drivers/scsi/sg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  History:
   4  *  Started: Aug 9 by Lawrence Foard ([email protected]),
   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 ([email protected]), 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 (1 << MINORBITS)
  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 *);
 100 static void sg_remove_device(struct device *);
 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 & FMODE_WRITE))
 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         struct scsi_device *device;
 289         Sg_device *sdp;
 290         Sg_fd *sfp;
 291         int retval;
 292
 293         nonseekable_open(inode, filp);
 294         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 295                 return -EPERM; /* Can't lock it with read only access */
 296         sdp = sg_get_dev(dev);
 297         if (IS_ERR(sdp))
 298                 return PTR_ERR(sdp);
 299
 300         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 301                                       "sg_open: flags=0x%x\n", flags));
 302
 303         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 304         /* Prevent the device driver from vanishing while we sleep */
 305         device = sdp->device;
 306         retval = scsi_device_get(device);
 307         if (retval)
 308                 goto sg_put;
 309
 310         /* scsi_block_when_processing_errors() may block so bypass
 311          * check if O_NONBLOCK. Permits SCSI commands to be issued
 312          * during error recovery. Tread carefully. */
 313         if (!((flags & O_NONBLOCK) ||
 314               scsi_block_when_processing_errors(device))) {
 315                 retval = -ENXIO;
 316                 /* we are in error recovery for this device */
 317                 goto sdp_put;
 318         }
 319
 320         mutex_lock(&sdp->open_rel_lock);
 321         if (flags & O_NONBLOCK) {
 322                 if (flags & O_EXCL) {
 323                         if (sdp->open_cnt > 0) {
 324                                 retval = -EBUSY;
 325                                 goto error_mutex_locked;
 326                         }
 327                 } else {
 328                         if (sdp->exclude) {
 329                                 retval = -EBUSY;
 330                                 goto error_mutex_locked;
 331                         }
 332                 }
 333         } else {
 334                 retval = open_wait(sdp, flags);
 335                 if (retval) /* -ERESTARTSYS or -ENODEV */
 336                         goto error_mutex_locked;
 337         }
 338
 339         /* N.B. at this point we are holding the open_rel_lock */
 340         if (flags & O_EXCL)
 341                 sdp->exclude = true;
 342
 343         if (sdp->open_cnt < 1) {  /* no existing opens */
 344                 sdp->sgdebug = 0;
 345                 q = device->request_queue;
 346                 sdp->sg_tablesize = queue_max_segments(q);
 347         }
 348         sfp = sg_add_sfp(sdp);
 349         if (IS_ERR(sfp)) {
 350                 retval = PTR_ERR(sfp);
 351                 goto out_undo;
 352         }
 353
 354         filp->private_data = sfp;
 355         sdp->open_cnt++;
 356         mutex_unlock(&sdp->open_rel_lock);
 357
 358         retval = 0;
 359 sg_put:
 360         kref_put(&sdp->d_ref, sg_device_destroy);
 361         return retval;
 362
 363 out_undo:
 364         if (flags & O_EXCL) {
 365                 sdp->exclude = false;   /* undo if error */
 366                 wake_up_interruptible(&sdp->open_wait);
 367         }
 368 error_mutex_locked:
 369         mutex_unlock(&sdp->open_rel_lock);
 370 sdp_put:
 371         kref_put(&sdp->d_ref, sg_device_destroy);
 372         scsi_device_put(device);
 373         return retval;
 374 }
 375
 376 /* Release resources associated with a successful sg_open()
 377  * Returns 0 on success, else a negated errno value */
 378 static int
 379 sg_release(struct inode *inode, struct file *filp)
 380 {
 381         Sg_device *sdp;
 382         Sg_fd *sfp;
 383
 384         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 385                 return -ENXIO;
 386         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 387
 388         mutex_lock(&sdp->open_rel_lock);
 389         sdp->open_cnt--;
 390
 391         /* possibly many open()s waiting on exlude clearing, start many;
 392          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 393         if (sdp->exclude) {
 394                 sdp->exclude = false;
 395                 wake_up_interruptible_all(&sdp->open_wait);
 396         } else if (0 == sdp->open_cnt) {
 397                 wake_up_interruptible(&sdp->open_wait);
 398         }
 399         mutex_unlock(&sdp->open_rel_lock);
 400         kref_put(&sfp->f_ref, sg_remove_sfp);
 401         return 0;
 402 }
 403
 404 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 405 {
 406         struct sg_header __user *old_hdr = buf;
 407         int reply_len;
 408
 409         if (count >= SZ_SG_HEADER) {
 410                 /* negative reply_len means v3 format, otherwise v1/v2 */
 411                 if (get_user(reply_len, &old_hdr->reply_len))
 412                         return -EFAULT;
 413
 414                 if (reply_len >= 0)
 415                         return get_user(*pack_id, &old_hdr->pack_id);
 416
 417                 if (in_compat_syscall() &&
 418                     count >= sizeof(struct compat_sg_io_hdr)) {
 419                         struct compat_sg_io_hdr __user *hp = buf;
 420
 421                         return get_user(*pack_id, &hp->pack_id);
 422                 }
 423
 424                 if (count >= sizeof(struct sg_io_hdr)) {
 425                         struct sg_io_hdr __user *hp = buf;
 426
 427                         return get_user(*pack_id, &hp->pack_id);
 428                 }
 429         }
 430
 431         /* no valid header was passed, so ignore the pack_id */
 432         *pack_id = -1;
 433         return 0;
 434 }
 435
 436 static ssize_t
 437 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 438 {
 439         Sg_device *sdp;
 440         Sg_fd *sfp;
 441         Sg_request *srp;
 442         int req_pack_id = -1;
 443         bool busy;
 444         sg_io_hdr_t *hp;
 445         struct sg_header *old_hdr;
 446         int retval;
 447
 448         /*
 449          * This could cause a response to be stranded. Close the associated
 450          * file descriptor to free up any resources being held.
 451          */
 452         retval = sg_check_file_access(filp, __func__);
 453         if (retval)
 454                 return retval;
 455
 456         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 457                 return -ENXIO;
 458         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 459                                       "sg_read: count=%d\n", (int) count));
 460
 461         if (sfp->force_packid)
 462                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 463         if (retval)
 464                 return retval;
 465
 466         srp = sg_get_rq_mark(sfp, req_pack_id, &busy);
 467         if (!srp) {             /* now wait on packet to arrive */
 468                 if (filp->f_flags & O_NONBLOCK)
 469                         return -EAGAIN;
 470                 retval = wait_event_interruptible(sfp->read_wait,
 471                         ((srp = sg_get_rq_mark(sfp, req_pack_id, &busy)) ||
 472                         (!busy && atomic_read(&sdp->detaching))));
 473                 if (!srp)
 474                         /* signal or detaching */
 475                         return retval ? retval : -ENODEV;
 476         }
 477         if (srp->header.interface_id != '\0')
 478                 return sg_new_read(sfp, buf, count, srp);
 479
 480         hp = &srp->header;
 481         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 482         if (!old_hdr)
 483                 return -ENOMEM;
 484
 485         old_hdr->reply_len = (int) hp->timeout;
 486         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 487         old_hdr->pack_id = hp->pack_id;
 488         old_hdr->twelve_byte =
 489             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 490         old_hdr->target_status = hp->masked_status;
 491         old_hdr->host_status = hp->host_status;
 492         old_hdr->driver_status = hp->driver_status;
 493         if ((CHECK_CONDITION & hp->masked_status) ||
 494             (srp->sense_b[0] & 0x70) == 0x70) {
 495                 old_hdr->driver_status = DRIVER_SENSE;
 496                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 497                        sizeof (old_hdr->sense_buffer));
 498         }
 499         switch (hp->host_status) {
 500         /* This setup of 'result' is for backward compatibility and is best
 501            ignored by the user who should use target, host + driver status */
 502         case DID_OK:
 503         case DID_PASSTHROUGH:
 504         case DID_SOFT_ERROR:
 505                 old_hdr->result = 0;
 506                 break;
 507         case DID_NO_CONNECT:
 508         case DID_BUS_BUSY:
 509         case DID_TIME_OUT:
 510                 old_hdr->result = EBUSY;
 511                 break;
 512         case DID_BAD_TARGET:
 513         case DID_ABORT:
 514         case DID_PARITY:
 515         case DID_RESET:
 516         case DID_BAD_INTR:
 517                 old_hdr->result = EIO;
 518                 break;
 519         case DID_ERROR:
 520                 old_hdr->result = (srp->sense_b[0] == 0 &&
 521                                   hp->masked_status == GOOD) ? 0 : EIO;
 522                 break;
 523         default:
 524                 old_hdr->result = EIO;
 525                 break;
 526         }
 527
 528         /* Now copy the result back to the user buffer.  */
 529         if (count >= SZ_SG_HEADER) {
 530                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 531                         retval = -EFAULT;
 532                         goto free_old_hdr;
 533                 }
 534                 buf += SZ_SG_HEADER;
 535                 if (count > old_hdr->reply_len)
 536                         count = old_hdr->reply_len;
 537                 if (count > SZ_SG_HEADER) {
 538                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 539                                 retval = -EFAULT;
 540                                 goto free_old_hdr;
 541                         }
 542                 }
 543         } else
 544                 count = (old_hdr->result == 0) ? 0 : -EIO;
 545         sg_finish_rem_req(srp);
 546         sg_remove_request(sfp, srp);
 547         retval = count;
 548 free_old_hdr:
 549         kfree(old_hdr);
 550         return retval;
 551 }
 552
 553 static ssize_t
 554 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 555 {
 556         sg_io_hdr_t *hp = &srp->header;
 557         int err = 0, err2;
 558         int len;
 559
 560         if (in_compat_syscall()) {
 561                 if (count < sizeof(struct compat_sg_io_hdr)) {
 562                         err = -EINVAL;
 563                         goto err_out;
 564                 }
 565         } else if (count < SZ_SG_IO_HDR) {
 566                 err = -EINVAL;
 567                 goto err_out;
 568         }
 569         hp->sb_len_wr = 0;
 570         if ((hp->mx_sb_len > 0) && hp->sbp) {
 571                 if ((CHECK_CONDITION & hp->masked_status) ||
 572                     (srp->sense_b[0] & 0x70) == 0x70) {
 573                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 574                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 575                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 576                         len = (len > sb_len) ? sb_len : len;
 577                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 578                                 err = -EFAULT;
 579                                 goto err_out;
 580                         }
 581                         hp->driver_status = DRIVER_SENSE;
 582                         hp->sb_len_wr = len;
 583                 }
 584         }
 585         if (hp->masked_status || hp->host_status || hp->driver_status)
 586                 hp->info |= SG_INFO_CHECK;
 587         err = put_sg_io_hdr(hp, buf);
 588 err_out:
 589         err2 = sg_finish_rem_req(srp);
 590         sg_remove_request(sfp, srp);
 591         return err ? : err2 ? : count;
 592 }
 593
 594 static ssize_t
 595 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 596 {
 597         int mxsize, cmd_size, k;
 598         int input_size, blocking;
 599         unsigned char opcode;
 600         Sg_device *sdp;
 601         Sg_fd *sfp;
 602         Sg_request *srp;
 603         struct sg_header old_hdr;
 604         sg_io_hdr_t *hp;
 605         unsigned char cmnd[SG_MAX_CDB_SIZE];
 606         int retval;
 607
 608         retval = sg_check_file_access(filp, __func__);
 609         if (retval)
 610                 return retval;
 611
 612         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 613                 return -ENXIO;
 614         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 615                                       "sg_write: count=%d\n", (int) count));
 616         if (atomic_read(&sdp->detaching))
 617                 return -ENODEV;
 618         if (!((filp->f_flags & O_NONBLOCK) ||
 619               scsi_block_when_processing_errors(sdp->device)))
 620                 return -ENXIO;
 621
 622         if (count < SZ_SG_HEADER)
 623                 return -EIO;
 624         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 625                 return -EFAULT;
 626         blocking = !(filp->f_flags & O_NONBLOCK);
 627         if (old_hdr.reply_len < 0)
 628                 return sg_new_write(sfp, filp, buf, count,
 629                                     blocking, 0, 0, NULL);
 630         if (count < (SZ_SG_HEADER + 6))
 631                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 632
 633         buf += SZ_SG_HEADER;
 634         if (get_user(opcode, buf))
 635                 return -EFAULT;
 636
 637         if (!(srp = sg_add_request(sfp))) {
 638                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 639                                               "sg_write: queue full\n"));
 640                 return -EDOM;
 641         }
 642         mutex_lock(&sfp->f_mutex);
 643         if (sfp->next_cmd_len > 0) {
 644                 cmd_size = sfp->next_cmd_len;
 645                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 646         } else {
 647                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 648                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 649                         cmd_size = 12;
 650         }
 651         mutex_unlock(&sfp->f_mutex);
 652         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 653                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 654 /* Determine buffer size.  */
 655         input_size = count - cmd_size;
 656         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 657         mxsize -= SZ_SG_HEADER;
 658         input_size -= SZ_SG_HEADER;
 659         if (input_size < 0) {
 660                 sg_remove_request(sfp, srp);
 661                 return -EIO;    /* User did not pass enough bytes for this command. */
 662         }
 663         hp = &srp->header;
 664         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 665         hp->cmd_len = (unsigned char) cmd_size;
 666         hp->iovec_count = 0;
 667         hp->mx_sb_len = 0;
 668         if (input_size > 0)
 669                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 670                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 671         else
 672                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 673         hp->dxfer_len = mxsize;
 674         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 675             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 676                 hp->dxferp = (char __user *)buf + cmd_size;
 677         else
 678                 hp->dxferp = NULL;
 679         hp->sbp = NULL;
 680         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 681         hp->flags = input_size; /* structure abuse ... */
 682         hp->pack_id = old_hdr.pack_id;
 683         hp->usr_ptr = NULL;
 684         if (copy_from_user(cmnd, buf, cmd_size)) {
 685                 sg_remove_request(sfp, srp);
 686                 return -EFAULT;
 687         }
 688         /*
 689          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 690          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 691          * is a non-zero input_size, so emit a warning.
 692          */
 693         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 694                 printk_ratelimited(KERN_WARNING
 695                                    "sg_write: data in/out %d/%d bytes "
 696                                    "for SCSI command 0x%x-- guessing "
 697                                    "data in;\n   program %s not setting "
 698                                    "count and/or reply_len properly\n",
 699                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 700                                    input_size, (unsigned int) cmnd[0],
 701                                    current->comm);
 702         }
 703         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 704         return (k < 0) ? k : count;
 705 }
 706
 707 static ssize_t
 708 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 709                  size_t count, int blocking, int read_only, int sg_io_owned,
 710                  Sg_request **o_srp)
 711 {
 712         int k;
 713         Sg_request *srp;
 714         sg_io_hdr_t *hp;
 715         unsigned char cmnd[SG_MAX_CDB_SIZE];
 716         int timeout;
 717         unsigned long ul_timeout;
 718
 719         if (count < SZ_SG_IO_HDR)
 720                 return -EINVAL;
 721
 722         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 723         if (!(srp = sg_add_request(sfp))) {
 724                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 725                                               "sg_new_write: queue full\n"));
 726                 return -EDOM;
 727         }
 728         srp->sg_io_owned = sg_io_owned;
 729         hp = &srp->header;
 730         if (get_sg_io_hdr(hp, buf)) {
 731                 sg_remove_request(sfp, srp);
 732                 return -EFAULT;
 733         }
 734         if (hp->interface_id != 'S') {
 735                 sg_remove_request(sfp, srp);
 736                 return -ENOSYS;
 737         }
 738         if (hp->flags & SG_FLAG_MMAP_IO) {
 739                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 740                         sg_remove_request(sfp, srp);
 741                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 742                 }
 743                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 744                         sg_remove_request(sfp, srp);
 745                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 746                 }
 747                 if (sfp->res_in_use) {
 748                         sg_remove_request(sfp, srp);
 749                         return -EBUSY;  /* reserve buffer already being used */
 750                 }
 751         }
 752         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 753         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 754         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 755                 sg_remove_request(sfp, srp);
 756                 return -EMSGSIZE;
 757         }
 758         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 759                 sg_remove_request(sfp, srp);
 760                 return -EFAULT;
 761         }
 762         if (read_only && sg_allow_access(file, cmnd)) {
 763                 sg_remove_request(sfp, srp);
 764                 return -EPERM;
 765         }
 766         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 767         if (k < 0)
 768                 return k;
 769         if (o_srp)
 770                 *o_srp = srp;
 771         return count;
 772 }
 773
 774 static int
 775 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 776                 unsigned char *cmnd, int timeout, int blocking)
 777 {
 778         int k, at_head;
 779         Sg_device *sdp = sfp->parentdp;
 780         sg_io_hdr_t *hp = &srp->header;
 781
 782         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 783         hp->status = 0;
 784         hp->masked_status = 0;
 785         hp->msg_status = 0;
 786         hp->info = 0;
 787         hp->host_status = 0;
 788         hp->driver_status = 0;
 789         hp->resid = 0;
 790         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 791                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 792                         (int) cmnd[0], (int) hp->cmd_len));
 793
 794         if (hp->dxfer_len >= SZ_256M) {
 795                 sg_remove_request(sfp, srp);
 796                 return -EINVAL;
 797         }
 798
 799         k = sg_start_req(srp, cmnd);
 800         if (k) {
 801                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 802                         "sg_common_write: start_req err=%d\n", k));
 803                 sg_finish_rem_req(srp);
 804                 sg_remove_request(sfp, srp);
 805                 return k;       /* probably out of space --> ENOMEM */
 806         }
 807         if (atomic_read(&sdp->detaching)) {
 808                 if (srp->bio) {
 809                         blk_mq_free_request(srp->rq);
 810                         srp->rq = NULL;
 811                 }
 812
 813                 sg_finish_rem_req(srp);
 814                 sg_remove_request(sfp, srp);
 815                 return -ENODEV;
 816         }
 817
 818         hp->duration = jiffies_to_msecs(jiffies);
 819         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 820             (SG_FLAG_Q_AT_TAIL & hp->flags))
 821                 at_head = 0;
 822         else
 823                 at_head = 1;
 824
 825         srp->rq->timeout = timeout;
 826         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 827         srp->rq->end_io = sg_rq_end_io;
 828         blk_execute_rq_nowait(srp->rq, at_head);
 829         return 0;
 830 }
 831
 832 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 833 {
 834         unsigned long flags;
 835         int ret;
 836
 837         read_lock_irqsave(&sfp->rq_list_lock, flags);
 838         ret = srp->done;
 839         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 840         return ret;
 841 }
 842
 843 static int max_sectors_bytes(struct request_queue *q)
 844 {
 845         unsigned int max_sectors = queue_max_sectors(q);
 846
 847         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 848
 849         return max_sectors << 9;
 850 }
 851
 852 static void
 853 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 854 {
 855         Sg_request *srp;
 856         int val;
 857         unsigned int ms;
 858
 859         val = 0;
 860         list_for_each_entry(srp, &sfp->rq_list, entry) {
 861                 if (val >= SG_MAX_QUEUE)
 862                         break;
 863                 rinfo[val].req_state = srp->done + 1;
 864                 rinfo[val].problem =
 865                         srp->header.masked_status &
 866                         srp->header.host_status &
 867                         srp->header.driver_status;
 868                 if (srp->done)
 869                         rinfo[val].duration =
 870                                 srp->header.duration;
 871                 else {
 872                         ms = jiffies_to_msecs(jiffies);
 873                         rinfo[val].duration =
 874                                 (ms > srp->header.duration) ?
 875                                 (ms - srp->header.duration) : 0;
 876                 }
 877                 rinfo[val].orphan = srp->orphan;
 878                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 879                 rinfo[val].pack_id = srp->header.pack_id;
 880                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 881                 val++;
 882         }
 883 }
 884
 885 #ifdef CONFIG_COMPAT
 886 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 887         char req_state;
 888         char orphan;
 889         char sg_io_owned;
 890         char problem;
 891         int pack_id;
 892         compat_uptr_t usr_ptr;
 893         unsigned int duration;
 894         int unused;
 895 };
 896
 897 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 898                                     struct sg_req_info *rinfo)
 899 {
 900         int i;
 901         for (i = 0; i < SG_MAX_QUEUE; i++) {
 902                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 903                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 904                     put_user(rinfo[i].duration, &o[i].duration) ||
 905                     put_user(rinfo[i].unused, &o[i].unused))
 906                         return -EFAULT;
 907         }
 908         return 0;
 909 }
 910 #endif
 911
 912 static long
 913 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 914                 unsigned int cmd_in, void __user *p)
 915 {
 916         int __user *ip = p;
 917         int result, val, read_only;
 918         Sg_request *srp;
 919         unsigned long iflags;
 920
 921         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 922                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 923         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 924
 925         switch (cmd_in) {
 926         case SG_IO:
 927                 if (atomic_read(&sdp->detaching))
 928                         return -ENODEV;
 929                 if (!scsi_block_when_processing_errors(sdp->device))
 930                         return -ENXIO;
 931                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 932                                  1, read_only, 1, &srp);
 933                 if (result < 0)
 934                         return result;
 935                 result = wait_event_interruptible(sfp->read_wait,
 936                         srp_done(sfp, srp));
 937                 write_lock_irq(&sfp->rq_list_lock);
 938                 if (srp->done) {
 939                         srp->done = 2;
 940                         write_unlock_irq(&sfp->rq_list_lock);
 941                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 942                         return (result < 0) ? result : 0;
 943                 }
 944                 srp->orphan = 1;
 945                 write_unlock_irq(&sfp->rq_list_lock);
 946                 return result;  /* -ERESTARTSYS because signal hit process */
 947         case SG_SET_TIMEOUT:
 948                 result = get_user(val, ip);
 949                 if (result)
 950                         return result;
 951                 if (val < 0)
 952                         return -EIO;
 953                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 954                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 955                                     INT_MAX);
 956                 sfp->timeout_user = val;
 957                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 958
 959                 return 0;
 960         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 961                                 /* strange ..., for backward compatibility */
 962                 return sfp->timeout_user;
 963         case SG_SET_FORCE_LOW_DMA:
 964                 /*
 965                  * N.B. This ioctl never worked properly, but failed to
 966                  * return an error value. So returning '0' to keep compability
 967                  * with legacy applications.
 968                  */
 969                 return 0;
 970         case SG_GET_LOW_DMA:
 971                 return put_user(0, ip);
 972         case SG_GET_SCSI_ID:
 973                 {
 974                         sg_scsi_id_t v;
 975
 976                         if (atomic_read(&sdp->detaching))
 977                                 return -ENODEV;
 978                         memset(&v, 0, sizeof(v));
 979                         v.host_no = sdp->device->host->host_no;
 980                         v.channel = sdp->device->channel;
 981                         v.scsi_id = sdp->device->id;
 982                         v.lun = sdp->device->lun;
 983                         v.scsi_type = sdp->device->type;
 984                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 985                         v.d_queue_depth = sdp->device->queue_depth;
 986                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 987                                 return -EFAULT;
 988                         return 0;
 989                 }
 990         case SG_SET_FORCE_PACK_ID:
 991                 result = get_user(val, ip);
 992                 if (result)
 993                         return result;
 994                 sfp->force_packid = val ? 1 : 0;
 995                 return 0;
 996         case SG_GET_PACK_ID:
 997                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
 998                 list_for_each_entry(srp, &sfp->rq_list, entry) {
 999                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1000                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1001                                                        iflags);
1002                                 return put_user(srp->header.pack_id, ip);
1003                         }
1004                 }
1005                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1006                 return put_user(-1, ip);
1007         case SG_GET_NUM_WAITING:
1008                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1009                 val = 0;
1010                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1011                         if ((1 == srp->done) && (!srp->sg_io_owned))
1012                                 ++val;
1013                 }
1014                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1015                 return put_user(val, ip);
1016         case SG_GET_SG_TABLESIZE:
1017                 return put_user(sdp->sg_tablesize, ip);
1018         case SG_SET_RESERVED_SIZE:
1019                 result = get_user(val, ip);
1020                 if (result)
1021                         return result;
1022                 if (val < 0)
1023                         return -EINVAL;
1024                 val = min_t(int, val,
1025                             max_sectors_bytes(sdp->device->request_queue));
1026                 mutex_lock(&sfp->f_mutex);
1027                 if (val != sfp->reserve.bufflen) {
1028                         if (sfp->mmap_called ||
1029                             sfp->res_in_use) {
1030                                 mutex_unlock(&sfp->f_mutex);
1031                                 return -EBUSY;
1032                         }
1033
1034                         sg_remove_scat(sfp, &sfp->reserve);
1035                         sg_build_reserve(sfp, val);
1036                 }
1037                 mutex_unlock(&sfp->f_mutex);
1038                 return 0;
1039         case SG_GET_RESERVED_SIZE:
1040                 val = min_t(int, sfp->reserve.bufflen,
1041                             max_sectors_bytes(sdp->device->request_queue));
1042                 return put_user(val, ip);
1043         case SG_SET_COMMAND_Q:
1044                 result = get_user(val, ip);
1045                 if (result)
1046                         return result;
1047                 sfp->cmd_q = val ? 1 : 0;
1048                 return 0;
1049         case SG_GET_COMMAND_Q:
1050                 return put_user((int) sfp->cmd_q, ip);
1051         case SG_SET_KEEP_ORPHAN:
1052                 result = get_user(val, ip);
1053                 if (result)
1054                         return result;
1055                 sfp->keep_orphan = val;
1056                 return 0;
1057         case SG_GET_KEEP_ORPHAN:
1058                 return put_user((int) sfp->keep_orphan, ip);
1059         case SG_NEXT_CMD_LEN:
1060                 result = get_user(val, ip);
1061                 if (result)
1062                         return result;
1063                 if (val > SG_MAX_CDB_SIZE)
1064                         return -ENOMEM;
1065                 sfp->next_cmd_len = (val > 0) ? val : 0;
1066                 return 0;
1067         case SG_GET_VERSION_NUM:
1068                 return put_user(sg_version_num, ip);
1069         case SG_GET_ACCESS_COUNT:
1070                 /* faked - we don't have a real access count anymore */
1071                 val = (sdp->device ? 1 : 0);
1072                 return put_user(val, ip);
1073         case SG_GET_REQUEST_TABLE:
1074                 {
1075                         sg_req_info_t *rinfo;
1076
1077                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1078                                         GFP_KERNEL);
1079                         if (!rinfo)
1080                                 return -ENOMEM;
1081                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1082                         sg_fill_request_table(sfp, rinfo);
1083                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1084         #ifdef CONFIG_COMPAT
1085                         if (in_compat_syscall())
1086                                 result = put_compat_request_table(p, rinfo);
1087                         else
1088         #endif
1089                                 result = copy_to_user(p, rinfo,
1090                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1091                         result = result ? -EFAULT : 0;
1092                         kfree(rinfo);
1093                         return result;
1094                 }
1095         case SG_EMULATED_HOST:
1096                 if (atomic_read(&sdp->detaching))
1097                         return -ENODEV;
1098                 return put_user(sdp->device->host->hostt->emulated, ip);
1099         case SCSI_IOCTL_SEND_COMMAND:
1100                 if (atomic_read(&sdp->detaching))
1101                         return -ENODEV;
1102                 return scsi_ioctl(sdp->device, filp->f_mode & FMODE_WRITE,
1103                                   cmd_in, p);
1104         case SG_SET_DEBUG:
1105                 result = get_user(val, ip);
1106                 if (result)
1107                         return result;
1108                 sdp->sgdebug = (char) val;
1109                 return 0;
1110         case BLKSECTGET:
1111                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1112                                 ip);
1113         case BLKTRACESETUP:
1114                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1115                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1116                                        NULL, p);
1117         case BLKTRACESTART:
1118                 return blk_trace_startstop(sdp->device->request_queue, 1);
1119         case BLKTRACESTOP:
1120                 return blk_trace_startstop(sdp->device->request_queue, 0);
1121         case BLKTRACETEARDOWN:
1122                 return blk_trace_remove(sdp->device->request_queue);
1123         case SCSI_IOCTL_GET_IDLUN:
1124         case SCSI_IOCTL_GET_BUS_NUMBER:
1125         case SCSI_IOCTL_PROBE_HOST:
1126         case SG_GET_TRANSFORM:
1127         case SG_SCSI_RESET:
1128                 if (atomic_read(&sdp->detaching))
1129                         return -ENODEV;
1130                 break;
1131         default:
1132                 if (read_only)
1133                         return -EPERM;  /* don't know so take safe approach */
1134                 break;
1135         }
1136
1137         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1138                         cmd_in, filp->f_flags & O_NDELAY);
1139         if (result)
1140                 return result;
1141
1142         return -ENOIOCTLCMD;
1143 }
1144
1145 static long
1146 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1147 {
1148         void __user *p = (void __user *)arg;
1149         Sg_device *sdp;
1150         Sg_fd *sfp;
1151         int ret;
1152
1153         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1154                 return -ENXIO;
1155
1156         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1157         if (ret != -ENOIOCTLCMD)
1158                 return ret;
1159         return scsi_ioctl(sdp->device, filp->f_mode & FMODE_WRITE, cmd_in, p);
1160 }
1161
1162 static __poll_t
1163 sg_poll(struct file *filp, poll_table * wait)
1164 {
1165         __poll_t res = 0;
1166         Sg_device *sdp;
1167         Sg_fd *sfp;
1168         Sg_request *srp;
1169         int count = 0;
1170         unsigned long iflags;
1171
1172         sfp = filp->private_data;
1173         if (!sfp)
1174                 return EPOLLERR;
1175         sdp = sfp->parentdp;
1176         if (!sdp)
1177                 return EPOLLERR;
1178         poll_wait(filp, &sfp->read_wait, wait);
1179         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1180         list_for_each_entry(srp, &sfp->rq_list, entry) {
1181                 /* if any read waiting, flag it */
1182                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1183                         res = EPOLLIN | EPOLLRDNORM;
1184                 ++count;
1185         }
1186         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1187
1188         if (atomic_read(&sdp->detaching))
1189                 res |= EPOLLHUP;
1190         else if (!sfp->cmd_q) {
1191                 if (0 == count)
1192                         res |= EPOLLOUT | EPOLLWRNORM;
1193         } else if (count < SG_MAX_QUEUE)
1194                 res |= EPOLLOUT | EPOLLWRNORM;
1195         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1196                                       "sg_poll: res=0x%x\n", (__force u32) res));
1197         return res;
1198 }
1199
1200 static int
1201 sg_fasync(int fd, struct file *filp, int mode)
1202 {
1203         Sg_device *sdp;
1204         Sg_fd *sfp;
1205
1206         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1207                 return -ENXIO;
1208         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1209                                       "sg_fasync: mode=%d\n", mode));
1210
1211         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1212 }
1213
1214 static vm_fault_t
1215 sg_vma_fault(struct vm_fault *vmf)
1216 {
1217         struct vm_area_struct *vma = vmf->vma;
1218         Sg_fd *sfp;
1219         unsigned long offset, len, sa;
1220         Sg_scatter_hold *rsv_schp;
1221         int k, length;
1222
1223         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1224                 return VM_FAULT_SIGBUS;
1225         rsv_schp = &sfp->reserve;
1226         offset = vmf->pgoff << PAGE_SHIFT;
1227         if (offset >= rsv_schp->bufflen)
1228                 return VM_FAULT_SIGBUS;
1229         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1230                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1231                                       offset, rsv_schp->k_use_sg));
1232         sa = vma->vm_start;
1233         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1234         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1235                 len = vma->vm_end - sa;
1236                 len = (len < length) ? len : length;
1237                 if (offset < len) {
1238                         struct page *page = nth_page(rsv_schp->pages[k],
1239                                                      offset >> PAGE_SHIFT);
1240                         get_page(page); /* increment page count */
1241                         vmf->page = page;
1242                         return 0; /* success */
1243                 }
1244                 sa += len;
1245                 offset -= len;
1246         }
1247
1248         return VM_FAULT_SIGBUS;
1249 }
1250
1251 static const struct vm_operations_struct sg_mmap_vm_ops = {
1252         .fault = sg_vma_fault,
1253 };
1254
1255 static int
1256 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1257 {
1258         Sg_fd *sfp;
1259         unsigned long req_sz, len, sa;
1260         Sg_scatter_hold *rsv_schp;
1261         int k, length;
1262         int ret = 0;
1263
1264         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1265                 return -ENXIO;
1266         req_sz = vma->vm_end - vma->vm_start;
1267         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1268                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1269                                       (void *) vma->vm_start, (int) req_sz));
1270         if (vma->vm_pgoff)
1271                 return -EINVAL; /* want no offset */
1272         rsv_schp = &sfp->reserve;
1273         mutex_lock(&sfp->f_mutex);
1274         if (req_sz > rsv_schp->bufflen) {
1275                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1276                 goto out;
1277         }
1278
1279         sa = vma->vm_start;
1280         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1281         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1282                 len = vma->vm_end - sa;
1283                 len = (len < length) ? len : length;
1284                 sa += len;
1285         }
1286
1287         sfp->mmap_called = 1;
1288         vm_flags_set(vma, VM_IO | VM_DONTEXPAND | VM_DONTDUMP);
1289         vma->vm_private_data = sfp;
1290         vma->vm_ops = &sg_mmap_vm_ops;
1291 out:
1292         mutex_unlock(&sfp->f_mutex);
1293         return ret;
1294 }
1295
1296 static void
1297 sg_rq_end_io_usercontext(struct work_struct *work)
1298 {
1299         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1300         struct sg_fd *sfp = srp->parentfp;
1301
1302         sg_finish_rem_req(srp);
1303         sg_remove_request(sfp, srp);
1304         kref_put(&sfp->f_ref, sg_remove_sfp);
1305 }
1306
1307 /*
1308  * This function is a "bottom half" handler that is called by the mid
1309  * level when a command is completed (or has failed).
1310  */
1311 static enum rq_end_io_ret
1312 sg_rq_end_io(struct request *rq, blk_status_t status)
1313 {
1314         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq);
1315         struct sg_request *srp = rq->end_io_data;
1316         Sg_device *sdp;
1317         Sg_fd *sfp;
1318         unsigned long iflags;
1319         unsigned int ms;
1320         char *sense;
1321         int result, resid, done = 1;
1322
1323         if (WARN_ON(srp->done != 0))
1324                 return RQ_END_IO_NONE;
1325
1326         sfp = srp->parentfp;
1327         if (WARN_ON(sfp == NULL))
1328                 return RQ_END_IO_NONE;
1329
1330         sdp = sfp->parentdp;
1331         if (unlikely(atomic_read(&sdp->detaching)))
1332                 pr_info("%s: device detaching\n", __func__);
1333
1334         sense = scmd->sense_buffer;
1335         result = scmd->result;
1336         resid = scmd->resid_len;
1337
1338         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1339                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1340                                       srp->header.pack_id, result));
1341         srp->header.resid = resid;
1342         ms = jiffies_to_msecs(jiffies);
1343         srp->header.duration = (ms > srp->header.duration) ?
1344                                 (ms - srp->header.duration) : 0;
1345         if (0 != result) {
1346                 struct scsi_sense_hdr sshdr;
1347
1348                 srp->header.status = 0xff & result;
1349                 srp->header.masked_status = sg_status_byte(result);
1350                 srp->header.msg_status = COMMAND_COMPLETE;
1351                 srp->header.host_status = host_byte(result);
1352                 srp->header.driver_status = driver_byte(result);
1353                 if ((sdp->sgdebug > 0) &&
1354                     ((CHECK_CONDITION == srp->header.masked_status) ||
1355                      (COMMAND_TERMINATED == srp->header.masked_status)))
1356                         __scsi_print_sense(sdp->device, __func__, sense,
1357                                            SCSI_SENSE_BUFFERSIZE);
1358
1359                 /* Following if statement is a patch supplied by Eric Youngdale */
1360                 if (driver_byte(result) != 0
1361                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1362                     && !scsi_sense_is_deferred(&sshdr)
1363                     && sshdr.sense_key == UNIT_ATTENTION
1364                     && sdp->device->removable) {
1365                         /* Detected possible disc change. Set the bit - this */
1366                         /* may be used if there are filesystems using this device */
1367                         sdp->device->changed = 1;
1368                 }
1369         }
1370
1371         if (scmd->sense_len)
1372                 memcpy(srp->sense_b, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
1373
1374         /* Rely on write phase to clean out srp status values, so no "else" */
1375
1376         /*
1377          * Free the request as soon as it is complete so that its resources
1378          * can be reused without waiting for userspace to read() the
1379          * result.  But keep the associated bio (if any) around until
1380          * blk_rq_unmap_user() can be called from user context.
1381          */
1382         srp->rq = NULL;
1383         blk_mq_free_request(rq);
1384
1385         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1386         if (unlikely(srp->orphan)) {
1387                 if (sfp->keep_orphan)
1388                         srp->sg_io_owned = 0;
1389                 else
1390                         done = 0;
1391         }
1392         srp->done = done;
1393         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1394
1395         if (likely(done)) {
1396                 /* Now wake up any sg_read() that is waiting for this
1397                  * packet.
1398                  */
1399                 wake_up_interruptible(&sfp->read_wait);
1400                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1401                 kref_put(&sfp->f_ref, sg_remove_sfp);
1402         } else {
1403                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1404                 schedule_work(&srp->ew.work);
1405         }
1406         return RQ_END_IO_NONE;
1407 }
1408
1409 static const struct file_operations sg_fops = {
1410         .owner = THIS_MODULE,
1411         .read = sg_read,
1412         .write = sg_write,
1413         .poll = sg_poll,
1414         .unlocked_ioctl = sg_ioctl,
1415         .compat_ioctl = compat_ptr_ioctl,
1416         .open = sg_open,
1417         .mmap = sg_mmap,
1418         .release = sg_release,
1419         .fasync = sg_fasync,
1420 };
1421
1422 static const struct class sg_sysfs_class = {
1423         .name = "scsi_generic"
1424 };
1425
1426 static int sg_sysfs_valid = 0;
1427
1428 static Sg_device *
1429 sg_alloc(struct scsi_device *scsidp)
1430 {
1431         struct request_queue *q = scsidp->request_queue;
1432         Sg_device *sdp;
1433         unsigned long iflags;
1434         int error;
1435         u32 k;
1436
1437         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1438         if (!sdp) {
1439                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1440                             "failure\n", __func__);
1441                 return ERR_PTR(-ENOMEM);
1442         }
1443
1444         idr_preload(GFP_KERNEL);
1445         write_lock_irqsave(&sg_index_lock, iflags);
1446
1447         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1448         if (error < 0) {
1449                 if (error == -ENOSPC) {
1450                         sdev_printk(KERN_WARNING, scsidp,
1451                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1452                                     scsidp->type, SG_MAX_DEVS - 1);
1453                         error = -ENODEV;
1454                 } else {
1455                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1456                                     "allocation Sg_device failure: %d\n",
1457                                     __func__, error);
1458                 }
1459                 goto out_unlock;
1460         }
1461         k = error;
1462
1463         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1464                                         "sg_alloc: dev=%d \n", k));
1465         sprintf(sdp->name, "sg%d", k);
1466         sdp->device = scsidp;
1467         mutex_init(&sdp->open_rel_lock);
1468         INIT_LIST_HEAD(&sdp->sfds);
1469         init_waitqueue_head(&sdp->open_wait);
1470         atomic_set(&sdp->detaching, 0);
1471         rwlock_init(&sdp->sfd_lock);
1472         sdp->sg_tablesize = queue_max_segments(q);
1473         sdp->index = k;
1474         kref_init(&sdp->d_ref);
1475         error = 0;
1476
1477 out_unlock:
1478         write_unlock_irqrestore(&sg_index_lock, iflags);
1479         idr_preload_end();
1480
1481         if (error) {
1482                 kfree(sdp);
1483                 return ERR_PTR(error);
1484         }
1485         return sdp;
1486 }
1487
1488 static int
1489 sg_add_device(struct device *cl_dev)
1490 {
1491         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1492         Sg_device *sdp = NULL;
1493         struct cdev * cdev = NULL;
1494         int error;
1495         unsigned long iflags;
1496
1497         if (!blk_get_queue(scsidp->request_queue)) {
1498                 pr_warn("%s: get scsi_device queue failed\n", __func__);
1499                 return -ENODEV;
1500         }
1501
1502         error = -ENOMEM;
1503         cdev = cdev_alloc();
1504         if (!cdev) {
1505                 pr_warn("%s: cdev_alloc failed\n", __func__);
1506                 goto out;
1507         }
1508         cdev->owner = THIS_MODULE;
1509         cdev->ops = &sg_fops;
1510
1511         sdp = sg_alloc(scsidp);
1512         if (IS_ERR(sdp)) {
1513                 pr_warn("%s: sg_alloc failed\n", __func__);
1514                 error = PTR_ERR(sdp);
1515                 goto out;
1516         }
1517
1518         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1519         if (error)
1520                 goto cdev_add_err;
1521
1522         sdp->cdev = cdev;
1523         if (sg_sysfs_valid) {
1524                 struct device *sg_class_member;
1525
1526                 sg_class_member = device_create(&sg_sysfs_class, cl_dev->parent,
1527                                                 MKDEV(SCSI_GENERIC_MAJOR,
1528                                                       sdp->index),
1529                                                 sdp, "%s", sdp->name);
1530                 if (IS_ERR(sg_class_member)) {
1531                         pr_err("%s: device_create failed\n", __func__);
1532                         error = PTR_ERR(sg_class_member);
1533                         goto cdev_add_err;
1534                 }
1535                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1536                                           &sg_class_member->kobj, "generic");
1537                 if (error)
1538                         pr_err("%s: unable to make symlink 'generic' back "
1539                                "to sg%d\n", __func__, sdp->index);
1540         } else
1541                 pr_warn("%s: sg_sys Invalid\n", __func__);
1542
1543         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1544                     "type %d\n", sdp->index, scsidp->type);
1545
1546         dev_set_drvdata(cl_dev, sdp);
1547
1548         return 0;
1549
1550 cdev_add_err:
1551         write_lock_irqsave(&sg_index_lock, iflags);
1552         idr_remove(&sg_index_idr, sdp->index);
1553         write_unlock_irqrestore(&sg_index_lock, iflags);
1554         kfree(sdp);
1555
1556 out:
1557         if (cdev)
1558                 cdev_del(cdev);
1559         blk_put_queue(scsidp->request_queue);
1560         return error;
1561 }
1562
1563 static void
1564 sg_device_destroy(struct kref *kref)
1565 {
1566         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1567         struct request_queue *q = sdp->device->request_queue;
1568         unsigned long flags;
1569
1570         /* CAUTION!  Note that the device can still be found via idr_find()
1571          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1572          * any other cleanup.
1573          */
1574
1575         blk_trace_remove(q);
1576         blk_put_queue(q);
1577
1578         write_lock_irqsave(&sg_index_lock, flags);
1579         idr_remove(&sg_index_idr, sdp->index);
1580         write_unlock_irqrestore(&sg_index_lock, flags);
1581
1582         SCSI_LOG_TIMEOUT(3,
1583                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1584
1585         kfree(sdp);
1586 }
1587
1588 static void
1589 sg_remove_device(struct device *cl_dev)
1590 {
1591         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1592         Sg_device *sdp = dev_get_drvdata(cl_dev);
1593         unsigned long iflags;
1594         Sg_fd *sfp;
1595         int val;
1596
1597         if (!sdp)
1598                 return;
1599         /* want sdp->detaching non-zero as soon as possible */
1600         val = atomic_inc_return(&sdp->detaching);
1601         if (val > 1)
1602                 return; /* only want to do following once per device */
1603
1604         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1605                                       "%s\n", __func__));
1606
1607         read_lock_irqsave(&sdp->sfd_lock, iflags);
1608         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1609                 wake_up_interruptible_all(&sfp->read_wait);
1610                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1611         }
1612         wake_up_interruptible_all(&sdp->open_wait);
1613         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1614
1615         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1616         device_destroy(&sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1617         cdev_del(sdp->cdev);
1618         sdp->cdev = NULL;
1619
1620         kref_put(&sdp->d_ref, sg_device_destroy);
1621 }
1622
1623 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1624 module_param_named(def_reserved_size, def_reserved_size, int,
1625                    S_IRUGO | S_IWUSR);
1626 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1627
1628 MODULE_AUTHOR("Douglas Gilbert");
1629 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1630 MODULE_LICENSE("GPL");
1631 MODULE_VERSION(SG_VERSION_STR);
1632 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1633
1634 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1635                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1636 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1637 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1638
1639 #ifdef CONFIG_SYSCTL
1640 #include <linux/sysctl.h>
1641
1642 static struct ctl_table sg_sysctls[] = {
1643         {
1644                 .procname       = "sg-big-buff",
1645                 .data           = &sg_big_buff,
1646                 .maxlen         = sizeof(int),
1647                 .mode           = 0444,
1648                 .proc_handler   = proc_dointvec,
1649         },
1650 };
1651
1652 static struct ctl_table_header *hdr;
1653 static void register_sg_sysctls(void)
1654 {
1655         if (!hdr)
1656                 hdr = register_sysctl("kernel", sg_sysctls);
1657 }
1658
1659 static void unregister_sg_sysctls(void)
1660 {
1661         if (hdr)
1662                 unregister_sysctl_table(hdr);
1663 }
1664 #else
1665 #define register_sg_sysctls() do { } while (0)
1666 #define unregister_sg_sysctls() do { } while (0)
1667 #endif /* CONFIG_SYSCTL */
1668
1669 static int __init
1670 init_sg(void)
1671 {
1672         int rc;
1673
1674         if (scatter_elem_sz < PAGE_SIZE) {
1675                 scatter_elem_sz = PAGE_SIZE;
1676                 scatter_elem_sz_prev = scatter_elem_sz;
1677         }
1678         if (def_reserved_size >= 0)
1679                 sg_big_buff = def_reserved_size;
1680         else
1681                 def_reserved_size = sg_big_buff;
1682
1683         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1684                                     SG_MAX_DEVS, "sg");
1685         if (rc)
1686                 return rc;
1687         rc = class_register(&sg_sysfs_class);
1688         if (rc)
1689                 goto err_out;
1690         sg_sysfs_valid = 1;
1691         rc = scsi_register_interface(&sg_interface);
1692         if (0 == rc) {
1693 #ifdef CONFIG_SCSI_PROC_FS
1694                 sg_proc_init();
1695 #endif                          /* CONFIG_SCSI_PROC_FS */
1696                 return 0;
1697         }
1698         class_unregister(&sg_sysfs_class);
1699         register_sg_sysctls();
1700 err_out:
1701         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1702         return rc;
1703 }
1704
1705 static void __exit
1706 exit_sg(void)
1707 {
1708         unregister_sg_sysctls();
1709 #ifdef CONFIG_SCSI_PROC_FS
1710         remove_proc_subtree("scsi/sg", NULL);
1711 #endif                          /* CONFIG_SCSI_PROC_FS */
1712         scsi_unregister_interface(&sg_interface);
1713         class_unregister(&sg_sysfs_class);
1714         sg_sysfs_valid = 0;
1715         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1716                                  SG_MAX_DEVS);
1717         idr_destroy(&sg_index_idr);
1718 }
1719
1720 static int
1721 sg_start_req(Sg_request *srp, unsigned char *cmd)
1722 {
1723         int res;
1724         struct request *rq;
1725         Sg_fd *sfp = srp->parentfp;
1726         sg_io_hdr_t *hp = &srp->header;
1727         int dxfer_len = (int) hp->dxfer_len;
1728         int dxfer_dir = hp->dxfer_direction;
1729         unsigned int iov_count = hp->iovec_count;
1730         Sg_scatter_hold *req_schp = &srp->data;
1731         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1732         struct request_queue *q = sfp->parentdp->device->request_queue;
1733         struct rq_map_data *md, map_data;
1734         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? ITER_SOURCE : ITER_DEST;
1735         struct scsi_cmnd *scmd;
1736
1737         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1738                                       "sg_start_req: dxfer_len=%d\n",
1739                                       dxfer_len));
1740
1741         /*
1742          * NOTE
1743          *
1744          * With scsi-mq enabled, there are a fixed number of preallocated
1745          * requests equal in number to shost->can_queue.  If all of the
1746          * preallocated requests are already in use, then scsi_alloc_request()
1747          * will sleep until an active command completes, freeing up a request.
1748          * Although waiting in an asynchronous interface is less than ideal, we
1749          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1750          * not expect an EWOULDBLOCK from this condition.
1751          */
1752         rq = scsi_alloc_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1753                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1754         if (IS_ERR(rq))
1755                 return PTR_ERR(rq);
1756         scmd = blk_mq_rq_to_pdu(rq);
1757
1758         if (hp->cmd_len > sizeof(scmd->cmnd)) {
1759                 blk_mq_free_request(rq);
1760                 return -EINVAL;
1761         }
1762
1763         memcpy(scmd->cmnd, cmd, hp->cmd_len);
1764         scmd->cmd_len = hp->cmd_len;
1765
1766         srp->rq = rq;
1767         rq->end_io_data = srp;
1768         scmd->allowed = SG_DEFAULT_RETRIES;
1769
1770         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1771                 return 0;
1772
1773         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1774             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1775             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1776                 md = NULL;
1777         else
1778                 md = &map_data;
1779
1780         if (md) {
1781                 mutex_lock(&sfp->f_mutex);
1782                 if (dxfer_len <= rsv_schp->bufflen &&
1783                     !sfp->res_in_use) {
1784                         sfp->res_in_use = 1;
1785                         sg_link_reserve(sfp, srp, dxfer_len);
1786                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1787                         res = -EBUSY; /* sfp->res_in_use == 1 */
1788                         if (dxfer_len > rsv_schp->bufflen)
1789                                 res = -ENOMEM;
1790                         mutex_unlock(&sfp->f_mutex);
1791                         return res;
1792                 } else {
1793                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1794                         if (res) {
1795                                 mutex_unlock(&sfp->f_mutex);
1796                                 return res;
1797                         }
1798                 }
1799                 mutex_unlock(&sfp->f_mutex);
1800
1801                 md->pages = req_schp->pages;
1802                 md->page_order = req_schp->page_order;
1803                 md->nr_entries = req_schp->k_use_sg;
1804                 md->offset = 0;
1805                 md->null_mapped = hp->dxferp ? 0 : 1;
1806                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1807                         md->from_user = 1;
1808                 else
1809                         md->from_user = 0;
1810         }
1811
1812         res = blk_rq_map_user_io(rq, md, hp->dxferp, hp->dxfer_len,
1813                         GFP_ATOMIC, iov_count, iov_count, 1, rw);
1814         if (!res) {
1815                 srp->bio = rq->bio;
1816
1817                 if (!md) {
1818                         req_schp->dio_in_use = 1;
1819                         hp->info |= SG_INFO_DIRECT_IO;
1820                 }
1821         }
1822         return res;
1823 }
1824
1825 static int
1826 sg_finish_rem_req(Sg_request *srp)
1827 {
1828         int ret = 0;
1829
1830         Sg_fd *sfp = srp->parentfp;
1831         Sg_scatter_hold *req_schp = &srp->data;
1832
1833         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1834                                       "sg_finish_rem_req: res_used=%d\n",
1835                                       (int) srp->res_used));
1836         if (srp->bio)
1837                 ret = blk_rq_unmap_user(srp->bio);
1838
1839         if (srp->rq)
1840                 blk_mq_free_request(srp->rq);
1841
1842         if (srp->res_used)
1843                 sg_unlink_reserve(sfp, srp);
1844         else
1845                 sg_remove_scat(sfp, req_schp);
1846
1847         return ret;
1848 }
1849
1850 static int
1851 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1852 {
1853         int sg_bufflen = tablesize * sizeof(struct page *);
1854         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1855
1856         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1857         if (!schp->pages)
1858                 return -ENOMEM;
1859         schp->sglist_len = sg_bufflen;
1860         return tablesize;       /* number of scat_gath elements allocated */
1861 }
1862
1863 static int
1864 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1865 {
1866         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1867         int sg_tablesize = sfp->parentdp->sg_tablesize;
1868         int blk_size = buff_size, order;
1869         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1870
1871         if (blk_size < 0)
1872                 return -EFAULT;
1873         if (0 == blk_size)
1874                 ++blk_size;     /* don't know why */
1875         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1876         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1877         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1878                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1879                 buff_size, blk_size));
1880
1881         /* N.B. ret_sz carried into this block ... */
1882         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1883         if (mx_sc_elems < 0)
1884                 return mx_sc_elems;     /* most likely -ENOMEM */
1885
1886         num = scatter_elem_sz;
1887         if (unlikely(num != scatter_elem_sz_prev)) {
1888                 if (num < PAGE_SIZE) {
1889                         scatter_elem_sz = PAGE_SIZE;
1890                         scatter_elem_sz_prev = PAGE_SIZE;
1891                 } else
1892                         scatter_elem_sz_prev = num;
1893         }
1894
1895         order = get_order(num);
1896 retry:
1897         ret_sz = 1 << (PAGE_SHIFT + order);
1898
1899         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1900              k++, rem_sz -= ret_sz) {
1901
1902                 num = (rem_sz > scatter_elem_sz_prev) ?
1903                         scatter_elem_sz_prev : rem_sz;
1904
1905                 schp->pages[k] = alloc_pages(gfp_mask, order);
1906                 if (!schp->pages[k])
1907                         goto out;
1908
1909                 if (num == scatter_elem_sz_prev) {
1910                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1911                                 scatter_elem_sz = ret_sz;
1912                                 scatter_elem_sz_prev = ret_sz;
1913                         }
1914                 }
1915
1916                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1917                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1918                                  k, num, ret_sz));
1919         }               /* end of for loop */
1920
1921         schp->page_order = order;
1922         schp->k_use_sg = k;
1923         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1924                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1925                          k, rem_sz));
1926
1927         schp->bufflen = blk_size;
1928         if (rem_sz > 0) /* must have failed */
1929                 return -ENOMEM;
1930         return 0;
1931 out:
1932         for (i = 0; i < k; i++)
1933                 __free_pages(schp->pages[i], order);
1934
1935         if (--order >= 0)
1936                 goto retry;
1937
1938         return -ENOMEM;
1939 }
1940
1941 static void
1942 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1943 {
1944         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1945                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1946         if (schp->pages && schp->sglist_len > 0) {
1947                 if (!schp->dio_in_use) {
1948                         int k;
1949
1950                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1951                                 SCSI_LOG_TIMEOUT(5,
1952                                         sg_printk(KERN_INFO, sfp->parentdp,
1953                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1954                                         k, schp->pages[k]));
1955                                 __free_pages(schp->pages[k], schp->page_order);
1956                         }
1957
1958                         kfree(schp->pages);
1959                 }
1960         }
1961         memset(schp, 0, sizeof (*schp));
1962 }
1963
1964 static int
1965 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1966 {
1967         Sg_scatter_hold *schp = &srp->data;
1968         int k, num;
1969
1970         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1971                          "sg_read_oxfer: num_read_xfer=%d\n",
1972                          num_read_xfer));
1973         if ((!outp) || (num_read_xfer <= 0))
1974                 return 0;
1975
1976         num = 1 << (PAGE_SHIFT + schp->page_order);
1977         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1978                 if (num > num_read_xfer) {
1979                         if (copy_to_user(outp, page_address(schp->pages[k]),
1980                                            num_read_xfer))
1981                                 return -EFAULT;
1982                         break;
1983                 } else {
1984                         if (copy_to_user(outp, page_address(schp->pages[k]),
1985                                            num))
1986                                 return -EFAULT;
1987                         num_read_xfer -= num;
1988                         if (num_read_xfer <= 0)
1989                                 break;
1990                         outp += num;
1991                 }
1992         }
1993
1994         return 0;
1995 }
1996
1997 static void
1998 sg_build_reserve(Sg_fd * sfp, int req_size)
1999 {
2000         Sg_scatter_hold *schp = &sfp->reserve;
2001
2002         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2003                          "sg_build_reserve: req_size=%d\n", req_size));
2004         do {
2005                 if (req_size < PAGE_SIZE)
2006                         req_size = PAGE_SIZE;
2007                 if (0 == sg_build_indirect(schp, sfp, req_size))
2008                         return;
2009                 else
2010                         sg_remove_scat(sfp, schp);
2011                 req_size >>= 1; /* divide by 2 */
2012         } while (req_size > (PAGE_SIZE / 2));
2013 }
2014
2015 static void
2016 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2017 {
2018         Sg_scatter_hold *req_schp = &srp->data;
2019         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2020         int k, num, rem;
2021
2022         srp->res_used = 1;
2023         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2024                          "sg_link_reserve: size=%d\n", size));
2025         rem = size;
2026
2027         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2028         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2029                 if (rem <= num) {
2030                         req_schp->k_use_sg = k + 1;
2031                         req_schp->sglist_len = rsv_schp->sglist_len;
2032                         req_schp->pages = rsv_schp->pages;
2033
2034                         req_schp->bufflen = size;
2035                         req_schp->page_order = rsv_schp->page_order;
2036                         break;
2037                 } else
2038                         rem -= num;
2039         }
2040
2041         if (k >= rsv_schp->k_use_sg)
2042                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2043                                  "sg_link_reserve: BAD size\n"));
2044 }
2045
2046 static void
2047 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2048 {
2049         Sg_scatter_hold *req_schp = &srp->data;
2050
2051         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2052                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2053                                       (int) req_schp->k_use_sg));
2054         req_schp->k_use_sg = 0;
2055         req_schp->bufflen = 0;
2056         req_schp->pages = NULL;
2057         req_schp->page_order = 0;
2058         req_schp->sglist_len = 0;
2059         srp->res_used = 0;
2060         /* Called without mutex lock to avoid deadlock */
2061         sfp->res_in_use = 0;
2062 }
2063
2064 static Sg_request *
2065 sg_get_rq_mark(Sg_fd * sfp, int pack_id, bool *busy)
2066 {
2067         Sg_request *resp;
2068         unsigned long iflags;
2069
2070         *busy = false;
2071         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2072         list_for_each_entry(resp, &sfp->rq_list, entry) {
2073                 /* look for requests that are not SG_IO owned */
2074                 if ((!resp->sg_io_owned) &&
2075                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2076                         switch (resp->done) {
2077                         case 0: /* request active */
2078                                 *busy = true;
2079                                 break;
2080                         case 1: /* request done; response ready to return */
2081                                 resp->done = 2; /* guard against other readers */
2082                                 write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2083                                 return resp;
2084                         case 2: /* response already being returned */
2085                                 break;
2086                         }
2087                 }
2088         }
2089         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2090         return NULL;
2091 }
2092
2093 /* always adds to end of list */
2094 static Sg_request *
2095 sg_add_request(Sg_fd * sfp)
2096 {
2097         int k;
2098         unsigned long iflags;
2099         Sg_request *rp = sfp->req_arr;
2100
2101         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2102         if (!list_empty(&sfp->rq_list)) {
2103                 if (!sfp->cmd_q)
2104                         goto out_unlock;
2105
2106                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2107                         if (!rp->parentfp)
2108                                 break;
2109                 }
2110                 if (k >= SG_MAX_QUEUE)
2111                         goto out_unlock;
2112         }
2113         memset(rp, 0, sizeof (Sg_request));
2114         rp->parentfp = sfp;
2115         rp->header.duration = jiffies_to_msecs(jiffies);
2116         list_add_tail(&rp->entry, &sfp->rq_list);
2117         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2118         return rp;
2119 out_unlock:
2120         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2121         return NULL;
2122 }
2123
2124 /* Return of 1 for found; 0 for not found */
2125 static int
2126 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2127 {
2128         unsigned long iflags;
2129         int res = 0;
2130
2131         if (!sfp || !srp || list_empty(&sfp->rq_list))
2132                 return res;
2133         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2134         if (!list_empty(&srp->entry)) {
2135                 list_del(&srp->entry);
2136                 srp->parentfp = NULL;
2137                 res = 1;
2138         }
2139         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2140
2141         /*
2142          * If the device is detaching, wakeup any readers in case we just
2143          * removed the last response, which would leave nothing for them to
2144          * return other than -ENODEV.
2145          */
2146         if (unlikely(atomic_read(&sfp->parentdp->detaching)))
2147                 wake_up_interruptible_all(&sfp->read_wait);
2148
2149         return res;
2150 }
2151
2152 static Sg_fd *
2153 sg_add_sfp(Sg_device * sdp)
2154 {
2155         Sg_fd *sfp;
2156         unsigned long iflags;
2157         int bufflen;
2158
2159         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2160         if (!sfp)
2161                 return ERR_PTR(-ENOMEM);
2162
2163         init_waitqueue_head(&sfp->read_wait);
2164         rwlock_init(&sfp->rq_list_lock);
2165         INIT_LIST_HEAD(&sfp->rq_list);
2166         kref_init(&sfp->f_ref);
2167         mutex_init(&sfp->f_mutex);
2168         sfp->timeout = SG_DEFAULT_TIMEOUT;
2169         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2170         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2171         sfp->cmd_q = SG_DEF_COMMAND_Q;
2172         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2173         sfp->parentdp = sdp;
2174         write_lock_irqsave(&sdp->sfd_lock, iflags);
2175         if (atomic_read(&sdp->detaching)) {
2176                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2177                 kfree(sfp);
2178                 return ERR_PTR(-ENODEV);
2179         }
2180         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2181         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2182         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2183                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2184         if (unlikely(sg_big_buff != def_reserved_size))
2185                 sg_big_buff = def_reserved_size;
2186
2187         bufflen = min_t(int, sg_big_buff,
2188                         max_sectors_bytes(sdp->device->request_queue));
2189         sg_build_reserve(sfp, bufflen);
2190         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2191                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2192                                       sfp->reserve.bufflen,
2193                                       sfp->reserve.k_use_sg));
2194
2195         kref_get(&sdp->d_ref);
2196         __module_get(THIS_MODULE);
2197         return sfp;
2198 }
2199
2200 static void
2201 sg_remove_sfp_usercontext(struct work_struct *work)
2202 {
2203         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2204         struct sg_device *sdp = sfp->parentdp;
2205         struct scsi_device *device = sdp->device;
2206         Sg_request *srp;
2207         unsigned long iflags;
2208
2209         /* Cleanup any responses which were never read(). */
2210         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2211         while (!list_empty(&sfp->rq_list)) {
2212                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2213                 sg_finish_rem_req(srp);
2214                 list_del(&srp->entry);
2215                 srp->parentfp = NULL;
2216         }
2217         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2218
2219         if (sfp->reserve.bufflen > 0) {
2220                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2221                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2222                                 (int) sfp->reserve.bufflen,
2223                                 (int) sfp->reserve.k_use_sg));
2224                 sg_remove_scat(sfp, &sfp->reserve);
2225         }
2226
2227         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2228                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2229         kfree(sfp);
2230
2231         kref_put(&sdp->d_ref, sg_device_destroy);
2232         scsi_device_put(device);
2233         module_put(THIS_MODULE);
2234 }
2235
2236 static void
2237 sg_remove_sfp(struct kref *kref)
2238 {
2239         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2240         struct sg_device *sdp = sfp->parentdp;
2241         unsigned long iflags;
2242
2243         write_lock_irqsave(&sdp->sfd_lock, iflags);
2244         list_del(&sfp->sfd_siblings);
2245         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2246
2247         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2248         schedule_work(&sfp->ew.work);
2249 }
2250
2251 #ifdef CONFIG_SCSI_PROC_FS
2252 static int
2253 sg_idr_max_id(int id, void *p, void *data)
2254 {
2255         int *k = data;
2256
2257         if (*k < id)
2258                 *k = id;
2259
2260         return 0;
2261 }
2262
2263 static int
2264 sg_last_dev(void)
2265 {
2266         int k = -1;
2267         unsigned long iflags;
2268
2269         read_lock_irqsave(&sg_index_lock, iflags);
2270         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2271         read_unlock_irqrestore(&sg_index_lock, iflags);
2272         return k + 1;           /* origin 1 */
2273 }
2274 #endif
2275
2276 /* must be called with sg_index_lock held */
2277 static Sg_device *sg_lookup_dev(int dev)
2278 {
2279         return idr_find(&sg_index_idr, dev);
2280 }
2281
2282 static Sg_device *
2283 sg_get_dev(int dev)
2284 {
2285         struct sg_device *sdp;
2286         unsigned long flags;
2287
2288         read_lock_irqsave(&sg_index_lock, flags);
2289         sdp = sg_lookup_dev(dev);
2290         if (!sdp)
2291                 sdp = ERR_PTR(-ENXIO);
2292         else if (atomic_read(&sdp->detaching)) {
2293                 /* If sdp->detaching, then the refcount may already be 0, in
2294                  * which case it would be a bug to do kref_get().
2295                  */
2296                 sdp = ERR_PTR(-ENODEV);
2297         } else
2298                 kref_get(&sdp->d_ref);
2299         read_unlock_irqrestore(&sg_index_lock, flags);
2300
2301         return sdp;
2302 }
2303
2304 #ifdef CONFIG_SCSI_PROC_FS
2305 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2306
2307 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2308 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2309                                   size_t count, loff_t *off);
2310 static const struct proc_ops adio_proc_ops = {
2311         .proc_open      = sg_proc_single_open_adio,
2312         .proc_read      = seq_read,
2313         .proc_lseek     = seq_lseek,
2314         .proc_write     = sg_proc_write_adio,
2315         .proc_release   = single_release,
2316 };
2317
2318 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2319 static ssize_t sg_proc_write_dressz(struct file *filp,
2320                 const char __user *buffer, size_t count, loff_t *off);
2321 static const struct proc_ops dressz_proc_ops = {
2322         .proc_open      = sg_proc_single_open_dressz,
2323         .proc_read      = seq_read,
2324         .proc_lseek     = seq_lseek,
2325         .proc_write     = sg_proc_write_dressz,
2326         .proc_release   = single_release,
2327 };
2328
2329 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2330 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2331 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2332 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2333 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2334 static void dev_seq_stop(struct seq_file *s, void *v);
2335 static const struct seq_operations dev_seq_ops = {
2336         .start = dev_seq_start,
2337         .next  = dev_seq_next,
2338         .stop  = dev_seq_stop,
2339         .show  = sg_proc_seq_show_dev,
2340 };
2341
2342 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2343 static const struct seq_operations devstrs_seq_ops = {
2344         .start = dev_seq_start,
2345         .next  = dev_seq_next,
2346         .stop  = dev_seq_stop,
2347         .show  = sg_proc_seq_show_devstrs,
2348 };
2349
2350 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2351 static const struct seq_operations debug_seq_ops = {
2352         .start = dev_seq_start,
2353         .next  = dev_seq_next,
2354         .stop  = dev_seq_stop,
2355         .show  = sg_proc_seq_show_debug,
2356 };
2357
2358 static int
2359 sg_proc_init(void)
2360 {
2361         struct proc_dir_entry *p;
2362
2363         p = proc_mkdir("scsi/sg", NULL);
2364         if (!p)
2365                 return 1;
2366
2367         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2368         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2369         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2370         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2371         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2372         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2373         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2374         return 0;
2375 }
2376
2377
2378 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2379 {
2380         seq_printf(s, "%d\n", *((int *)s->private));
2381         return 0;
2382 }
2383
2384 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2385 {
2386         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2387 }
2388
2389 static ssize_t
2390 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2391                    size_t count, loff_t *off)
2392 {
2393         int err;
2394         unsigned long num;
2395
2396         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2397                 return -EACCES;
2398         err = kstrtoul_from_user(buffer, count, 0, &num);
2399         if (err)
2400                 return err;
2401         sg_allow_dio = num ? 1 : 0;
2402         return count;
2403 }
2404
2405 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2406 {
2407         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2408 }
2409
2410 static ssize_t
2411 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2412                      size_t count, loff_t *off)
2413 {
2414         int err;
2415         unsigned long k = ULONG_MAX;
2416
2417         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2418                 return -EACCES;
2419
2420         err = kstrtoul_from_user(buffer, count, 0, &k);
2421         if (err)
2422                 return err;
2423         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2424                 sg_big_buff = k;
2425                 return count;
2426         }
2427         return -ERANGE;
2428 }
2429
2430 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2431 {
2432         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2433                    sg_version_date);
2434         return 0;
2435 }
2436
2437 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2438 {
2439         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2440         return 0;
2441 }
2442
2443 struct sg_proc_deviter {
2444         loff_t  index;
2445         size_t  max;
2446 };
2447
2448 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2449 {
2450         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2451
2452         s->private = it;
2453         if (! it)
2454                 return NULL;
2455
2456         it->index = *pos;
2457         it->max = sg_last_dev();
2458         if (it->index >= it->max)
2459                 return NULL;
2460         return it;
2461 }
2462
2463 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2464 {
2465         struct sg_proc_deviter * it = s->private;
2466
2467         *pos = ++it->index;
2468         return (it->index < it->max) ? it : NULL;
2469 }
2470
2471 static void dev_seq_stop(struct seq_file *s, void *v)
2472 {
2473         kfree(s->private);
2474 }
2475
2476 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2477 {
2478         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2479         Sg_device *sdp;
2480         struct scsi_device *scsidp;
2481         unsigned long iflags;
2482
2483         read_lock_irqsave(&sg_index_lock, iflags);
2484         sdp = it ? sg_lookup_dev(it->index) : NULL;
2485         if ((NULL == sdp) || (NULL == sdp->device) ||
2486             (atomic_read(&sdp->detaching)))
2487                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2488         else {
2489                 scsidp = sdp->device;
2490                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2491                               scsidp->host->host_no, scsidp->channel,
2492                               scsidp->id, scsidp->lun, (int) scsidp->type,
2493                               1,
2494                               (int) scsidp->queue_depth,
2495                               (int) scsi_device_busy(scsidp),
2496                               (int) scsi_device_online(scsidp));
2497         }
2498         read_unlock_irqrestore(&sg_index_lock, iflags);
2499         return 0;
2500 }
2501
2502 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2503 {
2504         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2505         Sg_device *sdp;
2506         struct scsi_device *scsidp;
2507         unsigned long iflags;
2508
2509         read_lock_irqsave(&sg_index_lock, iflags);
2510         sdp = it ? sg_lookup_dev(it->index) : NULL;
2511         scsidp = sdp ? sdp->device : NULL;
2512         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2513                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2514                            scsidp->vendor, scsidp->model, scsidp->rev);
2515         else
2516                 seq_puts(s, "<no active device>\n");
2517         read_unlock_irqrestore(&sg_index_lock, iflags);
2518         return 0;
2519 }
2520
2521 /* must be called while holding sg_index_lock */
2522 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2523 {
2524         int k, new_interface, blen, usg;
2525         Sg_request *srp;
2526         Sg_fd *fp;
2527         const sg_io_hdr_t *hp;
2528         const char * cp;
2529         unsigned int ms;
2530
2531         k = 0;
2532         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2533                 k++;
2534                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2535                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2536                            "(res)sgat=%d low_dma=%d\n", k,
2537                            jiffies_to_msecs(fp->timeout),
2538                            fp->reserve.bufflen,
2539                            (int) fp->reserve.k_use_sg, 0);
2540                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2541                            (int) fp->cmd_q, (int) fp->force_packid,
2542                            (int) fp->keep_orphan);
2543                 list_for_each_entry(srp, &fp->rq_list, entry) {
2544                         hp = &srp->header;
2545                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2546                         if (srp->res_used) {
2547                                 if (new_interface &&
2548                                     (SG_FLAG_MMAP_IO & hp->flags))
2549                                         cp = "     mmap>> ";
2550                                 else
2551                                         cp = "     rb>> ";
2552                         } else {
2553                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2554                                         cp = "     dio>> ";
2555                                 else
2556                                         cp = "     ";
2557                         }
2558                         seq_puts(s, cp);
2559                         blen = srp->data.bufflen;
2560                         usg = srp->data.k_use_sg;
2561                         seq_puts(s, srp->done ?
2562                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2563                                   : "act:");
2564                         seq_printf(s, " id=%d blen=%d",
2565                                    srp->header.pack_id, blen);
2566                         if (srp->done)
2567                                 seq_printf(s, " dur=%d", hp->duration);
2568                         else {
2569                                 ms = jiffies_to_msecs(jiffies);
2570                                 seq_printf(s, " t_o/elap=%d/%d",
2571                                         (new_interface ? hp->timeout :
2572                                                   jiffies_to_msecs(fp->timeout)),
2573                                         (ms > hp->duration ? ms - hp->duration : 0));
2574                         }
2575                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2576                                    (int) srp->data.cmd_opcode);
2577                 }
2578                 if (list_empty(&fp->rq_list))
2579                         seq_puts(s, "     No requests active\n");
2580                 read_unlock(&fp->rq_list_lock);
2581         }
2582 }
2583
2584 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2585 {
2586         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2587         Sg_device *sdp;
2588         unsigned long iflags;
2589
2590         if (it && (0 == it->index))
2591                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2592                            (int)it->max, sg_big_buff);
2593
2594         read_lock_irqsave(&sg_index_lock, iflags);
2595         sdp = it ? sg_lookup_dev(it->index) : NULL;
2596         if (NULL == sdp)
2597                 goto skip;
2598         read_lock(&sdp->sfd_lock);
2599         if (!list_empty(&sdp->sfds)) {
2600                 seq_printf(s, " >>> device=%s ", sdp->name);
2601                 if (atomic_read(&sdp->detaching))
2602                         seq_puts(s, "detaching pending close ");
2603                 else if (sdp->device) {
2604                         struct scsi_device *scsidp = sdp->device;
2605
2606                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2607                                    scsidp->host->host_no,
2608                                    scsidp->channel, scsidp->id,
2609                                    scsidp->lun,
2610                                    scsidp->host->hostt->emulated);
2611                 }
2612                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2613                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2614                 sg_proc_debug_helper(s, sdp);
2615         }
2616         read_unlock(&sdp->sfd_lock);
2617 skip:
2618         read_unlock_irqrestore(&sg_index_lock, iflags);
2619         return 0;
2620 }
2621
2622 #endif                          /* CONFIG_SCSI_PROC_FS */
2623
2624 module_init(init_sg);
2625 module_exit(exit_sg);