sg.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  History:
   4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   5  *           to allow user process control of SCSI devices.
   6  *  Development Sponsored by Killy Corp. NY NY
   7  *
   8  * Original driver (sg.c):
   9  *        Copyright (C) 1992 Lawrence Foard
  10  * Version 2 and 3 extensions to driver:
  11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  12  */
  13
  14 static int sg_version_num = 30536;      /* 2 digits for each component */
  15 #define SG_VERSION_STR "3.5.36"
  16
  17 /*
  18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  21  *        (otherwise the macros compile to empty statements).
  22  *
  23  */
  24 #include <linux/module.h>
  25
  26 #include <linux/fs.h>
  27 #include <linux/kernel.h>
  28 #include <linux/sched.h>
  29 #include <linux/string.h>
  30 #include <linux/mm.h>
  31 #include <linux/errno.h>
  32 #include <linux/mtio.h>
  33 #include <linux/ioctl.h>
  34 #include <linux/major.h>
  35 #include <linux/slab.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/init.h>
  38 #include <linux/poll.h>
  39 #include <linux/moduleparam.h>
  40 #include <linux/cdev.h>
  41 #include <linux/idr.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/delay.h>
  45 #include <linux/blktrace_api.h>
  46 #include <linux/mutex.h>
  47 #include <linux/atomic.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/uio.h>
  50 #include <linux/cred.h> /* for sg_check_file_access() */
  51
  52 #include <scsi/scsi.h>
  53 #include <scsi/scsi_cmnd.h>
  54 #include <scsi/scsi_dbg.h>
  55 #include <scsi/scsi_device.h>
  56 #include <scsi/scsi_driver.h>
  57 #include <scsi/scsi_eh.h>
  58 #include <scsi/scsi_host.h>
  59 #include <scsi/scsi_ioctl.h>
  60 #include <scsi/scsi_tcq.h>
  61 #include <scsi/sg.h>
  62
  63 #include "scsi_logging.h"
  64
  65 #ifdef CONFIG_SCSI_PROC_FS
  66 #include <linux/proc_fs.h>
  67 static char *sg_version_date = "20140603";
  68
  69 static int sg_proc_init(void);
  70 #endif
  71
  72 #define SG_ALLOW_DIO_DEF 0
  73
  74 #define SG_MAX_DEVS (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         retval = scsi_autopm_get_device(device);
 311         if (retval)
 312                 goto sdp_put;
 313
 314         /* scsi_block_when_processing_errors() may block so bypass
 315          * check if O_NONBLOCK. Permits SCSI commands to be issued
 316          * during error recovery. Tread carefully. */
 317         if (!((flags & O_NONBLOCK) ||
 318               scsi_block_when_processing_errors(device))) {
 319                 retval = -ENXIO;
 320                 /* we are in error recovery for this device */
 321                 goto error_out;
 322         }
 323
 324         mutex_lock(&sdp->open_rel_lock);
 325         if (flags & O_NONBLOCK) {
 326                 if (flags & O_EXCL) {
 327                         if (sdp->open_cnt > 0) {
 328                                 retval = -EBUSY;
 329                                 goto error_mutex_locked;
 330                         }
 331                 } else {
 332                         if (sdp->exclude) {
 333                                 retval = -EBUSY;
 334                                 goto error_mutex_locked;
 335                         }
 336                 }
 337         } else {
 338                 retval = open_wait(sdp, flags);
 339                 if (retval) /* -ERESTARTSYS or -ENODEV */
 340                         goto error_mutex_locked;
 341         }
 342
 343         /* N.B. at this point we are holding the open_rel_lock */
 344         if (flags & O_EXCL)
 345                 sdp->exclude = true;
 346
 347         if (sdp->open_cnt < 1) {  /* no existing opens */
 348                 sdp->sgdebug = 0;
 349                 q = device->request_queue;
 350                 sdp->sg_tablesize = queue_max_segments(q);
 351         }
 352         sfp = sg_add_sfp(sdp);
 353         if (IS_ERR(sfp)) {
 354                 retval = PTR_ERR(sfp);
 355                 goto out_undo;
 356         }
 357
 358         filp->private_data = sfp;
 359         sdp->open_cnt++;
 360         mutex_unlock(&sdp->open_rel_lock);
 361
 362         retval = 0;
 363 sg_put:
 364         kref_put(&sdp->d_ref, sg_device_destroy);
 365         return retval;
 366
 367 out_undo:
 368         if (flags & O_EXCL) {
 369                 sdp->exclude = false;   /* undo if error */
 370                 wake_up_interruptible(&sdp->open_wait);
 371         }
 372 error_mutex_locked:
 373         mutex_unlock(&sdp->open_rel_lock);
 374 error_out:
 375         scsi_autopm_put_device(device);
 376 sdp_put:
 377         kref_put(&sdp->d_ref, sg_device_destroy);
 378         scsi_device_put(device);
 379         return retval;
 380 }
 381
 382 /* Release resources associated with a successful sg_open()
 383  * Returns 0 on success, else a negated errno value */
 384 static int
 385 sg_release(struct inode *inode, struct file *filp)
 386 {
 387         Sg_device *sdp;
 388         Sg_fd *sfp;
 389
 390         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 391                 return -ENXIO;
 392         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 393
 394         mutex_lock(&sdp->open_rel_lock);
 395         scsi_autopm_put_device(sdp->device);
 396         kref_put(&sfp->f_ref, sg_remove_sfp);
 397         sdp->open_cnt--;
 398
 399         /* possibly many open()s waiting on exlude clearing, start many;
 400          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 401         if (sdp->exclude) {
 402                 sdp->exclude = false;
 403                 wake_up_interruptible_all(&sdp->open_wait);
 404         } else if (0 == sdp->open_cnt) {
 405                 wake_up_interruptible(&sdp->open_wait);
 406         }
 407         mutex_unlock(&sdp->open_rel_lock);
 408         return 0;
 409 }
 410
 411 static int get_sg_io_pack_id(int *pack_id, void __user *buf, size_t count)
 412 {
 413         struct sg_header __user *old_hdr = buf;
 414         int reply_len;
 415
 416         if (count >= SZ_SG_HEADER) {
 417                 /* negative reply_len means v3 format, otherwise v1/v2 */
 418                 if (get_user(reply_len, &old_hdr->reply_len))
 419                         return -EFAULT;
 420
 421                 if (reply_len >= 0)
 422                         return get_user(*pack_id, &old_hdr->pack_id);
 423
 424                 if (in_compat_syscall() &&
 425                     count >= sizeof(struct compat_sg_io_hdr)) {
 426                         struct compat_sg_io_hdr __user *hp = buf;
 427
 428                         return get_user(*pack_id, &hp->pack_id);
 429                 }
 430
 431                 if (count >= sizeof(struct sg_io_hdr)) {
 432                         struct sg_io_hdr __user *hp = buf;
 433
 434                         return get_user(*pack_id, &hp->pack_id);
 435                 }
 436         }
 437
 438         /* no valid header was passed, so ignore the pack_id */
 439         *pack_id = -1;
 440         return 0;
 441 }
 442
 443 static ssize_t
 444 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 445 {
 446         Sg_device *sdp;
 447         Sg_fd *sfp;
 448         Sg_request *srp;
 449         int req_pack_id = -1;
 450         bool busy;
 451         sg_io_hdr_t *hp;
 452         struct sg_header *old_hdr;
 453         int retval;
 454
 455         /*
 456          * This could cause a response to be stranded. Close the associated
 457          * file descriptor to free up any resources being held.
 458          */
 459         retval = sg_check_file_access(filp, __func__);
 460         if (retval)
 461                 return retval;
 462
 463         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 464                 return -ENXIO;
 465         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 466                                       "sg_read: count=%d\n", (int) count));
 467
 468         if (sfp->force_packid)
 469                 retval = get_sg_io_pack_id(&req_pack_id, buf, count);
 470         if (retval)
 471                 return retval;
 472
 473         srp = sg_get_rq_mark(sfp, req_pack_id, &busy);
 474         if (!srp) {             /* now wait on packet to arrive */
 475                 if (filp->f_flags & O_NONBLOCK)
 476                         return -EAGAIN;
 477                 retval = wait_event_interruptible(sfp->read_wait,
 478                         ((srp = sg_get_rq_mark(sfp, req_pack_id, &busy)) ||
 479                         (!busy && atomic_read(&sdp->detaching))));
 480                 if (!srp)
 481                         /* signal or detaching */
 482                         return retval ? retval : -ENODEV;
 483         }
 484         if (srp->header.interface_id != '\0')
 485                 return sg_new_read(sfp, buf, count, srp);
 486
 487         hp = &srp->header;
 488         old_hdr = kzalloc(SZ_SG_HEADER, GFP_KERNEL);
 489         if (!old_hdr)
 490                 return -ENOMEM;
 491
 492         old_hdr->reply_len = (int) hp->timeout;
 493         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 494         old_hdr->pack_id = hp->pack_id;
 495         old_hdr->twelve_byte =
 496             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 497         old_hdr->target_status = hp->masked_status;
 498         old_hdr->host_status = hp->host_status;
 499         old_hdr->driver_status = hp->driver_status;
 500         if ((CHECK_CONDITION & hp->masked_status) ||
 501             (srp->sense_b[0] & 0x70) == 0x70) {
 502                 old_hdr->driver_status = DRIVER_SENSE;
 503                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 504                        sizeof (old_hdr->sense_buffer));
 505         }
 506         switch (hp->host_status) {
 507         /* This setup of 'result' is for backward compatibility and is best
 508            ignored by the user who should use target, host + driver status */
 509         case DID_OK:
 510         case DID_PASSTHROUGH:
 511         case DID_SOFT_ERROR:
 512                 old_hdr->result = 0;
 513                 break;
 514         case DID_NO_CONNECT:
 515         case DID_BUS_BUSY:
 516         case DID_TIME_OUT:
 517                 old_hdr->result = EBUSY;
 518                 break;
 519         case DID_BAD_TARGET:
 520         case DID_ABORT:
 521         case DID_PARITY:
 522         case DID_RESET:
 523         case DID_BAD_INTR:
 524                 old_hdr->result = EIO;
 525                 break;
 526         case DID_ERROR:
 527                 old_hdr->result = (srp->sense_b[0] == 0 &&
 528                                   hp->masked_status == GOOD) ? 0 : EIO;
 529                 break;
 530         default:
 531                 old_hdr->result = EIO;
 532                 break;
 533         }
 534
 535         /* Now copy the result back to the user buffer.  */
 536         if (count >= SZ_SG_HEADER) {
 537                 if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 538                         retval = -EFAULT;
 539                         goto free_old_hdr;
 540                 }
 541                 buf += SZ_SG_HEADER;
 542                 if (count > old_hdr->reply_len)
 543                         count = old_hdr->reply_len;
 544                 if (count > SZ_SG_HEADER) {
 545                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 546                                 retval = -EFAULT;
 547                                 goto free_old_hdr;
 548                         }
 549                 }
 550         } else
 551                 count = (old_hdr->result == 0) ? 0 : -EIO;
 552         sg_finish_rem_req(srp);
 553         sg_remove_request(sfp, srp);
 554         retval = count;
 555 free_old_hdr:
 556         kfree(old_hdr);
 557         return retval;
 558 }
 559
 560 static ssize_t
 561 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 562 {
 563         sg_io_hdr_t *hp = &srp->header;
 564         int err = 0, err2;
 565         int len;
 566
 567         if (in_compat_syscall()) {
 568                 if (count < sizeof(struct compat_sg_io_hdr)) {
 569                         err = -EINVAL;
 570                         goto err_out;
 571                 }
 572         } else if (count < SZ_SG_IO_HDR) {
 573                 err = -EINVAL;
 574                 goto err_out;
 575         }
 576         hp->sb_len_wr = 0;
 577         if ((hp->mx_sb_len > 0) && hp->sbp) {
 578                 if ((CHECK_CONDITION & hp->masked_status) ||
 579                     (srp->sense_b[0] & 0x70) == 0x70) {
 580                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 581                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 582                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 583                         len = (len > sb_len) ? sb_len : len;
 584                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 585                                 err = -EFAULT;
 586                                 goto err_out;
 587                         }
 588                         hp->driver_status = DRIVER_SENSE;
 589                         hp->sb_len_wr = len;
 590                 }
 591         }
 592         if (hp->masked_status || hp->host_status || hp->driver_status)
 593                 hp->info |= SG_INFO_CHECK;
 594         err = put_sg_io_hdr(hp, buf);
 595 err_out:
 596         err2 = sg_finish_rem_req(srp);
 597         sg_remove_request(sfp, srp);
 598         return err ? : err2 ? : count;
 599 }
 600
 601 static ssize_t
 602 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 603 {
 604         int mxsize, cmd_size, k;
 605         int input_size, blocking;
 606         unsigned char opcode;
 607         Sg_device *sdp;
 608         Sg_fd *sfp;
 609         Sg_request *srp;
 610         struct sg_header old_hdr;
 611         sg_io_hdr_t *hp;
 612         unsigned char cmnd[SG_MAX_CDB_SIZE];
 613         int retval;
 614
 615         retval = sg_check_file_access(filp, __func__);
 616         if (retval)
 617                 return retval;
 618
 619         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 620                 return -ENXIO;
 621         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 622                                       "sg_write: count=%d\n", (int) count));
 623         if (atomic_read(&sdp->detaching))
 624                 return -ENODEV;
 625         if (!((filp->f_flags & O_NONBLOCK) ||
 626               scsi_block_when_processing_errors(sdp->device)))
 627                 return -ENXIO;
 628
 629         if (count < SZ_SG_HEADER)
 630                 return -EIO;
 631         if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 632                 return -EFAULT;
 633         blocking = !(filp->f_flags & O_NONBLOCK);
 634         if (old_hdr.reply_len < 0)
 635                 return sg_new_write(sfp, filp, buf, count,
 636                                     blocking, 0, 0, NULL);
 637         if (count < (SZ_SG_HEADER + 6))
 638                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 639
 640         buf += SZ_SG_HEADER;
 641         if (get_user(opcode, buf))
 642                 return -EFAULT;
 643
 644         if (!(srp = sg_add_request(sfp))) {
 645                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 646                                               "sg_write: queue full\n"));
 647                 return -EDOM;
 648         }
 649         mutex_lock(&sfp->f_mutex);
 650         if (sfp->next_cmd_len > 0) {
 651                 cmd_size = sfp->next_cmd_len;
 652                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 653         } else {
 654                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 655                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 656                         cmd_size = 12;
 657         }
 658         mutex_unlock(&sfp->f_mutex);
 659         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 660                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 661 /* Determine buffer size.  */
 662         input_size = count - cmd_size;
 663         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 664         mxsize -= SZ_SG_HEADER;
 665         input_size -= SZ_SG_HEADER;
 666         if (input_size < 0) {
 667                 sg_remove_request(sfp, srp);
 668                 return -EIO;    /* User did not pass enough bytes for this command. */
 669         }
 670         hp = &srp->header;
 671         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 672         hp->cmd_len = (unsigned char) cmd_size;
 673         hp->iovec_count = 0;
 674         hp->mx_sb_len = 0;
 675         if (input_size > 0)
 676                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 677                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 678         else
 679                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 680         hp->dxfer_len = mxsize;
 681         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 682             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 683                 hp->dxferp = (char __user *)buf + cmd_size;
 684         else
 685                 hp->dxferp = NULL;
 686         hp->sbp = NULL;
 687         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 688         hp->flags = input_size; /* structure abuse ... */
 689         hp->pack_id = old_hdr.pack_id;
 690         hp->usr_ptr = NULL;
 691         if (copy_from_user(cmnd, buf, cmd_size)) {
 692                 sg_remove_request(sfp, srp);
 693                 return -EFAULT;
 694         }
 695         /*
 696          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 697          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 698          * is a non-zero input_size, so emit a warning.
 699          */
 700         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 701                 printk_ratelimited(KERN_WARNING
 702                                    "sg_write: data in/out %d/%d bytes "
 703                                    "for SCSI command 0x%x-- guessing "
 704                                    "data in;\n   program %s not setting "
 705                                    "count and/or reply_len properly\n",
 706                                    old_hdr.reply_len - (int)SZ_SG_HEADER,
 707                                    input_size, (unsigned int) cmnd[0],
 708                                    current->comm);
 709         }
 710         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 711         return (k < 0) ? k : count;
 712 }
 713
 714 static ssize_t
 715 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 716                  size_t count, int blocking, int read_only, int sg_io_owned,
 717                  Sg_request **o_srp)
 718 {
 719         int k;
 720         Sg_request *srp;
 721         sg_io_hdr_t *hp;
 722         unsigned char cmnd[SG_MAX_CDB_SIZE];
 723         int timeout;
 724         unsigned long ul_timeout;
 725
 726         if (count < SZ_SG_IO_HDR)
 727                 return -EINVAL;
 728
 729         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 730         if (!(srp = sg_add_request(sfp))) {
 731                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 732                                               "sg_new_write: queue full\n"));
 733                 return -EDOM;
 734         }
 735         srp->sg_io_owned = sg_io_owned;
 736         hp = &srp->header;
 737         if (get_sg_io_hdr(hp, buf)) {
 738                 sg_remove_request(sfp, srp);
 739                 return -EFAULT;
 740         }
 741         if (hp->interface_id != 'S') {
 742                 sg_remove_request(sfp, srp);
 743                 return -ENOSYS;
 744         }
 745         if (hp->flags & SG_FLAG_MMAP_IO) {
 746                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 747                         sg_remove_request(sfp, srp);
 748                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 749                 }
 750                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 751                         sg_remove_request(sfp, srp);
 752                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 753                 }
 754                 if (sfp->res_in_use) {
 755                         sg_remove_request(sfp, srp);
 756                         return -EBUSY;  /* reserve buffer already being used */
 757                 }
 758         }
 759         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 760         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 761         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 762                 sg_remove_request(sfp, srp);
 763                 return -EMSGSIZE;
 764         }
 765         if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 766                 sg_remove_request(sfp, srp);
 767                 return -EFAULT;
 768         }
 769         if (read_only && sg_allow_access(file, cmnd)) {
 770                 sg_remove_request(sfp, srp);
 771                 return -EPERM;
 772         }
 773         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 774         if (k < 0)
 775                 return k;
 776         if (o_srp)
 777                 *o_srp = srp;
 778         return count;
 779 }
 780
 781 static int
 782 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 783                 unsigned char *cmnd, int timeout, int blocking)
 784 {
 785         int k, at_head;
 786         Sg_device *sdp = sfp->parentdp;
 787         sg_io_hdr_t *hp = &srp->header;
 788
 789         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 790         hp->status = 0;
 791         hp->masked_status = 0;
 792         hp->msg_status = 0;
 793         hp->info = 0;
 794         hp->host_status = 0;
 795         hp->driver_status = 0;
 796         hp->resid = 0;
 797         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 798                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 799                         (int) cmnd[0], (int) hp->cmd_len));
 800
 801         if (hp->dxfer_len >= SZ_256M) {
 802                 sg_remove_request(sfp, srp);
 803                 return -EINVAL;
 804         }
 805
 806         k = sg_start_req(srp, cmnd);
 807         if (k) {
 808                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 809                         "sg_common_write: start_req err=%d\n", k));
 810                 sg_finish_rem_req(srp);
 811                 sg_remove_request(sfp, srp);
 812                 return k;       /* probably out of space --> ENOMEM */
 813         }
 814         if (atomic_read(&sdp->detaching)) {
 815                 if (srp->bio) {
 816                         blk_mq_free_request(srp->rq);
 817                         srp->rq = NULL;
 818                 }
 819
 820                 sg_finish_rem_req(srp);
 821                 sg_remove_request(sfp, srp);
 822                 return -ENODEV;
 823         }
 824
 825         hp->duration = jiffies_to_msecs(jiffies);
 826         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 827             (SG_FLAG_Q_AT_TAIL & hp->flags))
 828                 at_head = 0;
 829         else
 830                 at_head = 1;
 831
 832         srp->rq->timeout = timeout;
 833         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 834         srp->rq->end_io = sg_rq_end_io;
 835         blk_execute_rq_nowait(srp->rq, at_head);
 836         return 0;
 837 }
 838
 839 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 840 {
 841         unsigned long flags;
 842         int ret;
 843
 844         read_lock_irqsave(&sfp->rq_list_lock, flags);
 845         ret = srp->done;
 846         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 847         return ret;
 848 }
 849
 850 static int max_sectors_bytes(struct request_queue *q)
 851 {
 852         unsigned int max_sectors = queue_max_sectors(q);
 853
 854         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 855
 856         return max_sectors << 9;
 857 }
 858
 859 static void
 860 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
 861 {
 862         Sg_request *srp;
 863         int val;
 864         unsigned int ms;
 865
 866         val = 0;
 867         list_for_each_entry(srp, &sfp->rq_list, entry) {
 868                 if (val >= SG_MAX_QUEUE)
 869                         break;
 870                 rinfo[val].req_state = srp->done + 1;
 871                 rinfo[val].problem =
 872                         srp->header.masked_status &
 873                         srp->header.host_status &
 874                         srp->header.driver_status;
 875                 if (srp->done)
 876                         rinfo[val].duration =
 877                                 srp->header.duration;
 878                 else {
 879                         ms = jiffies_to_msecs(jiffies);
 880                         rinfo[val].duration =
 881                                 (ms > srp->header.duration) ?
 882                                 (ms - srp->header.duration) : 0;
 883                 }
 884                 rinfo[val].orphan = srp->orphan;
 885                 rinfo[val].sg_io_owned = srp->sg_io_owned;
 886                 rinfo[val].pack_id = srp->header.pack_id;
 887                 rinfo[val].usr_ptr = srp->header.usr_ptr;
 888                 val++;
 889         }
 890 }
 891
 892 #ifdef CONFIG_COMPAT
 893 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
 894         char req_state;
 895         char orphan;
 896         char sg_io_owned;
 897         char problem;
 898         int pack_id;
 899         compat_uptr_t usr_ptr;
 900         unsigned int duration;
 901         int unused;
 902 };
 903
 904 static int put_compat_request_table(struct compat_sg_req_info __user *o,
 905                                     struct sg_req_info *rinfo)
 906 {
 907         int i;
 908         for (i = 0; i < SG_MAX_QUEUE; i++) {
 909                 if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
 910                     put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
 911                     put_user(rinfo[i].duration, &o[i].duration) ||
 912                     put_user(rinfo[i].unused, &o[i].unused))
 913                         return -EFAULT;
 914         }
 915         return 0;
 916 }
 917 #endif
 918
 919 static long
 920 sg_ioctl_common(struct file *filp, Sg_device *sdp, Sg_fd *sfp,
 921                 unsigned int cmd_in, void __user *p)
 922 {
 923         int __user *ip = p;
 924         int result, val, read_only;
 925         Sg_request *srp;
 926         unsigned long iflags;
 927
 928         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 929                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 930         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 931
 932         switch (cmd_in) {
 933         case SG_IO:
 934                 if (atomic_read(&sdp->detaching))
 935                         return -ENODEV;
 936                 if (!scsi_block_when_processing_errors(sdp->device))
 937                         return -ENXIO;
 938                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 939                                  1, read_only, 1, &srp);
 940                 if (result < 0)
 941                         return result;
 942                 result = wait_event_interruptible(sfp->read_wait,
 943                         srp_done(sfp, srp));
 944                 write_lock_irq(&sfp->rq_list_lock);
 945                 if (srp->done) {
 946                         srp->done = 2;
 947                         write_unlock_irq(&sfp->rq_list_lock);
 948                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 949                         return (result < 0) ? result : 0;
 950                 }
 951                 srp->orphan = 1;
 952                 write_unlock_irq(&sfp->rq_list_lock);
 953                 return result;  /* -ERESTARTSYS because signal hit process */
 954         case SG_SET_TIMEOUT:
 955                 result = get_user(val, ip);
 956                 if (result)
 957                         return result;
 958                 if (val < 0)
 959                         return -EIO;
 960                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 961                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 962                                     INT_MAX);
 963                 sfp->timeout_user = val;
 964                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 965
 966                 return 0;
 967         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 968                                 /* strange ..., for backward compatibility */
 969                 return sfp->timeout_user;
 970         case SG_SET_FORCE_LOW_DMA:
 971                 /*
 972                  * N.B. This ioctl never worked properly, but failed to
 973                  * return an error value. So returning '0' to keep compability
 974                  * with legacy applications.
 975                  */
 976                 return 0;
 977         case SG_GET_LOW_DMA:
 978                 return put_user(0, ip);
 979         case SG_GET_SCSI_ID:
 980                 {
 981                         sg_scsi_id_t v;
 982
 983                         if (atomic_read(&sdp->detaching))
 984                                 return -ENODEV;
 985                         memset(&v, 0, sizeof(v));
 986                         v.host_no = sdp->device->host->host_no;
 987                         v.channel = sdp->device->channel;
 988                         v.scsi_id = sdp->device->id;
 989                         v.lun = sdp->device->lun;
 990                         v.scsi_type = sdp->device->type;
 991                         v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
 992                         v.d_queue_depth = sdp->device->queue_depth;
 993                         if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
 994                                 return -EFAULT;
 995                         return 0;
 996                 }
 997         case SG_SET_FORCE_PACK_ID:
 998                 result = get_user(val, ip);
 999                 if (result)
1000                         return result;
1001                 sfp->force_packid = val ? 1 : 0;
1002                 return 0;
1003         case SG_GET_PACK_ID:
1004                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1005                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1006                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
1007                                 read_unlock_irqrestore(&sfp->rq_list_lock,
1008                                                        iflags);
1009                                 return put_user(srp->header.pack_id, ip);
1010                         }
1011                 }
1012                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1013                 return put_user(-1, ip);
1014         case SG_GET_NUM_WAITING:
1015                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
1016                 val = 0;
1017                 list_for_each_entry(srp, &sfp->rq_list, entry) {
1018                         if ((1 == srp->done) && (!srp->sg_io_owned))
1019                                 ++val;
1020                 }
1021                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1022                 return put_user(val, ip);
1023         case SG_GET_SG_TABLESIZE:
1024                 return put_user(sdp->sg_tablesize, ip);
1025         case SG_SET_RESERVED_SIZE:
1026                 result = get_user(val, ip);
1027                 if (result)
1028                         return result;
1029                 if (val < 0)
1030                         return -EINVAL;
1031                 val = min_t(int, val,
1032                             max_sectors_bytes(sdp->device->request_queue));
1033                 mutex_lock(&sfp->f_mutex);
1034                 if (val != sfp->reserve.bufflen) {
1035                         if (sfp->mmap_called ||
1036                             sfp->res_in_use) {
1037                                 mutex_unlock(&sfp->f_mutex);
1038                                 return -EBUSY;
1039                         }
1040
1041                         sg_remove_scat(sfp, &sfp->reserve);
1042                         sg_build_reserve(sfp, val);
1043                 }
1044                 mutex_unlock(&sfp->f_mutex);
1045                 return 0;
1046         case SG_GET_RESERVED_SIZE:
1047                 val = min_t(int, sfp->reserve.bufflen,
1048                             max_sectors_bytes(sdp->device->request_queue));
1049                 return put_user(val, ip);
1050         case SG_SET_COMMAND_Q:
1051                 result = get_user(val, ip);
1052                 if (result)
1053                         return result;
1054                 sfp->cmd_q = val ? 1 : 0;
1055                 return 0;
1056         case SG_GET_COMMAND_Q:
1057                 return put_user((int) sfp->cmd_q, ip);
1058         case SG_SET_KEEP_ORPHAN:
1059                 result = get_user(val, ip);
1060                 if (result)
1061                         return result;
1062                 sfp->keep_orphan = val;
1063                 return 0;
1064         case SG_GET_KEEP_ORPHAN:
1065                 return put_user((int) sfp->keep_orphan, ip);
1066         case SG_NEXT_CMD_LEN:
1067                 result = get_user(val, ip);
1068                 if (result)
1069                         return result;
1070                 if (val > SG_MAX_CDB_SIZE)
1071                         return -ENOMEM;
1072                 sfp->next_cmd_len = (val > 0) ? val : 0;
1073                 return 0;
1074         case SG_GET_VERSION_NUM:
1075                 return put_user(sg_version_num, ip);
1076         case SG_GET_ACCESS_COUNT:
1077                 /* faked - we don't have a real access count anymore */
1078                 val = (sdp->device ? 1 : 0);
1079                 return put_user(val, ip);
1080         case SG_GET_REQUEST_TABLE:
1081                 {
1082                         sg_req_info_t *rinfo;
1083
1084                         rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1085                                         GFP_KERNEL);
1086                         if (!rinfo)
1087                                 return -ENOMEM;
1088                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1089                         sg_fill_request_table(sfp, rinfo);
1090                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1091         #ifdef CONFIG_COMPAT
1092                         if (in_compat_syscall())
1093                                 result = put_compat_request_table(p, rinfo);
1094                         else
1095         #endif
1096                                 result = copy_to_user(p, rinfo,
1097                                                       SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1098                         result = result ? -EFAULT : 0;
1099                         kfree(rinfo);
1100                         return result;
1101                 }
1102         case SG_EMULATED_HOST:
1103                 if (atomic_read(&sdp->detaching))
1104                         return -ENODEV;
1105                 return put_user(sdp->device->host->hostt->emulated, ip);
1106         case SCSI_IOCTL_SEND_COMMAND:
1107                 if (atomic_read(&sdp->detaching))
1108                         return -ENODEV;
1109                 return scsi_ioctl(sdp->device, filp->f_mode & FMODE_WRITE,
1110                                   cmd_in, p);
1111         case SG_SET_DEBUG:
1112                 result = get_user(val, ip);
1113                 if (result)
1114                         return result;
1115                 sdp->sgdebug = (char) val;
1116                 return 0;
1117         case BLKSECTGET:
1118                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1119                                 ip);
1120         case BLKTRACESETUP:
1121                 return blk_trace_setup(sdp->device->request_queue, sdp->name,
1122                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1123                                        NULL, p);
1124         case BLKTRACESTART:
1125                 return blk_trace_startstop(sdp->device->request_queue, 1);
1126         case BLKTRACESTOP:
1127                 return blk_trace_startstop(sdp->device->request_queue, 0);
1128         case BLKTRACETEARDOWN:
1129                 return blk_trace_remove(sdp->device->request_queue);
1130         case SCSI_IOCTL_GET_IDLUN:
1131         case SCSI_IOCTL_GET_BUS_NUMBER:
1132         case SCSI_IOCTL_PROBE_HOST:
1133         case SG_GET_TRANSFORM:
1134         case SG_SCSI_RESET:
1135                 if (atomic_read(&sdp->detaching))
1136                         return -ENODEV;
1137                 break;
1138         default:
1139                 if (read_only)
1140                         return -EPERM;  /* don't know so take safe approach */
1141                 break;
1142         }
1143
1144         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1145                         cmd_in, filp->f_flags & O_NDELAY);
1146         if (result)
1147                 return result;
1148
1149         return -ENOIOCTLCMD;
1150 }
1151
1152 static long
1153 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1154 {
1155         void __user *p = (void __user *)arg;
1156         Sg_device *sdp;
1157         Sg_fd *sfp;
1158         int ret;
1159
1160         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1161                 return -ENXIO;
1162
1163         ret = sg_ioctl_common(filp, sdp, sfp, cmd_in, p);
1164         if (ret != -ENOIOCTLCMD)
1165                 return ret;
1166         return scsi_ioctl(sdp->device, filp->f_mode & FMODE_WRITE, cmd_in, p);
1167 }
1168
1169 static __poll_t
1170 sg_poll(struct file *filp, poll_table * wait)
1171 {
1172         __poll_t res = 0;
1173         Sg_device *sdp;
1174         Sg_fd *sfp;
1175         Sg_request *srp;
1176         int count = 0;
1177         unsigned long iflags;
1178
1179         sfp = filp->private_data;
1180         if (!sfp)
1181                 return EPOLLERR;
1182         sdp = sfp->parentdp;
1183         if (!sdp)
1184                 return EPOLLERR;
1185         poll_wait(filp, &sfp->read_wait, wait);
1186         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1187         list_for_each_entry(srp, &sfp->rq_list, entry) {
1188                 /* if any read waiting, flag it */
1189                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1190                         res = EPOLLIN | EPOLLRDNORM;
1191                 ++count;
1192         }
1193         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1194
1195         if (atomic_read(&sdp->detaching))
1196                 res |= EPOLLHUP;
1197         else if (!sfp->cmd_q) {
1198                 if (0 == count)
1199                         res |= EPOLLOUT | EPOLLWRNORM;
1200         } else if (count < SG_MAX_QUEUE)
1201                 res |= EPOLLOUT | EPOLLWRNORM;
1202         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1203                                       "sg_poll: res=0x%x\n", (__force u32) res));
1204         return res;
1205 }
1206
1207 static int
1208 sg_fasync(int fd, struct file *filp, int mode)
1209 {
1210         Sg_device *sdp;
1211         Sg_fd *sfp;
1212
1213         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1214                 return -ENXIO;
1215         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1216                                       "sg_fasync: mode=%d\n", mode));
1217
1218         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1219 }
1220
1221 static vm_fault_t
1222 sg_vma_fault(struct vm_fault *vmf)
1223 {
1224         struct vm_area_struct *vma = vmf->vma;
1225         Sg_fd *sfp;
1226         unsigned long offset, len, sa;
1227         Sg_scatter_hold *rsv_schp;
1228         int k, length;
1229
1230         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1231                 return VM_FAULT_SIGBUS;
1232         rsv_schp = &sfp->reserve;
1233         offset = vmf->pgoff << PAGE_SHIFT;
1234         if (offset >= rsv_schp->bufflen)
1235                 return VM_FAULT_SIGBUS;
1236         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1237                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1238                                       offset, rsv_schp->k_use_sg));
1239         sa = vma->vm_start;
1240         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1241         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1242                 len = vma->vm_end - sa;
1243                 len = (len < length) ? len : length;
1244                 if (offset < len) {
1245                         struct page *page = nth_page(rsv_schp->pages[k],
1246                                                      offset >> PAGE_SHIFT);
1247                         get_page(page); /* increment page count */
1248                         vmf->page = page;
1249                         return 0; /* success */
1250                 }
1251                 sa += len;
1252                 offset -= len;
1253         }
1254
1255         return VM_FAULT_SIGBUS;
1256 }
1257
1258 static const struct vm_operations_struct sg_mmap_vm_ops = {
1259         .fault = sg_vma_fault,
1260 };
1261
1262 static int
1263 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1264 {
1265         Sg_fd *sfp;
1266         unsigned long req_sz, len, sa;
1267         Sg_scatter_hold *rsv_schp;
1268         int k, length;
1269         int ret = 0;
1270
1271         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1272                 return -ENXIO;
1273         req_sz = vma->vm_end - vma->vm_start;
1274         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1275                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1276                                       (void *) vma->vm_start, (int) req_sz));
1277         if (vma->vm_pgoff)
1278                 return -EINVAL; /* want no offset */
1279         rsv_schp = &sfp->reserve;
1280         mutex_lock(&sfp->f_mutex);
1281         if (req_sz > rsv_schp->bufflen) {
1282                 ret = -ENOMEM;  /* cannot map more than reserved buffer */
1283                 goto out;
1284         }
1285
1286         sa = vma->vm_start;
1287         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1288         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1289                 len = vma->vm_end - sa;
1290                 len = (len < length) ? len : length;
1291                 sa += len;
1292         }
1293
1294         sfp->mmap_called = 1;
1295         vm_flags_set(vma, VM_IO | VM_DONTEXPAND | VM_DONTDUMP);
1296         vma->vm_private_data = sfp;
1297         vma->vm_ops = &sg_mmap_vm_ops;
1298 out:
1299         mutex_unlock(&sfp->f_mutex);
1300         return ret;
1301 }
1302
1303 static void
1304 sg_rq_end_io_usercontext(struct work_struct *work)
1305 {
1306         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1307         struct sg_fd *sfp = srp->parentfp;
1308
1309         sg_finish_rem_req(srp);
1310         sg_remove_request(sfp, srp);
1311         kref_put(&sfp->f_ref, sg_remove_sfp);
1312 }
1313
1314 /*
1315  * This function is a "bottom half" handler that is called by the mid
1316  * level when a command is completed (or has failed).
1317  */
1318 static enum rq_end_io_ret
1319 sg_rq_end_io(struct request *rq, blk_status_t status)
1320 {
1321         struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq);
1322         struct sg_request *srp = rq->end_io_data;
1323         Sg_device *sdp;
1324         Sg_fd *sfp;
1325         unsigned long iflags;
1326         unsigned int ms;
1327         char *sense;
1328         int result, resid, done = 1;
1329
1330         if (WARN_ON(srp->done != 0))
1331                 return RQ_END_IO_NONE;
1332
1333         sfp = srp->parentfp;
1334         if (WARN_ON(sfp == NULL))
1335                 return RQ_END_IO_NONE;
1336
1337         sdp = sfp->parentdp;
1338         if (unlikely(atomic_read(&sdp->detaching)))
1339                 pr_info("%s: device detaching\n", __func__);
1340
1341         sense = scmd->sense_buffer;
1342         result = scmd->result;
1343         resid = scmd->resid_len;
1344
1345         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1346                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1347                                       srp->header.pack_id, result));
1348         srp->header.resid = resid;
1349         ms = jiffies_to_msecs(jiffies);
1350         srp->header.duration = (ms > srp->header.duration) ?
1351                                 (ms - srp->header.duration) : 0;
1352         if (0 != result) {
1353                 struct scsi_sense_hdr sshdr;
1354
1355                 srp->header.status = 0xff & result;
1356                 srp->header.masked_status = sg_status_byte(result);
1357                 srp->header.msg_status = COMMAND_COMPLETE;
1358                 srp->header.host_status = host_byte(result);
1359                 srp->header.driver_status = driver_byte(result);
1360                 if ((sdp->sgdebug > 0) &&
1361                     ((CHECK_CONDITION == srp->header.masked_status) ||
1362                      (COMMAND_TERMINATED == srp->header.masked_status)))
1363                         __scsi_print_sense(sdp->device, __func__, sense,
1364                                            SCSI_SENSE_BUFFERSIZE);
1365
1366                 /* Following if statement is a patch supplied by Eric Youngdale */
1367                 if (driver_byte(result) != 0
1368                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1369                     && !scsi_sense_is_deferred(&sshdr)
1370                     && sshdr.sense_key == UNIT_ATTENTION
1371                     && sdp->device->removable) {
1372                         /* Detected possible disc change. Set the bit - this */
1373                         /* may be used if there are filesystems using this device */
1374                         sdp->device->changed = 1;
1375                 }
1376         }
1377
1378         if (scmd->sense_len)
1379                 memcpy(srp->sense_b, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
1380
1381         /* Rely on write phase to clean out srp status values, so no "else" */
1382
1383         /*
1384          * Free the request as soon as it is complete so that its resources
1385          * can be reused without waiting for userspace to read() the
1386          * result.  But keep the associated bio (if any) around until
1387          * blk_rq_unmap_user() can be called from user context.
1388          */
1389         srp->rq = NULL;
1390         blk_mq_free_request(rq);
1391
1392         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1393         if (unlikely(srp->orphan)) {
1394                 if (sfp->keep_orphan)
1395                         srp->sg_io_owned = 0;
1396                 else
1397                         done = 0;
1398         }
1399         srp->done = done;
1400         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1401
1402         if (likely(done)) {
1403                 /* Now wake up any sg_read() that is waiting for this
1404                  * packet.
1405                  */
1406                 wake_up_interruptible(&sfp->read_wait);
1407                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1408                 kref_put(&sfp->f_ref, sg_remove_sfp);
1409         } else {
1410                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1411                 schedule_work(&srp->ew.work);
1412         }
1413         return RQ_END_IO_NONE;
1414 }
1415
1416 static const struct file_operations sg_fops = {
1417         .owner = THIS_MODULE,
1418         .read = sg_read,
1419         .write = sg_write,
1420         .poll = sg_poll,
1421         .unlocked_ioctl = sg_ioctl,
1422         .compat_ioctl = compat_ptr_ioctl,
1423         .open = sg_open,
1424         .mmap = sg_mmap,
1425         .release = sg_release,
1426         .fasync = sg_fasync,
1427         .llseek = no_llseek,
1428 };
1429
1430 static struct class *sg_sysfs_class;
1431
1432 static int sg_sysfs_valid = 0;
1433
1434 static Sg_device *
1435 sg_alloc(struct scsi_device *scsidp)
1436 {
1437         struct request_queue *q = scsidp->request_queue;
1438         Sg_device *sdp;
1439         unsigned long iflags;
1440         int error;
1441         u32 k;
1442
1443         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1444         if (!sdp) {
1445                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1446                             "failure\n", __func__);
1447                 return ERR_PTR(-ENOMEM);
1448         }
1449
1450         idr_preload(GFP_KERNEL);
1451         write_lock_irqsave(&sg_index_lock, iflags);
1452
1453         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1454         if (error < 0) {
1455                 if (error == -ENOSPC) {
1456                         sdev_printk(KERN_WARNING, scsidp,
1457                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1458                                     scsidp->type, SG_MAX_DEVS - 1);
1459                         error = -ENODEV;
1460                 } else {
1461                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1462                                     "allocation Sg_device failure: %d\n",
1463                                     __func__, error);
1464                 }
1465                 goto out_unlock;
1466         }
1467         k = error;
1468
1469         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1470                                         "sg_alloc: dev=%d \n", k));
1471         sprintf(sdp->name, "sg%d", k);
1472         sdp->device = scsidp;
1473         mutex_init(&sdp->open_rel_lock);
1474         INIT_LIST_HEAD(&sdp->sfds);
1475         init_waitqueue_head(&sdp->open_wait);
1476         atomic_set(&sdp->detaching, 0);
1477         rwlock_init(&sdp->sfd_lock);
1478         sdp->sg_tablesize = queue_max_segments(q);
1479         sdp->index = k;
1480         kref_init(&sdp->d_ref);
1481         error = 0;
1482
1483 out_unlock:
1484         write_unlock_irqrestore(&sg_index_lock, iflags);
1485         idr_preload_end();
1486
1487         if (error) {
1488                 kfree(sdp);
1489                 return ERR_PTR(error);
1490         }
1491         return sdp;
1492 }
1493
1494 static int
1495 sg_add_device(struct device *cl_dev)
1496 {
1497         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1498         Sg_device *sdp = NULL;
1499         struct cdev * cdev = NULL;
1500         int error;
1501         unsigned long iflags;
1502
1503         if (!blk_get_queue(scsidp->request_queue)) {
1504                 pr_warn("%s: get scsi_device queue failed\n", __func__);
1505                 return -ENODEV;
1506         }
1507
1508         error = -ENOMEM;
1509         cdev = cdev_alloc();
1510         if (!cdev) {
1511                 pr_warn("%s: cdev_alloc failed\n", __func__);
1512                 goto out;
1513         }
1514         cdev->owner = THIS_MODULE;
1515         cdev->ops = &sg_fops;
1516
1517         sdp = sg_alloc(scsidp);
1518         if (IS_ERR(sdp)) {
1519                 pr_warn("%s: sg_alloc failed\n", __func__);
1520                 error = PTR_ERR(sdp);
1521                 goto out;
1522         }
1523
1524         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1525         if (error)
1526                 goto cdev_add_err;
1527
1528         sdp->cdev = cdev;
1529         if (sg_sysfs_valid) {
1530                 struct device *sg_class_member;
1531
1532                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1533                                                 MKDEV(SCSI_GENERIC_MAJOR,
1534                                                       sdp->index),
1535                                                 sdp, "%s", sdp->name);
1536                 if (IS_ERR(sg_class_member)) {
1537                         pr_err("%s: device_create failed\n", __func__);
1538                         error = PTR_ERR(sg_class_member);
1539                         goto cdev_add_err;
1540                 }
1541                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1542                                           &sg_class_member->kobj, "generic");
1543                 if (error)
1544                         pr_err("%s: unable to make symlink 'generic' back "
1545                                "to sg%d\n", __func__, sdp->index);
1546         } else
1547                 pr_warn("%s: sg_sys Invalid\n", __func__);
1548
1549         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1550                     "type %d\n", sdp->index, scsidp->type);
1551
1552         dev_set_drvdata(cl_dev, sdp);
1553
1554         return 0;
1555
1556 cdev_add_err:
1557         write_lock_irqsave(&sg_index_lock, iflags);
1558         idr_remove(&sg_index_idr, sdp->index);
1559         write_unlock_irqrestore(&sg_index_lock, iflags);
1560         kfree(sdp);
1561
1562 out:
1563         if (cdev)
1564                 cdev_del(cdev);
1565         blk_put_queue(scsidp->request_queue);
1566         return error;
1567 }
1568
1569 static void
1570 sg_device_destroy(struct kref *kref)
1571 {
1572         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1573         struct request_queue *q = sdp->device->request_queue;
1574         unsigned long flags;
1575
1576         /* CAUTION!  Note that the device can still be found via idr_find()
1577          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1578          * any other cleanup.
1579          */
1580
1581         blk_trace_remove(q);
1582         blk_put_queue(q);
1583
1584         write_lock_irqsave(&sg_index_lock, flags);
1585         idr_remove(&sg_index_idr, sdp->index);
1586         write_unlock_irqrestore(&sg_index_lock, flags);
1587
1588         SCSI_LOG_TIMEOUT(3,
1589                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1590
1591         kfree(sdp);
1592 }
1593
1594 static void
1595 sg_remove_device(struct device *cl_dev)
1596 {
1597         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1598         Sg_device *sdp = dev_get_drvdata(cl_dev);
1599         unsigned long iflags;
1600         Sg_fd *sfp;
1601         int val;
1602
1603         if (!sdp)
1604                 return;
1605         /* want sdp->detaching non-zero as soon as possible */
1606         val = atomic_inc_return(&sdp->detaching);
1607         if (val > 1)
1608                 return; /* only want to do following once per device */
1609
1610         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1611                                       "%s\n", __func__));
1612
1613         read_lock_irqsave(&sdp->sfd_lock, iflags);
1614         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1615                 wake_up_interruptible_all(&sfp->read_wait);
1616                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1617         }
1618         wake_up_interruptible_all(&sdp->open_wait);
1619         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1620
1621         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1622         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1623         cdev_del(sdp->cdev);
1624         sdp->cdev = NULL;
1625
1626         kref_put(&sdp->d_ref, sg_device_destroy);
1627 }
1628
1629 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1630 module_param_named(def_reserved_size, def_reserved_size, int,
1631                    S_IRUGO | S_IWUSR);
1632 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1633
1634 MODULE_AUTHOR("Douglas Gilbert");
1635 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1636 MODULE_LICENSE("GPL");
1637 MODULE_VERSION(SG_VERSION_STR);
1638 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1639
1640 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1641                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1642 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1643 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1644
1645 #ifdef CONFIG_SYSCTL
1646 #include <linux/sysctl.h>
1647
1648 static struct ctl_table sg_sysctls[] = {
1649         {
1650                 .procname       = "sg-big-buff",
1651                 .data           = &sg_big_buff,
1652                 .maxlen         = sizeof(int),
1653                 .mode           = 0444,
1654                 .proc_handler   = proc_dointvec,
1655         },
1656 };
1657
1658 static struct ctl_table_header *hdr;
1659 static void register_sg_sysctls(void)
1660 {
1661         if (!hdr)
1662                 hdr = register_sysctl("kernel", sg_sysctls);
1663 }
1664
1665 static void unregister_sg_sysctls(void)
1666 {
1667         if (hdr)
1668                 unregister_sysctl_table(hdr);
1669 }
1670 #else
1671 #define register_sg_sysctls() do { } while (0)
1672 #define unregister_sg_sysctls() do { } while (0)
1673 #endif /* CONFIG_SYSCTL */
1674
1675 static int __init
1676 init_sg(void)
1677 {
1678         int rc;
1679
1680         if (scatter_elem_sz < PAGE_SIZE) {
1681                 scatter_elem_sz = PAGE_SIZE;
1682                 scatter_elem_sz_prev = scatter_elem_sz;
1683         }
1684         if (def_reserved_size >= 0)
1685                 sg_big_buff = def_reserved_size;
1686         else
1687                 def_reserved_size = sg_big_buff;
1688
1689         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1690                                     SG_MAX_DEVS, "sg");
1691         if (rc)
1692                 return rc;
1693         sg_sysfs_class = class_create("scsi_generic");
1694         if ( IS_ERR(sg_sysfs_class) ) {
1695                 rc = PTR_ERR(sg_sysfs_class);
1696                 goto err_out;
1697         }
1698         sg_sysfs_valid = 1;
1699         rc = scsi_register_interface(&sg_interface);
1700         if (0 == rc) {
1701 #ifdef CONFIG_SCSI_PROC_FS
1702                 sg_proc_init();
1703 #endif                          /* CONFIG_SCSI_PROC_FS */
1704                 return 0;
1705         }
1706         class_destroy(sg_sysfs_class);
1707         register_sg_sysctls();
1708 err_out:
1709         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1710         return rc;
1711 }
1712
1713 static void __exit
1714 exit_sg(void)
1715 {
1716         unregister_sg_sysctls();
1717 #ifdef CONFIG_SCSI_PROC_FS
1718         remove_proc_subtree("scsi/sg", NULL);
1719 #endif                          /* CONFIG_SCSI_PROC_FS */
1720         scsi_unregister_interface(&sg_interface);
1721         class_destroy(sg_sysfs_class);
1722         sg_sysfs_valid = 0;
1723         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1724                                  SG_MAX_DEVS);
1725         idr_destroy(&sg_index_idr);
1726 }
1727
1728 static int
1729 sg_start_req(Sg_request *srp, unsigned char *cmd)
1730 {
1731         int res;
1732         struct request *rq;
1733         Sg_fd *sfp = srp->parentfp;
1734         sg_io_hdr_t *hp = &srp->header;
1735         int dxfer_len = (int) hp->dxfer_len;
1736         int dxfer_dir = hp->dxfer_direction;
1737         unsigned int iov_count = hp->iovec_count;
1738         Sg_scatter_hold *req_schp = &srp->data;
1739         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1740         struct request_queue *q = sfp->parentdp->device->request_queue;
1741         struct rq_map_data *md, map_data;
1742         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? ITER_SOURCE : ITER_DEST;
1743         struct scsi_cmnd *scmd;
1744
1745         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1746                                       "sg_start_req: dxfer_len=%d\n",
1747                                       dxfer_len));
1748
1749         /*
1750          * NOTE
1751          *
1752          * With scsi-mq enabled, there are a fixed number of preallocated
1753          * requests equal in number to shost->can_queue.  If all of the
1754          * preallocated requests are already in use, then scsi_alloc_request()
1755          * will sleep until an active command completes, freeing up a request.
1756          * Although waiting in an asynchronous interface is less than ideal, we
1757          * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1758          * not expect an EWOULDBLOCK from this condition.
1759          */
1760         rq = scsi_alloc_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1761                         REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
1762         if (IS_ERR(rq))
1763                 return PTR_ERR(rq);
1764         scmd = blk_mq_rq_to_pdu(rq);
1765
1766         if (hp->cmd_len > sizeof(scmd->cmnd)) {
1767                 blk_mq_free_request(rq);
1768                 return -EINVAL;
1769         }
1770
1771         memcpy(scmd->cmnd, cmd, hp->cmd_len);
1772         scmd->cmd_len = hp->cmd_len;
1773
1774         srp->rq = rq;
1775         rq->end_io_data = srp;
1776         scmd->allowed = SG_DEFAULT_RETRIES;
1777
1778         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1779                 return 0;
1780
1781         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1782             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1783             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1784                 md = NULL;
1785         else
1786                 md = &map_data;
1787
1788         if (md) {
1789                 mutex_lock(&sfp->f_mutex);
1790                 if (dxfer_len <= rsv_schp->bufflen &&
1791                     !sfp->res_in_use) {
1792                         sfp->res_in_use = 1;
1793                         sg_link_reserve(sfp, srp, dxfer_len);
1794                 } else if (hp->flags & SG_FLAG_MMAP_IO) {
1795                         res = -EBUSY; /* sfp->res_in_use == 1 */
1796                         if (dxfer_len > rsv_schp->bufflen)
1797                                 res = -ENOMEM;
1798                         mutex_unlock(&sfp->f_mutex);
1799                         return res;
1800                 } else {
1801                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1802                         if (res) {
1803                                 mutex_unlock(&sfp->f_mutex);
1804                                 return res;
1805                         }
1806                 }
1807                 mutex_unlock(&sfp->f_mutex);
1808
1809                 md->pages = req_schp->pages;
1810                 md->page_order = req_schp->page_order;
1811                 md->nr_entries = req_schp->k_use_sg;
1812                 md->offset = 0;
1813                 md->null_mapped = hp->dxferp ? 0 : 1;
1814                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1815                         md->from_user = 1;
1816                 else
1817                         md->from_user = 0;
1818         }
1819
1820         res = blk_rq_map_user_io(rq, md, hp->dxferp, hp->dxfer_len,
1821                         GFP_ATOMIC, iov_count, iov_count, 1, rw);
1822         if (!res) {
1823                 srp->bio = rq->bio;
1824
1825                 if (!md) {
1826                         req_schp->dio_in_use = 1;
1827                         hp->info |= SG_INFO_DIRECT_IO;
1828                 }
1829         }
1830         return res;
1831 }
1832
1833 static int
1834 sg_finish_rem_req(Sg_request *srp)
1835 {
1836         int ret = 0;
1837
1838         Sg_fd *sfp = srp->parentfp;
1839         Sg_scatter_hold *req_schp = &srp->data;
1840
1841         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1842                                       "sg_finish_rem_req: res_used=%d\n",
1843                                       (int) srp->res_used));
1844         if (srp->bio)
1845                 ret = blk_rq_unmap_user(srp->bio);
1846
1847         if (srp->rq)
1848                 blk_mq_free_request(srp->rq);
1849
1850         if (srp->res_used)
1851                 sg_unlink_reserve(sfp, srp);
1852         else
1853                 sg_remove_scat(sfp, req_schp);
1854
1855         return ret;
1856 }
1857
1858 static int
1859 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1860 {
1861         int sg_bufflen = tablesize * sizeof(struct page *);
1862         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1863
1864         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1865         if (!schp->pages)
1866                 return -ENOMEM;
1867         schp->sglist_len = sg_bufflen;
1868         return tablesize;       /* number of scat_gath elements allocated */
1869 }
1870
1871 static int
1872 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1873 {
1874         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1875         int sg_tablesize = sfp->parentdp->sg_tablesize;
1876         int blk_size = buff_size, order;
1877         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1878
1879         if (blk_size < 0)
1880                 return -EFAULT;
1881         if (0 == blk_size)
1882                 ++blk_size;     /* don't know why */
1883         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1884         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1885         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1886                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1887                 buff_size, blk_size));
1888
1889         /* N.B. ret_sz carried into this block ... */
1890         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1891         if (mx_sc_elems < 0)
1892                 return mx_sc_elems;     /* most likely -ENOMEM */
1893
1894         num = scatter_elem_sz;
1895         if (unlikely(num != scatter_elem_sz_prev)) {
1896                 if (num < PAGE_SIZE) {
1897                         scatter_elem_sz = PAGE_SIZE;
1898                         scatter_elem_sz_prev = PAGE_SIZE;
1899                 } else
1900                         scatter_elem_sz_prev = num;
1901         }
1902
1903         order = get_order(num);
1904 retry:
1905         ret_sz = 1 << (PAGE_SHIFT + order);
1906
1907         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1908              k++, rem_sz -= ret_sz) {
1909
1910                 num = (rem_sz > scatter_elem_sz_prev) ?
1911                         scatter_elem_sz_prev : rem_sz;
1912
1913                 schp->pages[k] = alloc_pages(gfp_mask, order);
1914                 if (!schp->pages[k])
1915                         goto out;
1916
1917                 if (num == scatter_elem_sz_prev) {
1918                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1919                                 scatter_elem_sz = ret_sz;
1920                                 scatter_elem_sz_prev = ret_sz;
1921                         }
1922                 }
1923
1924                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1925                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1926                                  k, num, ret_sz));
1927         }               /* end of for loop */
1928
1929         schp->page_order = order;
1930         schp->k_use_sg = k;
1931         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1932                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1933                          k, rem_sz));
1934
1935         schp->bufflen = blk_size;
1936         if (rem_sz > 0) /* must have failed */
1937                 return -ENOMEM;
1938         return 0;
1939 out:
1940         for (i = 0; i < k; i++)
1941                 __free_pages(schp->pages[i], order);
1942
1943         if (--order >= 0)
1944                 goto retry;
1945
1946         return -ENOMEM;
1947 }
1948
1949 static void
1950 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1951 {
1952         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1953                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1954         if (schp->pages && schp->sglist_len > 0) {
1955                 if (!schp->dio_in_use) {
1956                         int k;
1957
1958                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1959                                 SCSI_LOG_TIMEOUT(5,
1960                                         sg_printk(KERN_INFO, sfp->parentdp,
1961                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1962                                         k, schp->pages[k]));
1963                                 __free_pages(schp->pages[k], schp->page_order);
1964                         }
1965
1966                         kfree(schp->pages);
1967                 }
1968         }
1969         memset(schp, 0, sizeof (*schp));
1970 }
1971
1972 static int
1973 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1974 {
1975         Sg_scatter_hold *schp = &srp->data;
1976         int k, num;
1977
1978         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1979                          "sg_read_oxfer: num_read_xfer=%d\n",
1980                          num_read_xfer));
1981         if ((!outp) || (num_read_xfer <= 0))
1982                 return 0;
1983
1984         num = 1 << (PAGE_SHIFT + schp->page_order);
1985         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1986                 if (num > num_read_xfer) {
1987                         if (copy_to_user(outp, page_address(schp->pages[k]),
1988                                            num_read_xfer))
1989                                 return -EFAULT;
1990                         break;
1991                 } else {
1992                         if (copy_to_user(outp, page_address(schp->pages[k]),
1993                                            num))
1994                                 return -EFAULT;
1995                         num_read_xfer -= num;
1996                         if (num_read_xfer <= 0)
1997                                 break;
1998                         outp += num;
1999                 }
2000         }
2001
2002         return 0;
2003 }
2004
2005 static void
2006 sg_build_reserve(Sg_fd * sfp, int req_size)
2007 {
2008         Sg_scatter_hold *schp = &sfp->reserve;
2009
2010         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2011                          "sg_build_reserve: req_size=%d\n", req_size));
2012         do {
2013                 if (req_size < PAGE_SIZE)
2014                         req_size = PAGE_SIZE;
2015                 if (0 == sg_build_indirect(schp, sfp, req_size))
2016                         return;
2017                 else
2018                         sg_remove_scat(sfp, schp);
2019                 req_size >>= 1; /* divide by 2 */
2020         } while (req_size > (PAGE_SIZE / 2));
2021 }
2022
2023 static void
2024 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2025 {
2026         Sg_scatter_hold *req_schp = &srp->data;
2027         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2028         int k, num, rem;
2029
2030         srp->res_used = 1;
2031         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2032                          "sg_link_reserve: size=%d\n", size));
2033         rem = size;
2034
2035         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2036         for (k = 0; k < rsv_schp->k_use_sg; k++) {
2037                 if (rem <= num) {
2038                         req_schp->k_use_sg = k + 1;
2039                         req_schp->sglist_len = rsv_schp->sglist_len;
2040                         req_schp->pages = rsv_schp->pages;
2041
2042                         req_schp->bufflen = size;
2043                         req_schp->page_order = rsv_schp->page_order;
2044                         break;
2045                 } else
2046                         rem -= num;
2047         }
2048
2049         if (k >= rsv_schp->k_use_sg)
2050                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2051                                  "sg_link_reserve: BAD size\n"));
2052 }
2053
2054 static void
2055 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2056 {
2057         Sg_scatter_hold *req_schp = &srp->data;
2058
2059         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2060                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2061                                       (int) req_schp->k_use_sg));
2062         req_schp->k_use_sg = 0;
2063         req_schp->bufflen = 0;
2064         req_schp->pages = NULL;
2065         req_schp->page_order = 0;
2066         req_schp->sglist_len = 0;
2067         srp->res_used = 0;
2068         /* Called without mutex lock to avoid deadlock */
2069         sfp->res_in_use = 0;
2070 }
2071
2072 static Sg_request *
2073 sg_get_rq_mark(Sg_fd * sfp, int pack_id, bool *busy)
2074 {
2075         Sg_request *resp;
2076         unsigned long iflags;
2077
2078         *busy = false;
2079         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2080         list_for_each_entry(resp, &sfp->rq_list, entry) {
2081                 /* look for requests that are not SG_IO owned */
2082                 if ((!resp->sg_io_owned) &&
2083                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2084                         switch (resp->done) {
2085                         case 0: /* request active */
2086                                 *busy = true;
2087                                 break;
2088                         case 1: /* request done; response ready to return */
2089                                 resp->done = 2; /* guard against other readers */
2090                                 write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2091                                 return resp;
2092                         case 2: /* response already being returned */
2093                                 break;
2094                         }
2095                 }
2096         }
2097         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2098         return NULL;
2099 }
2100
2101 /* always adds to end of list */
2102 static Sg_request *
2103 sg_add_request(Sg_fd * sfp)
2104 {
2105         int k;
2106         unsigned long iflags;
2107         Sg_request *rp = sfp->req_arr;
2108
2109         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2110         if (!list_empty(&sfp->rq_list)) {
2111                 if (!sfp->cmd_q)
2112                         goto out_unlock;
2113
2114                 for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2115                         if (!rp->parentfp)
2116                                 break;
2117                 }
2118                 if (k >= SG_MAX_QUEUE)
2119                         goto out_unlock;
2120         }
2121         memset(rp, 0, sizeof (Sg_request));
2122         rp->parentfp = sfp;
2123         rp->header.duration = jiffies_to_msecs(jiffies);
2124         list_add_tail(&rp->entry, &sfp->rq_list);
2125         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2126         return rp;
2127 out_unlock:
2128         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2129         return NULL;
2130 }
2131
2132 /* Return of 1 for found; 0 for not found */
2133 static int
2134 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2135 {
2136         unsigned long iflags;
2137         int res = 0;
2138
2139         if (!sfp || !srp || list_empty(&sfp->rq_list))
2140                 return res;
2141         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2142         if (!list_empty(&srp->entry)) {
2143                 list_del(&srp->entry);
2144                 srp->parentfp = NULL;
2145                 res = 1;
2146         }
2147         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2148
2149         /*
2150          * If the device is detaching, wakeup any readers in case we just
2151          * removed the last response, which would leave nothing for them to
2152          * return other than -ENODEV.
2153          */
2154         if (unlikely(atomic_read(&sfp->parentdp->detaching)))
2155                 wake_up_interruptible_all(&sfp->read_wait);
2156
2157         return res;
2158 }
2159
2160 static Sg_fd *
2161 sg_add_sfp(Sg_device * sdp)
2162 {
2163         Sg_fd *sfp;
2164         unsigned long iflags;
2165         int bufflen;
2166
2167         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2168         if (!sfp)
2169                 return ERR_PTR(-ENOMEM);
2170
2171         init_waitqueue_head(&sfp->read_wait);
2172         rwlock_init(&sfp->rq_list_lock);
2173         INIT_LIST_HEAD(&sfp->rq_list);
2174         kref_init(&sfp->f_ref);
2175         mutex_init(&sfp->f_mutex);
2176         sfp->timeout = SG_DEFAULT_TIMEOUT;
2177         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2178         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2179         sfp->cmd_q = SG_DEF_COMMAND_Q;
2180         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2181         sfp->parentdp = sdp;
2182         write_lock_irqsave(&sdp->sfd_lock, iflags);
2183         if (atomic_read(&sdp->detaching)) {
2184                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2185                 kfree(sfp);
2186                 return ERR_PTR(-ENODEV);
2187         }
2188         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2189         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2190         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2191                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2192         if (unlikely(sg_big_buff != def_reserved_size))
2193                 sg_big_buff = def_reserved_size;
2194
2195         bufflen = min_t(int, sg_big_buff,
2196                         max_sectors_bytes(sdp->device->request_queue));
2197         sg_build_reserve(sfp, bufflen);
2198         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2199                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2200                                       sfp->reserve.bufflen,
2201                                       sfp->reserve.k_use_sg));
2202
2203         kref_get(&sdp->d_ref);
2204         __module_get(THIS_MODULE);
2205         return sfp;
2206 }
2207
2208 static void
2209 sg_remove_sfp_usercontext(struct work_struct *work)
2210 {
2211         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2212         struct sg_device *sdp = sfp->parentdp;
2213         struct scsi_device *device = sdp->device;
2214         Sg_request *srp;
2215         unsigned long iflags;
2216
2217         /* Cleanup any responses which were never read(). */
2218         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2219         while (!list_empty(&sfp->rq_list)) {
2220                 srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2221                 sg_finish_rem_req(srp);
2222                 list_del(&srp->entry);
2223                 srp->parentfp = NULL;
2224         }
2225         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2226
2227         if (sfp->reserve.bufflen > 0) {
2228                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2229                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2230                                 (int) sfp->reserve.bufflen,
2231                                 (int) sfp->reserve.k_use_sg));
2232                 sg_remove_scat(sfp, &sfp->reserve);
2233         }
2234
2235         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2236                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2237         kfree(sfp);
2238
2239         kref_put(&sdp->d_ref, sg_device_destroy);
2240         scsi_device_put(device);
2241         module_put(THIS_MODULE);
2242 }
2243
2244 static void
2245 sg_remove_sfp(struct kref *kref)
2246 {
2247         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2248         struct sg_device *sdp = sfp->parentdp;
2249         unsigned long iflags;
2250
2251         write_lock_irqsave(&sdp->sfd_lock, iflags);
2252         list_del(&sfp->sfd_siblings);
2253         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2254
2255         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2256         schedule_work(&sfp->ew.work);
2257 }
2258
2259 #ifdef CONFIG_SCSI_PROC_FS
2260 static int
2261 sg_idr_max_id(int id, void *p, void *data)
2262 {
2263         int *k = data;
2264
2265         if (*k < id)
2266                 *k = id;
2267
2268         return 0;
2269 }
2270
2271 static int
2272 sg_last_dev(void)
2273 {
2274         int k = -1;
2275         unsigned long iflags;
2276
2277         read_lock_irqsave(&sg_index_lock, iflags);
2278         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2279         read_unlock_irqrestore(&sg_index_lock, iflags);
2280         return k + 1;           /* origin 1 */
2281 }
2282 #endif
2283
2284 /* must be called with sg_index_lock held */
2285 static Sg_device *sg_lookup_dev(int dev)
2286 {
2287         return idr_find(&sg_index_idr, dev);
2288 }
2289
2290 static Sg_device *
2291 sg_get_dev(int dev)
2292 {
2293         struct sg_device *sdp;
2294         unsigned long flags;
2295
2296         read_lock_irqsave(&sg_index_lock, flags);
2297         sdp = sg_lookup_dev(dev);
2298         if (!sdp)
2299                 sdp = ERR_PTR(-ENXIO);
2300         else if (atomic_read(&sdp->detaching)) {
2301                 /* If sdp->detaching, then the refcount may already be 0, in
2302                  * which case it would be a bug to do kref_get().
2303                  */
2304                 sdp = ERR_PTR(-ENODEV);
2305         } else
2306                 kref_get(&sdp->d_ref);
2307         read_unlock_irqrestore(&sg_index_lock, flags);
2308
2309         return sdp;
2310 }
2311
2312 #ifdef CONFIG_SCSI_PROC_FS
2313 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2314
2315 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2316 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2317                                   size_t count, loff_t *off);
2318 static const struct proc_ops adio_proc_ops = {
2319         .proc_open      = sg_proc_single_open_adio,
2320         .proc_read      = seq_read,
2321         .proc_lseek     = seq_lseek,
2322         .proc_write     = sg_proc_write_adio,
2323         .proc_release   = single_release,
2324 };
2325
2326 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2327 static ssize_t sg_proc_write_dressz(struct file *filp,
2328                 const char __user *buffer, size_t count, loff_t *off);
2329 static const struct proc_ops dressz_proc_ops = {
2330         .proc_open      = sg_proc_single_open_dressz,
2331         .proc_read      = seq_read,
2332         .proc_lseek     = seq_lseek,
2333         .proc_write     = sg_proc_write_dressz,
2334         .proc_release   = single_release,
2335 };
2336
2337 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2338 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2339 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2340 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2341 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2342 static void dev_seq_stop(struct seq_file *s, void *v);
2343 static const struct seq_operations dev_seq_ops = {
2344         .start = dev_seq_start,
2345         .next  = dev_seq_next,
2346         .stop  = dev_seq_stop,
2347         .show  = sg_proc_seq_show_dev,
2348 };
2349
2350 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2351 static const struct seq_operations devstrs_seq_ops = {
2352         .start = dev_seq_start,
2353         .next  = dev_seq_next,
2354         .stop  = dev_seq_stop,
2355         .show  = sg_proc_seq_show_devstrs,
2356 };
2357
2358 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2359 static const struct seq_operations debug_seq_ops = {
2360         .start = dev_seq_start,
2361         .next  = dev_seq_next,
2362         .stop  = dev_seq_stop,
2363         .show  = sg_proc_seq_show_debug,
2364 };
2365
2366 static int
2367 sg_proc_init(void)
2368 {
2369         struct proc_dir_entry *p;
2370
2371         p = proc_mkdir("scsi/sg", NULL);
2372         if (!p)
2373                 return 1;
2374
2375         proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_proc_ops);
2376         proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2377         proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_proc_ops);
2378         proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2379         proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2380         proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2381         proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2382         return 0;
2383 }
2384
2385
2386 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2387 {
2388         seq_printf(s, "%d\n", *((int *)s->private));
2389         return 0;
2390 }
2391
2392 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2393 {
2394         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2395 }
2396
2397 static ssize_t
2398 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2399                    size_t count, loff_t *off)
2400 {
2401         int err;
2402         unsigned long num;
2403
2404         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2405                 return -EACCES;
2406         err = kstrtoul_from_user(buffer, count, 0, &num);
2407         if (err)
2408                 return err;
2409         sg_allow_dio = num ? 1 : 0;
2410         return count;
2411 }
2412
2413 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2414 {
2415         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2416 }
2417
2418 static ssize_t
2419 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2420                      size_t count, loff_t *off)
2421 {
2422         int err;
2423         unsigned long k = ULONG_MAX;
2424
2425         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2426                 return -EACCES;
2427
2428         err = kstrtoul_from_user(buffer, count, 0, &k);
2429         if (err)
2430                 return err;
2431         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2432                 sg_big_buff = k;
2433                 return count;
2434         }
2435         return -ERANGE;
2436 }
2437
2438 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2439 {
2440         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2441                    sg_version_date);
2442         return 0;
2443 }
2444
2445 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2446 {
2447         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2448         return 0;
2449 }
2450
2451 struct sg_proc_deviter {
2452         loff_t  index;
2453         size_t  max;
2454 };
2455
2456 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2457 {
2458         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2459
2460         s->private = it;
2461         if (! it)
2462                 return NULL;
2463
2464         it->index = *pos;
2465         it->max = sg_last_dev();
2466         if (it->index >= it->max)
2467                 return NULL;
2468         return it;
2469 }
2470
2471 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2472 {
2473         struct sg_proc_deviter * it = s->private;
2474
2475         *pos = ++it->index;
2476         return (it->index < it->max) ? it : NULL;
2477 }
2478
2479 static void dev_seq_stop(struct seq_file *s, void *v)
2480 {
2481         kfree(s->private);
2482 }
2483
2484 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2485 {
2486         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2487         Sg_device *sdp;
2488         struct scsi_device *scsidp;
2489         unsigned long iflags;
2490
2491         read_lock_irqsave(&sg_index_lock, iflags);
2492         sdp = it ? sg_lookup_dev(it->index) : NULL;
2493         if ((NULL == sdp) || (NULL == sdp->device) ||
2494             (atomic_read(&sdp->detaching)))
2495                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2496         else {
2497                 scsidp = sdp->device;
2498                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2499                               scsidp->host->host_no, scsidp->channel,
2500                               scsidp->id, scsidp->lun, (int) scsidp->type,
2501                               1,
2502                               (int) scsidp->queue_depth,
2503                               (int) scsi_device_busy(scsidp),
2504                               (int) scsi_device_online(scsidp));
2505         }
2506         read_unlock_irqrestore(&sg_index_lock, iflags);
2507         return 0;
2508 }
2509
2510 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2511 {
2512         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2513         Sg_device *sdp;
2514         struct scsi_device *scsidp;
2515         unsigned long iflags;
2516
2517         read_lock_irqsave(&sg_index_lock, iflags);
2518         sdp = it ? sg_lookup_dev(it->index) : NULL;
2519         scsidp = sdp ? sdp->device : NULL;
2520         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2521                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2522                            scsidp->vendor, scsidp->model, scsidp->rev);
2523         else
2524                 seq_puts(s, "<no active device>\n");
2525         read_unlock_irqrestore(&sg_index_lock, iflags);
2526         return 0;
2527 }
2528
2529 /* must be called while holding sg_index_lock */
2530 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2531 {
2532         int k, new_interface, blen, usg;
2533         Sg_request *srp;
2534         Sg_fd *fp;
2535         const sg_io_hdr_t *hp;
2536         const char * cp;
2537         unsigned int ms;
2538
2539         k = 0;
2540         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2541                 k++;
2542                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2543                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2544                            "(res)sgat=%d low_dma=%d\n", k,
2545                            jiffies_to_msecs(fp->timeout),
2546                            fp->reserve.bufflen,
2547                            (int) fp->reserve.k_use_sg, 0);
2548                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2549                            (int) fp->cmd_q, (int) fp->force_packid,
2550                            (int) fp->keep_orphan);
2551                 list_for_each_entry(srp, &fp->rq_list, entry) {
2552                         hp = &srp->header;
2553                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2554                         if (srp->res_used) {
2555                                 if (new_interface &&
2556                                     (SG_FLAG_MMAP_IO & hp->flags))
2557                                         cp = "     mmap>> ";
2558                                 else
2559                                         cp = "     rb>> ";
2560                         } else {
2561                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2562                                         cp = "     dio>> ";
2563                                 else
2564                                         cp = "     ";
2565                         }
2566                         seq_puts(s, cp);
2567                         blen = srp->data.bufflen;
2568                         usg = srp->data.k_use_sg;
2569                         seq_puts(s, srp->done ?
2570                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2571                                   : "act:");
2572                         seq_printf(s, " id=%d blen=%d",
2573                                    srp->header.pack_id, blen);
2574                         if (srp->done)
2575                                 seq_printf(s, " dur=%d", hp->duration);
2576                         else {
2577                                 ms = jiffies_to_msecs(jiffies);
2578                                 seq_printf(s, " t_o/elap=%d/%d",
2579                                         (new_interface ? hp->timeout :
2580                                                   jiffies_to_msecs(fp->timeout)),
2581                                         (ms > hp->duration ? ms - hp->duration : 0));
2582                         }
2583                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2584                                    (int) srp->data.cmd_opcode);
2585                 }
2586                 if (list_empty(&fp->rq_list))
2587                         seq_puts(s, "     No requests active\n");
2588                 read_unlock(&fp->rq_list_lock);
2589         }
2590 }
2591
2592 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2593 {
2594         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2595         Sg_device *sdp;
2596         unsigned long iflags;
2597
2598         if (it && (0 == it->index))
2599                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2600                            (int)it->max, sg_big_buff);
2601
2602         read_lock_irqsave(&sg_index_lock, iflags);
2603         sdp = it ? sg_lookup_dev(it->index) : NULL;
2604         if (NULL == sdp)
2605                 goto skip;
2606         read_lock(&sdp->sfd_lock);
2607         if (!list_empty(&sdp->sfds)) {
2608                 seq_printf(s, " >>> device=%s ", sdp->name);
2609                 if (atomic_read(&sdp->detaching))
2610                         seq_puts(s, "detaching pending close ");
2611                 else if (sdp->device) {
2612                         struct scsi_device *scsidp = sdp->device;
2613
2614                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2615                                    scsidp->host->host_no,
2616                                    scsidp->channel, scsidp->id,
2617                                    scsidp->lun,
2618                                    scsidp->host->hostt->emulated);
2619                 }
2620                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2621                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2622                 sg_proc_debug_helper(s, sdp);
2623         }
2624         read_unlock(&sdp->sfd_lock);
2625 skip:
2626         read_unlock_irqrestore(&sg_index_lock, iflags);
2627         return 0;
2628 }
2629
2630 #endif                          /* CONFIG_SCSI_PROC_FS */
2631
2632 module_init(init_sg);
2633 module_exit(exit_sg);