp);
 485
 486         hp = &srp->header;
 487         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 488         if (!old_hdr)
 489                 return -ENOMEM;
 490
 491         old_hdr->reply_len = (int) hp->timeout;
 492         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 493         old_hdr->pack_id = hp->pack_id;
 494         old_hdr->twelve_byte =
 495             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 496         old_hdr->target_status = hp->masked_status;
 497         old_hdr->host_status = hp->host_status;
 498         old_hdr->driver_status = hp->driver_status;
 499         if ((CHECK_CONDITION & hp->masked_status) ||
 500             (srp->sense_b[0] & 0x70) == 0x70) {
 501                 old_hdr->driver_status = DRIVER_SENSE;
 502                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 503                        sizeof (old_hdr->sense_buffer));
 504         }
 505         switch (hp->host_status) {
 506         /* This setup of 'result' is for backward compatibility and is best
 507            ignored by the user who should use target, host + driver status */
 508         case DID_OK:
 509         case DID_PASSTHROUGH:
 510         case DID_SOFT_ERROR:
 511                 old_hdr->result = 0;
 512                 break;
 513         case DID_NO_CONNECT:
 514         case DID_BUS_BUSY:
 515         case DID_TIME_OUT:
 516                 old_hdr->result = EBUSY;
 517                 break;
 518         case DID_BAD_TARGET:
 519         case DID_ABORT:
 520         case DID_PARITY:
 521         case DID_RESET:
 522         case DID_BAD_INTR:
 523                 old_hdr->result = EIO;
 524                 break;
 525         case DID_ERROR:
 526                 old_hdr->result = (srp->sense_b[0] == 0 &&
 527                                   hp->masked_status == GOOD) ? 0 : EIO;
 528                 break;
 529         default:
 530                 old_hdr->result = EIO;
 531                 break;
 532         }
 533
 534         /* Now copy the result back to the user buffer.  */
 535         if (count >= SZ_SG_HEADER) {
 536                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 537                         retval = -EFAULT;
 538                         goto free_old_hdr;
 539                 }
 540                 buf += SZ_SG_HEADER;
 541                 if (count > old_hdr->reply_len)
 542                         count = old_hdr->reply_len;
 543                 if (count > SZ_SG_HEADER) {
 544                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 545                                 retval = -EFAULT;
 546                                 goto free_old_hdr;
 547                         }
 548                 }
 549         } else
 550                 count = (old_hdr->result == 0) ? 0 : -EIO;
 551         sg_finish_rem_req(srp);
 552         sg_remove_request(sfp, srp);
 553         retval = count;
 554 free_old_hdr:
 555         kfree(old_hdr);
 556         return retval;
 557 }
 558
 559 static ssize_t
 560 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 561 {
 562         sg_io_hdr_t *hp = &srp->header;
 563         int err = 0, err2;
 564         int len;
 565
 566         if (in_compat_syscall()) {
 567                 if (count < sizeof(struct compat_sg_io_hdr)) {
 568                         err = -EINVAL;
 569                         goto err_out;
 570                 }
 571         } else if (count < SZ_SG_IO_HDR) {
 572                 err = -EINVAL;
 573                 goto err_out;
 574         }
 575         hp->sb_len_wr = 0;
 576         if ((hp->mx_sb_len > 0) && hp->sbp) {
 577                 if ((CHECK_CONDITION & hp->masked_status) ||
 578                     (srp->sense_b[0] & 0x70) == 0x70) {
 579                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 580                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 581                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 582                         len = (len > sb_len) ? sb_len : len;
 583                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 584                                 err = -EFAULT;
 585                                 goto err_out;
 586                         }
 587                         hp->driver_status = DRIVER_SENSE;
 588                         hp->sb_len_wr = len;
 589                 }
 590         }
 591         if (hp->masked_status || hp->host_status || hp->driver_status)
 592                 hp->info |= SG_INFO_CHECK;
 593         err = put_sg_io_hdr(hp, buf);
 594 err_out:
 595         err2 = sg_finish_rem_req(srp);
 596         sg_remove_request(sfp, srp);
 597         return err ? : err2 ? : count;
 598 }
 599
 600 static ssize_t
 601 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 602 {
 603         int mxsize, cmd_size, k;
 604         int input_size, blocking;
 605         unsigned char opcode;
 606         Sg_device *sdp;
 607         Sg_fd *sfp;
 608         Sg_request *srp;
 609         struct sg_header old_hdr;
 610         sg_io_hdr_t *hp;
 611         unsigned char cmnd[SG_MAX_CDB_SIZE];
 612         int retval;
 613
 614         retval = sg_check_file_access(filp, __func__);
 615         if (retval)
 616                 return retval;
 617
 618         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 619                 return -ENXIO;
 620         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 621                                       "sg_write: count=%d\n", (int) count));
 622         if (atomic_read(&sdp->detaching))
 623                 return -ENODEV;
 624         if (!((filp->f_flags & O_NONBLOCK) ||
 625               scsi_block_when_processing_errors(sdp->device)))
 626                 return -ENXIO;
 627
 628         if (count < SZ_SG_HEADER)
 629                 return -EIO;
 630         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 631                 return -EFAULT;
 632         blocking = !(filp->f_flags & O_NONBLOCK);
 633         if (old_hdr.reply_len < 0)
 634                 return sg_new_write(sfp, filp, buf, count,
 635                                     blocking, 0, 0, NULL);
 636         if (count < (SZ_SG_HEADER + 6))
 637                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 638
 639         buf += SZ_SG_HEADER;
 640         if (get_user(opcode, buf))
 641                 return -EFAULT;
 642
 643         if (!(srp = sg_add_request(sfp))) {
 644                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 645                                               "sg_write: queue full\n"));
 646                 return -EDOM;
 647         }
 648         mutex_lock(&sfp->f_mutex);
 649         if (sfp->next_cmd_len > 0) {
 650                 cmd_size = sfp->next_cmd_len;
 651                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 652         } else {
 653                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 654                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 655                         cmd_size = 12;
 656         }
 657         mutex_unlock(&sfp->f_mutex);
 658         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 659                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 660 /* Determine buffer size.  */
 661         input_size = count - cmd_size;
 662         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 663         mxsize -= SZ_SG_HEADER;
 664         input_size -= SZ_SG_HEADER;
 665         if (input_size < 0) {
 666                 sg_remove_request(sfp, srp);
 667                 return -EIO;    /* User did not pass enough bytes for this command. */
 668         }
 669         hp = &srp->header;
 670         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 671         hp->cmd_len = (unsigned char) cmd_size;
 672         hp->iovec_count = 0;
 673         hp->mx_sb_len = 0;
 674         if (input_size > 0)
 675                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 676                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 677         else
 678                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 679         hp->dxfer_len = mxsize;
 680         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 681             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 682                 hp->dxferp = (char __user *)buf + cmd_size;
 683         else
 684                 hp->dxferp = NULL;
 685         hp->sbp = NULL;
 686         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 687         hp->flags = input_size; /* structure abuse ... */
 688         hp->pack_id = old_hdr.pack_id;
 689         hp->usr_ptr = NULL;
 690         if (copy_from_user(cmnd, buf, cmd_size)) {
 691                 sg_remove_request(sfp, srp);
 692                 return -EFAULT;
 693         }
 694         /*
 695          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 696          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 697          * is a non-zero input_size, so emit a warning.
 698          */
 699         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 700                 printk_ratelimited(KERN_WARNING
 701                                    "sg_write: data in/out %d/%d bytes "
 702                                    "for SCSI command 0x%x-- guessing "
 703                                    "data in;\n   program %s not setting "
 704                                    "count and/or reply_len properly\n",
 705                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 706                                    input_size, (unsigned int) cmnd[0],
 707                                    current->comm);
 708         }
 709         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 710         return (k < 0) ? k : count;
 711 }
 712
 713 static ssize_t
 714 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 715                  size_t count, int blocking, int read_only, int sg_io_owned,
 716                  Sg_request **o_srp)
 717 {
 718         int k;
 719         Sg_request *srp;
 720         sg_io_hdr_t *hp;
 721         unsigned char cmnd[SG_MAX_CDB_SIZE];
 722         int timeout;
 723         unsigned long ul_timeout;
 724
 725         if (count < SZ_SG_IO_HDR)
 726                 return -EINVAL;
 727
 728         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 729         if (!(srp = sg_add_request(sfp))) {
 730                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 731                                               "sg_new_write: queue full\n"));
 732                 return -EDOM;
 733         }
 734         srp->sg_io_owned = sg_io_owned;
 735         hp = &srp->header;
 736         if (get_sg_io_hdr(hp, buf)) {
 737                 sg_remove_request(sfp, srp);
 738                 return -EFAULT;
 739         }
 740         if (hp->interface_id != 'S') {
 741                 sg_remove_request(sfp, srp);
 742                 return -ENOSYS;
 743         }
 744         if (hp->flags & SG_FLAG_MMAP_IO) {
 745                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 746                         sg_remove_request(sfp, srp);
 747                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 748                 }
 749                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 750                         sg_remove_request(sfp, srp);
 751                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 752                 }
 753                 if (sfp->res_in_use) {
 754                         sg_remove_request(sfp, srp);
 755                         return -EBUSY;  /* reserve buffer already being used */
 756                 }
 757         }
 758         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 759         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 760         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 761                 sg_remove_request(sfp, srp);
 762                 return -EMSGSIZE;
 763         }
 764         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 765                 sg_remove_request(sfp, srp);
 766                 return -EFAULT;
 767         }
 768         if (read_only && sg_allow_access(file, cmnd)) {
 769                 sg_remove_request(sfp, srp);
 770                 return -EPERM;
 771         }
 772         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 773         if (k < 0)
 774                 return k;
 775         if (o_srp)
 776                 *o_srp = srp;
 777         return count;
 778 }
 779
 780 static int
 781 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 782                 unsigned char *cmnd, int timeout, int blocking)
 783 {
 784         int k, at_head;
 785         Sg_device *sdp = sfp->parentdp;
 786         sg_io_hdr_t *hp = &srp->header;
 787
 788         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 789         hp->status = 0;
 790         hp->masked_status = 0;
 791         hp->msg_status = 0;
 792         hp->info = 0;
 793         hp->host_status = 0;
 794         hp->driver_status = 0;
 795         hp->resid = 0;
 796         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 797                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 798                         (int) cmnd[0], (int) hp->cmd_len));
 799
 800         if (hp->dxfer_len >= SZ_256M) {
 801                 sg_remove_request(sfp, srp);
 802                 return -EINVAL;
 803         }
 804
 805         k = sg_start_req(srp, cmnd);
 806         if (k) {
 807                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 808                         "sg_common_write: start_req err=%d\n", k));
 809                 sg_finish_rem_req(srp);
 810                 sg_remove_request(sfp, srp);
 811                 return k;       /* probably out of space --> ENOMEM */
 812         }
 813         if (atomic_read(&sdp->detaching)) {
 814                 if (srp->bio) {
 815                         scsi_req_free_cmd(scsi_req(srp->rq));
 816                         blk_put_request(srp->rq);
 817                         srp->rq = NULL;
 818                 }
 819
 820                 sg_finish_rem_req(srp);
 821                 sg_remove_request(sfp, srp);
 822                 return -ENODEV;
 823         }
 824
 825         hp->duration = jiffies_to_msecs(jiffies);
 826         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 827             (SG_FLAG_Q_AT_TAIL & hp->flags))
 828                 at_head = 0;
 829         else
 830                 at_head = 1;
 831
 832         srp->rq->timeout = timeout;
 833         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 834         blk_execute_rq_nowait(NULL, srp->rq, at_head, sg_rq_end_io);
 835         return 0;
 836 }
 837
 838 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 839 {
 840         unsigned long flags;
 841         int ret;
 842
 843         read_lock_irqsave(&sfp->rq_list_lock, flags);
 844         ret = srp->done;
 845         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 846         return ret;
 847 }
 848
 849 static int max_sectors_bytes(struct request_queue *q)
 850 {
 851         unsigned int max_sectors = queue_max_sectors(q);
 852
 853         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 854
 855         return max_sectors << 9;
 856 }
 857
 858 static void
 859 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 860 {
 861         Sg_request *srp;
 862         int val;
 863         unsigned int ms;
 864
 865         val = 0;
 866         list_for_each_entry(srp, &sfp->rq_list, entry) {
 867                 if (val >= SG_MAX_QUEUE)
 868                         break;
 869                 rinfo[val].req_state = srp->done + 1;
 870                 rinfo[val].problem =
 871                         srp->header.masked_status &
 872                         srp->header.host_status &
 873                         srp->header.driver_status;
 874                 if (srp->done)
 875                         rinfo[val].duration =
 876                                 srp->header.duration;
 877                 else {
 878                         ms = jiffies_to_msecs(jiffies);
 879                         rinfo[val].duration =
 880                                 (ms > srp->header.duration) ?
 881                                 (ms - srp->header.duration) : 0;
 882                 }
 883                 rinfo[val].orphan = srp->orphan;
 884                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 885                 rinfo[val].pack_id = srp->header.pack_id;
 886                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 887                 val++;
 888         }
 889 }
 890
 891 #ifdef CONFIG_COMPAT
 892 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 893         char req_state;
 894         char orphan;
 895         char sg_io_owned;
 896         char problem;
 897         int pack_id;
 898         compat_uptr_t usr_ptr;
 899         unsigned int duration;
 900         int unused;
 901 };
 902
 903 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 904                                     struct sg_req_info *rinfo)
 905 {
 906         int i;
 907         for (i = 0; i < SG_MAX_QUEUE; i++) {
 908                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 909                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 910                     put_user(rinfo[i].duration, &o[i].duration) ||
 911                     put_user(rinfo[i].unused, &o[i].unused))
 912                         return -EFAULT;
 913         }
 914         return 0;
 915 }
 916 #endif
 917
 918 static long
 919 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 920                 unsigned int cmd_in, void __user *p)
 921 {
 922         int __user *ip = p;
 923         int result, val, read_only;
 924         Sg_request *srp;
 925         unsigned long iflags;
 926
 927         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 928                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 929         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 930
 931         switch (cmd_in) {
 932         case SG_IO:
 933                 if (atomic_read(&sdp->detaching))
 934                         return -ENODEV;
 935                 if (!scsi_block_when_processing_errors(sdp->device))
 936                         return -ENXIO;
 937                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 938                                  1, read_only, 1, &srp);
 939                 if (result < 0)
 940                         return result;
 941                 result = wait_event_interruptible(sfp->read_wait,
 942                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 943                 if (atomic_read(&sdp->detaching))
 944                         return -ENODEV;
 945                 write_lock_irq(&sfp->rq_list_lock);
 946                 if (srp->done) {
 947                         srp->done = 2;
 948                         write_unlock_irq(&sfp->rq_list_lock);
 949                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 950                         return (result < 0) ? result : 0;
 951                 }
 952                 srp->orphan = 1;
 953                 write_unlock_irq(&sfp->rq_list_lock);
 954                 return result;  /* -ERESTARTSYS because signal hit process */
 955         case SG_SET_TIMEOUT:
 956                 result = get_user(val, ip);
 957                 if (result)
 958                         return result;
 959                 if (val < 0)
 960                         return -EIO;
 961                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 962                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 963                                     INT_MAX);
 964                 sfp->timeout_user = val;
 965                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 966
 967                 return 0;
 968         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 969                                 /* strange ..., for backward compatibility */
 970                 return sfp->timeout_user;
 971         case SG_SET_FORCE_LOW_DMA:
 972                 /*
 973                  * N.B. This ioctl never worked properly, but failed to
 974                  * return an error value. So returning '0' to keep compability
 975                  * with legacy applications.
 976                  */
 977                 return 0;
 978         case SG_GET_LOW_DMA:
 979                 return put_user(0, ip);
 980         case SG_GET_SCSI_ID:
 981                 {
 982                         sg_scsi_id_t v;
 983
 984                         if (atomic_read(&sdp->detaching))
 985                                 return -ENODEV;
 986                         memset(&v, 0, sizeof(v));
 987                         v.host_no = sdp->device->host->host_no;
 988                         v.channel = sdp->device->channel;
 989                         v.scsi_id = sdp->device->id;
 990                         v.lun = sdp->device->lun;
 991                         v.scsi_type = sdp->device->type;
 992                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 993                         v.d_queue_depth = sdp->device->queue_depth;
 994                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 995                                 return -EFAULT;
 996                         return 0;
 997                 }
 998         case SG_SET_FORCE_PACK_ID:
 999                 result = get_user(val, ip);
1000                 if (result)
1001                         return result;
1002                 sfp->force_packid = val ? 1 : 0;
1003                 return 0;
1004         case SG_GET_PACK_ID:
1005                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1006                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1007                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1008                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1009                                                        iflags);
1010                                 return put_user(srp->header.pack_id, ip);
1011                         }
1012                 }
1013                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1014                 return put_user(-1, ip);
1015         case SG_GET_NUM_WAITING:
1016                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1017                 val = 0;
1018                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1019                         if ((1 == srp->done) && (!srp->sg_io_owned))
1020                                 ++val;
1021                 }
1022                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1023                 return put_user(val, ip);
1024         case SG_GET_SG_TABLESIZE:
1025                 return put_user(sdp->sg_tablesize, ip);
1026         case SG_SET_RESERVED_SIZE:
1027                 result = get_user(val, ip);
1028                 if (result)
1029                         return result;
1030                 if (val < 0)
1031                         return -EINVAL;
1032                 val = min_t(int, val,
1033                             max_sectors_bytes(sdp->device->request_queue));
1034                 mutex_lock(&sfp->f_mutex);
1035                 if (val != sfp->reserve.bufflen) {
1036                         if (sfp->mmap_called ||
1037                             sfp->res_in_use) {
1038                                 mutex_unlock(&sfp->f_mutex);
1039                                 return -EBUSY;
1040                         }
1041
1042                         sg_remove_scat(sfp, &sfp->reserve);
1043                         sg_build_reserve(sfp, val);
1044                 }
1045                 mutex_unlock(&sfp->f_mutex);
1046                 return 0;
1047         case SG_GET_RESERVED_SIZE:
1048                 val = min_t(int, sfp->reserve.bufflen,
1049                             max_sectors_bytes(sdp->device->request_queue));
1050                 return put_user(val, ip);
1051         case SG_SET_COMMAND_Q:
1052                 result = get_user(val, ip);
1053                 if (result)
1054                         return result;
1055                 sfp->cmd_q = val ? 1 : 0;
1056                 return 0;
1057         case SG_GET_COMMAND_Q:
1058                 return put_user((int) sfp->cmd_q, ip);
1059         case SG_SET_KEEP_ORPHAN:
1060                 result = get_user(val, ip);
1061                 if (result)
1062                         return result;
1063                 sfp->keep_orphan = val;
1064                 return 0;
1065         case SG_GET_KEEP_ORPHAN:
1066                 return put_user((int) sfp->keep_orphan, ip);
1067         case SG_NEXT_CMD_LEN:
1068                 result = get_user(val, ip);
1069                 if (result)
1070                         return result;
1071                 if (val > SG_MAX_CDB_SIZE)
1072                         return -ENOMEM;
1073                 sfp->next_cmd_len = (val > 0) ? val : 0;
1074                 return 0;
1075         case SG_GET_VERSION_NUM:
1076                 return put_user(sg_version_num, ip);
1077         case SG_GET_ACCESS_COUNT:
1078                 /* faked - we don't have a real access count anymore */
1079                 val = (sdp->device ? 1 : 0);
1080                 return put_user(val, ip);
1081         case SG_GET_REQUEST_TABLE:
1082                 {
1083                         sg_req_info_t *rinfo;
1084
1085                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1086                                         GFP_KERNEL);
1087                         if (!rinfo)
1088                                 return -ENOMEM;
1089                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1090                         sg_fill_request_table(sfp, rinfo);
1091                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1092         #ifdef CONFIG_COMPAT
1093                         if (in_compat_syscall())
1094                                 result = put_compat_request_table(p, rinfo);
1095                         else
1096         #endif
1097                                 result = copy_to_user(p, rinfo,
1098                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1099                         result = result ? -EFAULT : 0;
1100                         kfree(rinfo);
1101                         return result;
1102                 }
1103         case SG_EMULATED_HOST:
1104                 if (atomic_read(&sdp->detaching))
1105                         return -ENODEV;
1106                 return put_user(sdp->device->host->hostt->emulated, ip);
1107         case SCSI_IOCTL_SEND_COMMAND:
1108                 if (atomic_read(&sdp->detaching))
1109                         return -ENODEV;
1110                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1111         case SG_SET_DEBUG:
1112                 result = get_user(val, ip);
1113                 if (result)
1114                         return result;
1115                 sdp->sgdebug = (char) val;
1116                 return 0;
1117         case BLKSECTGET:
1118                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1119                                 ip);
1120         case BLKTRACESETUP:
1121                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1122                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1123                                        NULL, p);
1124         case BLKTRACESTART:
1125                 return blk_trace_startstop(sdp->device->request_queue, 1);
1126         case BLKTRACESTOP:
1127                 return blk_trace_startstop(sdp->device->request_queue, 0);
1128         case BLKTRACETEARDOWN:
1129                 return blk_trace_remove(sdp->device->request_queue);
1130         case SCSI_IOCTL_GET_IDLUN:
1131         case SCSI_IOCTL_GET_BUS_NUMBER:
1132         case SCSI_IOCTL_PROBE_HOST:
1133         case SG_GET_TRANSFORM:
1134         case SG_SCSI_RESET:
1135                 if (atomic_read(&sdp->detaching))
1136                         return -ENODEV;
1137                 break;
1138         default:
1139                 if (read_only)
1140                         return -EPERM;  /* don't know so take safe approach */
1141                 break;
1142         }
1143
1144         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1145                         cmd_in, filp->f_flags & O_NDELAY);
1146         if (result)
1147                 return result;
1148
1149         return -ENOIOCTLCMD;
1150 }
1151
1152 static long
1153 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1154 {
1155         void __user *p = (void __user *)arg;
1156         Sg_device *sdp;
1157         Sg_fd *sfp;
1158         int ret;
1159
1160         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1161                 return -ENXIO;
1162
1163         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1164         if (ret != -ENOIOCTLCMD)
1165                 return ret;
1166
1167         return scsi_ioctl(sdp->device, cmd_in, p);
1168 }
1169
1170 #ifdef CONFIG_COMPAT
1171 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1172 {
1173         void __user *p = compat_ptr(arg);
1174         Sg_device *sdp;
1175         Sg_fd *sfp;
1176         int ret;
1177
1178         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1179                 return -ENXIO;
1180
1181         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1182         if (ret != -ENOIOCTLCMD)
1183                 return ret;
1184
1185         return scsi_compat_ioctl(sdp->device, cmd_in, p);
1186 }
1187 #endif
1188
1189 static __poll_t
1190 sg_poll(struct file *filp, poll_table * wait)
1191 {
1192         __poll_t res = 0;
1193         Sg_device *sdp;
1194         Sg_fd *sfp;
1195         Sg_request *srp;
1196         int count = 0;
1197         unsigned long iflags;
1198
1199         sfp = filp->private_data;
1200         if (!sfp)
1201                 return EPOLLERR;
1202         sdp = sfp->parentdp;
1203         if (!sdp)
1204                 return EPOLLERR;
1205         poll_wait(filp, &sfp->read_wait, wait);
1206         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1207         list_for_each_entry(srp, &sfp->rq_list, entry) {
1208                 /* if any read waiting, flag it */
1209                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1210                         res = EPOLLIN | EPOLLRDNORM;
1211                 ++count;
1212         }
1213         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1214
1215         if (atomic_read(&sdp->detaching))
1216                 res |= EPOLLHUP;
1217         else if (!sfp->cmd_q) {
1218                 if (0 == count)
1219                         res |= EPOLLOUT | EPOLLWRNORM;
1220         } else if (count < SG_MAX_QUEUE)
1221                 res |= EPOLLOUT | EPOLLWRNORM;
1222         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1223                                       "sg_poll: res=0x%x\n", (__force u32) res));
1224         return res;
1225 }
1226
1227 static int
1228 sg_fasync(int fd, struct file *filp, int mode)
1229 {
1230         Sg_device *sdp;
1231         Sg_fd *sfp;
1232
1233         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1234                 return -ENXIO;
1235         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1236                                       "sg_fasync: mode=%d\n", mode));
1237
1238         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1239 }
1240
1241 static vm_fault_t
1242 sg_vma_fault(struct vm_fault *vmf)
1243 {
1244         struct vm_area_struct *vma = vmf->vma;
1245         Sg_fd *sfp;
1246         unsigned long offset, len, sa;
1247         Sg_scatter_hold *rsv_schp;
1248         int k, length;
1249
1250         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1251                 return VM_FAULT_SIGBUS;
1252         rsv_schp = &sfp->reserve;
1253         offset = vmf->pgoff << PAGE_SHIFT;
1254         if (offset >= rsv_schp->bufflen)
1255                 return VM_FAULT_SIGBUS;
1256         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1257                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1258                                       offset, rsv_schp->k_use_sg));
1259         sa = vma->vm_start;
1260         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1261         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1262                 len = vma->vm_end - sa;
1263                 len = (len < length) ? len : length;
1264                 if (offset < len) {
1265                         struct page *page = nth_page(rsv_schp->pages[k],
1266                                                      offset >> PAGE_SHIFT);
1267                         get_page(page); /* increment page count */
1268                         vmf->page = page;
1269                         return 0; /* success */
1270                 }
1271                 sa += len;
1272                 offset -= len;
1273         }
1274
1275         return VM_FAULT_SIGBUS;
1276 }
1277
1278 static const struct vm_operations_struct sg_mmap_vm_ops = {
1279         .fault = sg_vma_fault,
1280 };
1281
1282 static int
1283 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1284 {
1285         Sg_fd *sfp;
1286         unsigned long req_sz, len, sa;
1287         Sg_scatter_hold *rsv_schp;
1288         int k, length;
1289         int ret = 0;
1290
1291         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1292                 return -ENXIO;
1293         req_sz = vma->vm_end - vma->vm_start;
1294         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1295                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1296                                       (void *) vma->vm_start, (int) req_sz));
1297         if (vma->vm_pgoff)
1298                 return -EINVAL; /* want no offset */
1299         rsv_schp = &sfp->reserve;
1300         mutex_lock(&sfp->f_mutex);
1301         if (req_sz > rsv_schp->bufflen) {
1302                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1303                 goto out;
1304         }
1305
1306         sa = vma->vm_start;
1307         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1308         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1309                 len = vma->vm_end - sa;
1310                 len = (len < length) ? len : length;
1311                 sa += len;
1312         }
1313
1314         sfp->mmap_called = 1;
1315         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1316         vma->vm_private_data = sfp;
1317         vma->vm_ops = &sg_mmap_vm_ops;
1318 out:
1319         mutex_unlock(&sfp->f_mutex);
1320         return ret;
1321 }
1322
1323 static void
1324 sg_rq_end_io_usercontext(struct work_struct *work)
1325 {
1326         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1327         struct sg_fd *sfp = srp->parentfp;
1328
1329         sg_finish_rem_req(srp);
1330         sg_remove_request(sfp, srp);
1331         kref_put(&sfp->f_ref, sg_remove_sfp);
1332 }
1333
1334 /*
1335  * This function is a "bottom half" handler that is called by the mid
1336  * level when a command is completed (or has failed).
1337  */
1338 static void
1339 sg_rq_end_io(struct request *rq, blk_status_t status)
1340 {
1341         struct sg_request *srp = rq->end_io_data;
1342         struct scsi_request *req = scsi_req(rq);
1343         Sg_device *sdp;
1344         Sg_fd *sfp;
1345         unsigned long iflags;
1346         unsigned int ms;
1347         char *sense;
1348         int result, resid, done = 1;
1349
1350         if (WARN_ON(srp->done != 0))
1351                 return;
1352
1353         sfp = srp->parentfp;
1354         if (WARN_ON(sfp == NULL))
1355                 return;
1356
1357         sdp = sfp->parentdp;
1358         if (unlikely(atomic_read(&sdp->detaching)))
1359                 pr_info("%s: device detaching\n", __func__);
1360
1361         sense = req->sense;
1362         result = req->result;
1363         resid = req->resid_len;
1364
1365         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1366                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1367                                       srp->header.pack_id, result));
1368         srp->header.resid = resid;
1369         ms = jiffies_to_msecs(jiffies);
1370         srp->header.duration = (ms > srp->header.duration) ?
1371                                 (ms - srp->header.duration) : 0;
1372         if (0 != result) {
1373                 struct scsi_sense_hdr sshdr;
1374
1375                 srp->header.status = 0xff & result;
1376                 srp->header.masked_status = status_byte(result);
1377                 srp->header.msg_status = COMMAND_COMPLETE;
1378                 srp->header.host_status = host_byte(result);
1379                 srp->header.driver_status = driver_byte(result);
1380                 if ((sdp->sgdebug > 0) &&
1381                     ((CHECK_CONDITION == srp->header.masked_status) ||
1382                      (COMMAND_TERMINATED == srp->header.masked_status)))
1383                         __scsi_print_sense(sdp->device, __func__, sense,
1384                                            SCSI_SENSE_BUFFERSIZE);
1385
1386                 /* Following if statement is a patch supplied by Eric Youngdale */
1387                 if (driver_byte(result) != 0
1388                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1389                     && !scsi_sense_is_deferred(&sshdr)
1390                     && sshdr.sense_key == UNIT_ATTENTION
1391                     && sdp->device->removable) {
1392                         /* Detected possible disc change. Set the bit - this */
1393                         /* may be used if there are filesystems using this device */
1394                         sdp->device->changed = 1;
1395                 }
1396         }
1397
1398         if (req->sense_len)
1399                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1400
1401         /* Rely on write phase to clean out srp status values, so no "else" */
1402
1403         /*
1404          * Free the request as soon as it is complete so that its resources
1405          * can be reused without waiting for userspace to read() the
1406          * result.  But keep the associated bio (if any) around until
1407          * blk_rq_unmap_user() can be called from user context.
1408          */
1409         srp->rq = NULL;
1410         scsi_req_free_cmd(scsi_req(rq));
1411         blk_put_request(rq);
1412
1413         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1414         if (unlikely(srp->orphan)) {
1415                 if (sfp->keep_orphan)
1416                         srp->sg_io_owned = 0;
1417                 else
1418                         done = 0;
1419         }
1420         srp->done = done;
1421         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1422
1423         if (likely(done)) {
1424                 /* Now wake up any sg_read() that is waiting for this
1425                  * packet.
1426                  */
1427                 wake_up_interruptible(&sfp->read_wait);
1428                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1429                 kref_put(&sfp->f_ref, sg_remove_sfp);
1430         } else {
1431                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1432                 schedule_work(&srp->ew.work);
1433         }
1434 }
1435
1436 static const struct file_operations sg_fops = {
1437         .owner = THIS_MODULE,
1438         .read = sg_read,
1439         .write = sg_write,
1440         .poll = sg_poll,
1441         .unlocked_ioctl = sg_ioctl,
1442 #ifdef CONFIG_COMPAT
1443         .compat_ioctl = sg_compat_ioctl,
1444 #endif
1445         .open = sg_open,
1446         .mmap = sg_mmap,
1447         .release = sg_release,
1448         .fasync = sg_fasync,
1449         .llseek = no_llseek,
1450 };
1451
1452 static struct class *sg_sysfs_class;
1453
1454 static int sg_sysfs_valid = 0;
1455
1456 static Sg_device *
1457 sg_alloc(struct scsi_device *scsidp)
1458 {
1459         struct request_queue *q = scsidp->request_queue;
1460         Sg_device *sdp;
1461         unsigned long iflags;
1462         int error;
1463         u32 k;
1464
1465         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1466         if (!sdp) {
1467                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1468                             "failure\n", __func__);
1469                 return ERR_PTR(-ENOMEM);
1470         }
1471
1472         idr_preload(GFP_KERNEL);
1473         write_lock_irqsave(&sg_index_lock, iflags);
1474
1475         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1476         if (error < 0) {
1477                 if (error == -ENOSPC) {
1478                         sdev_printk(KERN_WARNING, scsidp,
1479                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1480                                     scsidp->type, SG_MAX_DEVS - 1);
1481                         error = -ENODEV;
1482                 } else {
1483                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1484                                     "allocation Sg_device failure: %d\n",
1485                                     __func__, error);
1486                 }
1487                 goto out_unlock;
1488         }
1489         k = error;
1490
1491         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1492                                         "sg_alloc: dev=%d \n", k));
1493         sprintf(sdp->name, "sg%d", k);
1494         sdp->device = scsidp;
1495         mutex_init(&sdp->open_rel_lock);
1496         INIT_LIST_HEAD(&sdp->sfds);
1497         init_waitqueue_head(&sdp->open_wait);
1498         atomic_set(&sdp->detaching, 0);
1499         rwlock_init(&sdp->sfd_lock);
1500         sdp->sg_tablesize = queue_max_segments(q);
1501         sdp->index = k;
1502         kref_init(&sdp->d_ref);
1503         error = 0;
1504
1505 out_unlock:
1506         write_unlock_irqrestore(&sg_index_lock, iflags);
1507         idr_preload_end();
1508
1509         if (error) {
1510                 kfree(sdp);
1511                 return ERR_PTR(error);
1512         }
1513         return sdp;
1514 }
1515
1516 static int
1517 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1518 {
1519         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1520         Sg_device *sdp = NULL;
1521         struct cdev * cdev = NULL;
1522         int error;
1523         unsigned long iflags;
1524
1525         error = -ENOMEM;
1526         cdev = cdev_alloc();
1527         if (!cdev) {
1528                 pr_warn("%s: cdev_alloc failed\n", __func__);
1529                 goto out;
1530         }
1531         cdev->owner = THIS_MODULE;
1532         cdev->ops = &sg_fops;
1533
1534         sdp = sg_alloc(scsidp);
1535         if (IS_ERR(sdp)) {
1536                 pr_warn("%s: sg_alloc failed\n", __func__);
1537                 error = PTR_ERR(sdp);
1538                 goto out;
1539         }
1540
1541         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1542         if (error)
1543                 goto cdev_add_err;
1544
1545         sdp->cdev = cdev;
1546         if (sg_sysfs_valid) {
1547                 struct device *sg_class_member;
1548
1549                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1550                                                 MKDEV(SCSI_GENERIC_MAJOR,
1551                                                       sdp->index),
1552                                                 sdp, "%s", sdp->name);
1553                 if (IS_ERR(sg_class_member)) {
1554                         pr_err("%s: device_create failed\n", __func__);
1555                         error = PTR_ERR(sg_class_member);
1556                         goto cdev_add_err;
1557                 }
1558                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1559                                           &sg_class_member->kobj, "generic");
1560                 if (error)
1561                         pr_err("%s: unable to make symlink 'generic' back "
1562                                "to sg%d\n", __func__, sdp->index);
1563         } else
1564                 pr_warn("%s: sg_sys Invalid\n", __func__);
1565
1566         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1567                     "type %d\n", sdp->index, scsidp->type);
1568
1569         dev_set_drvdata(cl_dev, sdp);
1570
1571         return 0;
1572
1573 cdev_add_err:
1574         write_lock_irqsave(&sg_index_lock, iflags);
1575         idr_remove(&sg_index_idr, sdp->index);
1576         write_unlock_irqrestore(&sg_index_lock, iflags);
1577         kfree(sdp);
1578
1579 out:
1580         if (cdev)
1581                 cdev_del(cdev);
1582         return error;
1583 }
1584
1585 static void
1586 sg_device_destroy(struct kref *kref)
1587 {
1588         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1589         unsigned long flags;
1590
1591         /* CAUTION!  Note that the device can still be found via idr_find()
1592          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1593          * any other cleanup.
1594          */
1595
1596         write_lock_irqsave(&sg_index_lock, flags);
1597         idr_remove(&sg_index_idr, sdp->index);
1598         write_unlock_irqrestore(&sg_index_lock, flags);
1599
1600         SCSI_LOG_TIMEOUT(3,
1601                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1602
1603         kfree(sdp);
1604 }
1605
1606 static void
1607 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1608 {
1609         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1610         Sg_device *sdp = dev_get_drvdata(cl_dev);
1611         unsigned long iflags;
1612         Sg_fd *sfp;
1613         int val;
1614
1615         if (!sdp)
1616                 return;
1617         /* want sdp->detaching non-zero as soon as possible */
1618         val = atomic_inc_return(&sdp->detaching);
1619         if (val > 1)
1620                 return; /* only want to do following once per device */
1621
1622         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1623                                       "%s\n", __func__));
1624
1625         read_lock_irqsave(&sdp->sfd_lock, iflags);
1626         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1627                 wake_up_interruptible_all(&sfp->read_wait);
1628                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1629         }
1630         wake_up_interruptible_all(&sdp->open_wait);
1631         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1632
1633         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1634         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1635         cdev_del(sdp->cdev);
1636         sdp->cdev = NULL;
1637
1638         kref_put(&sdp->d_ref, sg_device_destroy);
1639 }
1640
1641 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1642 module_param_named(def_reserved_size, def_reserved_size, int,
1643                    S_IRUGO | S_IWUSR);
1644 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1645
1646 MODULE_AUTHOR("Douglas Gilbert");
1647 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1648 MODULE_LICENSE("GPL");
1649 MODULE_VERSION(SG_VERSION_STR);
1650 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1651
1652 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1653                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1654 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1655 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1656
1657 static int __init
1658 init_sg(void)
1659 {
1660         int rc;
1661
1662         if (scatter_elem_sz < PAGE_SIZE) {
1663                 scatter_elem_sz = PAGE_SIZE;
1664                 scatter_elem_sz_prev = scatter_elem_sz;
1665         }
1666         if (def_reserved_size >= 0)
1667                 sg_big_buff = def_reserved_size;
1668         else
1669                 def_reserved_size = sg_big_buff;
1670
1671         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1672                                     SG_MAX_DEVS, "sg");
1673         if (rc)
1674                 return rc;
1675         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1676         if ( IS_ERR(sg_sysfs_class) ) {
1677                 rc = PTR_ERR(sg_sysfs_class);
1678                 goto err_out;
1679         }
1680         sg_sysfs_valid = 1;
1681         rc = scsi_register_interface(&sg_interface);
1682         if (0 == rc) {
1683 #ifdef CONFIG_SCSI_PROC_FS
1684                 sg_proc_init();
1685 #endif                          /* CONFIG_SCSI_PROC_FS */
1686                 return 0;
1687         }
1688         class_destroy(sg_sysfs_class);
1689 err_out:
1690         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1691         return rc;
1692 }
1693
1694 static void __exit
1695 exit_sg(void)
1696 {
1697 #ifdef CONFIG_SCSI_PROC_FS
1698         remove_proc_subtree("scsi/sg", NULL);
1699 #endif                          /* CONFIG_SCSI_PROC_FS */
1700         scsi_unregister_interface(&sg_interface);
1701         class_destroy(sg_sysfs_class);
1702         sg_sysfs_valid = 0;
1703         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1704                                  SG_MAX_DEVS);
1705         idr_destroy(&sg_index_idr);
1706 }
1707
1708 static int
1709 sg_start_req(Sg_request *srp, unsigned char *cmd)
1710 {
1711         int res;
1712         struct request *rq;
1713         struct scsi_request *req;
1714         Sg_fd *sfp = srp->parentfp;
1715         sg_io_hdr_t *hp = &srp->header;
1716         int dxfer_len = (int) hp->dxfer_len;
1717         int dxfer_dir = hp->dxfer_direction;
1718         unsigned int iov_count = hp->iovec_count;
1719         Sg_scatter_hold *req_schp = &srp->data;
1720         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1721         struct request_queue *q = sfp->parentdp->device->request_queue;
1722         struct rq_map_data *md, map_data;
1723         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1724         unsigned char *long_cmdp = NULL;
1725
1726         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1727                                       "sg_start_req: dxfer_len=%d\n",
1728                                       dxfer_len));
1729
1730         if (hp->cmd_len > BLK_MAX_CDB) {
1731                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1732                 if (!long_cmdp)
1733                         return -ENOMEM;
1734         }
1735
1736         /*
1737          * NOTE
1738          *
1739          * With scsi-mq enabled, there are a fixed number of preallocated
1740          * requests equal in number to shost->can_queue.  If all of the
1741          * preallocated requests are already in use, then blk_get_request()
1742          * will sleep until an active command completes, freeing up a request.
1743          * Although waiting in an asynchronous interface is less than ideal, we
1744          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1745          * not expect an EWOULDBLOCK from this condition.
1746          */
1747         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1748                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1749         if (IS_ERR(rq)) {
1750                 kfree(long_cmdp);
1751                 return PTR_ERR(rq);
1752         }
1753         req = scsi_req(rq);
1754
1755         if (hp->cmd_len > BLK_MAX_CDB)
1756                 req->cmd = long_cmdp;
1757         memcpy(req->cmd, cmd, hp->cmd_len);
1758         req->cmd_len = hp->cmd_len;
1759
1760         srp->rq = rq;
1761         rq->end_io_data = srp;
1762         req->retries = SG_DEFAULT_RETRIES;
1763
1764         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1765                 return 0;
1766
1767         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1768             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1769             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1770                 md = NULL;
1771         else
1772                 md = &map_data;
1773
1774         if (md) {
1775                 mutex_lock(&sfp->f_mutex);
1776                 if (dxfer_len <= rsv_schp->bufflen &&
1777                     !sfp->res_in_use) {
1778                         sfp->res_in_use = 1;
1779                         sg_link_reserve(sfp, srp, dxfer_len);
1780                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1781                         res = -EBUSY; /* sfp->res_in_use == 1 */
1782                         if (dxfer_len > rsv_schp->bufflen)
1783                                 res = -ENOMEM;
1784                         mutex_unlock(&sfp->f_mutex);
1785                         return res;
1786                 } else {
1787                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1788                         if (res) {
1789                                 mutex_unlock(&sfp->f_mutex);
1790                                 return res;
1791                         }
1792                 }
1793                 mutex_unlock(&sfp->f_mutex);
1794
1795                 md->pages = req_schp->pages;
1796                 md->page_order = req_schp->page_order;
1797                 md->nr_entries = req_schp->k_use_sg;
1798                 md->offset = 0;
1799                 md->null_mapped = hp->dxferp ? 0 : 1;
1800                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1801                         md->from_user = 1;
1802                 else
1803                         md->from_user = 0;
1804         }
1805
1806         if (iov_count) {
1807                 struct iovec *iov = NULL;
1808                 struct iov_iter i;
1809
1810                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1811                 if (res < 0)
1812                         return res;
1813
1814                 iov_iter_truncate(&i, hp->dxfer_len);
1815                 if (!iov_iter_count(&i)) {
1816                         kfree(iov);
1817                         return -EINVAL;
1818                 }
1819
1820                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1821                 kfree(iov);
1822         } else
1823                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1824                                       hp->dxfer_len, GFP_ATOMIC);
1825
1826         if (!res) {
1827                 srp->bio = rq->bio;
1828
1829                 if (!md) {
1830                         req_schp->dio_in_use = 1;
1831                         hp->info |= SG_INFO_DIRECT_IO;
1832                 }
1833         }
1834         return res;
1835 }
1836
1837 static int
1838 sg_finish_rem_req(Sg_request *srp)
1839 {
1840         int ret = 0;
1841
1842         Sg_fd *sfp = srp->parentfp;
1843         Sg_scatter_hold *req_schp = &srp->data;
1844
1845         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1846                                       "sg_finish_rem_req: res_used=%d\n",
1847                                       (int) srp->res_used));
1848         if (srp->bio)
1849                 ret = blk_rq_unmap_user(srp->bio);
1850
1851         if (srp->rq) {
1852                 scsi_req_free_cmd(scsi_req(srp->rq));
1853                 blk_put_request(srp->rq);
1854         }
1855
1856         if (srp->res_used)
1857                 sg_unlink_reserve(sfp, srp);
1858         else
1859                 sg_remove_scat(sfp, req_schp);
1860
1861         return ret;
1862 }
1863
1864 static int
1865 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1866 {
1867         int sg_bufflen = tablesize * sizeof(struct page *);
1868         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1869
1870         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1871         if (!schp->pages)
1872                 return -ENOMEM;
1873         schp->sglist_len = sg_bufflen;
1874         return tablesize;       /* number of scat_gath elements allocated */
1875 }
1876
1877 static int
1878 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1879 {
1880         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1881         int sg_tablesize = sfp->parentdp->sg_tablesize;
1882         int blk_size = buff_size, order;
1883         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1884
1885         if (blk_size < 0)
1886                 return -EFAULT;
1887         if (0 == blk_size)
1888                 ++blk_size;     /* don't know why */
1889         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1890         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1891         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1892                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1893                 buff_size, blk_size));
1894
1895         /* N.B. ret_sz carried into this block ... */
1896         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1897         if (mx_sc_elems < 0)
1898                 return mx_sc_elems;     /* most likely -ENOMEM */
1899
1900         num = scatter_elem_sz;
1901         if (unlikely(num != scatter_elem_sz_prev)) {
1902                 if (num < PAGE_SIZE) {
1903                         scatter_elem_sz = PAGE_SIZE;
1904                         scatter_elem_sz_prev = PAGE_SIZE;
1905                 } else
1906                         scatter_elem_sz_prev = num;
1907         }
1908
1909         order = get_order(num);
1910 retry:
1911         ret_sz = 1 << (PAGE_SHIFT + order);
1912
1913         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1914              k++, rem_sz -= ret_sz) {
1915
1916                 num = (rem_sz > scatter_elem_sz_prev) ?
1917                         scatter_elem_sz_prev : rem_sz;
1918
1919                 schp->pages[k] = alloc_pages(gfp_mask, order);
1920                 if (!schp->pages[k])
1921                         goto out;
1922
1923                 if (num == scatter_elem_sz_prev) {
1924                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1925                                 scatter_elem_sz = ret_sz;
1926                                 scatter_elem_sz_prev = ret_sz;
1927                         }
1928                 }
1929
1930                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1931                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1932                                  k, num, ret_sz));
1933         }               /* end of for loop */
1934
1935         schp->page_order = order;
1936         schp->k_use_sg = k;
1937         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1938                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1939                          k, rem_sz));
1940
1941         schp->bufflen = blk_size;
1942         if (rem_sz > 0) /* must have failed */
1943                 return -ENOMEM;
1944         return 0;
1945 out:
1946         for (i = 0; i < k; i++)
1947                 __free_pages(schp->pages[i], order);
1948
1949         if (--order >= 0)
1950                 goto retry;
1951
1952         return -ENOMEM;
1953 }
1954
1955 static void
1956 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1957 {
1958         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1959                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1960         if (schp->pages && schp->sglist_len > 0) {
1961                 if (!schp->dio_in_use) {
1962                         int k;
1963
1964                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1965                                 SCSI_LOG_TIMEOUT(5,
1966                                         sg_printk(KERN_INFO, sfp->parentdp,
1967                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1968                                         k, schp->pages[k]));
1969                                 __free_pages(schp->pages[k], schp->page_order);
1970                         }
1971
1972                         kfree(schp->pages);
1973                 }
1974         }
1975         memset(schp, 0, sizeof (*schp));
1976 }
1977
1978 static int
1979 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1980 {
1981         Sg_scatter_hold *schp = &srp->data;
1982         int k, num;
1983
1984         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1985                          "sg_read_oxfer: num_read_xfer=%d\n",
1986                          num_read_xfer));
1987         if ((!outp) || (num_read_xfer <= 0))
1988                 return 0;
1989
1990         num = 1 << (PAGE_SHIFT + schp->page_order);
1991         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1992                 if (num > num_read_xfer) {
1993                         if (copy_to_user(outp, page_address(schp->pages[k]),
1994                                            num_read_xfer))
1995                                 return -EFAULT;
1996                         break;
1997                 } else {
1998                         if (copy_to_user(outp, page_address(schp->pages[k]),
1999                                            num))
2000                                 return -EFAULT;
2001                         num_read_xfer -= num;
2002                         if (num_read_xfer <= 0)
2003                                 break;
2004                         outp += num;
2005                 }
2006         }
2007
2008         return 0;
2009 }
2010
2011 static void
2012 sg_build_reserve(Sg_fd * sfp, int req_size)
2013 {
2014         Sg_scatter_hold *schp = &sfp->reserve;
2015
2016         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2017                          "sg_build_reserve: req_size=%d\n", req_size));
2018         do {
2019                 if (req_size < PAGE_SIZE)
2020                         req_size = PAGE_SIZE;
2021                 if (0 == sg_build_indirect(schp, sfp, req_size))
2022                         return;
2023                 else
2024                         sg_remove_scat(sfp, schp);
2025                 req_size >>= 1; /* divide by 2 */
2026         } while (req_size > (PAGE_SIZE / 2));
2027 }
2028
2029 static void
2030 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2031 {
2032         Sg_scatter_hold *req_schp = &srp->data;
2033         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2034         int k, num, rem;
2035
2036         srp->res_used = 1;
2037         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2038                          "sg_link_reserve: size=%d\n", size));
2039         rem = size;
2040
2041         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2042         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2043                 if (rem <= num) {
2044                         req_schp->k_use_sg = k + 1;
2045                         req_schp->sglist_len = rsv_schp->sglist_len;
2046                         req_schp->pages = rsv_schp->pages;
2047
2048                         req_schp->bufflen = size;
2049                         req_schp->page_order = rsv_schp->page_order;
2050                         break;
2051                 } else
2052                         rem -= num;
2053         }
2054
2055         if (k >= rsv_schp->k_use_sg)
2056                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2057                                  "sg_link_reserve: BAD size\n"));
2058 }
2059
2060 static void
2061 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2062 {
2063         Sg_scatter_hold *req_schp = &srp->data;
2064
2065         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2066                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2067                                       (int) req_schp->k_use_sg));
2068         req_schp->k_use_sg = 0;
2069         req_schp->bufflen = 0;
2070         req_schp->pages = NULL;
2071         req_schp->page_order = 0;
2072         req_schp->sglist_len = 0;
2073         srp->res_used = 0;
2074         /* Called without mutex lock to avoid deadlock */
2075         sfp->res_in_use = 0;
2076 }
2077
2078 static Sg_request *
2079 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2080 {
2081         Sg_request *resp;
2082         unsigned long iflags;
2083
2084         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2085         list_for_each_entry(resp, &sfp->rq_list, entry) {
2086                 /* look for requests that are ready + not SG_IO owned */
2087                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2088                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2089                         resp->done = 2; /* guard against other readers */
2090                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2091                         return resp;
2092                 }
2093         }
2094         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2095         return NULL;
2096 }
2097
2098 /* always adds to end of list */
2099 static Sg_request *
2100 sg_add_request(Sg_fd * sfp)
2101 {
2102         int k;
2103         unsigned long iflags;
2104         Sg_request *rp = sfp->req_arr;
2105
2106         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2107         if (!list_empty(&sfp->rq_list)) {
2108                 if (!sfp->cmd_q)
2109                         goto out_unlock;
2110
2111                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2112                         if (!rp->parentfp)
2113                                 break;
2114                 }
2115                 if (k >= SG_MAX_QUEUE)
2116                         goto out_unlock;
2117         }
2118         memset(rp, 0, sizeof (Sg_request));
2119         rp->parentfp = sfp;
2120         rp->header.duration = jiffies_to_msecs(jiffies);
2121         list_add_tail(&rp->entry, &sfp->rq_list);
2122         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2123         return rp;
2124 out_unlock:
2125         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2126         return NULL;
2127 }
2128
2129 /* Return of 1 for found; 0 for not found */
2130 static int
2131 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2132 {
2133         unsigned long iflags;
2134         int res = 0;
2135
2136         if (!sfp || !srp || list_empty(&sfp->rq_list))
2137                 return res;
2138         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2139         if (!list_empty(&srp->entry)) {
2140                 list_del(&srp->entry);
2141                 srp->parentfp = NULL;
2142                 res = 1;
2143         }
2144         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2145         return res;
2146 }
2147
2148 static Sg_fd *
2149 sg_add_sfp(Sg_device * sdp)
2150 {
2151         Sg_fd *sfp;
2152         unsigned long iflags;
2153         int bufflen;
2154
2155         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2156         if (!sfp)
2157                 return ERR_PTR(-ENOMEM);
2158
2159         init_waitqueue_head(&sfp->read_wait);
2160         rwlock_init(&sfp->rq_list_lock);
2161         INIT_LIST_HEAD(&sfp->rq_list);
2162         kref_init(&sfp->f_ref);
2163         mutex_init(&sfp->f_mutex);
2164         sfp->timeout = SG_DEFAULT_TIMEOUT;
2165         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2166         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2167         sfp->cmd_q = SG_DEF_COMMAND_Q;
2168         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2169         sfp->parentdp = sdp;
2170         write_lock_irqsave(&sdp->sfd_lock, iflags);
2171         if (atomic_read(&sdp->detaching)) {
2172                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2173                 kfree(sfp);
2174                 return ERR_PTR(-ENODEV);
2175         }
2176         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2177         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2178         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2179                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2180         if (unlikely(sg_big_buff != def_reserved_size))
2181                 sg_big_buff = def_reserved_size;
2182
2183         bufflen = min_t(int, sg_big_buff,
2184                         max_sectors_bytes(sdp->device->request_queue));
2185         sg_build_reserve(sfp, bufflen);
2186         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2187                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2188                                       sfp->reserve.bufflen,
2189                                       sfp->reserve.k_use_sg));
2190
2191         kref_get(&sdp->d_ref);
2192         __module_get(THIS_MODULE);
2193         return sfp;
2194 }
2195
2196 static void
2197 sg_remove_sfp_usercontext(struct work_struct *work)
2198 {
2199         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2200         struct sg_device *sdp = sfp->parentdp;
2201         Sg_request *srp;
2202         unsigned long iflags;
2203
2204         /* Cleanup any responses which were never read(). */
2205         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2206         while (!list_empty(&sfp->rq_list)) {
2207                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2208                 sg_finish_rem_req(srp);
2209                 list_del(&srp->entry);
2210                 srp->parentfp = NULL;
2211         }
2212         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2213
2214         if (sfp->reserve.bufflen > 0) {
2215                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2216                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2217                                 (int) sfp->reserve.bufflen,
2218                                 (int) sfp->reserve.k_use_sg));
2219                 sg_remove_scat(sfp, &sfp->reserve);
2220         }
2221
2222         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2223                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2224         kfree(sfp);
2225
2226         scsi_device_put(sdp->device);
2227         kref_put(&sdp->d_ref, sg_device_destroy);
2228         module_put(THIS_MODULE);
2229 }
2230
2231 static void
2232 sg_remove_sfp(struct kref *kref)
2233 {
2234         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2235         struct sg_device *sdp = sfp->parentdp;
2236         unsigned long iflags;
2237
2238         write_lock_irqsave(&sdp->sfd_lock, iflags);
2239         list_del(&sfp->sfd_siblings);
2240         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2241
2242         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2243         schedule_work(&sfp->ew.work);
2244 }
2245
2246 #ifdef CONFIG_SCSI_PROC_FS
2247 static int
2248 sg_idr_max_id(int id, void *p, void *data)
2249 {
2250         int *k = data;
2251
2252         if (*k < id)
2253                 *k = id;
2254
2255         return 0;
2256 }
2257
2258 static int
2259 sg_last_dev(void)
2260 {
2261         int k = -1;
2262         unsigned long iflags;
2263
2264         read_lock_irqsave(&sg_index_lock, iflags);
2265         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2266         read_unlock_irqrestore(&sg_index_lock, iflags);
2267         return k + 1;           /* origin 1 */
2268 }
2269 #endif
2270
2271 /* must be called with sg_index_lock held */
2272 static Sg_device *sg_lookup_dev(int dev)
2273 {
2274         return idr_find(&sg_index_idr, dev);
2275 }
2276
2277 static Sg_device *
2278 sg_get_dev(int dev)
2279 {
2280         struct sg_device *sdp;
2281         unsigned long flags;
2282
2283         read_lock_irqsave(&sg_index_lock, flags);
2284         sdp = sg_lookup_dev(dev);
2285         if (!sdp)
2286                 sdp = ERR_PTR(-ENXIO);
2287         else if (atomic_read(&sdp->detaching)) {
2288                 /* If sdp->detaching, then the refcount may already be 0, in
2289                  * which case it would be a bug to do kref_get().
2290                  */
2291                 sdp = ERR_PTR(-ENODEV);
2292         } else
2293                 kref_get(&sdp->d_ref);
2294         read_unlock_irqrestore(&sg_index_lock, flags);
2295
2296         return sdp;
2297 }
2298
2299 #ifdef CONFIG_SCSI_PROC_FS
2300 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2301
2302 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2303 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2304                                   size_t count, loff_t *off);
2305 static const struct proc_ops adio_proc_ops = {
2306         .proc_open      = sg_proc_single_open_adio,
2307         .proc_read      = seq_read,
2308         .proc_lseek     = seq_lseek,
2309         .proc_write     = sg_proc_write_adio,
2310         .proc_release   = single_release,
2311 };
2312
2313 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2314 static ssize_t sg_proc_write_dressz(struct file *filp,
2315                 const char __user *buffer, size_t count, loff_t *off);
2316 static const struct proc_ops dressz_proc_ops = {
2317         .proc_open      = sg_proc_single_open_dressz,
2318         .proc_read      = seq_read,
2319         .proc_lseek     = seq_lseek,
2320         .proc_write     = sg_proc_write_dressz,
2321         .proc_release   = single_release,
2322 };
2323
2324 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2325 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2326 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2327 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2328 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2329 static void dev_seq_stop(struct seq_file *s, void *v);
2330 static const struct seq_operations dev_seq_ops = {
2331         .start = dev_seq_start,
2332         .next  = dev_seq_next,
2333         .stop  = dev_seq_stop,
2334         .show  = sg_proc_seq_show_dev,
2335 };
2336
2337 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2338 static const struct seq_operations devstrs_seq_ops = {
2339         .start = dev_seq_start,
2340         .next  = dev_seq_next,
2341         .stop  = dev_seq_stop,
2342         .show  = sg_proc_seq_show_devstrs,
2343 };
2344
2345 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2346 static const struct seq_operations debug_seq_ops = {
2347         .start = dev_seq_start,
2348         .next  = dev_seq_next,
2349         .stop  = dev_seq_stop,
2350         .show  = sg_proc_seq_show_debug,
2351 };
2352
2353 static int
2354 sg_proc_init(void)
2355 {
2356         struct proc_dir_entry *p;
2357
2358         p = proc_mkdir("scsi/sg", NULL);
2359         if (!p)
2360                 return 1;
2361
2362         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2363         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2364         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2365         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2366         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2367         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2368         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2369         return 0;
2370 }
2371
2372
2373 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2374 {
2375         seq_printf(s, "%d\n", *((int *)s->private));
2376         return 0;
2377 }
2378
2379 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2380 {
2381         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2382 }
2383
2384 static ssize_t
2385 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2386                    size_t count, loff_t *off)
2387 {
2388         int err;
2389         unsigned long num;
2390
2391         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2392                 return -EACCES;
2393         err = kstrtoul_from_user(buffer, count, 0, &num);
2394         if (err)
2395                 return err;
2396         sg_allow_dio = num ? 1 : 0;
2397         return count;
2398 }
2399
2400 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2401 {
2402         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2403 }
2404
2405 static ssize_t
2406 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2407                      size_t count, loff_t *off)
2408 {
2409         int err;
2410         unsigned long k = ULONG_MAX;
2411
2412         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2413                 return -EACCES;
2414
2415         err = kstrtoul_from_user(buffer, count, 0, &k);
2416         if (err)
2417                 return err;
2418         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2419                 sg_big_buff = k;
2420                 return count;
2421         }
2422         return -ERANGE;
2423 }
2424
2425 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2426 {
2427         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2428                    sg_version_date);
2429         return 0;
2430 }
2431
2432 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2433 {
2434         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2435         return 0;
2436 }
2437
2438 struct sg_proc_deviter {
2439         loff_t  index;
2440         size_t  max;
2441 };
2442
2443 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2444 {
2445         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2446
2447         s->private = it;
2448         if (! it)
2449                 return NULL;
2450
2451         it->index = *pos;
2452         it->max = sg_last_dev();
2453         if (it->index >= it->max)
2454                 return NULL;
2455         return it;
2456 }
2457
2458 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2459 {
2460         struct sg_proc_deviter * it = s->private;
2461
2462         *pos = ++it->index;
2463         return (it->index < it->max) ? it : NULL;
2464 }
2465
2466 static void dev_seq_stop(struct seq_file *s, void *v)
2467 {
2468         kfree(s->private);
2469 }
2470
2471 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2472 {
2473         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2474         Sg_device *sdp;
2475         struct scsi_device *scsidp;
2476         unsigned long iflags;
2477
2478         read_lock_irqsave(&sg_index_lock, iflags);
2479         sdp = it ? sg_lookup_dev(it->index) : NULL;
2480         if ((NULL == sdp) || (NULL == sdp->device) ||
2481             (atomic_read(&sdp->detaching)))
2482                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2483         else {
2484                 scsidp = sdp->device;
2485                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2486                               scsidp->host->host_no, scsidp->channel,
2487                               scsidp->id, scsidp->lun, (int) scsidp->type,
2488                               1,
2489                               (int) scsidp->queue_depth,
2490                               (int) scsi_device_busy(scsidp),
2491                               (int) scsi_device_online(scsidp));
2492         }
2493         read_unlock_irqrestore(&sg_index_lock, iflags);
2494         return 0;
2495 }
2496
2497 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2498 {
2499         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2500         Sg_device *sdp;
2501         struct scsi_device *scsidp;
2502         unsigned long iflags;
2503
2504         read_lock_irqsave(&sg_index_lock, iflags);
2505         sdp = it ? sg_lookup_dev(it->index) : NULL;
2506         scsidp = sdp ? sdp->device : NULL;
2507         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2508                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2509                            scsidp->vendor, scsidp->model, scsidp->rev);
2510         else
2511                 seq_puts(s, "<no active device>\n");
2512         read_unlock_irqrestore(&sg_index_lock, iflags);
2513         return 0;
2514 }
2515
2516 /* must be called while holding sg_index_lock */
2517 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2518 {
2519         int k, new_interface, blen, usg;
2520         Sg_request *srp;
2521         Sg_fd *fp;
2522         const sg_io_hdr_t *hp;
2523         const char * cp;
2524         unsigned int ms;
2525
2526         k = 0;
2527         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2528                 k++;
2529                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2530                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2531                            "(res)sgat=%d low_dma=%d\n", k,
2532                            jiffies_to_msecs(fp->timeout),
2533                            fp->reserve.bufflen,
2534                            (int) fp->reserve.k_use_sg, 0);
2535                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2536                            (int) fp->cmd_q, (int) fp->force_packid,
2537                            (int) fp->keep_orphan);
2538                 list_for_each_entry(srp, &fp->rq_list, entry) {
2539                         hp = &srp->header;
2540                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2541                         if (srp->res_used) {
2542                                 if (new_interface &&
2543                                     (SG_FLAG_MMAP_IO & hp->flags))
2544                                         cp = "     mmap>> ";
2545                                 else
2546                                         cp = "     rb>> ";
2547                         } else {
2548                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2549                                         cp = "     dio>> ";
2550                                 else
2551                                         cp = "     ";
2552                         }
2553                         seq_puts(s, cp);
2554                         blen = srp->data.bufflen;
2555                         usg = srp->data.k_use_sg;
2556                         seq_puts(s, srp->done ?
2557                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2558                                   : "act:");
2559                         seq_printf(s, " id=%d blen=%d",
2560                                    srp->header.pack_id, blen);
2561                         if (srp->done)
2562                                 seq_printf(s, " dur=%d", hp->duration);
2563                         else {
2564                                 ms = jiffies_to_msecs(jiffies);
2565                                 seq_printf(s, " t_o/elap=%d/%d",
2566                                         (new_interface ? hp->timeout :
2567                                                   jiffies_to_msecs(fp->timeout)),
2568                                         (ms > hp->duration ? ms - hp->duration : 0));
2569                         }
2570                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2571                                    (int) srp->data.cmd_opcode);
2572                 }
2573                 if (list_empty(&fp->rq_list))
2574                         seq_puts(s, "     No requests active\n");
2575                 read_unlock(&fp->rq_list_lock);
2576         }
2577 }
2578
2579 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2580 {
2581         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2582         Sg_device *sdp;
2583         unsigned long iflags;
2584
2585         if (it && (0 == it->index))
2586                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2587                            (int)it->max, sg_big_buff);
2588
2589         read_lock_irqsave(&sg_index_lock, iflags);
2590         sdp = it ? sg_lookup_dev(it->index) : NULL;
2591         if (NULL == sdp)
2592                 goto skip;
2593         read_lock(&sdp->sfd_lock);
2594         if (!list_empty(&sdp->sfds)) {
2595                 seq_printf(s, " >>> device=%s ", sdp->name);
2596                 if (atomic_read(&sdp->detaching))
2597                         seq_puts(s, "detaching pending close ");
2598                 else if (sdp->device) {
2599                         struct scsi_device *scsidp = sdp->device;
2600
2601                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2602                                    scsidp->host->host_no,
2603                                    scsidp->channel, scsidp->id,
2604                                    scsidp->lun,
2605                                    scsidp->host->hostt->emulated);
2606                 }
2607                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2608                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2609                 sg_proc_debug_helper(s, sdp);
2610         }
2611         read_unlock(&sdp->sfd_lock);
2612 skip:
2613         read_unlock_irqrestore(&sg_index_lock, iflags);
2614         return 0;
2615 }
2616
2617 #endif                          /* CONFIG_SCSI_PROC_FS */
2618
2619 module_init(init_sg);
2620 module_exit(exit_sg);