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/major.h>
  35 #include <linux/slab.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/init.h>
  38 #include <linux/poll.h>
  39 #include <linux/moduleparam.h>
  40 #include <linux/cdev.h>
  41 #include <linux/idr.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/delay.h>
  45 #include <linux/blktrace_api.h>
  46 #include <linux/mutex.h>
  47 #include <linux/atomic.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/uio.h>
  50 #include <linux/cred.h> /* for sg_check_file_access() */
  51
  52 #include <scsi/scsi.h>
  53 #include <scsi/scsi_cmnd.h>
  54 #include <scsi/scsi_dbg.h>
  55 #include <scsi/scsi_device.h>
  56 #include <scsi/scsi_driver.h>
  57 #include <scsi/scsi_eh.h>
  58 #include <scsi/scsi_host.h>
  59 #include <scsi/scsi_ioctl.h>
  60 #include <scsi/scsi_tcq.h>
  61 #include <scsi/sg.h>
  62
  63 #include "scsi_logging.h"
  64
  65 #ifdef CONFIG_SCSI_PROC_FS
  66 #include <linux/proc_fs.h>
  67 static char *sg_version_date = "20140603";
  68
  69 static int sg_proc_init(void);
  70 #endif
  71
  72 #define SG_ALLOW_DIO_DEF 0
  73
  74 #define SG_MAX_DEVS 32768
  75
  76 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  77  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  78  * than 16 bytes are "variable length" whose length is a multiple of 4
  79  */
  80 #define SG_MAX_CDB_SIZE 252
  81
  82 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
  83
  84 static int sg_big_buff = SG_DEF_RESERVED_SIZE;
  85 /* N.B. This variable is readable and writeable via
  86    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
  87    of this size (or less if there is not enough memory) will be reserved
  88    for use by this file descriptor. [Deprecated usage: this variable is also
  89    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
  90    the kernel (i.e. it is not a module).] */
  91 static int def_reserved_size = -1;      /* picks up init parameter */
  92 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
  93
  94 static int scatter_elem_sz = SG_SCATTER_SZ;
  95 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
  96
  97 #define SG_SECTOR_SZ 512
  98
  99 static int sg_add_device(struct device *, struct class_interface *);
 100 static void sg_remove_device(struct device *, struct class_interface *);
 101
 102 static DEFINE_IDR(sg_index_idr);
 103 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
 104                                                            file descriptor list for device */
 105
 106 static struct class_interface sg_interface = {
 107         .add_dev        = sg_add_device,
 108         .remove_dev     = sg_remove_device,
 109 };
 110
 111 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
 112         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
 113         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
 114         unsigned bufflen;       /* Size of (aggregate) data buffer */
 115         struct page **pages;
 116         int page_order;
 117         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
 118         unsigned char cmd_opcode; /* first byte of command */
 119 } Sg_scatter_hold;
 120
 121 struct sg_device;               /* forward declarations */
 122 struct sg_fd;
 123
 124 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
 125         struct list_head entry; /* list entry */
 126         struct sg_fd *parentfp; /* NULL -> not in use */
 127         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
 128         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
 129         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
 130         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
 131         char orphan;            /* 1 -> drop on sight, 0 -> normal */
 132         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
 133         /* done protected by rq_list_lock */
 134         char done;              /* 0->before bh, 1->before read, 2->read */
 135         struct request *rq;
 136         struct bio *bio;
 137         struct execute_work ew;
 138 } Sg_request;
 139
 140 typedef struct sg_fd {          /* holds the state of a file descriptor */
 141         struct list_head sfd_siblings;  /* protected by device's sfd_lock */
 142         struct sg_device *parentdp;     /* owning device */
 143         wait_queue_head_t read_wait;    /* queue read until command done */
 144         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
 145         struct mutex f_mutex;   /* protect against changes in this fd */
 146         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 147         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 148         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 149         struct list_head rq_list; /* head of request list */
 150         struct fasync_struct *async_qp; /* used by asynchronous notification */
 151         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
 152         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 153         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 154         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 155         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 156         char mmap_called;       /* 0 -> mmap() never called on this fd */
 157         char res_in_use;        /* 1 -> 'reserve' array in use */
 158         struct kref f_ref;
 159         struct execute_work ew;
 160 } Sg_fd;
 161
 162 typedef struct sg_device { /* holds the state of each scsi generic device */
 163         struct scsi_device *device;
 164         wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
 165         struct mutex open_rel_lock;     /* held when in open() or release() */
 166         int sg_tablesize;       /* adapter's max scatter-gather table size */
 167         u32 index;              /* device index number */
 168         struct list_head sfds;
 169         rwlock_t sfd_lock;      /* protect access to sfd list */
 170         atomic_t detaching;     /* 0->device usable, 1->device detaching */
 171         bool exclude;           /* 1->open(O_EXCL) succeeded and is active */
 172         int open_cnt;           /* count of opens (perhaps < num(sfds) ) */
 173         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
 174         char name[DISK_NAME_LEN];
 175         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
 176         struct kref d_ref;
 177 } Sg_device;
 178
 179 /* tasklet or soft irq callback */
 180 static void sg_rq_end_io(struct request *rq, blk_status_t status);
 181 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
 182 static int sg_finish_rem_req(Sg_request * srp);
 183 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
 184 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
 185                            Sg_request * srp);
 186 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
 187                         const char __user *buf, size_t count, int blocking,
 188                         int read_only, int sg_io_owned, Sg_request **o_srp);
 189 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
 190                            unsigned char *cmnd, int timeout, int blocking);
 191 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
 192 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
 193 static void sg_build_reserve(Sg_fd * sfp, int req_size);
 194 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 195 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 196 static Sg_fd *sg_add_sfp(Sg_device * sdp);
 197 static void sg_remove_sfp(struct kref *);
 198 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 199 static Sg_request *sg_add_request(Sg_fd * sfp);
 200 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 201 static Sg_device *sg_get_dev(int dev);
 202 static void sg_device_destroy(struct kref *kref);
 203
 204 #define SZ_SG_HEADER sizeof(struct sg_header)
 205 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 206 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 207 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 208
 209 #define sg_printk(prefix, sdp, fmt, a...) \
 210         sdev_prefix_printk(prefix, (sdp)->device, (sdp)->name, fmt, ##a)
 211
 212 /*
 213  * The SCSI interfaces that use read() and write() as an asynchronous variant of
 214  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
 215  * to trigger read() and write() calls from various contexts with elevated
 216  * privileges. This can lead to kernel memory corruption (e.g. if these
 217  * interfaces are called through splice()) and privilege escalation inside
 218  * userspace (e.g. if a process with access to such a device passes a file
 219  * descriptor to a SUID binary as stdin/stdout/stderr).
 220  *
 221  * This function provides protection for the legacy API by restricting the
 222  * calling context.
 223  */
 224 static int sg_check_file_access(struct file *filp, const char *caller)
 225 {
 226         if (filp->f_cred != current_real_cred()) {
 227                 pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
 228                         caller, task_tgid_vnr(current), current->comm);
 229                 return -EPERM;
 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         if (!scsi_cmd_allowed(cmd, filp->f_mode))
 241                 return -EPERM;
 242         return 0;
 243 }
 244
 245 static int
 246 open_wait(Sg_device *sdp, int flags)
 247 {
 248         int retval = 0;
 249
 250         if (flags & O_EXCL) {
 251                 while (sdp->open_cnt > 0) {
 252                         mutex_unlock(&sdp->open_rel_lock);
 253                         retval = wait_event_interruptible(sdp->open_wait,
 254                                         (atomic_read(&sdp->detaching) ||
 255                                          !sdp->open_cnt));
 256                         mutex_lock(&sdp->open_rel_lock);
 257
 258                         if (retval) /* -ERESTARTSYS */
 259                                 return retval;
 260                         if (atomic_read(&sdp->detaching))
 261                                 return -ENODEV;
 262                 }
 263         } else {
 264                 while (sdp->exclude) {
 265                         mutex_unlock(&sdp->open_rel_lock);
 266                         retval = wait_event_interruptible(sdp->open_wait,
 267                                         (atomic_read(&sdp->detaching) ||
 268                                          !sdp->exclude));
 269                         mutex_lock(&sdp->open_rel_lock);
 270
 271                         if (retval) /* -ERESTARTSYS */
 272                                 return retval;
 273                         if (atomic_read(&sdp->detaching))
 274                                 return -ENODEV;
 275                 }
 276         }
 277
 278         return retval;
 279 }
 280
 281 /* Returns 0 on success, else a negated errno value */
 282 static int
 283 sg_open(struct inode *inode, struct file *filp)
 284 {
 285         int dev = iminor(inode);
 286         int flags = filp->f_flags;
 287         struct request_queue *q;
 288         Sg_device *sdp;
 289         Sg_fd *sfp;
 290         int retval;
 291
 292         nonseekable_open(inode, filp);
 293         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 294                 return -EPERM; /* Can't lock it with read only access */
 295         sdp = sg_get_dev(dev);
 296         if (IS_ERR(sdp))
 297                 return PTR_ERR(sdp);
 298
 299         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 300                                       "sg_open: flags=0x%x\n", flags));
 301
 302         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 303         /* Prevent the device driver from vanishing while we sleep */
 304         retval = scsi_device_get(sdp->device);
 305         if (retval)
 306                 goto sg_put;
 307
 308         retval = scsi_autopm_get_device(sdp->device);
 309         if (retval)
 310                 goto sdp_put;
 311
 312         /* scsi_block_when_processing_errors() may block so bypass
 313          * check if O_NONBLOCK. Permits SCSI commands to be issued
 314          * during error recovery. Tread carefully. */
 315         if (!((flags & O_NONBLOCK) ||
 316               scsi_block_when_processing_errors(sdp->device))) {
 317                 retval = -ENXIO;
 318                 /* we are in error recovery for this device */
 319                 goto error_out;
 320         }
 321
 322         mutex_lock(&sdp->open_rel_lock);
 323         if (flags & O_NONBLOCK) {
 324                 if (flags & O_EXCL) {
 325                         if (sdp->open_cnt > 0) {
 326                                 retval = -EBUSY;
 327                                 goto error_mutex_locked;
 328                         }
 329                 } else {
 330                         if (sdp->exclude) {
 331                                 retval = -EBUSY;
 332                                 goto error_mutex_locked;
 333                         }
 334                 }
 335         } else {
 336                 retval = open_wait(sdp, flags);
 337                 if (retval) /* -ERESTARTSYS or -ENODEV */
 338                         goto error_mutex_locked;
 339         }
 340
 341         /* N.B. at this point we are holding the open_rel_lock */
 342         if (flags & O_EXCL)
 343                 sdp->exclude = true;
 344
 345         if (sdp->open_cnt < 1) {  /* no existing opens */
 346                 sdp->sgdebug = 0;
 347                 q = sdp->device->request_queue;
 348                 sdp->sg_tablesize = queue_max_segments(q);
 349         }
 350         sfp = sg_add_sfp(sdp);
 351         if (IS_ERR(sfp)) {
 352                 retval = PTR_ERR(sfp);
 353                 goto out_undo;
 354         }
 355
 356         filp->private_data = sfp;
 357         sdp->open_cnt++;
 358         mutex_unlock(&sdp->open_rel_lock);
 359
 360         retval = 0;
 361 sg_put:
 362         kref_put(&sdp->d_ref, sg_device_destroy);
 363         return retval;
 364
 365 out_undo:
 366         if (flags & O_EXCL) {
 367                 sdp->exclude = false;   /* undo if error */
 368                 wake_up_interruptible(&sdp->open_wait);
 369         }
 370 error_mutex_locked:
 371         mutex_unlock(&sdp->open_rel_lock);
 372 error_out:
 373         scsi_autopm_put_device(sdp->device);
 374 sdp_put:
 375         scsi_device_put(sdp->device);
 376         goto sg_put;
 377 }
 378
 379 /* Release resources associated with a successful sg_open()
 380  * Returns 0 on success, else a negated errno value */
 381 static int
 382 sg_release(struct inode *inode, struct file *filp)
 383 {
 384         Sg_device *sdp;
 385         Sg_fd *sfp;
 386
 387         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 388                 return -ENXIO;
 389         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 390
 391         mutex_lock(&sdp->open_rel_lock);
 392         scsi_autopm_put_device(sdp->device);
 393         kref_put(&sfp->f_ref, sg_remove_sfp);
 394         sdp->open_cnt--;
 395
 396         /* possibly many open()s waiting on exlude clearing, start many;
 397          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 398         if (sdp->exclude) {
 399                 sdp->exclude = false;
 400                 wake_up_interruptible_all(&sdp->open_wait);
 401         } else if (0 == sdp->open_cnt) {
 402                 wake_up_interruptible(&sdp->open_wait);
 403         }
 404         mutex_unlock(&sdp->open_rel_lock);
 405         return 0;
 406 }
 407
 408 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 409 {
 410         struct sg_header __user *old_hdr = buf;
 411         int reply_len;
 412
 413         if (count >= SZ_SG_HEADER) {
 414                 /* negative reply_len means v3 format, otherwise v1/v2 */
 415                 if (get_user(reply_len, &old_hdr->reply_len))
 416                         return -EFAULT;
 417
 418                 if (reply_len >= 0)
 419                         return get_user(*pack_id, &old_hdr->pack_id);
 420
 421                 if (in_compat_syscall() &&
 422                     count >= sizeof(struct compat_sg_io_hdr)) {
 423                         struct compat_sg_io_hdr __user *hp = buf;
 424
 425                         return get_user(*pack_id, &hp->pack_id);
 426                 }
 427
 428                 if (count >= sizeof(struct sg_io_hdr)) {
 429                         struct sg_io_hdr __user *hp = buf;
 430
 431                         return get_user(*pack_id, &hp->pack_id);
 432                 }
 433         }
 434
 435         /* no valid header was passed, so ignore the pack_id */
 436         *pack_id = -1;
 437         return 0;
 438 }
 439
 440 static ssize_t
 441 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 442 {
 443         Sg_device *sdp;
 444         Sg_fd *sfp;
 445         Sg_request *srp;
 446         int req_pack_id = -1;
 447         sg_io_hdr_t *hp;
 448         struct sg_header *old_hdr;
 449         int retval;
 450
 451         /*
 452          * This could cause a response to be stranded. Close the associated
 453          * file descriptor to free up any resources being held.
 454          */
 455         retval = sg_check_file_access(filp, __func__);
 456         if (retval)
 457                 return retval;
 458
 459         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 460                 return -ENXIO;
 461         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 462                                       "sg_read: count=%d\n", (int) count));
 463
 464         if (sfp->force_packid)
 465                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 466         if (retval)
 467                 return retval;
 468
 469         srp = sg_get_rq_mark(sfp, req_pack_id);
 470         if (!srp) {             /* now wait on packet to arrive */
 471                 if (atomic_read(&sdp->detaching))
 472                         return -ENODEV;
 473                 if (filp->f_flags & O_NONBLOCK)
 474                         return -EAGAIN;
 475                 retval = wait_event_interruptible(sfp->read_wait,
 476                         (atomic_read(&sdp->detaching) ||
 477                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 478                 if (atomic_read(&sdp->detaching))
 479                         return -ENODEV;
 480                 if (retval)
 481                         /* -ERESTARTSYS as signal hit process */
 482                         return retval;
 483         }
 484         if (srp->header.interface_id != '\0')
 485                 return sg_new_read(sfp, buf, count, srp);
 486
 487         hp = &srp->header;
 488         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 489         if (!old_hdr)
 490                 return -ENOMEM;
 491
 492         old_hdr->reply_len = (int) hp->timeout;
 493         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 494         old_hdr->pack_id = hp->pack_id;
 495         old_hdr->twelve_byte =
 496             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 497         old_hdr->target_status = hp->masked_status;
 498         old_hdr->host_status = hp->host_status;
 499         old_hdr->driver_status = hp->driver_status;
 500         if ((CHECK_CONDITION & hp->masked_status) ||
 501             (srp->sense_b[0] & 0x70) == 0x70) {
 502                 old_hdr->driver_status = DRIVER_SENSE;
 503                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 504                        sizeof (old_hdr->sense_buffer));
 505         }
 506         switch (hp->host_status) {
 507         /* This setup of 'result' is for backward compatibility and is best
 508            ignored by the user who should use target, host + driver status */
 509         case DID_OK:
 510         case DID_PASSTHROUGH:
 511         case DID_SOFT_ERROR:
 512                 old_hdr->result = 0;
 513                 break;
 514         case DID_NO_CONNECT:
 515         case DID_BUS_BUSY:
 516         case DID_TIME_OUT:
 517                 old_hdr->result = EBUSY;
 518                 break;
 519         case DID_BAD_TARGET:
 520         case DID_ABORT:
 521         case DID_PARITY:
 522         case DID_RESET:
 523         case DID_BAD_INTR:
 524                 old_hdr->result = EIO;
 525                 break;
 526         case DID_ERROR:
 527                 old_hdr->result = (srp->sense_b[0] == 0 &&
 528                                   hp->masked_status == GOOD) ? 0 : EIO;
 529                 break;
 530         default:
 531                 old_hdr->result = EIO;
 532                 break;
 533         }
 534
 535         /* Now copy the result back to the user buffer.  */
 536         if (count >= SZ_SG_HEADER) {
 537                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 538                         retval = -EFAULT;
 539                         goto free_old_hdr;
 540                 }
 541                 buf += SZ_SG_HEADER;
 542                 if (count > old_hdr->reply_len)
 543                         count = old_hdr->reply_len;
 544                 if (count > SZ_SG_HEADER) {
 545                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 546                                 retval = -EFAULT;
 547                                 goto free_old_hdr;
 548                         }
 549                 }
 550         } else
 551                 count = (old_hdr->result == 0) ? 0 : -EIO;
 552         sg_finish_rem_req(srp);
 553         sg_remove_request(sfp, srp);
 554         retval = count;
 555 free_old_hdr:
 556         kfree(old_hdr);
 557         return retval;
 558 }
 559
 560 static ssize_t
 561 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 562 {
 563         sg_io_hdr_t *hp = &srp->header;
 564         int err = 0, err2;
 565         int len;
 566
 567         if (in_compat_syscall()) {
 568                 if (count < sizeof(struct compat_sg_io_hdr)) {
 569                         err = -EINVAL;
 570                         goto err_out;
 571                 }
 572         } else if (count < SZ_SG_IO_HDR) {
 573                 err = -EINVAL;
 574                 goto err_out;
 575         }
 576         hp->sb_len_wr = 0;
 577         if ((hp->mx_sb_len > 0) && hp->sbp) {
 578                 if ((CHECK_CONDITION & hp->masked_status) ||
 579                     (srp->sense_b[0] & 0x70) == 0x70) {
 580                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 581                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 582                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 583                         len = (len > sb_len) ? sb_len : len;
 584                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 585                                 err = -EFAULT;
 586                                 goto err_out;
 587                         }
 588                         hp->driver_status = DRIVER_SENSE;
 589                         hp->sb_len_wr = len;
 590                 }
 591         }
 592         if (hp->masked_status || hp->host_status || hp->driver_status)
 593                 hp->info |= SG_INFO_CHECK;
 594         err = put_sg_io_hdr(hp, buf);
 595 err_out:
 596         err2 = sg_finish_rem_req(srp);
 597         sg_remove_request(sfp, srp);
 598         return err ? : err2 ? : count;
 599 }
 600
 601 static ssize_t
 602 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 603 {
 604         int mxsize, cmd_size, k;
 605         int input_size, blocking;
 606         unsigned char opcode;
 607         Sg_device *sdp;
 608         Sg_fd *sfp;
 609         Sg_request *srp;
 610         struct sg_header old_hdr;
 611         sg_io_hdr_t *hp;
 612         unsigned char cmnd[SG_MAX_CDB_SIZE];
 613         int retval;
 614
 615         retval = sg_check_file_access(filp, __func__);
 616         if (retval)
 617                 return retval;
 618
 619         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 620                 return -ENXIO;
 621         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 622                                       "sg_write: count=%d\n", (int) count));
 623         if (atomic_read(&sdp->detaching))
 624                 return -ENODEV;
 625         if (!((filp->f_flags & O_NONBLOCK) ||
 626               scsi_block_when_processing_errors(sdp->device)))
 627                 return -ENXIO;
 628
 629         if (count < SZ_SG_HEADER)
 630                 return -EIO;
 631         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 632                 return -EFAULT;
 633         blocking = !(filp->f_flags & O_NONBLOCK);
 634         if (old_hdr.reply_len < 0)
 635                 return sg_new_write(sfp, filp, buf, count,
 636                                     blocking, 0, 0, NULL);
 637         if (count < (SZ_SG_HEADER + 6))
 638                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 639
 640         buf += SZ_SG_HEADER;
 641         if (get_user(opcode, buf))
 642                 return -EFAULT;
 643
 644         if (!(srp = sg_add_request(sfp))) {
 645                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 646                                               "sg_write: queue full\n"));
 647                 return -EDOM;
 648         }
 649         mutex_lock(&sfp->f_mutex);
 650         if (sfp->next_cmd_len > 0) {
 651                 cmd_size = sfp->next_cmd_len;
 652                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 653         } else {
 654                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 655                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 656                         cmd_size = 12;
 657         }
 658         mutex_unlock(&sfp->f_mutex);
 659         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 660                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 661 /* Determine buffer size.  */
 662         input_size = count - cmd_size;
 663         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 664         mxsize -= SZ_SG_HEADER;
 665         input_size -= SZ_SG_HEADER;
 666         if (input_size < 0) {
 667                 sg_remove_request(sfp, srp);
 668                 return -EIO;    /* User did not pass enough bytes for this command. */
 669         }
 670         hp = &srp->header;
 671         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 672         hp->cmd_len = (unsigned char) cmd_size;
 673         hp->iovec_count = 0;
 674         hp->mx_sb_len = 0;
 675         if (input_size > 0)
 676                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 677                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 678         else
 679                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 680         hp->dxfer_len = mxsize;
 681         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 682             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 683                 hp->dxferp = (char __user *)buf + cmd_size;
 684         else
 685                 hp->dxferp = NULL;
 686         hp->sbp = NULL;
 687         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 688         hp->flags = input_size; /* structure abuse ... */
 689         hp->pack_id = old_hdr.pack_id;
 690         hp->usr_ptr = NULL;
 691         if (copy_from_user(cmnd, buf, cmd_size)) {
 692                 sg_remove_request(sfp, srp);
 693                 return -EFAULT;
 694         }
 695         /*
 696          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 697          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 698          * is a non-zero input_size, so emit a warning.
 699          */
 700         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 701                 printk_ratelimited(KERN_WARNING
 702                                    "sg_write: data in/out %d/%d bytes "
 703                                    "for SCSI command 0x%x-- guessing "
 704                                    "data in;\n   program %s not setting "
 705                                    "count and/or reply_len properly\n",
 706                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 707                                    input_size, (unsigned int) cmnd[0],
 708                                    current->comm);
 709         }
 710         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 711         return (k < 0) ? k : count;
 712 }
 713
 714 static ssize_t
 715 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 716                  size_t count, int blocking, int read_only, int sg_io_owned,
 717                  Sg_request **o_srp)
 718 {
 719         int k;
 720         Sg_request *srp;
 721         sg_io_hdr_t *hp;
 722         unsigned char cmnd[SG_MAX_CDB_SIZE];
 723         int timeout;
 724         unsigned long ul_timeout;
 725
 726         if (count < SZ_SG_IO_HDR)
 727                 return -EINVAL;
 728
 729         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 730         if (!(srp = sg_add_request(sfp))) {
 731                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 732                                               "sg_new_write: queue full\n"));
 733                 return -EDOM;
 734         }
 735         srp->sg_io_owned = sg_io_owned;
 736         hp = &srp->header;
 737         if (get_sg_io_hdr(hp, buf)) {
 738                 sg_remove_request(sfp, srp);
 739                 return -EFAULT;
 740         }
 741         if (hp->interface_id != 'S') {
 742                 sg_remove_request(sfp, srp);
 743                 return -ENOSYS;
 744         }
 745         if (hp->flags & SG_FLAG_MMAP_IO) {
 746                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 747                         sg_remove_request(sfp, srp);
 748                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 749                 }
 750                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 751                         sg_remove_request(sfp, srp);
 752                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 753                 }
 754                 if (sfp->res_in_use) {
 755                         sg_remove_request(sfp, srp);
 756                         return -EBUSY;  /* reserve buffer already being used */
 757                 }
 758         }
 759         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 760         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 761         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 762                 sg_remove_request(sfp, srp);
 763                 return -EMSGSIZE;
 764         }
 765         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 766                 sg_remove_request(sfp, srp);
 767                 return -EFAULT;
 768         }
 769         if (read_only && sg_allow_access(file, cmnd)) {
 770                 sg_remove_request(sfp, srp);
 771                 return -EPERM;
 772         }
 773         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 774         if (k < 0)
 775                 return k;
 776         if (o_srp)
 777                 *o_srp = srp;
 778         return count;
 779 }
 780
 781 static int
 782 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 783                 unsigned char *cmnd, int timeout, int blocking)
 784 {
 785         int k, at_head;
 786         Sg_device *sdp = sfp->parentdp;
 787         sg_io_hdr_t *hp = &srp->header;
 788
 789         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 790         hp->status = 0;
 791         hp->masked_status = 0;
 792         hp->msg_status = 0;
 793         hp->info = 0;
 794         hp->host_status = 0;
 795         hp->driver_status = 0;
 796         hp->resid = 0;
 797         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 798                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 799                         (int) cmnd[0], (int) hp->cmd_len));
 800
 801         if (hp->dxfer_len >= SZ_256M) {
 802                 sg_remove_request(sfp, srp);
 803                 return -EINVAL;
 804         }
 805
 806         k = sg_start_req(srp, cmnd);
 807         if (k) {
 808                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 809                         "sg_common_write: start_req err=%d\n", k));
 810                 sg_finish_rem_req(srp);
 811                 sg_remove_request(sfp, srp);
 812                 return k;       /* probably out of space --> ENOMEM */
 813         }
 814         if (atomic_read(&sdp->detaching)) {
 815                 if (srp->bio) {
 816                         blk_mq_free_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         srp->rq->end_io = sg_rq_end_io;
 835         blk_execute_rq_nowait(srp->rq, at_head);
 836         return 0;
 837 }
 838
 839 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 840 {
 841         unsigned long flags;
 842         int ret;
 843
 844         read_lock_irqsave(&sfp->rq_list_lock, flags);
 845         ret = srp->done;
 846         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 847         return ret;
 848 }
 849
 850 static int max_sectors_bytes(struct request_queue *q)
 851 {
 852         unsigned int max_sectors = queue_max_sectors(q);
 853
 854         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 855
 856         return max_sectors << 9;
 857 }
 858
 859 static void
 860 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 861 {
 862         Sg_request *srp;
 863         int val;
 864         unsigned int ms;
 865
 866         val = 0;
 867         list_for_each_entry(srp, &sfp->rq_list, entry) {
 868                 if (val >= SG_MAX_QUEUE)
 869                         break;
 870                 rinfo[val].req_state = srp->done + 1;
 871                 rinfo[val].problem =
 872                         srp->header.masked_status &
 873                         srp->header.host_status &
 874                         srp->header.driver_status;
 875                 if (srp->done)
 876                         rinfo[val].duration =
 877                                 srp->header.duration;
 878                 else {
 879                         ms = jiffies_to_msecs(jiffies);
 880                         rinfo[val].duration =
 881                                 (ms > srp->header.duration) ?
 882                                 (ms - srp->header.duration) : 0;
 883                 }
 884                 rinfo[val].orphan = srp->orphan;
 885                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 886                 rinfo[val].pack_id = srp->header.pack_id;
 887                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 888                 val++;
 889         }
 890 }
 891
 892 #ifdef CONFIG_COMPAT
 893 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 894         char req_state;
 895         char orphan;
 896         char sg_io_owned;
 897         char problem;
 898         int pack_id;
 899         compat_uptr_t usr_ptr;
 900         unsigned int duration;
 901         int unused;
 902 };
 903
 904 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 905                                     struct sg_req_info *rinfo)
 906 {
 907         int i;
 908         for (i = 0; i < SG_MAX_QUEUE; i++) {
 909                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 910                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 911                     put_user(rinfo[i].duration, &o[i].duration) ||
 912                     put_user(rinfo[i].unused, &o[i].unused))
 913                         return -EFAULT;
 914         }
 915         return 0;
 916 }
 917 #endif
 918
 919 static long
 920 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 921                 unsigned int cmd_in, void __user *p)
 922 {
 923         int __user *ip = p;
 924         int result, val, read_only;
 925         Sg_request *srp;
 926         unsigned long iflags;
 927
 928         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 929                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 930         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 931
 932         switch (cmd_in) {
 933         case SG_IO:
 934                 if (atomic_read(&sdp->detaching))
 935                         return -ENODEV;
 936                 if (!scsi_block_when_processing_errors(sdp->device))
 937                         return -ENXIO;
 938                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 939                                  1, read_only, 1, &srp);
 940                 if (result < 0)
 941                         return result;
 942                 result = wait_event_interruptible(sfp->read_wait,
 943                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 944                 if (atomic_read(&sdp->detaching))
 945                         return -ENODEV;
 946                 write_lock_irq(&sfp->rq_list_lock);
 947                 if (srp->done) {
 948                         srp->done = 2;
 949                         write_unlock_irq(&sfp->rq_list_lock);
 950                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 951                         return (result < 0) ? result : 0;
 952                 }
 953                 srp->orphan = 1;
 954                 write_unlock_irq(&sfp->rq_list_lock);
 955                 return result;  /* -ERESTARTSYS because signal hit process */
 956         case SG_SET_TIMEOUT:
 957                 result = get_user(val, ip);
 958                 if (result)
 959                         return result;
 960                 if (val < 0)
 961                         return -EIO;
 962                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 963                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 964                                     INT_MAX);
 965                 sfp->timeout_user = val;
 966                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 967
 968                 return 0;
 969         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 970                                 /* strange ..., for backward compatibility */
 971                 return sfp->timeout_user;
 972         case SG_SET_FORCE_LOW_DMA:
 973                 /*
 974                  * N.B. This ioctl never worked properly, but failed to
 975                  * return an error value. So returning '0' to keep compability
 976                  * with legacy applications.
 977                  */
 978                 return 0;
 979         case SG_GET_LOW_DMA:
 980                 return put_user(0, ip);
 981         case SG_GET_SCSI_ID:
 982                 {
 983                         sg_scsi_id_t v;
 984
 985                         if (atomic_read(&sdp->detaching))
 986                                 return -ENODEV;
 987                         memset(&v, 0, sizeof(v));
 988                         v.host_no = sdp->device->host->host_no;
 989                         v.channel = sdp->device->channel;
 990                         v.scsi_id = sdp->device->id;
 991                         v.lun = sdp->device->lun;
 992                         v.scsi_type = sdp->device->type;
 993                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 994                         v.d_queue_depth = sdp->device->queue_depth;
 995                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 996                                 return -EFAULT;
 997                         return 0;
 998                 }
 999         case SG_SET_FORCE_PACK_ID:
1000                 result = get_user(val, ip);
1001                 if (result)
1002                         return result;
1003                 sfp->force_packid = val ? 1 : 0;
1004                 return 0;
1005         case SG_GET_PACK_ID:
1006                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1007                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1008                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1009                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1010                                                        iflags);
1011                                 return put_user(srp->header.pack_id, ip);
1012                         }
1013                 }
1014                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1015                 return put_user(-1, ip);
1016         case SG_GET_NUM_WAITING:
1017                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1018                 val = 0;
1019                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1020                         if ((1 == srp->done) && (!srp->sg_io_owned))
1021                                 ++val;
1022                 }
1023                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1024                 return put_user(val, ip);
1025         case SG_GET_SG_TABLESIZE:
1026                 return put_user(sdp->sg_tablesize, ip);
1027         case SG_SET_RESERVED_SIZE:
1028                 result = get_user(val, ip);
1029                 if (result)
1030                         return result;
1031                 if (val < 0)
1032                         return -EINVAL;
1033                 val = min_t(int, val,
1034                             max_sectors_bytes(sdp->device->request_queue));
1035                 mutex_lock(&sfp->f_mutex);
1036                 if (val != sfp->reserve.bufflen) {
1037                         if (sfp->mmap_called ||
1038                             sfp->res_in_use) {
1039                                 mutex_unlock(&sfp->f_mutex);
1040                                 return -EBUSY;
1041                         }
1042
1043                         sg_remove_scat(sfp, &sfp->reserve);
1044                         sg_build_reserve(sfp, val);
1045                 }
1046                 mutex_unlock(&sfp->f_mutex);
1047                 return 0;
1048         case SG_GET_RESERVED_SIZE:
1049                 val = min_t(int, sfp->reserve.bufflen,
1050                             max_sectors_bytes(sdp->device->request_queue));
1051                 return put_user(val, ip);
1052         case SG_SET_COMMAND_Q:
1053                 result = get_user(val, ip);
1054                 if (result)
1055                         return result;
1056                 sfp->cmd_q = val ? 1 : 0;
1057                 return 0;
1058         case SG_GET_COMMAND_Q:
1059                 return put_user((int) sfp->cmd_q, ip);
1060         case SG_SET_KEEP_ORPHAN:
1061                 result = get_user(val, ip);
1062                 if (result)
1063                         return result;
1064                 sfp->keep_orphan = val;
1065                 return 0;
1066         case SG_GET_KEEP_ORPHAN:
1067                 return put_user((int) sfp->keep_orphan, ip);
1068         case SG_NEXT_CMD_LEN:
1069                 result = get_user(val, ip);
1070                 if (result)
1071                         return result;
1072                 if (val > SG_MAX_CDB_SIZE)
1073                         return -ENOMEM;
1074                 sfp->next_cmd_len = (val > 0) ? val : 0;
1075                 return 0;
1076         case SG_GET_VERSION_NUM:
1077                 return put_user(sg_version_num, ip);
1078         case SG_GET_ACCESS_COUNT:
1079                 /* faked - we don't have a real access count anymore */
1080                 val = (sdp->device ? 1 : 0);
1081                 return put_user(val, ip);
1082         case SG_GET_REQUEST_TABLE:
1083                 {
1084                         sg_req_info_t *rinfo;
1085
1086                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1087                                         GFP_KERNEL);
1088                         if (!rinfo)
1089                                 return -ENOMEM;
1090                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1091                         sg_fill_request_table(sfp, rinfo);
1092                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1093         #ifdef CONFIG_COMPAT
1094                         if (in_compat_syscall())
1095                                 result = put_compat_request_table(p, rinfo);
1096                         else
1097         #endif
1098                                 result = copy_to_user(p, rinfo,
1099                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1100                         result = result ? -EFAULT : 0;
1101                         kfree(rinfo);
1102                         return result;
1103                 }
1104         case SG_EMULATED_HOST:
1105                 if (atomic_read(&sdp->detaching))
1106                         return -ENODEV;
1107                 return put_user(sdp->device->host->hostt->emulated, ip);
1108         case SCSI_IOCTL_SEND_COMMAND:
1109                 if (atomic_read(&sdp->detaching))
1110                         return -ENODEV;
1111                 return scsi_ioctl(sdp->device, filp->f_mode, cmd_in, p);
1112         case SG_SET_DEBUG:
1113                 result = get_user(val, ip);
1114                 if (result)
1115                         return result;
1116                 sdp->sgdebug = (char) val;
1117                 return 0;
1118         case BLKSECTGET:
1119                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1120                                 ip);
1121         case BLKTRACESETUP:
1122                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1123                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1124                                        NULL, p);
1125         case BLKTRACESTART:
1126                 return blk_trace_startstop(sdp->device->request_queue, 1);
1127         case BLKTRACESTOP:
1128                 return blk_trace_startstop(sdp->device->request_queue, 0);
1129         case BLKTRACETEARDOWN:
1130                 return blk_trace_remove(sdp->device->request_queue);
1131         case SCSI_IOCTL_GET_IDLUN:
1132         case SCSI_IOCTL_GET_BUS_NUMBER:
1133         case SCSI_IOCTL_PROBE_HOST:
1134         case SG_GET_TRANSFORM:
1135         case SG_SCSI_RESET:
1136                 if (atomic_read(&sdp->detaching))
1137                         return -ENODEV;
1138                 break;
1139         default:
1140                 if (read_only)
1141                         return -EPERM;  /* don't know so take safe approach */
1142                 break;
1143         }
1144
1145         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1146                         cmd_in, filp->f_flags & O_NDELAY);
1147         if (result)
1148                 return result;
1149
1150         return -ENOIOCTLCMD;
1151 }
1152
1153 static long
1154 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1155 {
1156         void __user *p = (void __user *)arg;
1157         Sg_device *sdp;
1158         Sg_fd *sfp;
1159         int ret;
1160
1161         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1162                 return -ENXIO;
1163
1164         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1165         if (ret != -ENOIOCTLCMD)
1166                 return ret;
1167         return scsi_ioctl(sdp->device, filp->f_mode, cmd_in, p);
1168 }
1169
1170 static __poll_t
1171 sg_poll(struct file *filp, poll_table * wait)
1172 {
1173         __poll_t res = 0;
1174         Sg_device *sdp;
1175         Sg_fd *sfp;
1176         Sg_request *srp;
1177         int count = 0;
1178         unsigned long iflags;
1179
1180         sfp = filp->private_data;
1181         if (!sfp)
1182                 return EPOLLERR;
1183         sdp = sfp->parentdp;
1184         if (!sdp)
1185                 return EPOLLERR;
1186         poll_wait(filp, &sfp->read_wait, wait);
1187         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1188         list_for_each_entry(srp, &sfp->rq_list, entry) {
1189                 /* if any read waiting, flag it */
1190                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1191                         res = EPOLLIN | EPOLLRDNORM;
1192                 ++count;
1193         }
1194         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1195
1196         if (atomic_read(&sdp->detaching))
1197                 res |= EPOLLHUP;
1198         else if (!sfp->cmd_q) {
1199                 if (0 == count)
1200                         res |= EPOLLOUT | EPOLLWRNORM;
1201         } else if (count < SG_MAX_QUEUE)
1202                 res |= EPOLLOUT | EPOLLWRNORM;
1203         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1204                                       "sg_poll: res=0x%x\n", (__force u32) res));
1205         return res;
1206 }
1207
1208 static int
1209 sg_fasync(int fd, struct file *filp, int mode)
1210 {
1211         Sg_device *sdp;
1212         Sg_fd *sfp;
1213
1214         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1215                 return -ENXIO;
1216         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1217                                       "sg_fasync: mode=%d\n", mode));
1218
1219         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1220 }
1221
1222 static vm_fault_t
1223 sg_vma_fault(struct vm_fault *vmf)
1224 {
1225         struct vm_area_struct *vma = vmf->vma;
1226         Sg_fd *sfp;
1227         unsigned long offset, len, sa;
1228         Sg_scatter_hold *rsv_schp;
1229         int k, length;
1230
1231         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1232                 return VM_FAULT_SIGBUS;
1233         rsv_schp = &sfp->reserve;
1234         offset = vmf->pgoff << PAGE_SHIFT;
1235         if (offset >= rsv_schp->bufflen)
1236                 return VM_FAULT_SIGBUS;
1237         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1238                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1239                                       offset, rsv_schp->k_use_sg));
1240         sa = vma->vm_start;
1241         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1242         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1243                 len = vma->vm_end - sa;
1244                 len = (len < length) ? len : length;
1245                 if (offset < len) {
1246                         struct page *page = nth_page(rsv_schp->pages[k],
1247                                                      offset >> PAGE_SHIFT);
1248                         get_page(page); /* increment page count */
1249                         vmf->page = page;
1250                         return 0; /* success */
1251                 }
1252                 sa += len;
1253                 offset -= len;
1254         }
1255
1256         return VM_FAULT_SIGBUS;
1257 }
1258
1259 static const struct vm_operations_struct sg_mmap_vm_ops = {
1260         .fault = sg_vma_fault,
1261 };
1262
1263 static int
1264 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1265 {
1266         Sg_fd *sfp;
1267         unsigned long req_sz, len, sa;
1268         Sg_scatter_hold *rsv_schp;
1269         int k, length;
1270         int ret = 0;
1271
1272         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1273                 return -ENXIO;
1274         req_sz = vma->vm_end - vma->vm_start;
1275         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1276                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1277                                       (void *) vma->vm_start, (int) req_sz));
1278         if (vma->vm_pgoff)
1279                 return -EINVAL; /* want no offset */
1280         rsv_schp = &sfp->reserve;
1281         mutex_lock(&sfp->f_mutex);
1282         if (req_sz > rsv_schp->bufflen) {
1283                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1284                 goto out;
1285         }
1286
1287         sa = vma->vm_start;
1288         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1289         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1290                 len = vma->vm_end - sa;
1291                 len = (len < length) ? len : length;
1292                 sa += len;
1293         }
1294
1295         sfp->mmap_called = 1;
1296         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1297         vma->vm_private_data = sfp;
1298         vma->vm_ops = &sg_mmap_vm_ops;
1299 out:
1300         mutex_unlock(&sfp->f_mutex);
1301         return ret;
1302 }
1303
1304 static void
1305 sg_rq_end_io_usercontext(struct work_struct *work)
1306 {
1307         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1308         struct sg_fd *sfp = srp->parentfp;
1309
1310         sg_finish_rem_req(srp);
1311         sg_remove_request(sfp, srp);
1312         kref_put(&sfp->f_ref, sg_remove_sfp);
1313 }
1314
1315 /*
1316  * This function is a "bottom half" handler that is called by the mid
1317  * level when a command is completed (or has failed).
1318  */
1319 static void
1320 sg_rq_end_io(struct request *rq, blk_status_t status)
1321 {
1322         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq);
1323         struct sg_request *srp = rq->end_io_data;
1324         Sg_device *sdp;
1325         Sg_fd *sfp;
1326         unsigned long iflags;
1327         unsigned int ms;
1328         char *sense;
1329         int result, resid, done = 1;
1330
1331         if (WARN_ON(srp->done != 0))
1332                 return;
1333
1334         sfp = srp->parentfp;
1335         if (WARN_ON(sfp == NULL))
1336                 return;
1337
1338         sdp = sfp->parentdp;
1339         if (unlikely(atomic_read(&sdp->detaching)))
1340                 pr_info("%s: device detaching\n", __func__);
1341
1342         sense = scmd->sense_buffer;
1343         result = scmd->result;
1344         resid = scmd->resid_len;
1345
1346         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1347                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1348                                       srp->header.pack_id, result));
1349         srp->header.resid = resid;
1350         ms = jiffies_to_msecs(jiffies);
1351         srp->header.duration = (ms > srp->header.duration) ?
1352                                 (ms - srp->header.duration) : 0;
1353         if (0 != result) {
1354                 struct scsi_sense_hdr sshdr;
1355
1356                 srp->header.status = 0xff & result;
1357                 srp->header.masked_status = status_byte(result);
1358                 srp->header.msg_status = COMMAND_COMPLETE;
1359                 srp->header.host_status = host_byte(result);
1360                 srp->header.driver_status = driver_byte(result);
1361                 if ((sdp->sgdebug > 0) &&
1362                     ((CHECK_CONDITION == srp->header.masked_status) ||
1363                      (COMMAND_TERMINATED == srp->header.masked_status)))
1364                         __scsi_print_sense(sdp->device, __func__, sense,
1365                                            SCSI_SENSE_BUFFERSIZE);
1366
1367                 /* Following if statement is a patch supplied by Eric Youngdale */
1368                 if (driver_byte(result) != 0
1369                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1370                     && !scsi_sense_is_deferred(&sshdr)
1371                     && sshdr.sense_key == UNIT_ATTENTION
1372                     && sdp->device->removable) {
1373                         /* Detected possible disc change. Set the bit - this */
1374                         /* may be used if there are filesystems using this device */
1375                         sdp->device->changed = 1;
1376                 }
1377         }
1378
1379         if (scmd->sense_len)
1380                 memcpy(srp->sense_b, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
1381
1382         /* Rely on write phase to clean out srp status values, so no "else" */
1383
1384         /*
1385          * Free the request as soon as it is complete so that its resources
1386          * can be reused without waiting for userspace to read() the
1387          * result.  But keep the associated bio (if any) around until
1388          * blk_rq_unmap_user() can be called from user context.
1389          */
1390         srp->rq = NULL;
1391         blk_mq_free_request(rq);
1392
1393         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1394         if (unlikely(srp->orphan)) {
1395                 if (sfp->keep_orphan)
1396                         srp->sg_io_owned = 0;
1397                 else
1398                         done = 0;
1399         }
1400         srp->done = done;
1401         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1402
1403         if (likely(done)) {
1404                 /* Now wake up any sg_read() that is waiting for this
1405                  * packet.
1406                  */
1407                 wake_up_interruptible(&sfp->read_wait);
1408                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1409                 kref_put(&sfp->f_ref, sg_remove_sfp);
1410         } else {
1411                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1412                 schedule_work(&srp->ew.work);
1413         }
1414 }
1415
1416 static const struct file_operations sg_fops = {
1417         .owner = THIS_MODULE,
1418         .read = sg_read,
1419         .write = sg_write,
1420         .poll = sg_poll,
1421         .unlocked_ioctl = sg_ioctl,
1422         .compat_ioctl = compat_ptr_ioctl,
1423         .open = sg_open,
1424         .mmap = sg_mmap,
1425         .release = sg_release,
1426         .fasync = sg_fasync,
1427         .llseek = no_llseek,
1428 };
1429
1430 static struct class *sg_sysfs_class;
1431
1432 static int sg_sysfs_valid = 0;
1433
1434 static Sg_device *
1435 sg_alloc(struct scsi_device *scsidp)
1436 {
1437         struct request_queue *q = scsidp->request_queue;
1438         Sg_device *sdp;
1439         unsigned long iflags;
1440         int error;
1441         u32 k;
1442
1443         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1444         if (!sdp) {
1445                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1446                             "failure\n", __func__);
1447                 return ERR_PTR(-ENOMEM);
1448         }
1449
1450         idr_preload(GFP_KERNEL);
1451         write_lock_irqsave(&sg_index_lock, iflags);
1452
1453         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1454         if (error < 0) {
1455                 if (error == -ENOSPC) {
1456                         sdev_printk(KERN_WARNING, scsidp,
1457                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1458                                     scsidp->type, SG_MAX_DEVS - 1);
1459                         error = -ENODEV;
1460                 } else {
1461                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1462                                     "allocation Sg_device failure: %d\n",
1463                                     __func__, error);
1464                 }
1465                 goto out_unlock;
1466         }
1467         k = error;
1468
1469         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1470                                         "sg_alloc: dev=%d \n", k));
1471         sprintf(sdp->name, "sg%d", k);
1472         sdp->device = scsidp;
1473         mutex_init(&sdp->open_rel_lock);
1474         INIT_LIST_HEAD(&sdp->sfds);
1475         init_waitqueue_head(&sdp->open_wait);
1476         atomic_set(&sdp->detaching, 0);
1477         rwlock_init(&sdp->sfd_lock);
1478         sdp->sg_tablesize = queue_max_segments(q);
1479         sdp->index = k;
1480         kref_init(&sdp->d_ref);
1481         error = 0;
1482
1483 out_unlock:
1484         write_unlock_irqrestore(&sg_index_lock, iflags);
1485         idr_preload_end();
1486
1487         if (error) {
1488                 kfree(sdp);
1489                 return ERR_PTR(error);
1490         }
1491         return sdp;
1492 }
1493
1494 static int
1495 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1496 {
1497         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1498         Sg_device *sdp = NULL;
1499         struct cdev * cdev = NULL;
1500         int error;
1501         unsigned long iflags;
1502
1503         error = -ENOMEM;
1504         cdev = cdev_alloc();
1505         if (!cdev) {
1506                 pr_warn("%s: cdev_alloc failed\n", __func__);
1507                 goto out;
1508         }
1509         cdev->owner = THIS_MODULE;
1510         cdev->ops = &sg_fops;
1511
1512         sdp = sg_alloc(scsidp);
1513         if (IS_ERR(sdp)) {
1514                 pr_warn("%s: sg_alloc failed\n", __func__);
1515                 error = PTR_ERR(sdp);
1516                 goto out;
1517         }
1518
1519         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1520         if (error)
1521                 goto cdev_add_err;
1522
1523         sdp->cdev = cdev;
1524         if (sg_sysfs_valid) {
1525                 struct device *sg_class_member;
1526
1527                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1528                                                 MKDEV(SCSI_GENERIC_MAJOR,
1529                                                       sdp->index),
1530                                                 sdp, "%s", sdp->name);
1531                 if (IS_ERR(sg_class_member)) {
1532                         pr_err("%s: device_create failed\n", __func__);
1533                         error = PTR_ERR(sg_class_member);
1534                         goto cdev_add_err;
1535                 }
1536                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1537                                           &sg_class_member->kobj, "generic");
1538                 if (error)
1539                         pr_err("%s: unable to make symlink 'generic' back "
1540                                "to sg%d\n", __func__, sdp->index);
1541         } else
1542                 pr_warn("%s: sg_sys Invalid\n", __func__);
1543
1544         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1545                     "type %d\n", sdp->index, scsidp->type);
1546
1547         dev_set_drvdata(cl_dev, sdp);
1548
1549         return 0;
1550
1551 cdev_add_err:
1552         write_lock_irqsave(&sg_index_lock, iflags);
1553         idr_remove(&sg_index_idr, sdp->index);
1554         write_unlock_irqrestore(&sg_index_lock, iflags);
1555         kfree(sdp);
1556
1557 out:
1558         if (cdev)
1559                 cdev_del(cdev);
1560         return error;
1561 }
1562
1563 static void
1564 sg_device_destroy(struct kref *kref)
1565 {
1566         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1567         unsigned long flags;
1568
1569         /* CAUTION!  Note that the device can still be found via idr_find()
1570          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1571          * any other cleanup.
1572          */
1573
1574         write_lock_irqsave(&sg_index_lock, flags);
1575         idr_remove(&sg_index_idr, sdp->index);
1576         write_unlock_irqrestore(&sg_index_lock, flags);
1577
1578         SCSI_LOG_TIMEOUT(3,
1579                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1580
1581         kfree(sdp);
1582 }
1583
1584 static void
1585 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1586 {
1587         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1588         Sg_device *sdp = dev_get_drvdata(cl_dev);
1589         unsigned long iflags;
1590         Sg_fd *sfp;
1591         int val;
1592
1593         if (!sdp)
1594                 return;
1595         /* want sdp->detaching non-zero as soon as possible */
1596         val = atomic_inc_return(&sdp->detaching);
1597         if (val > 1)
1598                 return; /* only want to do following once per device */
1599
1600         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1601                                       "%s\n", __func__));
1602
1603         read_lock_irqsave(&sdp->sfd_lock, iflags);
1604         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1605                 wake_up_interruptible_all(&sfp->read_wait);
1606                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1607         }
1608         wake_up_interruptible_all(&sdp->open_wait);
1609         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1610
1611         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1612         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1613         cdev_del(sdp->cdev);
1614         sdp->cdev = NULL;
1615
1616         kref_put(&sdp->d_ref, sg_device_destroy);
1617 }
1618
1619 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1620 module_param_named(def_reserved_size, def_reserved_size, int,
1621                    S_IRUGO | S_IWUSR);
1622 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1623
1624 MODULE_AUTHOR("Douglas Gilbert");
1625 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1626 MODULE_LICENSE("GPL");
1627 MODULE_VERSION(SG_VERSION_STR);
1628 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1629
1630 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1631                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1632 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1633 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1634
1635 #ifdef CONFIG_SYSCTL
1636 #include <linux/sysctl.h>
1637
1638 static struct ctl_table sg_sysctls[] = {
1639         {
1640                 .procname       = "sg-big-buff",
1641                 .data           = &sg_big_buff,
1642                 .maxlen         = sizeof(int),
1643                 .mode           = 0444,
1644                 .proc_handler   = proc_dointvec,
1645         },
1646         {}
1647 };
1648
1649 static struct ctl_table_header *hdr;
1650 static void register_sg_sysctls(void)
1651 {
1652         if (!hdr)
1653                 hdr = register_sysctl("kernel", sg_sysctls);
1654 }
1655
1656 static void unregister_sg_sysctls(void)
1657 {
1658         if (hdr)
1659                 unregister_sysctl_table(hdr);
1660 }
1661 #else
1662 #define register_sg_sysctls() do { } while (0)
1663 #define unregister_sg_sysctls() do { } while (0)
1664 #endif /* CONFIG_SYSCTL */
1665
1666 static int __init
1667 init_sg(void)
1668 {
1669         int rc;
1670
1671         if (scatter_elem_sz < PAGE_SIZE) {
1672                 scatter_elem_sz = PAGE_SIZE;
1673                 scatter_elem_sz_prev = scatter_elem_sz;
1674         }
1675         if (def_reserved_size >= 0)
1676                 sg_big_buff = def_reserved_size;
1677         else
1678                 def_reserved_size = sg_big_buff;
1679
1680         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1681                                     SG_MAX_DEVS, "sg");
1682         if (rc)
1683                 return rc;
1684         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1685         if ( IS_ERR(sg_sysfs_class) ) {
1686                 rc = PTR_ERR(sg_sysfs_class);
1687                 goto err_out;
1688         }
1689         sg_sysfs_valid = 1;
1690         rc = scsi_register_interface(&sg_interface);
1691         if (0 == rc) {
1692 #ifdef CONFIG_SCSI_PROC_FS
1693                 sg_proc_init();
1694 #endif                          /* CONFIG_SCSI_PROC_FS */
1695                 return 0;
1696         }
1697         class_destroy(sg_sysfs_class);
1698         register_sg_sysctls();
1699 err_out:
1700         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1701         return rc;
1702 }
1703
1704 static void __exit
1705 exit_sg(void)
1706 {
1707         unregister_sg_sysctls();
1708 #ifdef CONFIG_SCSI_PROC_FS
1709         remove_proc_subtree("scsi/sg", NULL);
1710 #endif                          /* CONFIG_SCSI_PROC_FS */
1711         scsi_unregister_interface(&sg_interface);
1712         class_destroy(sg_sysfs_class);
1713         sg_sysfs_valid = 0;
1714         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1715                                  SG_MAX_DEVS);
1716         idr_destroy(&sg_index_idr);
1717 }
1718
1719 static int
1720 sg_start_req(Sg_request *srp, unsigned char *cmd)
1721 {
1722         int res;
1723         struct request *rq;
1724         Sg_fd *sfp = srp->parentfp;
1725         sg_io_hdr_t *hp = &srp->header;
1726         int dxfer_len = (int) hp->dxfer_len;
1727         int dxfer_dir = hp->dxfer_direction;
1728         unsigned int iov_count = hp->iovec_count;
1729         Sg_scatter_hold *req_schp = &srp->data;
1730         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1731         struct request_queue *q = sfp->parentdp->device->request_queue;
1732         struct rq_map_data *md, map_data;
1733         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1734         struct scsi_cmnd *scmd;
1735
1736         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1737                                       "sg_start_req: dxfer_len=%d\n",
1738                                       dxfer_len));
1739
1740         /*
1741          * NOTE
1742          *
1743          * With scsi-mq enabled, there are a fixed number of preallocated
1744          * requests equal in number to shost->can_queue.  If all of the
1745          * preallocated requests are already in use, then scsi_alloc_request()
1746          * will sleep until an active command completes, freeing up a request.
1747          * Although waiting in an asynchronous interface is less than ideal, we
1748          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1749          * not expect an EWOULDBLOCK from this condition.
1750          */
1751         rq = scsi_alloc_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1752                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1753         if (IS_ERR(rq))
1754                 return PTR_ERR(rq);
1755         scmd = blk_mq_rq_to_pdu(rq);
1756
1757         if (hp->cmd_len > sizeof(scmd->cmnd)) {
1758                 blk_mq_free_request(rq);
1759                 return -EINVAL;
1760         }
1761
1762         memcpy(scmd->cmnd, cmd, hp->cmd_len);
1763         scmd->cmd_len = hp->cmd_len;
1764
1765         srp->rq = rq;
1766         rq->end_io_data = srp;
1767         scmd->allowed = SG_DEFAULT_RETRIES;
1768
1769         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1770                 return 0;
1771
1772         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1773             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1774             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1775                 md = NULL;
1776         else
1777                 md = &map_data;
1778
1779         if (md) {
1780                 mutex_lock(&sfp->f_mutex);
1781                 if (dxfer_len <= rsv_schp->bufflen &&
1782                     !sfp->res_in_use) {
1783                         sfp->res_in_use = 1;
1784                         sg_link_reserve(sfp, srp, dxfer_len);
1785                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1786                         res = -EBUSY; /* sfp->res_in_use == 1 */
1787                         if (dxfer_len > rsv_schp->bufflen)
1788                                 res = -ENOMEM;
1789                         mutex_unlock(&sfp->f_mutex);
1790                         return res;
1791                 } else {
1792                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1793                         if (res) {
1794                                 mutex_unlock(&sfp->f_mutex);
1795                                 return res;
1796                         }
1797                 }
1798                 mutex_unlock(&sfp->f_mutex);
1799
1800                 md->pages = req_schp->pages;
1801                 md->page_order = req_schp->page_order;
1802                 md->nr_entries = req_schp->k_use_sg;
1803                 md->offset = 0;
1804                 md->null_mapped = hp->dxferp ? 0 : 1;
1805                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1806                         md->from_user = 1;
1807                 else
1808                         md->from_user = 0;
1809         }
1810
1811         if (iov_count) {
1812                 struct iovec *iov = NULL;
1813                 struct iov_iter i;
1814
1815                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1816                 if (res < 0)
1817                         return res;
1818
1819                 iov_iter_truncate(&i, hp->dxfer_len);
1820                 if (!iov_iter_count(&i)) {
1821                         kfree(iov);
1822                         return -EINVAL;
1823                 }
1824
1825                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1826                 kfree(iov);
1827         } else
1828                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1829                                       hp->dxfer_len, GFP_ATOMIC);
1830
1831         if (!res) {
1832                 srp->bio = rq->bio;
1833
1834                 if (!md) {
1835                         req_schp->dio_in_use = 1;
1836                         hp->info |= SG_INFO_DIRECT_IO;
1837                 }
1838         }
1839         return res;
1840 }
1841
1842 static int
1843 sg_finish_rem_req(Sg_request *srp)
1844 {
1845         int ret = 0;
1846
1847         Sg_fd *sfp = srp->parentfp;
1848         Sg_scatter_hold *req_schp = &srp->data;
1849
1850         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1851                                       "sg_finish_rem_req: res_used=%d\n",
1852                                       (int) srp->res_used));
1853         if (srp->bio)
1854                 ret = blk_rq_unmap_user(srp->bio);
1855
1856         if (srp->rq)
1857                 blk_mq_free_request(srp->rq);
1858
1859         if (srp->res_used)
1860                 sg_unlink_reserve(sfp, srp);
1861         else
1862                 sg_remove_scat(sfp, req_schp);
1863
1864         return ret;
1865 }
1866
1867 static int
1868 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1869 {
1870         int sg_bufflen = tablesize * sizeof(struct page *);
1871         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1872
1873         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1874         if (!schp->pages)
1875                 return -ENOMEM;
1876         schp->sglist_len = sg_bufflen;
1877         return tablesize;       /* number of scat_gath elements allocated */
1878 }
1879
1880 static int
1881 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1882 {
1883         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1884         int sg_tablesize = sfp->parentdp->sg_tablesize;
1885         int blk_size = buff_size, order;
1886         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1887
1888         if (blk_size < 0)
1889                 return -EFAULT;
1890         if (0 == blk_size)
1891                 ++blk_size;     /* don't know why */
1892         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1893         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1894         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1895                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1896                 buff_size, blk_size));
1897
1898         /* N.B. ret_sz carried into this block ... */
1899         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1900         if (mx_sc_elems < 0)
1901                 return mx_sc_elems;     /* most likely -ENOMEM */
1902
1903         num = scatter_elem_sz;
1904         if (unlikely(num != scatter_elem_sz_prev)) {
1905                 if (num < PAGE_SIZE) {
1906                         scatter_elem_sz = PAGE_SIZE;
1907                         scatter_elem_sz_prev = PAGE_SIZE;
1908                 } else
1909                         scatter_elem_sz_prev = num;
1910         }
1911
1912         order = get_order(num);
1913 retry:
1914         ret_sz = 1 << (PAGE_SHIFT + order);
1915
1916         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1917              k++, rem_sz -= ret_sz) {
1918
1919                 num = (rem_sz > scatter_elem_sz_prev) ?
1920                         scatter_elem_sz_prev : rem_sz;
1921
1922                 schp->pages[k] = alloc_pages(gfp_mask, order);
1923                 if (!schp->pages[k])
1924                         goto out;
1925
1926                 if (num == scatter_elem_sz_prev) {
1927                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1928                                 scatter_elem_sz = ret_sz;
1929                                 scatter_elem_sz_prev = ret_sz;
1930                         }
1931                 }
1932
1933                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1934                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1935                                  k, num, ret_sz));
1936         }               /* end of for loop */
1937
1938         schp->page_order = order;
1939         schp->k_use_sg = k;
1940         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1941                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1942                          k, rem_sz));
1943
1944         schp->bufflen = blk_size;
1945         if (rem_sz > 0) /* must have failed */
1946                 return -ENOMEM;
1947         return 0;
1948 out:
1949         for (i = 0; i < k; i++)
1950                 __free_pages(schp->pages[i], order);
1951
1952         if (--order >= 0)
1953                 goto retry;
1954
1955         return -ENOMEM;
1956 }
1957
1958 static void
1959 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1960 {
1961         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1962                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1963         if (schp->pages && schp->sglist_len > 0) {
1964                 if (!schp->dio_in_use) {
1965                         int k;
1966
1967                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1968                                 SCSI_LOG_TIMEOUT(5,
1969                                         sg_printk(KERN_INFO, sfp->parentdp,
1970                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1971                                         k, schp->pages[k]));
1972                                 __free_pages(schp->pages[k], schp->page_order);
1973                         }
1974
1975                         kfree(schp->pages);
1976                 }
1977         }
1978         memset(schp, 0, sizeof (*schp));
1979 }
1980
1981 static int
1982 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1983 {
1984         Sg_scatter_hold *schp = &srp->data;
1985         int k, num;
1986
1987         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1988                          "sg_read_oxfer: num_read_xfer=%d\n",
1989                          num_read_xfer));
1990         if ((!outp) || (num_read_xfer <= 0))
1991                 return 0;
1992
1993         num = 1 << (PAGE_SHIFT + schp->page_order);
1994         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1995                 if (num > num_read_xfer) {
1996                         if (copy_to_user(outp, page_address(schp->pages[k]),
1997                                            num_read_xfer))
1998                                 return -EFAULT;
1999                         break;
2000                 } else {
2001                         if (copy_to_user(outp, page_address(schp->pages[k]),
2002                                            num))
2003                                 return -EFAULT;
2004                         num_read_xfer -= num;
2005                         if (num_read_xfer <= 0)
2006                                 break;
2007                         outp += num;
2008                 }
2009         }
2010
2011         return 0;
2012 }
2013
2014 static void
2015 sg_build_reserve(Sg_fd * sfp, int req_size)
2016 {
2017         Sg_scatter_hold *schp = &sfp->reserve;
2018
2019         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2020                          "sg_build_reserve: req_size=%d\n", req_size));
2021         do {
2022                 if (req_size < PAGE_SIZE)
2023                         req_size = PAGE_SIZE;
2024                 if (0 == sg_build_indirect(schp, sfp, req_size))
2025                         return;
2026                 else
2027                         sg_remove_scat(sfp, schp);
2028                 req_size >>= 1; /* divide by 2 */
2029         } while (req_size > (PAGE_SIZE / 2));
2030 }
2031
2032 static void
2033 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2034 {
2035         Sg_scatter_hold *req_schp = &srp->data;
2036         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2037         int k, num, rem;
2038
2039         srp->res_used = 1;
2040         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2041                          "sg_link_reserve: size=%d\n", size));
2042         rem = size;
2043
2044         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2045         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2046                 if (rem <= num) {
2047                         req_schp->k_use_sg = k + 1;
2048                         req_schp->sglist_len = rsv_schp->sglist_len;
2049                         req_schp->pages = rsv_schp->pages;
2050
2051                         req_schp->bufflen = size;
2052                         req_schp->page_order = rsv_schp->page_order;
2053                         break;
2054                 } else
2055                         rem -= num;
2056         }
2057
2058         if (k >= rsv_schp->k_use_sg)
2059                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2060                                  "sg_link_reserve: BAD size\n"));
2061 }
2062
2063 static void
2064 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2065 {
2066         Sg_scatter_hold *req_schp = &srp->data;
2067
2068         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2069                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2070                                       (int) req_schp->k_use_sg));
2071         req_schp->k_use_sg = 0;
2072         req_schp->bufflen = 0;
2073         req_schp->pages = NULL;
2074         req_schp->page_order = 0;
2075         req_schp->sglist_len = 0;
2076         srp->res_used = 0;
2077         /* Called without mutex lock to avoid deadlock */
2078         sfp->res_in_use = 0;
2079 }
2080
2081 static Sg_request *
2082 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2083 {
2084         Sg_request *resp;
2085         unsigned long iflags;
2086
2087         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2088         list_for_each_entry(resp, &sfp->rq_list, entry) {
2089                 /* look for requests that are ready + not SG_IO owned */
2090                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2091                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2092                         resp->done = 2; /* guard against other readers */
2093                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2094                         return resp;
2095                 }
2096         }
2097         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2098         return NULL;
2099 }
2100
2101 /* always adds to end of list */
2102 static Sg_request *
2103 sg_add_request(Sg_fd * sfp)
2104 {
2105         int k;
2106         unsigned long iflags;
2107         Sg_request *rp = sfp->req_arr;
2108
2109         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2110         if (!list_empty(&sfp->rq_list)) {
2111                 if (!sfp->cmd_q)
2112                         goto out_unlock;
2113
2114                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2115                         if (!rp->parentfp)
2116                                 break;
2117                 }
2118                 if (k >= SG_MAX_QUEUE)
2119                         goto out_unlock;
2120         }
2121         memset(rp, 0, sizeof (Sg_request));
2122         rp->parentfp = sfp;
2123         rp->header.duration = jiffies_to_msecs(jiffies);
2124         list_add_tail(&rp->entry, &sfp->rq_list);
2125         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2126         return rp;
2127 out_unlock:
2128         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2129         return NULL;
2130 }
2131
2132 /* Return of 1 for found; 0 for not found */
2133 static int
2134 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2135 {
2136         unsigned long iflags;
2137         int res = 0;
2138
2139         if (!sfp || !srp || list_empty(&sfp->rq_list))
2140                 return res;
2141         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2142         if (!list_empty(&srp->entry)) {
2143                 list_del(&srp->entry);
2144                 srp->parentfp = NULL;
2145                 res = 1;
2146         }
2147         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2148         return res;
2149 }
2150
2151 static Sg_fd *
2152 sg_add_sfp(Sg_device * sdp)
2153 {
2154         Sg_fd *sfp;
2155         unsigned long iflags;
2156         int bufflen;
2157
2158         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2159         if (!sfp)
2160                 return ERR_PTR(-ENOMEM);
2161
2162         init_waitqueue_head(&sfp->read_wait);
2163         rwlock_init(&sfp->rq_list_lock);
2164         INIT_LIST_HEAD(&sfp->rq_list);
2165         kref_init(&sfp->f_ref);
2166         mutex_init(&sfp->f_mutex);
2167         sfp->timeout = SG_DEFAULT_TIMEOUT;
2168         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2169         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2170         sfp->cmd_q = SG_DEF_COMMAND_Q;
2171         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2172         sfp->parentdp = sdp;
2173         write_lock_irqsave(&sdp->sfd_lock, iflags);
2174         if (atomic_read(&sdp->detaching)) {
2175                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2176                 kfree(sfp);
2177                 return ERR_PTR(-ENODEV);
2178         }
2179         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2180         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2181         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2182                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2183         if (unlikely(sg_big_buff != def_reserved_size))
2184                 sg_big_buff = def_reserved_size;
2185
2186         bufflen = min_t(int, sg_big_buff,
2187                         max_sectors_bytes(sdp->device->request_queue));
2188         sg_build_reserve(sfp, bufflen);
2189         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2190                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2191                                       sfp->reserve.bufflen,
2192                                       sfp->reserve.k_use_sg));
2193
2194         kref_get(&sdp->d_ref);
2195         __module_get(THIS_MODULE);
2196         return sfp;
2197 }
2198
2199 static void
2200 sg_remove_sfp_usercontext(struct work_struct *work)
2201 {
2202         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2203         struct sg_device *sdp = sfp->parentdp;
2204         Sg_request *srp;
2205         unsigned long iflags;
2206
2207         /* Cleanup any responses which were never read(). */
2208         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2209         while (!list_empty(&sfp->rq_list)) {
2210                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2211                 sg_finish_rem_req(srp);
2212                 list_del(&srp->entry);
2213                 srp->parentfp = NULL;
2214         }
2215         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2216
2217         if (sfp->reserve.bufflen > 0) {
2218                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2219                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2220                                 (int) sfp->reserve.bufflen,
2221                                 (int) sfp->reserve.k_use_sg));
2222                 sg_remove_scat(sfp, &sfp->reserve);
2223         }
2224
2225         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2226                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2227         kfree(sfp);
2228
2229         scsi_device_put(sdp->device);
2230         kref_put(&sdp->d_ref, sg_device_destroy);
2231         module_put(THIS_MODULE);
2232 }
2233
2234 static void
2235 sg_remove_sfp(struct kref *kref)
2236 {
2237         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2238         struct sg_device *sdp = sfp->parentdp;
2239         unsigned long iflags;
2240
2241         write_lock_irqsave(&sdp->sfd_lock, iflags);
2242         list_del(&sfp->sfd_siblings);
2243         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2244
2245         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2246         schedule_work(&sfp->ew.work);
2247 }
2248
2249 #ifdef CONFIG_SCSI_PROC_FS
2250 static int
2251 sg_idr_max_id(int id, void *p, void *data)
2252 {
2253         int *k = data;
2254
2255         if (*k < id)
2256                 *k = id;
2257
2258         return 0;
2259 }
2260
2261 static int
2262 sg_last_dev(void)
2263 {
2264         int k = -1;
2265         unsigned long iflags;
2266
2267         read_lock_irqsave(&sg_index_lock, iflags);
2268         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2269         read_unlock_irqrestore(&sg_index_lock, iflags);
2270         return k + 1;           /* origin 1 */
2271 }
2272 #endif
2273
2274 /* must be called with sg_index_lock held */
2275 static Sg_device *sg_lookup_dev(int dev)
2276 {
2277         return idr_find(&sg_index_idr, dev);
2278 }
2279
2280 static Sg_device *
2281 sg_get_dev(int dev)
2282 {
2283         struct sg_device *sdp;
2284         unsigned long flags;
2285
2286         read_lock_irqsave(&sg_index_lock, flags);
2287         sdp = sg_lookup_dev(dev);
2288         if (!sdp)
2289                 sdp = ERR_PTR(-ENXIO);
2290         else if (atomic_read(&sdp->detaching)) {
2291                 /* If sdp->detaching, then the refcount may already be 0, in
2292                  * which case it would be a bug to do kref_get().
2293                  */
2294                 sdp = ERR_PTR(-ENODEV);
2295         } else
2296                 kref_get(&sdp->d_ref);
2297         read_unlock_irqrestore(&sg_index_lock, flags);
2298
2299         return sdp;
2300 }
2301
2302 #ifdef CONFIG_SCSI_PROC_FS
2303 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2304
2305 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2306 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2307                                   size_t count, loff_t *off);
2308 static const struct proc_ops adio_proc_ops = {
2309         .proc_open      = sg_proc_single_open_adio,
2310         .proc_read      = seq_read,
2311         .proc_lseek     = seq_lseek,
2312         .proc_write     = sg_proc_write_adio,
2313         .proc_release   = single_release,
2314 };
2315
2316 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2317 static ssize_t sg_proc_write_dressz(struct file *filp,
2318                 const char __user *buffer, size_t count, loff_t *off);
2319 static const struct proc_ops dressz_proc_ops = {
2320         .proc_open      = sg_proc_single_open_dressz,
2321         .proc_read      = seq_read,
2322         .proc_lseek     = seq_lseek,
2323         .proc_write     = sg_proc_write_dressz,
2324         .proc_release   = single_release,
2325 };
2326
2327 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2328 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2329 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2330 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2331 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2332 static void dev_seq_stop(struct seq_file *s, void *v);
2333 static const struct seq_operations dev_seq_ops = {
2334         .start = dev_seq_start,
2335         .next  = dev_seq_next,
2336         .stop  = dev_seq_stop,
2337         .show  = sg_proc_seq_show_dev,
2338 };
2339
2340 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2341 static const struct seq_operations devstrs_seq_ops = {
2342         .start = dev_seq_start,
2343         .next  = dev_seq_next,
2344         .stop  = dev_seq_stop,
2345         .show  = sg_proc_seq_show_devstrs,
2346 };
2347
2348 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2349 static const struct seq_operations debug_seq_ops = {
2350         .start = dev_seq_start,
2351         .next  = dev_seq_next,
2352         .stop  = dev_seq_stop,
2353         .show  = sg_proc_seq_show_debug,
2354 };
2355
2356 static int
2357 sg_proc_init(void)
2358 {
2359         struct proc_dir_entry *p;
2360
2361         p = proc_mkdir("scsi/sg", NULL);
2362         if (!p)
2363                 return 1;
2364
2365         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2366         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2367         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2368         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2369         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2370         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2371         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2372         return 0;
2373 }
2374
2375
2376 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2377 {
2378         seq_printf(s, "%d\n", *((int *)s->private));
2379         return 0;
2380 }
2381
2382 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2383 {
2384         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2385 }
2386
2387 static ssize_t
2388 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2389                    size_t count, loff_t *off)
2390 {
2391         int err;
2392         unsigned long num;
2393
2394         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2395                 return -EACCES;
2396         err = kstrtoul_from_user(buffer, count, 0, &num);
2397         if (err)
2398                 return err;
2399         sg_allow_dio = num ? 1 : 0;
2400         return count;
2401 }
2402
2403 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2404 {
2405         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2406 }
2407
2408 static ssize_t
2409 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2410                      size_t count, loff_t *off)
2411 {
2412         int err;
2413         unsigned long k = ULONG_MAX;
2414
2415         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2416                 return -EACCES;
2417
2418         err = kstrtoul_from_user(buffer, count, 0, &k);
2419         if (err)
2420                 return err;
2421         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2422                 sg_big_buff = k;
2423                 return count;
2424         }
2425         return -ERANGE;
2426 }
2427
2428 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2429 {
2430         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2431                    sg_version_date);
2432         return 0;
2433 }
2434
2435 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2436 {
2437         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2438         return 0;
2439 }
2440
2441 struct sg_proc_deviter {
2442         loff_t  index;
2443         size_t  max;
2444 };
2445
2446 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2447 {
2448         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2449
2450         s->private = it;
2451         if (! it)
2452                 return NULL;
2453
2454         it->index = *pos;
2455         it->max = sg_last_dev();
2456         if (it->index >= it->max)
2457                 return NULL;
2458         return it;
2459 }
2460
2461 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2462 {
2463         struct sg_proc_deviter * it = s->private;
2464
2465         *pos = ++it->index;
2466         return (it->index < it->max) ? it : NULL;
2467 }
2468
2469 static void dev_seq_stop(struct seq_file *s, void *v)
2470 {
2471         kfree(s->private);
2472 }
2473
2474 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2475 {
2476         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2477         Sg_device *sdp;
2478         struct scsi_device *scsidp;
2479         unsigned long iflags;
2480
2481         read_lock_irqsave(&sg_index_lock, iflags);
2482         sdp = it ? sg_lookup_dev(it->index) : NULL;
2483         if ((NULL == sdp) || (NULL == sdp->device) ||
2484             (atomic_read(&sdp->detaching)))
2485                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2486         else {
2487                 scsidp = sdp->device;
2488                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2489                               scsidp->host->host_no, scsidp->channel,
2490                               scsidp->id, scsidp->lun, (int) scsidp->type,
2491                               1,
2492                               (int) scsidp->queue_depth,
2493                               (int) scsi_device_busy(scsidp),
2494                               (int) scsi_device_online(scsidp));
2495         }
2496         read_unlock_irqrestore(&sg_index_lock, iflags);
2497         return 0;
2498 }
2499
2500 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2501 {
2502         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2503         Sg_device *sdp;
2504         struct scsi_device *scsidp;
2505         unsigned long iflags;
2506
2507         read_lock_irqsave(&sg_index_lock, iflags);
2508         sdp = it ? sg_lookup_dev(it->index) : NULL;
2509         scsidp = sdp ? sdp->device : NULL;
2510         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2511                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2512                            scsidp->vendor, scsidp->model, scsidp->rev);
2513         else
2514                 seq_puts(s, "<no active device>\n");
2515         read_unlock_irqrestore(&sg_index_lock, iflags);
2516         return 0;
2517 }
2518
2519 /* must be called while holding sg_index_lock */
2520 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2521 {
2522         int k, new_interface, blen, usg;
2523         Sg_request *srp;
2524         Sg_fd *fp;
2525         const sg_io_hdr_t *hp;
2526         const char * cp;
2527         unsigned int ms;
2528
2529         k = 0;
2530         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2531                 k++;
2532                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2533                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2534                            "(res)sgat=%d low_dma=%d\n", k,
2535                            jiffies_to_msecs(fp->timeout),
2536                            fp->reserve.bufflen,
2537                            (int) fp->reserve.k_use_sg, 0);
2538                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2539                            (int) fp->cmd_q, (int) fp->force_packid,
2540                            (int) fp->keep_orphan);
2541                 list_for_each_entry(srp, &fp->rq_list, entry) {
2542                         hp = &srp->header;
2543                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2544                         if (srp->res_used) {
2545                                 if (new_interface &&
2546                                     (SG_FLAG_MMAP_IO & hp->flags))
2547                                         cp = "     mmap>> ";
2548                                 else
2549                                         cp = "     rb>> ";
2550                         } else {
2551                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2552                                         cp = "     dio>> ";
2553                                 else
2554                                         cp = "     ";
2555                         }
2556                         seq_puts(s, cp);
2557                         blen = srp->data.bufflen;
2558                         usg = srp->data.k_use_sg;
2559                         seq_puts(s, srp->done ?
2560                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2561                                   : "act:");
2562                         seq_printf(s, " id=%d blen=%d",
2563                                    srp->header.pack_id, blen);
2564                         if (srp->done)
2565                                 seq_printf(s, " dur=%d", hp->duration);
2566                         else {
2567                                 ms = jiffies_to_msecs(jiffies);
2568                                 seq_printf(s, " t_o/elap=%d/%d",
2569                                         (new_interface ? hp->timeout :
2570                                                   jiffies_to_msecs(fp->timeout)),
2571                                         (ms > hp->duration ? ms - hp->duration : 0));
2572                         }
2573                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2574                                    (int) srp->data.cmd_opcode);
2575                 }
2576                 if (list_empty(&fp->rq_list))
2577                         seq_puts(s, "     No requests active\n");
2578                 read_unlock(&fp->rq_list_lock);
2579         }
2580 }
2581
2582 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2583 {
2584         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2585         Sg_device *sdp;
2586         unsigned long iflags;
2587
2588         if (it && (0 == it->index))
2589                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2590                            (int)it->max, sg_big_buff);
2591
2592         read_lock_irqsave(&sg_index_lock, iflags);
2593         sdp = it ? sg_lookup_dev(it->index) : NULL;
2594         if (NULL == sdp)
2595                 goto skip;
2596         read_lock(&sdp->sfd_lock);
2597         if (!list_empty(&sdp->sfds)) {
2598                 seq_printf(s, " >>> device=%s ", sdp->name);
2599                 if (atomic_read(&sdp->detaching))
2600                         seq_puts(s, "detaching pending close ");
2601                 else if (sdp->device) {
2602                         struct scsi_device *scsidp = sdp->device;
2603
2604                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2605                                    scsidp->host->host_no,
2606                                    scsidp->channel, scsidp->id,
2607                                    scsidp->lun,
2608                                    scsidp->host->hostt->emulated);
2609                 }
2610                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2611                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2612                 sg_proc_debug_helper(s, sdp);
2613         }
2614         read_unlock(&sdp->sfd_lock);
2615 skip:
2616         read_unlock_irqrestore(&sg_index_lock, iflags);
2617         return 0;
2618 }
2619
2620 #endif                          /* CONFIG_SCSI_PROC_FS */
2621
2622 module_init(init_sg);
2623 module_exit(exit_sg);