[SCSI] sg: fix races during device removal
authorTony Battersby <tonyb@cybernetics.com>
Wed, 21 Jan 2009 19:45:50 +0000 (14:45 -0500)
committerJames Bottomley <James.Bottomley@HansenPartnership.com>
Thu, 12 Mar 2009 17:58:04 +0000 (12:58 -0500)
sg has the following problems related to device removal:

* opening a sg fd races with removing a device
* closing a sg fd races with removing a device
* /proc/scsi/sg/* access races with removing a device
* command completion races with removing a device
* command completion races with closing a sg fd
* can rmmod sg with active commands

These problems can cause kernel oopses, memory-use-after-free, or
double-free errors.  This patch fixes these problems by using krefs
to manage the lifetime of sg_device and sg_fd.

Each command submitted to the midlevel holds a reference to sg_fd
until the completion callback.  This ensures that sg_fd doesn't go
away if the fd is closed with commands still outstanding.

sg_fd gets the reference of sg_device (with scsi_device) and also
makes sure that the sg module doesn't go away.

/proc/scsi/sg/* functions don't play nicely with krefs because they
give information about sg_fds which have been closed but not yet
freed due to still having outstanding commands and sg_devices which
have been removed but not yet freed due to still being referenced
by one or more sg_fds.  To deal with this safely without removing
functionality, /proc functions now access sg_device and sg_fd while
holding a lock instead of using kref_get()/kref_put().

Signed-off-by: Tony Battersby <tonyb@cybernetics.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
drivers/scsi/sg.c

index 516925d8b570c3fab05d5d1b0a6b729e0ebce395..b447527555a70c438a1c5c5d9c123bbed6067c7a 100644 (file)
@@ -101,6 +101,7 @@ static int scatter_elem_sz_prev = SG_SCATTER_SZ;
 #define SG_SECTOR_MSK (SG_SECTOR_SZ - 1)
 
 static int sg_add(struct device *, struct class_interface *);
+static void sg_device_destroy(struct kref *kref);
 static void sg_remove(struct device *, struct class_interface *);
 
 static DEFINE_IDR(sg_index_idr);
@@ -158,6 +159,8 @@ typedef struct sg_fd {              /* holds the state of a file descriptor */
        char next_cmd_len;      /* 0 -> automatic (def), >0 -> use on next write() */
        char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
        char mmap_called;       /* 0 -> mmap() never called on this fd */
+       struct kref f_ref;
+       struct execute_work ew;
 } Sg_fd;
 
 typedef struct sg_device { /* holds the state of each scsi generic device */
@@ -171,6 +174,7 @@ typedef struct sg_device { /* holds the state of each scsi generic device */
        char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
        struct gendisk *disk;
        struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
+       struct kref d_ref;
 } Sg_device;
 
 static int sg_fasync(int fd, struct file *filp, int mode);
@@ -194,13 +198,14 @@ static void sg_build_reserve(Sg_fd * sfp, int req_size);
 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
 static Sg_fd *sg_add_sfp(Sg_device * sdp, int dev);
-static int sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
-static void __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
+static void sg_remove_sfp(struct kref *);
 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
 static Sg_request *sg_add_request(Sg_fd * sfp);
 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
 static int sg_res_in_use(Sg_fd * sfp);
+static Sg_device *sg_lookup_dev(int dev);
 static Sg_device *sg_get_dev(int dev);
+static void sg_put_dev(Sg_device *sdp);
 #ifdef CONFIG_SCSI_PROC_FS
 static int sg_last_dev(void);
 #endif
@@ -237,22 +242,17 @@ sg_open(struct inode *inode, struct file *filp)
        nonseekable_open(inode, filp);
        SCSI_LOG_TIMEOUT(3, printk("sg_open: dev=%d, flags=0x%x\n", dev, flags));
        sdp = sg_get_dev(dev);
-       if ((!sdp) || (!sdp->device)) {
-               unlock_kernel();
-               return -ENXIO;
-       }
-       if (sdp->detached) {
-               unlock_kernel();
-               return -ENODEV;
+       if (IS_ERR(sdp)) {
+               retval = PTR_ERR(sdp);
+               sdp = NULL;
+               goto sg_put;
        }
 
        /* This driver's module count bumped by fops_get in <linux/fs.h> */
        /* Prevent the device driver from vanishing while we sleep */
        retval = scsi_device_get(sdp->device);
-       if (retval) {
-               unlock_kernel();
-               return retval;
-       }
+       if (retval)
+               goto sg_put;
 
        if (!((flags & O_NONBLOCK) ||
              scsi_block_when_processing_errors(sdp->device))) {
@@ -303,16 +303,20 @@ sg_open(struct inode *inode, struct file *filp)
        if ((sfp = sg_add_sfp(sdp, dev)))
                filp->private_data = sfp;
        else {
-               if (flags & O_EXCL)
+               if (flags & O_EXCL) {
                        sdp->exclude = 0;       /* undo if error */
+                       wake_up_interruptible(&sdp->o_excl_wait);
+               }
                retval = -ENOMEM;
                goto error_out;
        }
-       unlock_kernel();
-       return 0;
-
-      error_out:
-       scsi_device_put(sdp->device);
+       retval = 0;
+error_out:
+       if (retval)
+               scsi_device_put(sdp->device);
+sg_put:
+       if (sdp)
+               sg_put_dev(sdp);
        unlock_kernel();
        return retval;
 }
@@ -327,13 +331,13 @@ sg_release(struct inode *inode, struct file *filp)
        if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
                return -ENXIO;
        SCSI_LOG_TIMEOUT(3, printk("sg_release: %s\n", sdp->disk->disk_name));
-       if (0 == sg_remove_sfp(sdp, sfp)) {     /* Returns 1 when sdp gone */
-               if (!sdp->detached) {
-                       scsi_device_put(sdp->device);
-               }
-               sdp->exclude = 0;
-               wake_up_interruptible(&sdp->o_excl_wait);
-       }
+
+       sfp->closed = 1;
+
+       sdp->exclude = 0;
+       wake_up_interruptible(&sdp->o_excl_wait);
+
+       kref_put(&sfp->f_ref, sg_remove_sfp);
        return 0;
 }
 
@@ -755,6 +759,7 @@ sg_common_write(Sg_fd * sfp, Sg_request * srp,
        hp->duration = jiffies_to_msecs(jiffies);
 
        srp->rq->timeout = timeout;
+       kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
        blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
                              srp->rq, 1, sg_rq_end_io);
        return 0;
@@ -1247,24 +1252,23 @@ sg_mmap(struct file *filp, struct vm_area_struct *vma)
 static void sg_rq_end_io(struct request *rq, int uptodate)
 {
        struct sg_request *srp = rq->end_io_data;
-       Sg_device *sdp = NULL;
+       Sg_device *sdp;
        Sg_fd *sfp;
        unsigned long iflags;
        unsigned int ms;
        char *sense;
-       int result, resid;
+       int result, resid, done = 1;
 
-       if (NULL == srp) {
-               printk(KERN_ERR "sg_cmd_done: NULL request\n");
+       if (WARN_ON(srp->done != 0))
                return;
-       }
+
        sfp = srp->parentfp;
-       if (sfp)
-               sdp = sfp->parentdp;
-       if ((NULL == sdp) || sdp->detached) {
-               printk(KERN_INFO "sg_cmd_done: device detached\n");
+       if (WARN_ON(sfp == NULL))
                return;
-       }
+
+       sdp = sfp->parentdp;
+       if (unlikely(sdp->detached))
+               printk(KERN_INFO "sg_rq_end_io: device detached\n");
 
        sense = rq->sense;
        result = rq->errors;
@@ -1303,33 +1307,26 @@ static void sg_rq_end_io(struct request *rq, int uptodate)
        }
        /* Rely on write phase to clean out srp status values, so no "else" */
 
-       if (sfp->closed) {      /* whoops this fd already released, cleanup */
-               SCSI_LOG_TIMEOUT(1, printk("sg_cmd_done: already closed, freeing ...\n"));
-               sg_finish_rem_req(srp);
-               srp = NULL;
-               if (NULL == sfp->headrp) {
-                       SCSI_LOG_TIMEOUT(1, printk("sg_cmd_done: already closed, final cleanup\n"));
-                       if (0 == sg_remove_sfp(sdp, sfp)) {     /* device still present */
-                               scsi_device_put(sdp->device);
-                       }
-                       sfp = NULL;
-               }
-       } else if (srp && srp->orphan) {
+       write_lock_irqsave(&sfp->rq_list_lock, iflags);
+       if (unlikely(srp->orphan)) {
                if (sfp->keep_orphan)
                        srp->sg_io_owned = 0;
-               else {
-                       sg_finish_rem_req(srp);
-                       srp = NULL;
-               }
+               else
+                       done = 0;
        }
-       if (sfp && srp) {
-               /* Now wake up any sg_read() that is waiting for this packet. */
-               kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
-               write_lock_irqsave(&sfp->rq_list_lock, iflags);
-               srp->done = 1;
+       srp->done = done;
+       write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
+
+       if (likely(done)) {
+               /* Now wake up any sg_read() that is waiting for this
+                * packet.
+                */
                wake_up_interruptible(&sfp->read_wait);
-               write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
-       }
+               kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
+       } else
+               sg_finish_rem_req(srp); /* call with srp->done == 0 */
+
+       kref_put(&sfp->f_ref, sg_remove_sfp);
 }
 
 static struct file_operations sg_fops = {
@@ -1364,17 +1361,18 @@ static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
                printk(KERN_WARNING "kmalloc Sg_device failure\n");
                return ERR_PTR(-ENOMEM);
        }
-       error = -ENOMEM;
+
        if (!idr_pre_get(&sg_index_idr, GFP_KERNEL)) {
                printk(KERN_WARNING "idr expansion Sg_device failure\n");
+               error = -ENOMEM;
                goto out;
        }
 
        write_lock_irqsave(&sg_index_lock, iflags);
-       error = idr_get_new(&sg_index_idr, sdp, &k);
-       write_unlock_irqrestore(&sg_index_lock, iflags);
 
+       error = idr_get_new(&sg_index_idr, sdp, &k);
        if (error) {
+               write_unlock_irqrestore(&sg_index_lock, iflags);
                printk(KERN_WARNING "idr allocation Sg_device failure: %d\n",
                       error);
                goto out;
@@ -1391,6 +1389,9 @@ static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
        init_waitqueue_head(&sdp->o_excl_wait);
        sdp->sg_tablesize = min(q->max_hw_segments, q->max_phys_segments);
        sdp->index = k;
+       kref_init(&sdp->d_ref);
+
+       write_unlock_irqrestore(&sg_index_lock, iflags);
 
        error = 0;
  out:
@@ -1401,6 +1402,8 @@ static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
        return sdp;
 
  overflow:
+       idr_remove(&sg_index_idr, k);
+       write_unlock_irqrestore(&sg_index_lock, iflags);
        sdev_printk(KERN_WARNING, scsidp,
                    "Unable to attach sg device type=%d, minor "
                    "number exceeds %d\n", scsidp->type, SG_MAX_DEVS - 1);
@@ -1488,49 +1491,46 @@ out:
        return error;
 }
 
-static void
-sg_remove(struct device *cl_dev, struct class_interface *cl_intf)
+static void sg_device_destroy(struct kref *kref)
+{
+       struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
+       unsigned long flags;
+
+       /* CAUTION!  Note that the device can still be found via idr_find()
+        * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
+        * any other cleanup.
+        */
+
+       write_lock_irqsave(&sg_index_lock, flags);
+       idr_remove(&sg_index_idr, sdp->index);
+       write_unlock_irqrestore(&sg_index_lock, flags);
+
+       SCSI_LOG_TIMEOUT(3,
+               printk("sg_device_destroy: %s\n",
+                       sdp->disk->disk_name));
+
+       put_disk(sdp->disk);
+       kfree(sdp);
+}
+
+static void sg_remove(struct device *cl_dev, struct class_interface *cl_intf)
 {
        struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
        Sg_device *sdp = dev_get_drvdata(cl_dev);
        unsigned long iflags;
        Sg_fd *sfp;
-       Sg_fd *tsfp;
-       Sg_request *srp;
-       Sg_request *tsrp;
-       int delay;
 
-       if (!sdp)
+       if (!sdp || sdp->detached)
                return;
 
-       delay = 0;
+       SCSI_LOG_TIMEOUT(3, printk("sg_remove: %s\n", sdp->disk->disk_name));
+
+       /* Need a write lock to set sdp->detached. */
        write_lock_irqsave(&sg_index_lock, iflags);
-       if (sdp->headfp) {
-               sdp->detached = 1;
-               for (sfp = sdp->headfp; sfp; sfp = tsfp) {
-                       tsfp = sfp->nextfp;
-                       for (srp = sfp->headrp; srp; srp = tsrp) {
-                               tsrp = srp->nextrp;
-                               if (sfp->closed || (0 == sg_srp_done(srp, sfp)))
-                                       sg_finish_rem_req(srp);
-                       }
-                       if (sfp->closed) {
-                               scsi_device_put(sdp->device);
-                               __sg_remove_sfp(sdp, sfp);
-                       } else {
-                               delay = 1;
-                               wake_up_interruptible(&sfp->read_wait);
-                               kill_fasync(&sfp->async_qp, SIGPOLL,
-                                           POLL_HUP);
-                       }
-               }
-               SCSI_LOG_TIMEOUT(3, printk("sg_remove: dev=%d, dirty\n", sdp->index));
-               if (NULL == sdp->headfp) {
-                       idr_remove(&sg_index_idr, sdp->index);
-               }
-       } else {        /* nothing active, simple case */
-               SCSI_LOG_TIMEOUT(3, printk("sg_remove: dev=%d\n", sdp->index));
-               idr_remove(&sg_index_idr, sdp->index);
+       sdp->detached = 1;
+       for (sfp = sdp->headfp; sfp; sfp = sfp->nextfp) {
+               wake_up_interruptible(&sfp->read_wait);
+               kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
        }
        write_unlock_irqrestore(&sg_index_lock, iflags);
 
@@ -1538,13 +1538,8 @@ sg_remove(struct device *cl_dev, struct class_interface *cl_intf)
        device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
        cdev_del(sdp->cdev);
        sdp->cdev = NULL;
-       put_disk(sdp->disk);
-       sdp->disk = NULL;
-       if (NULL == sdp->headfp)
-               kfree(sdp);
 
-       if (delay)
-               msleep(10);     /* dirty detach so delay device destruction */
+       sg_put_dev(sdp);
 }
 
 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
@@ -1941,22 +1936,6 @@ sg_get_rq_mark(Sg_fd * sfp, int pack_id)
        return resp;
 }
 
-#ifdef CONFIG_SCSI_PROC_FS
-static Sg_request *
-sg_get_nth_request(Sg_fd * sfp, int nth)
-{
-       Sg_request *resp;
-       unsigned long iflags;
-       int k;
-
-       read_lock_irqsave(&sfp->rq_list_lock, iflags);
-       for (k = 0, resp = sfp->headrp; resp && (k < nth);
-            ++k, resp = resp->nextrp) ;
-       read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
-       return resp;
-}
-#endif
-
 /* always adds to end of list */
 static Sg_request *
 sg_add_request(Sg_fd * sfp)
@@ -2032,22 +2011,6 @@ sg_remove_request(Sg_fd * sfp, Sg_request * srp)
        return res;
 }
 
-#ifdef CONFIG_SCSI_PROC_FS
-static Sg_fd *
-sg_get_nth_sfp(Sg_device * sdp, int nth)
-{
-       Sg_fd *resp;
-       unsigned long iflags;
-       int k;
-
-       read_lock_irqsave(&sg_index_lock, iflags);
-       for (k = 0, resp = sdp->headfp; resp && (k < nth);
-            ++k, resp = resp->nextfp) ;
-       read_unlock_irqrestore(&sg_index_lock, iflags);
-       return resp;
-}
-#endif
-
 static Sg_fd *
 sg_add_sfp(Sg_device * sdp, int dev)
 {
@@ -2062,6 +2025,7 @@ sg_add_sfp(Sg_device * sdp, int dev)
        init_waitqueue_head(&sfp->read_wait);
        rwlock_init(&sfp->rq_list_lock);
 
+       kref_init(&sfp->f_ref);
        sfp->timeout = SG_DEFAULT_TIMEOUT;
        sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
        sfp->force_packid = SG_DEF_FORCE_PACK_ID;
@@ -2089,15 +2053,54 @@ sg_add_sfp(Sg_device * sdp, int dev)
        sg_build_reserve(sfp, bufflen);
        SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp:   bufflen=%d, k_use_sg=%d\n",
                           sfp->reserve.bufflen, sfp->reserve.k_use_sg));
+
+       kref_get(&sdp->d_ref);
+       __module_get(THIS_MODULE);
        return sfp;
 }
 
-static void
-__sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
+static void sg_remove_sfp_usercontext(struct work_struct *work)
+{
+       struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
+       struct sg_device *sdp = sfp->parentdp;
+
+       /* Cleanup any responses which were never read(). */
+       while (sfp->headrp)
+               sg_finish_rem_req(sfp->headrp);
+
+       if (sfp->reserve.bufflen > 0) {
+               SCSI_LOG_TIMEOUT(6,
+                       printk("sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
+                               (int) sfp->reserve.bufflen,
+                               (int) sfp->reserve.k_use_sg));
+               sg_remove_scat(&sfp->reserve);
+       }
+
+       SCSI_LOG_TIMEOUT(6,
+               printk("sg_remove_sfp: %s, sfp=0x%p\n",
+                       sdp->disk->disk_name,
+                       sfp));
+       kfree(sfp);
+
+       scsi_device_put(sdp->device);
+       sg_put_dev(sdp);
+       module_put(THIS_MODULE);
+}
+
+static void sg_remove_sfp(struct kref *kref)
 {
+       struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
+       struct sg_device *sdp = sfp->parentdp;
        Sg_fd *fp;
        Sg_fd *prev_fp;
+       unsigned long iflags;
+
+       /* CAUTION!  Note that sfp can still be found by walking sdp->headfp
+        * even though the refcount is now 0.  Therefore, unlink sfp from
+        * sdp->headfp BEFORE doing any other cleanup.
+        */
 
+       write_lock_irqsave(&sg_index_lock, iflags);
        prev_fp = sdp->headfp;
        if (sfp == prev_fp)
                sdp->headfp = prev_fp->nextfp;
@@ -2110,54 +2113,10 @@ __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
                        prev_fp = fp;
                }
        }
-       if (sfp->reserve.bufflen > 0) {
-               SCSI_LOG_TIMEOUT(6, 
-                       printk("__sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
-                       (int) sfp->reserve.bufflen, (int) sfp->reserve.k_use_sg));
-               sg_remove_scat(&sfp->reserve);
-       }
-       sfp->parentdp = NULL;
-       SCSI_LOG_TIMEOUT(6, printk("__sg_remove_sfp:    sfp=0x%p\n", sfp));
-       kfree(sfp);
-}
-
-/* Returns 0 in normal case, 1 when detached and sdp object removed */
-static int
-sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
-{
-       Sg_request *srp;
-       Sg_request *tsrp;
-       int dirty = 0;
-       int res = 0;
-
-       for (srp = sfp->headrp; srp; srp = tsrp) {
-               tsrp = srp->nextrp;
-               if (sg_srp_done(srp, sfp))
-                       sg_finish_rem_req(srp);
-               else
-                       ++dirty;
-       }
-       if (0 == dirty) {
-               unsigned long iflags;
+       write_unlock_irqrestore(&sg_index_lock, iflags);
+       wake_up_interruptible(&sdp->o_excl_wait);
 
-               write_lock_irqsave(&sg_index_lock, iflags);
-               __sg_remove_sfp(sdp, sfp);
-               if (sdp->detached && (NULL == sdp->headfp)) {
-                       idr_remove(&sg_index_idr, sdp->index);
-                       kfree(sdp);
-                       res = 1;
-               }
-               write_unlock_irqrestore(&sg_index_lock, iflags);
-       } else {
-               /* MOD_INC's to inhibit unloading sg and associated adapter driver */
-               /* only bump the access_count if we actually succeeded in
-                * throwing another counter on the host module */
-               scsi_device_get(sdp->device);   /* XXX: retval ignored? */      
-               sfp->closed = 1;        /* flag dirty state on this fd */
-               SCSI_LOG_TIMEOUT(1, printk("sg_remove_sfp: worrisome, %d writes pending\n",
-                                 dirty));
-       }
-       return res;
+       execute_in_process_context(sg_remove_sfp_usercontext, &sfp->ew);
 }
 
 static int
@@ -2199,19 +2158,38 @@ sg_last_dev(void)
 }
 #endif
 
-static Sg_device *
-sg_get_dev(int dev)
+/* must be called with sg_index_lock held */
+static Sg_device *sg_lookup_dev(int dev)
 {
-       Sg_device *sdp;
-       unsigned long iflags;
+       return idr_find(&sg_index_idr, dev);
+}
 
-       read_lock_irqsave(&sg_index_lock, iflags);
-       sdp = idr_find(&sg_index_idr, dev);
-       read_unlock_irqrestore(&sg_index_lock, iflags);
+static Sg_device *sg_get_dev(int dev)
+{
+       struct sg_device *sdp;
+       unsigned long flags;
+
+       read_lock_irqsave(&sg_index_lock, flags);
+       sdp = sg_lookup_dev(dev);
+       if (!sdp)
+               sdp = ERR_PTR(-ENXIO);
+       else if (sdp->detached) {
+               /* If sdp->detached, then the refcount may already be 0, in
+                * which case it would be a bug to do kref_get().
+                */
+               sdp = ERR_PTR(-ENODEV);
+       } else
+               kref_get(&sdp->d_ref);
+       read_unlock_irqrestore(&sg_index_lock, flags);
 
        return sdp;
 }
 
+static void sg_put_dev(struct sg_device *sdp)
+{
+       kref_put(&sdp->d_ref, sg_device_destroy);
+}
+
 #ifdef CONFIG_SCSI_PROC_FS
 
 static struct proc_dir_entry *sg_proc_sgp = NULL;
@@ -2468,8 +2446,10 @@ static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
        struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
        Sg_device *sdp;
        struct scsi_device *scsidp;
+       unsigned long iflags;
 
-       sdp = it ? sg_get_dev(it->index) : NULL;
+       read_lock_irqsave(&sg_index_lock, iflags);
+       sdp = it ? sg_lookup_dev(it->index) : NULL;
        if (sdp && (scsidp = sdp->device) && (!sdp->detached))
                seq_printf(s, "%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\n",
                              scsidp->host->host_no, scsidp->channel,
@@ -2480,6 +2460,7 @@ static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
                              (int) scsi_device_online(scsidp));
        else
                seq_printf(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
+       read_unlock_irqrestore(&sg_index_lock, iflags);
        return 0;
 }
 
@@ -2493,16 +2474,20 @@ static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
        struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
        Sg_device *sdp;
        struct scsi_device *scsidp;
+       unsigned long iflags;
 
-       sdp = it ? sg_get_dev(it->index) : NULL;
+       read_lock_irqsave(&sg_index_lock, iflags);
+       sdp = it ? sg_lookup_dev(it->index) : NULL;
        if (sdp && (scsidp = sdp->device) && (!sdp->detached))
                seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
                           scsidp->vendor, scsidp->model, scsidp->rev);
        else
                seq_printf(s, "<no active device>\n");
+       read_unlock_irqrestore(&sg_index_lock, iflags);
        return 0;
 }
 
+/* must be called while holding sg_index_lock */
 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
 {
        int k, m, new_interface, blen, usg;
@@ -2512,7 +2497,8 @@ static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
        const char * cp;
        unsigned int ms;
 
-       for (k = 0; (fp = sg_get_nth_sfp(sdp, k)); ++k) {
+       for (k = 0, fp = sdp->headfp; fp != NULL; ++k, fp = fp->nextfp) {
+               read_lock(&fp->rq_list_lock); /* irqs already disabled */
                seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
                           "(res)sgat=%d low_dma=%d\n", k + 1,
                           jiffies_to_msecs(fp->timeout),
@@ -2522,7 +2508,9 @@ static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
                seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=%d\n",
                           (int) fp->cmd_q, (int) fp->force_packid,
                           (int) fp->keep_orphan, (int) fp->closed);
-               for (m = 0; (srp = sg_get_nth_request(fp, m)); ++m) {
+               for (m = 0, srp = fp->headrp;
+                               srp != NULL;
+                               ++m, srp = srp->nextrp) {
                        hp = &srp->header;
                        new_interface = (hp->interface_id == '\0') ? 0 : 1;
                        if (srp->res_used) {
@@ -2559,6 +2547,7 @@ static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
                }
                if (0 == m)
                        seq_printf(s, "     No requests active\n");
+               read_unlock(&fp->rq_list_lock);
        }
 }
 
@@ -2571,39 +2560,34 @@ static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
 {
        struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
        Sg_device *sdp;
+       unsigned long iflags;
 
        if (it && (0 == it->index)) {
                seq_printf(s, "max_active_device=%d(origin 1)\n",
                           (int)it->max);
                seq_printf(s, " def_reserved_size=%d\n", sg_big_buff);
        }
-       sdp = it ? sg_get_dev(it->index) : NULL;
-       if (sdp) {
-               struct scsi_device *scsidp = sdp->device;
 
-               if (NULL == scsidp) {
-                       seq_printf(s, "device %d detached ??\n", 
-                                  (int)it->index);
-                       return 0;
-               }
+       read_lock_irqsave(&sg_index_lock, iflags);
+       sdp = it ? sg_lookup_dev(it->index) : NULL;
+       if (sdp && sdp->headfp) {
+               struct scsi_device *scsidp = sdp->device;
 
-               if (sg_get_nth_sfp(sdp, 0)) {
-                       seq_printf(s, " >>> device=%s ",
-                               sdp->disk->disk_name);
-                       if (sdp->detached)
-                               seq_printf(s, "detached pending close ");
-                       else
-                               seq_printf
-                                   (s, "scsi%d chan=%d id=%d lun=%d   em=%d",
-                                    scsidp->host->host_no,
-                                    scsidp->channel, scsidp->id,
-                                    scsidp->lun,
-                                    scsidp->host->hostt->emulated);
-                       seq_printf(s, " sg_tablesize=%d excl=%d\n",
-                                  sdp->sg_tablesize, sdp->exclude);
-               }
+               seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
+               if (sdp->detached)
+                       seq_printf(s, "detached pending close ");
+               else
+                       seq_printf
+                           (s, "scsi%d chan=%d id=%d lun=%d   em=%d",
+                            scsidp->host->host_no,
+                            scsidp->channel, scsidp->id,
+                            scsidp->lun,
+                            scsidp->host->hostt->emulated);
+               seq_printf(s, " sg_tablesize=%d excl=%d\n",
+                          sdp->sg_tablesize, sdp->exclude);
                sg_proc_debug_helper(s, sdp);
        }
+       read_unlock_irqrestore(&sg_index_lock, iflags);
        return 0;
 }