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 enum rq_end_io_ret 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, bool *busy);
 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         bool busy;
 448         sg_io_hdr_t *hp;
 449         struct sg_header *old_hdr;
 450         int retval;
 451
 452         /*
 453          * This could cause a response to be stranded. Close the associated
 454          * file descriptor to free up any resources being held.
 455          */
 456         retval = sg_check_file_access(filp, __func__);
 457         if (retval)
 458                 return retval;
 459
 460         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 461                 return -ENXIO;
 462         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 463                                       "sg_read: count=%d\n", (int) count));
 464
 465         if (sfp->force_packid)
 466                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 467         if (retval)
 468                 return retval;
 469
 470         srp = sg_get_rq_mark(sfp, req_pack_id, &busy);
 471         if (!srp) {             /* now wait on packet to arrive */
 472                 if (filp->f_flags & O_NONBLOCK)
 473                         return -EAGAIN;
 474                 retval = wait_event_interruptible(sfp->read_wait,
 475                         ((srp = sg_get_rq_mark(sfp, req_pack_id, &busy)) ||
 476                         (!busy && atomic_read(&sdp->detaching))));
 477                 if (!srp)
 478                         /* signal or detaching */
 479                         return retval ? retval : -ENODEV;
 480         }
 481         if (srp->header.interface_id != '\0')
 482                 return sg_new_read(sfp, buf, count, srp);
 483
 484         hp = &srp->header;
 485         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 486         if (!old_hdr)
 487                 return -ENOMEM;
 488
 489         old_hdr->reply_len = (int) hp->timeout;
 490         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 491         old_hdr->pack_id = hp->pack_id;
 492         old_hdr->twelve_byte =
 493             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 494         old_hdr->target_status = hp->masked_status;
 495         old_hdr->host_status = hp->host_status;
 496         old_hdr->driver_status = hp->driver_status;
 497         if ((CHECK_CONDITION & hp->masked_status) ||
 498             (srp->sense_b[0] & 0x70) == 0x70) {
 499                 old_hdr->driver_status = DRIVER_SENSE;
 500                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 501                        sizeof (old_hdr->sense_buffer));
 502         }
 503         switch (hp->host_status) {
 504         /* This setup of 'result' is for backward compatibility and is best
 505            ignored by the user who should use target, host + driver status */
 506         case DID_OK:
 507         case DID_PASSTHROUGH:
 508         case DID_SOFT_ERROR:
 509                 old_hdr->result = 0;
 510                 break;
 511         case DID_NO_CONNECT:
 512         case DID_BUS_BUSY:
 513         case DID_TIME_OUT:
 514                 old_hdr->result = EBUSY;
 515                 break;
 516         case DID_BAD_TARGET:
 517         case DID_ABORT:
 518         case DID_PARITY:
 519         case DID_RESET:
 520         case DID_BAD_INTR:
 521                 old_hdr->result = EIO;
 522                 break;
 523         case DID_ERROR:
 524                 old_hdr->result = (srp->sense_b[0] == 0 &&
 525                                   hp->masked_status == GOOD) ? 0 : EIO;
 526                 break;
 527         default:
 528                 old_hdr->result = EIO;
 529                 break;
 530         }
 531
 532         /* Now copy the result back to the user buffer.  */
 533         if (count >= SZ_SG_HEADER) {
 534                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 535                         retval = -EFAULT;
 536                         goto free_old_hdr;
 537                 }
 538                 buf += SZ_SG_HEADER;
 539                 if (count > old_hdr->reply_len)
 540                         count = old_hdr->reply_len;
 541                 if (count > SZ_SG_HEADER) {
 542                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 543                                 retval = -EFAULT;
 544                                 goto free_old_hdr;
 545                         }
 546                 }
 547         } else
 548                 count = (old_hdr->result == 0) ? 0 : -EIO;
 549         sg_finish_rem_req(srp);
 550         sg_remove_request(sfp, srp);
 551         retval = count;
 552 free_old_hdr:
 553         kfree(old_hdr);
 554         return retval;
 555 }
 556
 557 static ssize_t
 558 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 559 {
 560         sg_io_hdr_t *hp = &srp->header;
 561         int err = 0, err2;
 562         int len;
 563
 564         if (in_compat_syscall()) {
 565                 if (count < sizeof(struct compat_sg_io_hdr)) {
 566                         err = -EINVAL;
 567                         goto err_out;
 568                 }
 569         } else if (count < SZ_SG_IO_HDR) {
 570                 err = -EINVAL;
 571                 goto err_out;
 572         }
 573         hp->sb_len_wr = 0;
 574         if ((hp->mx_sb_len > 0) && hp->sbp) {
 575                 if ((CHECK_CONDITION & hp->masked_status) ||
 576                     (srp->sense_b[0] & 0x70) == 0x70) {
 577                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 578                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 579                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 580                         len = (len > sb_len) ? sb_len : len;
 581                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 582                                 err = -EFAULT;
 583                                 goto err_out;
 584                         }
 585                         hp->driver_status = DRIVER_SENSE;
 586                         hp->sb_len_wr = len;
 587                 }
 588         }
 589         if (hp->masked_status || hp->host_status || hp->driver_status)
 590                 hp->info |= SG_INFO_CHECK;
 591         err = put_sg_io_hdr(hp, buf);
 592 err_out:
 593         err2 = sg_finish_rem_req(srp);
 594         sg_remove_request(sfp, srp);
 595         return err ? : err2 ? : count;
 596 }
 597
 598 static ssize_t
 599 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 600 {
 601         int mxsize, cmd_size, k;
 602         int input_size, blocking;
 603         unsigned char opcode;
 604         Sg_device *sdp;
 605         Sg_fd *sfp;
 606         Sg_request *srp;
 607         struct sg_header old_hdr;
 608         sg_io_hdr_t *hp;
 609         unsigned char cmnd[SG_MAX_CDB_SIZE];
 610         int retval;
 611
 612         retval = sg_check_file_access(filp, __func__);
 613         if (retval)
 614                 return retval;
 615
 616         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 617                 return -ENXIO;
 618         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 619                                       "sg_write: count=%d\n", (int) count));
 620         if (atomic_read(&sdp->detaching))
 621                 return -ENODEV;
 622         if (!((filp->f_flags & O_NONBLOCK) ||
 623               scsi_block_when_processing_errors(sdp->device)))
 624                 return -ENXIO;
 625
 626         if (count < SZ_SG_HEADER)
 627                 return -EIO;
 628         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 629                 return -EFAULT;
 630         blocking = !(filp->f_flags & O_NONBLOCK);
 631         if (old_hdr.reply_len < 0)
 632                 return sg_new_write(sfp, filp, buf, count,
 633                                     blocking, 0, 0, NULL);
 634         if (count < (SZ_SG_HEADER + 6))
 635                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 636
 637         buf += SZ_SG_HEADER;
 638         if (get_user(opcode, buf))
 639                 return -EFAULT;
 640
 641         if (!(srp = sg_add_request(sfp))) {
 642                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 643                                               "sg_write: queue full\n"));
 644                 return -EDOM;
 645         }
 646         mutex_lock(&sfp->f_mutex);
 647         if (sfp->next_cmd_len > 0) {
 648                 cmd_size = sfp->next_cmd_len;
 649                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 650         } else {
 651                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 652                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 653                         cmd_size = 12;
 654         }
 655         mutex_unlock(&sfp->f_mutex);
 656         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 657                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 658 /* Determine buffer size.  */
 659         input_size = count - cmd_size;
 660         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 661         mxsize -= SZ_SG_HEADER;
 662         input_size -= SZ_SG_HEADER;
 663         if (input_size < 0) {
 664                 sg_remove_request(sfp, srp);
 665                 return -EIO;    /* User did not pass enough bytes for this command. */
 666         }
 667         hp = &srp->header;
 668         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 669         hp->cmd_len = (unsigned char) cmd_size;
 670         hp->iovec_count = 0;
 671         hp->mx_sb_len = 0;
 672         if (input_size > 0)
 673                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 674                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 675         else
 676                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 677         hp->dxfer_len = mxsize;
 678         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 679             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 680                 hp->dxferp = (char __user *)buf + cmd_size;
 681         else
 682                 hp->dxferp = NULL;
 683         hp->sbp = NULL;
 684         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 685         hp->flags = input_size; /* structure abuse ... */
 686         hp->pack_id = old_hdr.pack_id;
 687         hp->usr_ptr = NULL;
 688         if (copy_from_user(cmnd, buf, cmd_size)) {
 689                 sg_remove_request(sfp, srp);
 690                 return -EFAULT;
 691         }
 692         /*
 693          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 694          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 695          * is a non-zero input_size, so emit a warning.
 696          */
 697         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 698                 printk_ratelimited(KERN_WARNING
 699                                    "sg_write: data in/out %d/%d bytes "
 700                                    "for SCSI command 0x%x-- guessing "
 701                                    "data in;\n   program %s not setting "
 702                                    "count and/or reply_len properly\n",
 703                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 704                                    input_size, (unsigned int) cmnd[0],
 705                                    current->comm);
 706         }
 707         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 708         return (k < 0) ? k : count;
 709 }
 710
 711 static ssize_t
 712 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 713                  size_t count, int blocking, int read_only, int sg_io_owned,
 714                  Sg_request **o_srp)
 715 {
 716         int k;
 717         Sg_request *srp;
 718         sg_io_hdr_t *hp;
 719         unsigned char cmnd[SG_MAX_CDB_SIZE];
 720         int timeout;
 721         unsigned long ul_timeout;
 722
 723         if (count < SZ_SG_IO_HDR)
 724                 return -EINVAL;
 725
 726         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 727         if (!(srp = sg_add_request(sfp))) {
 728                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 729                                               "sg_new_write: queue full\n"));
 730                 return -EDOM;
 731         }
 732         srp->sg_io_owned = sg_io_owned;
 733         hp = &srp->header;
 734         if (get_sg_io_hdr(hp, buf)) {
 735                 sg_remove_request(sfp, srp);
 736                 return -EFAULT;
 737         }
 738         if (hp->interface_id != 'S') {
 739                 sg_remove_request(sfp, srp);
 740                 return -ENOSYS;
 741         }
 742         if (hp->flags & SG_FLAG_MMAP_IO) {
 743                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 744                         sg_remove_request(sfp, srp);
 745                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 746                 }
 747                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 748                         sg_remove_request(sfp, srp);
 749                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 750                 }
 751                 if (sfp->res_in_use) {
 752                         sg_remove_request(sfp, srp);
 753                         return -EBUSY;  /* reserve buffer already being used */
 754                 }
 755         }
 756         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 757         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 758         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 759                 sg_remove_request(sfp, srp);
 760                 return -EMSGSIZE;
 761         }
 762         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 763                 sg_remove_request(sfp, srp);
 764                 return -EFAULT;
 765         }
 766         if (read_only && sg_allow_access(file, cmnd)) {
 767                 sg_remove_request(sfp, srp);
 768                 return -EPERM;
 769         }
 770         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 771         if (k < 0)
 772                 return k;
 773         if (o_srp)
 774                 *o_srp = srp;
 775         return count;
 776 }
 777
 778 static int
 779 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 780                 unsigned char *cmnd, int timeout, int blocking)
 781 {
 782         int k, at_head;
 783         Sg_device *sdp = sfp->parentdp;
 784         sg_io_hdr_t *hp = &srp->header;
 785
 786         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 787         hp->status = 0;
 788         hp->masked_status = 0;
 789         hp->msg_status = 0;
 790         hp->info = 0;
 791         hp->host_status = 0;
 792         hp->driver_status = 0;
 793         hp->resid = 0;
 794         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 795                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 796                         (int) cmnd[0], (int) hp->cmd_len));
 797
 798         if (hp->dxfer_len >= SZ_256M) {
 799                 sg_remove_request(sfp, srp);
 800                 return -EINVAL;
 801         }
 802
 803         k = sg_start_req(srp, cmnd);
 804         if (k) {
 805                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 806                         "sg_common_write: start_req err=%d\n", k));
 807                 sg_finish_rem_req(srp);
 808                 sg_remove_request(sfp, srp);
 809                 return k;       /* probably out of space --> ENOMEM */
 810         }
 811         if (atomic_read(&sdp->detaching)) {
 812                 if (srp->bio) {
 813                         blk_mq_free_request(srp->rq);
 814                         srp->rq = NULL;
 815                 }
 816
 817                 sg_finish_rem_req(srp);
 818                 sg_remove_request(sfp, srp);
 819                 return -ENODEV;
 820         }
 821
 822         hp->duration = jiffies_to_msecs(jiffies);
 823         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 824             (SG_FLAG_Q_AT_TAIL & hp->flags))
 825                 at_head = 0;
 826         else
 827                 at_head = 1;
 828
 829         srp->rq->timeout = timeout;
 830         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 831         srp->rq->end_io = sg_rq_end_io;
 832         blk_execute_rq_nowait(srp->rq, at_head);
 833         return 0;
 834 }
 835
 836 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 837 {
 838         unsigned long flags;
 839         int ret;
 840
 841         read_lock_irqsave(&sfp->rq_list_lock, flags);
 842         ret = srp->done;
 843         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 844         return ret;
 845 }
 846
 847 static int max_sectors_bytes(struct request_queue *q)
 848 {
 849         unsigned int max_sectors = queue_max_sectors(q);
 850
 851         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 852
 853         return max_sectors << 9;
 854 }
 855
 856 static void
 857 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 858 {
 859         Sg_request *srp;
 860         int val;
 861         unsigned int ms;
 862
 863         val = 0;
 864         list_for_each_entry(srp, &sfp->rq_list, entry) {
 865                 if (val >= SG_MAX_QUEUE)
 866                         break;
 867                 rinfo[val].req_state = srp->done + 1;
 868                 rinfo[val].problem =
 869                         srp->header.masked_status &
 870                         srp->header.host_status &
 871                         srp->header.driver_status;
 872                 if (srp->done)
 873                         rinfo[val].duration =
 874                                 srp->header.duration;
 875                 else {
 876                         ms = jiffies_to_msecs(jiffies);
 877                         rinfo[val].duration =
 878                                 (ms > srp->header.duration) ?
 879                                 (ms - srp->header.duration) : 0;
 880                 }
 881                 rinfo[val].orphan = srp->orphan;
 882                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 883                 rinfo[val].pack_id = srp->header.pack_id;
 884                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 885                 val++;
 886         }
 887 }
 888
 889 #ifdef CONFIG_COMPAT
 890 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 891         char req_state;
 892         char orphan;
 893         char sg_io_owned;
 894         char problem;
 895         int pack_id;
 896         compat_uptr_t usr_ptr;
 897         unsigned int duration;
 898         int unused;
 899 };
 900
 901 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 902                                     struct sg_req_info *rinfo)
 903 {
 904         int i;
 905         for (i = 0; i < SG_MAX_QUEUE; i++) {
 906                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 907                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 908                     put_user(rinfo[i].duration, &o[i].duration) ||
 909                     put_user(rinfo[i].unused, &o[i].unused))
 910                         return -EFAULT;
 911         }
 912         return 0;
 913 }
 914 #endif
 915
 916 static long
 917 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 918                 unsigned int cmd_in, void __user *p)
 919 {
 920         int __user *ip = p;
 921         int result, val, read_only;
 922         Sg_request *srp;
 923         unsigned long iflags;
 924
 925         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 926                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 927         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 928
 929         switch (cmd_in) {
 930         case SG_IO:
 931                 if (atomic_read(&sdp->detaching))
 932                         return -ENODEV;
 933                 if (!scsi_block_when_processing_errors(sdp->device))
 934                         return -ENXIO;
 935                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 936                                  1, read_only, 1, &srp);
 937                 if (result < 0)
 938                         return result;
 939                 result = wait_event_interruptible(sfp->read_wait,
 940                         srp_done(sfp, srp));
 941                 write_lock_irq(&sfp->rq_list_lock);
 942                 if (srp->done) {
 943                         srp->done = 2;
 944                         write_unlock_irq(&sfp->rq_list_lock);
 945                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 946                         return (result < 0) ? result : 0;
 947                 }
 948                 srp->orphan = 1;
 949                 write_unlock_irq(&sfp->rq_list_lock);
 950                 return result;  /* -ERESTARTSYS because signal hit process */
 951         case SG_SET_TIMEOUT:
 952                 result = get_user(val, ip);
 953                 if (result)
 954                         return result;
 955                 if (val < 0)
 956                         return -EIO;
 957                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 958                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 959                                     INT_MAX);
 960                 sfp->timeout_user = val;
 961                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 962
 963                 return 0;
 964         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 965                                 /* strange ..., for backward compatibility */
 966                 return sfp->timeout_user;
 967         case SG_SET_FORCE_LOW_DMA:
 968                 /*
 969                  * N.B. This ioctl never worked properly, but failed to
 970                  * return an error value. So returning '0' to keep compability
 971                  * with legacy applications.
 972                  */
 973                 return 0;
 974         case SG_GET_LOW_DMA:
 975                 return put_user(0, ip);
 976         case SG_GET_SCSI_ID:
 977                 {
 978                         sg_scsi_id_t v;
 979
 980                         if (atomic_read(&sdp->detaching))
 981                                 return -ENODEV;
 982                         memset(&v, 0, sizeof(v));
 983                         v.host_no = sdp->device->host->host_no;
 984                         v.channel = sdp->device->channel;
 985                         v.scsi_id = sdp->device->id;
 986                         v.lun = sdp->device->lun;
 987                         v.scsi_type = sdp->device->type;
 988                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 989                         v.d_queue_depth = sdp->device->queue_depth;
 990                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 991                                 return -EFAULT;
 992                         return 0;
 993                 }
 994         case SG_SET_FORCE_PACK_ID:
 995                 result = get_user(val, ip);
 996                 if (result)
 997                         return result;
 998                 sfp->force_packid = val ? 1 : 0;
 999                 return 0;
1000         case SG_GET_PACK_ID:
1001                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1002                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1003                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1004                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1005                                                        iflags);
1006                                 return put_user(srp->header.pack_id, ip);
1007                         }
1008                 }
1009                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1010                 return put_user(-1, ip);
1011         case SG_GET_NUM_WAITING:
1012                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1013                 val = 0;
1014                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1015                         if ((1 == srp->done) && (!srp->sg_io_owned))
1016                                 ++val;
1017                 }
1018                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1019                 return put_user(val, ip);
1020         case SG_GET_SG_TABLESIZE:
1021                 return put_user(sdp->sg_tablesize, ip);
1022         case SG_SET_RESERVED_SIZE:
1023                 result = get_user(val, ip);
1024                 if (result)
1025                         return result;
1026                 if (val < 0)
1027                         return -EINVAL;
1028                 val = min_t(int, val,
1029                             max_sectors_bytes(sdp->device->request_queue));
1030                 mutex_lock(&sfp->f_mutex);
1031                 if (val != sfp->reserve.bufflen) {
1032                         if (sfp->mmap_called ||
1033                             sfp->res_in_use) {
1034                                 mutex_unlock(&sfp->f_mutex);
1035                                 return -EBUSY;
1036                         }
1037
1038                         sg_remove_scat(sfp, &sfp->reserve);
1039                         sg_build_reserve(sfp, val);
1040                 }
1041                 mutex_unlock(&sfp->f_mutex);
1042                 return 0;
1043         case SG_GET_RESERVED_SIZE:
1044                 val = min_t(int, sfp->reserve.bufflen,
1045                             max_sectors_bytes(sdp->device->request_queue));
1046                 return put_user(val, ip);
1047         case SG_SET_COMMAND_Q:
1048                 result = get_user(val, ip);
1049                 if (result)
1050                         return result;
1051                 sfp->cmd_q = val ? 1 : 0;
1052                 return 0;
1053         case SG_GET_COMMAND_Q:
1054                 return put_user((int) sfp->cmd_q, ip);
1055         case SG_SET_KEEP_ORPHAN:
1056                 result = get_user(val, ip);
1057                 if (result)
1058                         return result;
1059                 sfp->keep_orphan = val;
1060                 return 0;
1061         case SG_GET_KEEP_ORPHAN:
1062                 return put_user((int) sfp->keep_orphan, ip);
1063         case SG_NEXT_CMD_LEN:
1064                 result = get_user(val, ip);
1065                 if (result)
1066                         return result;
1067                 if (val > SG_MAX_CDB_SIZE)
1068                         return -ENOMEM;
1069                 sfp->next_cmd_len = (val > 0) ? val : 0;
1070                 return 0;
1071         case SG_GET_VERSION_NUM:
1072                 return put_user(sg_version_num, ip);
1073         case SG_GET_ACCESS_COUNT:
1074                 /* faked - we don't have a real access count anymore */
1075                 val = (sdp->device ? 1 : 0);
1076                 return put_user(val, ip);
1077         case SG_GET_REQUEST_TABLE:
1078                 {
1079                         sg_req_info_t *rinfo;
1080
1081                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1082                                         GFP_KERNEL);
1083                         if (!rinfo)
1084                                 return -ENOMEM;
1085                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1086                         sg_fill_request_table(sfp, rinfo);
1087                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1088         #ifdef CONFIG_COMPAT
1089                         if (in_compat_syscall())
1090                                 result = put_compat_request_table(p, rinfo);
1091                         else
1092         #endif
1093                                 result = copy_to_user(p, rinfo,
1094                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1095                         result = result ? -EFAULT : 0;
1096                         kfree(rinfo);
1097                         return result;
1098                 }
1099         case SG_EMULATED_HOST:
1100                 if (atomic_read(&sdp->detaching))
1101                         return -ENODEV;
1102                 return put_user(sdp->device->host->hostt->emulated, ip);
1103         case SCSI_IOCTL_SEND_COMMAND:
1104                 if (atomic_read(&sdp->detaching))
1105                         return -ENODEV;
1106                 return scsi_ioctl(sdp->device, filp->f_mode, cmd_in, p);
1107         case SG_SET_DEBUG:
1108                 result = get_user(val, ip);
1109                 if (result)
1110                         return result;
1111                 sdp->sgdebug = (char) val;
1112                 return 0;
1113         case BLKSECTGET:
1114                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1115                                 ip);
1116         case BLKTRACESETUP:
1117                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1118                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1119                                        NULL, p);
1120         case BLKTRACESTART:
1121                 return blk_trace_startstop(sdp->device->request_queue, 1);
1122         case BLKTRACESTOP:
1123                 return blk_trace_startstop(sdp->device->request_queue, 0);
1124         case BLKTRACETEARDOWN:
1125                 return blk_trace_remove(sdp->device->request_queue);
1126         case SCSI_IOCTL_GET_IDLUN:
1127         case SCSI_IOCTL_GET_BUS_NUMBER:
1128         case SCSI_IOCTL_PROBE_HOST:
1129         case SG_GET_TRANSFORM:
1130         case SG_SCSI_RESET:
1131                 if (atomic_read(&sdp->detaching))
1132                         return -ENODEV;
1133                 break;
1134         default:
1135                 if (read_only)
1136                         return -EPERM;  /* don't know so take safe approach */
1137                 break;
1138         }
1139
1140         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1141                         cmd_in, filp->f_flags & O_NDELAY);
1142         if (result)
1143                 return result;
1144
1145         return -ENOIOCTLCMD;
1146 }
1147
1148 static long
1149 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1150 {
1151         void __user *p = (void __user *)arg;
1152         Sg_device *sdp;
1153         Sg_fd *sfp;
1154         int ret;
1155
1156         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1157                 return -ENXIO;
1158
1159         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1160         if (ret != -ENOIOCTLCMD)
1161                 return ret;
1162         return scsi_ioctl(sdp->device, filp->f_mode, cmd_in, p);
1163 }
1164
1165 static __poll_t
1166 sg_poll(struct file *filp, poll_table * wait)
1167 {
1168         __poll_t res = 0;
1169         Sg_device *sdp;
1170         Sg_fd *sfp;
1171         Sg_request *srp;
1172         int count = 0;
1173         unsigned long iflags;
1174
1175         sfp = filp->private_data;
1176         if (!sfp)
1177                 return EPOLLERR;
1178         sdp = sfp->parentdp;
1179         if (!sdp)
1180                 return EPOLLERR;
1181         poll_wait(filp, &sfp->read_wait, wait);
1182         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1183         list_for_each_entry(srp, &sfp->rq_list, entry) {
1184                 /* if any read waiting, flag it */
1185                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1186                         res = EPOLLIN | EPOLLRDNORM;
1187                 ++count;
1188         }
1189         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1190
1191         if (atomic_read(&sdp->detaching))
1192                 res |= EPOLLHUP;
1193         else if (!sfp->cmd_q) {
1194                 if (0 == count)
1195                         res |= EPOLLOUT | EPOLLWRNORM;
1196         } else if (count < SG_MAX_QUEUE)
1197                 res |= EPOLLOUT | EPOLLWRNORM;
1198         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1199                                       "sg_poll: res=0x%x\n", (__force u32) res));
1200         return res;
1201 }
1202
1203 static int
1204 sg_fasync(int fd, struct file *filp, int mode)
1205 {
1206         Sg_device *sdp;
1207         Sg_fd *sfp;
1208
1209         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1210                 return -ENXIO;
1211         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1212                                       "sg_fasync: mode=%d\n", mode));
1213
1214         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1215 }
1216
1217 static vm_fault_t
1218 sg_vma_fault(struct vm_fault *vmf)
1219 {
1220         struct vm_area_struct *vma = vmf->vma;
1221         Sg_fd *sfp;
1222         unsigned long offset, len, sa;
1223         Sg_scatter_hold *rsv_schp;
1224         int k, length;
1225
1226         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1227                 return VM_FAULT_SIGBUS;
1228         rsv_schp = &sfp->reserve;
1229         offset = vmf->pgoff << PAGE_SHIFT;
1230         if (offset >= rsv_schp->bufflen)
1231                 return VM_FAULT_SIGBUS;
1232         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1233                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1234                                       offset, rsv_schp->k_use_sg));
1235         sa = vma->vm_start;
1236         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1237         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1238                 len = vma->vm_end - sa;
1239                 len = (len < length) ? len : length;
1240                 if (offset < len) {
1241                         struct page *page = nth_page(rsv_schp->pages[k],
1242                                                      offset >> PAGE_SHIFT);
1243                         get_page(page); /* increment page count */
1244                         vmf->page = page;
1245                         return 0; /* success */
1246                 }
1247                 sa += len;
1248                 offset -= len;
1249         }
1250
1251         return VM_FAULT_SIGBUS;
1252 }
1253
1254 static const struct vm_operations_struct sg_mmap_vm_ops = {
1255         .fault = sg_vma_fault,
1256 };
1257
1258 static int
1259 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1260 {
1261         Sg_fd *sfp;
1262         unsigned long req_sz, len, sa;
1263         Sg_scatter_hold *rsv_schp;
1264         int k, length;
1265         int ret = 0;
1266
1267         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1268                 return -ENXIO;
1269         req_sz = vma->vm_end - vma->vm_start;
1270         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1271                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1272                                       (void *) vma->vm_start, (int) req_sz));
1273         if (vma->vm_pgoff)
1274                 return -EINVAL; /* want no offset */
1275         rsv_schp = &sfp->reserve;
1276         mutex_lock(&sfp->f_mutex);
1277         if (req_sz > rsv_schp->bufflen) {
1278                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1279                 goto out;
1280         }
1281
1282         sa = vma->vm_start;
1283         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1284         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1285                 len = vma->vm_end - sa;
1286                 len = (len < length) ? len : length;
1287                 sa += len;
1288         }
1289
1290         sfp->mmap_called = 1;
1291         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1292         vma->vm_private_data = sfp;
1293         vma->vm_ops = &sg_mmap_vm_ops;
1294 out:
1295         mutex_unlock(&sfp->f_mutex);
1296         return ret;
1297 }
1298
1299 static void
1300 sg_rq_end_io_usercontext(struct work_struct *work)
1301 {
1302         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1303         struct sg_fd *sfp = srp->parentfp;
1304
1305         sg_finish_rem_req(srp);
1306         sg_remove_request(sfp, srp);
1307         kref_put(&sfp->f_ref, sg_remove_sfp);
1308 }
1309
1310 /*
1311  * This function is a "bottom half" handler that is called by the mid
1312  * level when a command is completed (or has failed).
1313  */
1314 static enum rq_end_io_ret
1315 sg_rq_end_io(struct request *rq, blk_status_t status)
1316 {
1317         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq);
1318         struct sg_request *srp = rq->end_io_data;
1319         Sg_device *sdp;
1320         Sg_fd *sfp;
1321         unsigned long iflags;
1322         unsigned int ms;
1323         char *sense;
1324         int result, resid, done = 1;
1325
1326         if (WARN_ON(srp->done != 0))
1327                 return RQ_END_IO_NONE;
1328
1329         sfp = srp->parentfp;
1330         if (WARN_ON(sfp == NULL))
1331                 return RQ_END_IO_NONE;
1332
1333         sdp = sfp->parentdp;
1334         if (unlikely(atomic_read(&sdp->detaching)))
1335                 pr_info("%s: device detaching\n", __func__);
1336
1337         sense = scmd->sense_buffer;
1338         result = scmd->result;
1339         resid = scmd->resid_len;
1340
1341         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1342                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1343                                       srp->header.pack_id, result));
1344         srp->header.resid = resid;
1345         ms = jiffies_to_msecs(jiffies);
1346         srp->header.duration = (ms > srp->header.duration) ?
1347                                 (ms - srp->header.duration) : 0;
1348         if (0 != result) {
1349                 struct scsi_sense_hdr sshdr;
1350
1351                 srp->header.status = 0xff & result;
1352                 srp->header.masked_status = status_byte(result);
1353                 srp->header.msg_status = COMMAND_COMPLETE;
1354                 srp->header.host_status = host_byte(result);
1355                 srp->header.driver_status = driver_byte(result);
1356                 if ((sdp->sgdebug > 0) &&
1357                     ((CHECK_CONDITION == srp->header.masked_status) ||
1358                      (COMMAND_TERMINATED == srp->header.masked_status)))
1359                         __scsi_print_sense(sdp->device, __func__, sense,
1360                                            SCSI_SENSE_BUFFERSIZE);
1361
1362                 /* Following if statement is a patch supplied by Eric Youngdale */
1363                 if (driver_byte(result) != 0
1364                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1365                     && !scsi_sense_is_deferred(&sshdr)
1366                     && sshdr.sense_key == UNIT_ATTENTION
1367                     && sdp->device->removable) {
1368                         /* Detected possible disc change. Set the bit - this */
1369                         /* may be used if there are filesystems using this device */
1370                         sdp->device->changed = 1;
1371                 }
1372         }
1373
1374         if (scmd->sense_len)
1375                 memcpy(srp->sense_b, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
1376
1377         /* Rely on write phase to clean out srp status values, so no "else" */
1378
1379         /*
1380          * Free the request as soon as it is complete so that its resources
1381          * can be reused without waiting for userspace to read() the
1382          * result.  But keep the associated bio (if any) around until
1383          * blk_rq_unmap_user() can be called from user context.
1384          */
1385         srp->rq = NULL;
1386         blk_mq_free_request(rq);
1387
1388         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1389         if (unlikely(srp->orphan)) {
1390                 if (sfp->keep_orphan)
1391                         srp->sg_io_owned = 0;
1392                 else
1393                         done = 0;
1394         }
1395         srp->done = done;
1396         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1397
1398         if (likely(done)) {
1399                 /* Now wake up any sg_read() that is waiting for this
1400                  * packet.
1401                  */
1402                 wake_up_interruptible(&sfp->read_wait);
1403                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1404                 kref_put(&sfp->f_ref, sg_remove_sfp);
1405         } else {
1406                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1407                 schedule_work(&srp->ew.work);
1408         }
1409         return RQ_END_IO_NONE;
1410 }
1411
1412 static const struct file_operations sg_fops = {
1413         .owner = THIS_MODULE,
1414         .read = sg_read,
1415         .write = sg_write,
1416         .poll = sg_poll,
1417         .unlocked_ioctl = sg_ioctl,
1418         .compat_ioctl = compat_ptr_ioctl,
1419         .open = sg_open,
1420         .mmap = sg_mmap,
1421         .release = sg_release,
1422         .fasync = sg_fasync,
1423         .llseek = no_llseek,
1424 };
1425
1426 static struct class *sg_sysfs_class;
1427
1428 static int sg_sysfs_valid = 0;
1429
1430 static Sg_device *
1431 sg_alloc(struct scsi_device *scsidp)
1432 {
1433         struct request_queue *q = scsidp->request_queue;
1434         Sg_device *sdp;
1435         unsigned long iflags;
1436         int error;
1437         u32 k;
1438
1439         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1440         if (!sdp) {
1441                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1442                             "failure\n", __func__);
1443                 return ERR_PTR(-ENOMEM);
1444         }
1445
1446         idr_preload(GFP_KERNEL);
1447         write_lock_irqsave(&sg_index_lock, iflags);
1448
1449         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1450         if (error < 0) {
1451                 if (error == -ENOSPC) {
1452                         sdev_printk(KERN_WARNING, scsidp,
1453                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1454                                     scsidp->type, SG_MAX_DEVS - 1);
1455                         error = -ENODEV;
1456                 } else {
1457                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1458                                     "allocation Sg_device failure: %d\n",
1459                                     __func__, error);
1460                 }
1461                 goto out_unlock;
1462         }
1463         k = error;
1464
1465         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1466                                         "sg_alloc: dev=%d \n", k));
1467         sprintf(sdp->name, "sg%d", k);
1468         sdp->device = scsidp;
1469         mutex_init(&sdp->open_rel_lock);
1470         INIT_LIST_HEAD(&sdp->sfds);
1471         init_waitqueue_head(&sdp->open_wait);
1472         atomic_set(&sdp->detaching, 0);
1473         rwlock_init(&sdp->sfd_lock);
1474         sdp->sg_tablesize = queue_max_segments(q);
1475         sdp->index = k;
1476         kref_init(&sdp->d_ref);
1477         error = 0;
1478
1479 out_unlock:
1480         write_unlock_irqrestore(&sg_index_lock, iflags);
1481         idr_preload_end();
1482
1483         if (error) {
1484                 kfree(sdp);
1485                 return ERR_PTR(error);
1486         }
1487         return sdp;
1488 }
1489
1490 static int
1491 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1492 {
1493         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1494         Sg_device *sdp = NULL;
1495         struct cdev * cdev = NULL;
1496         int error;
1497         unsigned long iflags;
1498
1499         error = -ENOMEM;
1500         cdev = cdev_alloc();
1501         if (!cdev) {
1502                 pr_warn("%s: cdev_alloc failed\n", __func__);
1503                 goto out;
1504         }
1505         cdev->owner = THIS_MODULE;
1506         cdev->ops = &sg_fops;
1507
1508         sdp = sg_alloc(scsidp);
1509         if (IS_ERR(sdp)) {
1510                 pr_warn("%s: sg_alloc failed\n", __func__);
1511                 error = PTR_ERR(sdp);
1512                 goto out;
1513         }
1514
1515         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1516         if (error)
1517                 goto cdev_add_err;
1518
1519         sdp->cdev = cdev;
1520         if (sg_sysfs_valid) {
1521                 struct device *sg_class_member;
1522
1523                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1524                                                 MKDEV(SCSI_GENERIC_MAJOR,
1525                                                       sdp->index),
1526                                                 sdp, "%s", sdp->name);
1527                 if (IS_ERR(sg_class_member)) {
1528                         pr_err("%s: device_create failed\n", __func__);
1529                         error = PTR_ERR(sg_class_member);
1530                         goto cdev_add_err;
1531                 }
1532                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1533                                           &sg_class_member->kobj, "generic");
1534                 if (error)
1535                         pr_err("%s: unable to make symlink 'generic' back "
1536                                "to sg%d\n", __func__, sdp->index);
1537         } else
1538                 pr_warn("%s: sg_sys Invalid\n", __func__);
1539
1540         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1541                     "type %d\n", sdp->index, scsidp->type);
1542
1543         dev_set_drvdata(cl_dev, sdp);
1544
1545         return 0;
1546
1547 cdev_add_err:
1548         write_lock_irqsave(&sg_index_lock, iflags);
1549         idr_remove(&sg_index_idr, sdp->index);
1550         write_unlock_irqrestore(&sg_index_lock, iflags);
1551         kfree(sdp);
1552
1553 out:
1554         if (cdev)
1555                 cdev_del(cdev);
1556         return error;
1557 }
1558
1559 static void
1560 sg_device_destroy(struct kref *kref)
1561 {
1562         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1563         unsigned long flags;
1564
1565         /* CAUTION!  Note that the device can still be found via idr_find()
1566          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1567          * any other cleanup.
1568          */
1569
1570         write_lock_irqsave(&sg_index_lock, flags);
1571         idr_remove(&sg_index_idr, sdp->index);
1572         write_unlock_irqrestore(&sg_index_lock, flags);
1573
1574         SCSI_LOG_TIMEOUT(3,
1575                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1576
1577         kfree(sdp);
1578 }
1579
1580 static void
1581 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1582 {
1583         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1584         Sg_device *sdp = dev_get_drvdata(cl_dev);
1585         unsigned long iflags;
1586         Sg_fd *sfp;
1587         int val;
1588
1589         if (!sdp)
1590                 return;
1591         /* want sdp->detaching non-zero as soon as possible */
1592         val = atomic_inc_return(&sdp->detaching);
1593         if (val > 1)
1594                 return; /* only want to do following once per device */
1595
1596         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1597                                       "%s\n", __func__));
1598
1599         read_lock_irqsave(&sdp->sfd_lock, iflags);
1600         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1601                 wake_up_interruptible_all(&sfp->read_wait);
1602                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1603         }
1604         wake_up_interruptible_all(&sdp->open_wait);
1605         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1606
1607         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1608         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1609         cdev_del(sdp->cdev);
1610         sdp->cdev = NULL;
1611
1612         kref_put(&sdp->d_ref, sg_device_destroy);
1613 }
1614
1615 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1616 module_param_named(def_reserved_size, def_reserved_size, int,
1617                    S_IRUGO | S_IWUSR);
1618 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1619
1620 MODULE_AUTHOR("Douglas Gilbert");
1621 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1622 MODULE_LICENSE("GPL");
1623 MODULE_VERSION(SG_VERSION_STR);
1624 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1625
1626 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1627                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1628 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1629 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1630
1631 #ifdef CONFIG_SYSCTL
1632 #include <linux/sysctl.h>
1633
1634 static struct ctl_table sg_sysctls[] = {
1635         {
1636                 .procname       = "sg-big-buff",
1637                 .data           = &sg_big_buff,
1638                 .maxlen         = sizeof(int),
1639                 .mode           = 0444,
1640                 .proc_handler   = proc_dointvec,
1641         },
1642         {}
1643 };
1644
1645 static struct ctl_table_header *hdr;
1646 static void register_sg_sysctls(void)
1647 {
1648         if (!hdr)
1649                 hdr = register_sysctl("kernel", sg_sysctls);
1650 }
1651
1652 static void unregister_sg_sysctls(void)
1653 {
1654         if (hdr)
1655                 unregister_sysctl_table(hdr);
1656 }
1657 #else
1658 #define register_sg_sysctls() do { } while (0)
1659 #define unregister_sg_sysctls() do { } while (0)
1660 #endif /* CONFIG_SYSCTL */
1661
1662 static int __init
1663 init_sg(void)
1664 {
1665         int rc;
1666
1667         if (scatter_elem_sz < PAGE_SIZE) {
1668                 scatter_elem_sz = PAGE_SIZE;
1669                 scatter_elem_sz_prev = scatter_elem_sz;
1670         }
1671         if (def_reserved_size >= 0)
1672                 sg_big_buff = def_reserved_size;
1673         else
1674                 def_reserved_size = sg_big_buff;
1675
1676         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1677                                     SG_MAX_DEVS, "sg");
1678         if (rc)
1679                 return rc;
1680         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1681         if ( IS_ERR(sg_sysfs_class) ) {
1682                 rc = PTR_ERR(sg_sysfs_class);
1683                 goto err_out;
1684         }
1685         sg_sysfs_valid = 1;
1686         rc = scsi_register_interface(&sg_interface);
1687         if (0 == rc) {
1688 #ifdef CONFIG_SCSI_PROC_FS
1689                 sg_proc_init();
1690 #endif                          /* CONFIG_SCSI_PROC_FS */
1691                 return 0;
1692         }
1693         class_destroy(sg_sysfs_class);
1694         register_sg_sysctls();
1695 err_out:
1696         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1697         return rc;
1698 }
1699
1700 static void __exit
1701 exit_sg(void)
1702 {
1703         unregister_sg_sysctls();
1704 #ifdef CONFIG_SCSI_PROC_FS
1705         remove_proc_subtree("scsi/sg", NULL);
1706 #endif                          /* CONFIG_SCSI_PROC_FS */
1707         scsi_unregister_interface(&sg_interface);
1708         class_destroy(sg_sysfs_class);
1709         sg_sysfs_valid = 0;
1710         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1711                                  SG_MAX_DEVS);
1712         idr_destroy(&sg_index_idr);
1713 }
1714
1715 static int
1716 sg_start_req(Sg_request *srp, unsigned char *cmd)
1717 {
1718         int res;
1719         struct request *rq;
1720         Sg_fd *sfp = srp->parentfp;
1721         sg_io_hdr_t *hp = &srp->header;
1722         int dxfer_len = (int) hp->dxfer_len;
1723         int dxfer_dir = hp->dxfer_direction;
1724         unsigned int iov_count = hp->iovec_count;
1725         Sg_scatter_hold *req_schp = &srp->data;
1726         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1727         struct request_queue *q = sfp->parentdp->device->request_queue;
1728         struct rq_map_data *md, map_data;
1729         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1730         struct scsi_cmnd *scmd;
1731
1732         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1733                                       "sg_start_req: dxfer_len=%d\n",
1734                                       dxfer_len));
1735
1736         /*
1737          * NOTE
1738          *
1739          * With scsi-mq enabled, there are a fixed number of preallocated
1740          * requests equal in number to shost->can_queue.  If all of the
1741          * preallocated requests are already in use, then scsi_alloc_request()
1742          * will sleep until an active command completes, freeing up a request.
1743          * Although waiting in an asynchronous interface is less than ideal, we
1744          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1745          * not expect an EWOULDBLOCK from this condition.
1746          */
1747         rq = scsi_alloc_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1748                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1749         if (IS_ERR(rq))
1750                 return PTR_ERR(rq);
1751         scmd = blk_mq_rq_to_pdu(rq);
1752
1753         if (hp->cmd_len > sizeof(scmd->cmnd)) {
1754                 blk_mq_free_request(rq);
1755                 return -EINVAL;
1756         }
1757
1758         memcpy(scmd->cmnd, cmd, hp->cmd_len);
1759         scmd->cmd_len = hp->cmd_len;
1760
1761         srp->rq = rq;
1762         rq->end_io_data = srp;
1763         scmd->allowed = SG_DEFAULT_RETRIES;
1764
1765         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1766                 return 0;
1767
1768         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1769             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1770             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1771                 md = NULL;
1772         else
1773                 md = &map_data;
1774
1775         if (md) {
1776                 mutex_lock(&sfp->f_mutex);
1777                 if (dxfer_len <= rsv_schp->bufflen &&
1778                     !sfp->res_in_use) {
1779                         sfp->res_in_use = 1;
1780                         sg_link_reserve(sfp, srp, dxfer_len);
1781                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1782                         res = -EBUSY; /* sfp->res_in_use == 1 */
1783                         if (dxfer_len > rsv_schp->bufflen)
1784                                 res = -ENOMEM;
1785                         mutex_unlock(&sfp->f_mutex);
1786                         return res;
1787                 } else {
1788                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1789                         if (res) {
1790                                 mutex_unlock(&sfp->f_mutex);
1791                                 return res;
1792                         }
1793                 }
1794                 mutex_unlock(&sfp->f_mutex);
1795
1796                 md->pages = req_schp->pages;
1797                 md->page_order = req_schp->page_order;
1798                 md->nr_entries = req_schp->k_use_sg;
1799                 md->offset = 0;
1800                 md->null_mapped = hp->dxferp ? 0 : 1;
1801                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1802                         md->from_user = 1;
1803                 else
1804                         md->from_user = 0;
1805         }
1806
1807         res = blk_rq_map_user_io(rq, md, hp->dxferp, hp->dxfer_len,
1808                         GFP_ATOMIC, iov_count, iov_count, 1, rw);
1809         if (!res) {
1810                 srp->bio = rq->bio;
1811
1812                 if (!md) {
1813                         req_schp->dio_in_use = 1;
1814                         hp->info |= SG_INFO_DIRECT_IO;
1815                 }
1816         }
1817         return res;
1818 }
1819
1820 static int
1821 sg_finish_rem_req(Sg_request *srp)
1822 {
1823         int ret = 0;
1824
1825         Sg_fd *sfp = srp->parentfp;
1826         Sg_scatter_hold *req_schp = &srp->data;
1827
1828         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1829                                       "sg_finish_rem_req: res_used=%d\n",
1830                                       (int) srp->res_used));
1831         if (srp->bio)
1832                 ret = blk_rq_unmap_user(srp->bio);
1833
1834         if (srp->rq)
1835                 blk_mq_free_request(srp->rq);
1836
1837         if (srp->res_used)
1838                 sg_unlink_reserve(sfp, srp);
1839         else
1840                 sg_remove_scat(sfp, req_schp);
1841
1842         return ret;
1843 }
1844
1845 static int
1846 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1847 {
1848         int sg_bufflen = tablesize * sizeof(struct page *);
1849         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1850
1851         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1852         if (!schp->pages)
1853                 return -ENOMEM;
1854         schp->sglist_len = sg_bufflen;
1855         return tablesize;       /* number of scat_gath elements allocated */
1856 }
1857
1858 static int
1859 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1860 {
1861         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1862         int sg_tablesize = sfp->parentdp->sg_tablesize;
1863         int blk_size = buff_size, order;
1864         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1865
1866         if (blk_size < 0)
1867                 return -EFAULT;
1868         if (0 == blk_size)
1869                 ++blk_size;     /* don't know why */
1870         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1871         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1872         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1873                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1874                 buff_size, blk_size));
1875
1876         /* N.B. ret_sz carried into this block ... */
1877         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1878         if (mx_sc_elems < 0)
1879                 return mx_sc_elems;     /* most likely -ENOMEM */
1880
1881         num = scatter_elem_sz;
1882         if (unlikely(num != scatter_elem_sz_prev)) {
1883                 if (num < PAGE_SIZE) {
1884                         scatter_elem_sz = PAGE_SIZE;
1885                         scatter_elem_sz_prev = PAGE_SIZE;
1886                 } else
1887                         scatter_elem_sz_prev = num;
1888         }
1889
1890         order = get_order(num);
1891 retry:
1892         ret_sz = 1 << (PAGE_SHIFT + order);
1893
1894         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1895              k++, rem_sz -= ret_sz) {
1896
1897                 num = (rem_sz > scatter_elem_sz_prev) ?
1898                         scatter_elem_sz_prev : rem_sz;
1899
1900                 schp->pages[k] = alloc_pages(gfp_mask, order);
1901                 if (!schp->pages[k])
1902                         goto out;
1903
1904                 if (num == scatter_elem_sz_prev) {
1905                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1906                                 scatter_elem_sz = ret_sz;
1907                                 scatter_elem_sz_prev = ret_sz;
1908                         }
1909                 }
1910
1911                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1912                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1913                                  k, num, ret_sz));
1914         }               /* end of for loop */
1915
1916         schp->page_order = order;
1917         schp->k_use_sg = k;
1918         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1919                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1920                          k, rem_sz));
1921
1922         schp->bufflen = blk_size;
1923         if (rem_sz > 0) /* must have failed */
1924                 return -ENOMEM;
1925         return 0;
1926 out:
1927         for (i = 0; i < k; i++)
1928                 __free_pages(schp->pages[i], order);
1929
1930         if (--order >= 0)
1931                 goto retry;
1932
1933         return -ENOMEM;
1934 }
1935
1936 static void
1937 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1938 {
1939         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1940                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1941         if (schp->pages && schp->sglist_len > 0) {
1942                 if (!schp->dio_in_use) {
1943                         int k;
1944
1945                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1946                                 SCSI_LOG_TIMEOUT(5,
1947                                         sg_printk(KERN_INFO, sfp->parentdp,
1948                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1949                                         k, schp->pages[k]));
1950                                 __free_pages(schp->pages[k], schp->page_order);
1951                         }
1952
1953                         kfree(schp->pages);
1954                 }
1955         }
1956         memset(schp, 0, sizeof (*schp));
1957 }
1958
1959 static int
1960 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1961 {
1962         Sg_scatter_hold *schp = &srp->data;
1963         int k, num;
1964
1965         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1966                          "sg_read_oxfer: num_read_xfer=%d\n",
1967                          num_read_xfer));
1968         if ((!outp) || (num_read_xfer <= 0))
1969                 return 0;
1970
1971         num = 1 << (PAGE_SHIFT + schp->page_order);
1972         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1973                 if (num > num_read_xfer) {
1974                         if (copy_to_user(outp, page_address(schp->pages[k]),
1975                                            num_read_xfer))
1976                                 return -EFAULT;
1977                         break;
1978                 } else {
1979                         if (copy_to_user(outp, page_address(schp->pages[k]),
1980                                            num))
1981                                 return -EFAULT;
1982                         num_read_xfer -= num;
1983                         if (num_read_xfer <= 0)
1984                                 break;
1985                         outp += num;
1986                 }
1987         }
1988
1989         return 0;
1990 }
1991
1992 static void
1993 sg_build_reserve(Sg_fd * sfp, int req_size)
1994 {
1995         Sg_scatter_hold *schp = &sfp->reserve;
1996
1997         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1998                          "sg_build_reserve: req_size=%d\n", req_size));
1999         do {
2000                 if (req_size < PAGE_SIZE)
2001                         req_size = PAGE_SIZE;
2002                 if (0 == sg_build_indirect(schp, sfp, req_size))
2003                         return;
2004                 else
2005                         sg_remove_scat(sfp, schp);
2006                 req_size >>= 1; /* divide by 2 */
2007         } while (req_size > (PAGE_SIZE / 2));
2008 }
2009
2010 static void
2011 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2012 {
2013         Sg_scatter_hold *req_schp = &srp->data;
2014         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2015         int k, num, rem;
2016
2017         srp->res_used = 1;
2018         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2019                          "sg_link_reserve: size=%d\n", size));
2020         rem = size;
2021
2022         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2023         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2024                 if (rem <= num) {
2025                         req_schp->k_use_sg = k + 1;
2026                         req_schp->sglist_len = rsv_schp->sglist_len;
2027                         req_schp->pages = rsv_schp->pages;
2028
2029                         req_schp->bufflen = size;
2030                         req_schp->page_order = rsv_schp->page_order;
2031                         break;
2032                 } else
2033                         rem -= num;
2034         }
2035
2036         if (k >= rsv_schp->k_use_sg)
2037                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2038                                  "sg_link_reserve: BAD size\n"));
2039 }
2040
2041 static void
2042 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2043 {
2044         Sg_scatter_hold *req_schp = &srp->data;
2045
2046         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2047                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2048                                       (int) req_schp->k_use_sg));
2049         req_schp->k_use_sg = 0;
2050         req_schp->bufflen = 0;
2051         req_schp->pages = NULL;
2052         req_schp->page_order = 0;
2053         req_schp->sglist_len = 0;
2054         srp->res_used = 0;
2055         /* Called without mutex lock to avoid deadlock */
2056         sfp->res_in_use = 0;
2057 }
2058
2059 static Sg_request *
2060 sg_get_rq_mark(Sg_fd * sfp, int pack_id, bool *busy)
2061 {
2062         Sg_request *resp;
2063         unsigned long iflags;
2064
2065         *busy = false;
2066         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2067         list_for_each_entry(resp, &sfp->rq_list, entry) {
2068                 /* look for requests that are not SG_IO owned */
2069                 if ((!resp->sg_io_owned) &&
2070                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2071                         switch (resp->done) {
2072                         case 0: /* request active */
2073                                 *busy = true;
2074                                 break;
2075                         case 1: /* request done; response ready to return */
2076                                 resp->done = 2; /* guard against other readers */
2077                                 write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2078                                 return resp;
2079                         case 2: /* response already being returned */
2080                                 break;
2081                         }
2082                 }
2083         }
2084         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2085         return NULL;
2086 }
2087
2088 /* always adds to end of list */
2089 static Sg_request *
2090 sg_add_request(Sg_fd * sfp)
2091 {
2092         int k;
2093         unsigned long iflags;
2094         Sg_request *rp = sfp->req_arr;
2095
2096         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2097         if (!list_empty(&sfp->rq_list)) {
2098                 if (!sfp->cmd_q)
2099                         goto out_unlock;
2100
2101                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2102                         if (!rp->parentfp)
2103                                 break;
2104                 }
2105                 if (k >= SG_MAX_QUEUE)
2106                         goto out_unlock;
2107         }
2108         memset(rp, 0, sizeof (Sg_request));
2109         rp->parentfp = sfp;
2110         rp->header.duration = jiffies_to_msecs(jiffies);
2111         list_add_tail(&rp->entry, &sfp->rq_list);
2112         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2113         return rp;
2114 out_unlock:
2115         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2116         return NULL;
2117 }
2118
2119 /* Return of 1 for found; 0 for not found */
2120 static int
2121 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2122 {
2123         unsigned long iflags;
2124         int res = 0;
2125
2126         if (!sfp || !srp || list_empty(&sfp->rq_list))
2127                 return res;
2128         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2129         if (!list_empty(&srp->entry)) {
2130                 list_del(&srp->entry);
2131                 srp->parentfp = NULL;
2132                 res = 1;
2133         }
2134         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2135
2136         /*
2137          * If the device is detaching, wakeup any readers in case we just
2138          * removed the last response, which would leave nothing for them to
2139          * return other than -ENODEV.
2140          */
2141         if (unlikely(atomic_read(&sfp->parentdp->detaching)))
2142                 wake_up_interruptible_all(&sfp->read_wait);
2143
2144         return res;
2145 }
2146
2147 static Sg_fd *
2148 sg_add_sfp(Sg_device * sdp)
2149 {
2150         Sg_fd *sfp;
2151         unsigned long iflags;
2152         int bufflen;
2153
2154         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2155         if (!sfp)
2156                 return ERR_PTR(-ENOMEM);
2157
2158         init_waitqueue_head(&sfp->read_wait);
2159         rwlock_init(&sfp->rq_list_lock);
2160         INIT_LIST_HEAD(&sfp->rq_list);
2161         kref_init(&sfp->f_ref);
2162         mutex_init(&sfp->f_mutex);
2163         sfp->timeout = SG_DEFAULT_TIMEOUT;
2164         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2165         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2166         sfp->cmd_q = SG_DEF_COMMAND_Q;
2167         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2168         sfp->parentdp = sdp;
2169         write_lock_irqsave(&sdp->sfd_lock, iflags);
2170         if (atomic_read(&sdp->detaching)) {
2171                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2172                 kfree(sfp);
2173                 return ERR_PTR(-ENODEV);
2174         }
2175         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2176         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2177         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2178                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2179         if (unlikely(sg_big_buff != def_reserved_size))
2180                 sg_big_buff = def_reserved_size;
2181
2182         bufflen = min_t(int, sg_big_buff,
2183                         max_sectors_bytes(sdp->device->request_queue));
2184         sg_build_reserve(sfp, bufflen);
2185         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2186                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2187                                       sfp->reserve.bufflen,
2188                                       sfp->reserve.k_use_sg));
2189
2190         kref_get(&sdp->d_ref);
2191         __module_get(THIS_MODULE);
2192         return sfp;
2193 }
2194
2195 static void
2196 sg_remove_sfp_usercontext(struct work_struct *work)
2197 {
2198         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2199         struct sg_device *sdp = sfp->parentdp;
2200         Sg_request *srp;
2201         unsigned long iflags;
2202
2203         /* Cleanup any responses which were never read(). */
2204         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2205         while (!list_empty(&sfp->rq_list)) {
2206                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2207                 sg_finish_rem_req(srp);
2208                 list_del(&srp->entry);
2209                 srp->parentfp = NULL;
2210         }
2211         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2212
2213         if (sfp->reserve.bufflen > 0) {
2214                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2215                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2216                                 (int) sfp->reserve.bufflen,
2217                                 (int) sfp->reserve.k_use_sg));
2218                 sg_remove_scat(sfp, &sfp->reserve);
2219         }
2220
2221         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2222                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2223         kfree(sfp);
2224
2225         scsi_device_put(sdp->device);
2226         kref_put(&sdp->d_ref, sg_device_destroy);
2227         module_put(THIS_MODULE);
2228 }
2229
2230 static void
2231 sg_remove_sfp(struct kref *kref)
2232 {
2233         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2234         struct sg_device *sdp = sfp->parentdp;
2235         unsigned long iflags;
2236
2237         write_lock_irqsave(&sdp->sfd_lock, iflags);
2238         list_del(&sfp->sfd_siblings);
2239         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2240
2241         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2242         schedule_work(&sfp->ew.work);
2243 }
2244
2245 #ifdef CONFIG_SCSI_PROC_FS
2246 static int
2247 sg_idr_max_id(int id, void *p, void *data)
2248 {
2249         int *k = data;
2250
2251         if (*k < id)
2252                 *k = id;
2253
2254         return 0;
2255 }
2256
2257 static int
2258 sg_last_dev(void)
2259 {
2260         int k = -1;
2261         unsigned long iflags;
2262
2263         read_lock_irqsave(&sg_index_lock, iflags);
2264         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2265         read_unlock_irqrestore(&sg_index_lock, iflags);
2266         return k + 1;           /* origin 1 */
2267 }
2268 #endif
2269
2270 /* must be called with sg_index_lock held */
2271 static Sg_device *sg_lookup_dev(int dev)
2272 {
2273         return idr_find(&sg_index_idr, dev);
2274 }
2275
2276 static Sg_device *
2277 sg_get_dev(int dev)
2278 {
2279         struct sg_device *sdp;
2280         unsigned long flags;
2281
2282         read_lock_irqsave(&sg_index_lock, flags);
2283         sdp = sg_lookup_dev(dev);
2284         if (!sdp)
2285                 sdp = ERR_PTR(-ENXIO);
2286         else if (atomic_read(&sdp->detaching)) {
2287                 /* If sdp->detaching, then the refcount may already be 0, in
2288                  * which case it would be a bug to do kref_get().
2289                  */
2290                 sdp = ERR_PTR(-ENODEV);
2291         } else
2292                 kref_get(&sdp->d_ref);
2293         read_unlock_irqrestore(&sg_index_lock, flags);
2294
2295         return sdp;
2296 }
2297
2298 #ifdef CONFIG_SCSI_PROC_FS
2299 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2300
2301 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2302 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2303                                   size_t count, loff_t *off);
2304 static const struct proc_ops adio_proc_ops = {
2305         .proc_open      = sg_proc_single_open_adio,
2306         .proc_read      = seq_read,
2307         .proc_lseek     = seq_lseek,
2308         .proc_write     = sg_proc_write_adio,
2309         .proc_release   = single_release,
2310 };
2311
2312 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2313 static ssize_t sg_proc_write_dressz(struct file *filp,
2314                 const char __user *buffer, size_t count, loff_t *off);
2315 static const struct proc_ops dressz_proc_ops = {
2316         .proc_open      = sg_proc_single_open_dressz,
2317         .proc_read      = seq_read,
2318         .proc_lseek     = seq_lseek,
2319         .proc_write     = sg_proc_write_dressz,
2320         .proc_release   = single_release,
2321 };
2322
2323 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2324 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2325 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2326 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2327 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2328 static void dev_seq_stop(struct seq_file *s, void *v);
2329 static const struct seq_operations dev_seq_ops = {
2330         .start = dev_seq_start,
2331         .next  = dev_seq_next,
2332         .stop  = dev_seq_stop,
2333         .show  = sg_proc_seq_show_dev,
2334 };
2335
2336 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2337 static const struct seq_operations devstrs_seq_ops = {
2338         .start = dev_seq_start,
2339         .next  = dev_seq_next,
2340         .stop  = dev_seq_stop,
2341         .show  = sg_proc_seq_show_devstrs,
2342 };
2343
2344 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2345 static const struct seq_operations debug_seq_ops = {
2346         .start = dev_seq_start,
2347         .next  = dev_seq_next,
2348         .stop  = dev_seq_stop,
2349         .show  = sg_proc_seq_show_debug,
2350 };
2351
2352 static int
2353 sg_proc_init(void)
2354 {
2355         struct proc_dir_entry *p;
2356
2357         p = proc_mkdir("scsi/sg", NULL);
2358         if (!p)
2359                 return 1;
2360
2361         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2362         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2363         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2364         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2365         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2366         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2367         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2368         return 0;
2369 }
2370
2371
2372 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2373 {
2374         seq_printf(s, "%d\n", *((int *)s->private));
2375         return 0;
2376 }
2377
2378 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2379 {
2380         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2381 }
2382
2383 static ssize_t
2384 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2385                    size_t count, loff_t *off)
2386 {
2387         int err;
2388         unsigned long num;
2389
2390         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2391                 return -EACCES;
2392         err = kstrtoul_from_user(buffer, count, 0, &num);
2393         if (err)
2394                 return err;
2395         sg_allow_dio = num ? 1 : 0;
2396         return count;
2397 }
2398
2399 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2400 {
2401         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2402 }
2403
2404 static ssize_t
2405 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2406                      size_t count, loff_t *off)
2407 {
2408         int err;
2409         unsigned long k = ULONG_MAX;
2410
2411         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2412                 return -EACCES;
2413
2414         err = kstrtoul_from_user(buffer, count, 0, &k);
2415         if (err)
2416                 return err;
2417         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2418                 sg_big_buff = k;
2419                 return count;
2420         }
2421         return -ERANGE;
2422 }
2423
2424 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2425 {
2426         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2427                    sg_version_date);
2428         return 0;
2429 }
2430
2431 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2432 {
2433         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2434         return 0;
2435 }
2436
2437 struct sg_proc_deviter {
2438         loff_t  index;
2439         size_t  max;
2440 };
2441
2442 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2443 {
2444         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2445
2446         s->private = it;
2447         if (! it)
2448                 return NULL;
2449
2450         it->index = *pos;
2451         it->max = sg_last_dev();
2452         if (it->index >= it->max)
2453                 return NULL;
2454         return it;
2455 }
2456
2457 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2458 {
2459         struct sg_proc_deviter * it = s->private;
2460
2461         *pos = ++it->index;
2462         return (it->index < it->max) ? it : NULL;
2463 }
2464
2465 static void dev_seq_stop(struct seq_file *s, void *v)
2466 {
2467         kfree(s->private);
2468 }
2469
2470 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2471 {
2472         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2473         Sg_device *sdp;
2474         struct scsi_device *scsidp;
2475         unsigned long iflags;
2476
2477         read_lock_irqsave(&sg_index_lock, iflags);
2478         sdp = it ? sg_lookup_dev(it->index) : NULL;
2479         if ((NULL == sdp) || (NULL == sdp->device) ||
2480             (atomic_read(&sdp->detaching)))
2481                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2482         else {
2483                 scsidp = sdp->device;
2484                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2485                               scsidp->host->host_no, scsidp->channel,
2486                               scsidp->id, scsidp->lun, (int) scsidp->type,
2487                               1,
2488                               (int) scsidp->queue_depth,
2489                               (int) scsi_device_busy(scsidp),
2490                               (int) scsi_device_online(scsidp));
2491         }
2492         read_unlock_irqrestore(&sg_index_lock, iflags);
2493         return 0;
2494 }
2495
2496 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2497 {
2498         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2499         Sg_device *sdp;
2500         struct scsi_device *scsidp;
2501         unsigned long iflags;
2502
2503         read_lock_irqsave(&sg_index_lock, iflags);
2504         sdp = it ? sg_lookup_dev(it->index) : NULL;
2505         scsidp = sdp ? sdp->device : NULL;
2506         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2507                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2508                            scsidp->vendor, scsidp->model, scsidp->rev);
2509         else
2510                 seq_puts(s, "<no active device>\n");
2511         read_unlock_irqrestore(&sg_index_lock, iflags);
2512         return 0;
2513 }
2514
2515 /* must be called while holding sg_index_lock */
2516 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2517 {
2518         int k, new_interface, blen, usg;
2519         Sg_request *srp;
2520         Sg_fd *fp;
2521         const sg_io_hdr_t *hp;
2522         const char * cp;
2523         unsigned int ms;
2524
2525         k = 0;
2526         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2527                 k++;
2528                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2529                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2530                            "(res)sgat=%d low_dma=%d\n", k,
2531                            jiffies_to_msecs(fp->timeout),
2532                            fp->reserve.bufflen,
2533                            (int) fp->reserve.k_use_sg, 0);
2534                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2535                            (int) fp->cmd_q, (int) fp->force_packid,
2536                            (int) fp->keep_orphan);
2537                 list_for_each_entry(srp, &fp->rq_list, entry) {
2538                         hp = &srp->header;
2539                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2540                         if (srp->res_used) {
2541                                 if (new_interface &&
2542                                     (SG_FLAG_MMAP_IO & hp->flags))
2543                                         cp = "     mmap>> ";
2544                                 else
2545                                         cp = "     rb>> ";
2546                         } else {
2547                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2548                                         cp = "     dio>> ";
2549                                 else
2550                                         cp = "     ";
2551                         }
2552                         seq_puts(s, cp);
2553                         blen = srp->data.bufflen;
2554                         usg = srp->data.k_use_sg;
2555                         seq_puts(s, srp->done ?
2556                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2557                                   : "act:");
2558                         seq_printf(s, " id=%d blen=%d",
2559                                    srp->header.pack_id, blen);
2560                         if (srp->done)
2561                                 seq_printf(s, " dur=%d", hp->duration);
2562                         else {
2563                                 ms = jiffies_to_msecs(jiffies);
2564                                 seq_printf(s, " t_o/elap=%d/%d",
2565                                         (new_interface ? hp->timeout :
2566                                                   jiffies_to_msecs(fp->timeout)),
2567                                         (ms > hp->duration ? ms - hp->duration : 0));
2568                         }
2569                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2570                                    (int) srp->data.cmd_opcode);
2571                 }
2572                 if (list_empty(&fp->rq_list))
2573                         seq_puts(s, "     No requests active\n");
2574                 read_unlock(&fp->rq_list_lock);
2575         }
2576 }
2577
2578 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2579 {
2580         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2581         Sg_device *sdp;
2582         unsigned long iflags;
2583
2584         if (it && (0 == it->index))
2585                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2586                            (int)it->max, sg_big_buff);
2587
2588         read_lock_irqsave(&sg_index_lock, iflags);
2589         sdp = it ? sg_lookup_dev(it->index) : NULL;
2590         if (NULL == sdp)
2591                 goto skip;
2592         read_lock(&sdp->sfd_lock);
2593         if (!list_empty(&sdp->sfds)) {
2594                 seq_printf(s, " >>> device=%s ", sdp->name);
2595                 if (atomic_read(&sdp->detaching))
2596                         seq_puts(s, "detaching pending close ");
2597                 else if (sdp->device) {
2598                         struct scsi_device *scsidp = sdp->device;
2599
2600                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2601                                    scsidp->host->host_no,
2602                                    scsidp->channel, scsidp->id,
2603                                    scsidp->lun,
2604                                    scsidp->host->hostt->emulated);
2605                 }
2606                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2607                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2608                 sg_proc_debug_helper(s, sdp);
2609         }
2610         read_unlock(&sdp->sfd_lock);
2611 skip:
2612         read_unlock_irqrestore(&sg_index_lock, iflags);
2613         return 0;
2614 }
2615
2616 #endif                          /* CONFIG_SCSI_PROC_FS */
2617
2618 module_init(init_sg);
2619 module_exit(exit_sg);