[SCSI] Revert "sg: push file descriptor list locking down to per-device locking"
authorJames Bottomley <JBottomley@Parallels.com>
Fri, 25 Oct 2013 09:21:57 +0000 (10:21 +0100)
committerJames Bottomley <JBottomley@Parallels.com>
Fri, 25 Oct 2013 09:58:07 +0000 (10:58 +0100)
This reverts commit 1f962ebcdfa15cede59e9edb299d1330949eec92.

This is one of four patches that was causing this bug

[  205.372823] ================================================
[  205.372901] [ BUG: lock held when returning to user space! ]
[  205.372979] 3.12.0-rc6-hw-debug-pagealloc+ #67 Not tainted
[  205.373055] ------------------------------------------------
[  205.373132] megarc.bin/5283 is leaving the kernel with locks still held!
[  205.373212] 1 lock held by megarc.bin/5283:
[  205.373285]  #0:  (&sdp->o_sem){.+.+..}, at: [<ffffffff8161e650>] sg_open+0x3a0/0x4d0

Cc: Vaughan Cao <vaughan.cao@oracle.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
drivers/scsi/sg.c

index 5cbc4bb..64df1ab 100644 (file)
@@ -106,7 +106,8 @@ static int sg_add(struct device *, struct class_interface *);
 static void sg_remove(struct device *, struct class_interface *);
 
 static DEFINE_IDR(sg_index_idr);
-static DEFINE_RWLOCK(sg_index_lock);
+static DEFINE_RWLOCK(sg_index_lock);   /* Also used to lock
+                                                          file descriptor list for device */
 
 static struct class_interface sg_interface = {
        .add_dev        = sg_add,
@@ -143,7 +144,8 @@ typedef struct sg_request { /* SG_MAX_QUEUE requests outstanding per file */
 } Sg_request;
 
 typedef struct sg_fd {         /* holds the state of a file descriptor */
-       struct list_head sfd_siblings; /* protected by sfd_lock of device */
+       /* sfd_siblings is protected by sg_index_lock */
+       struct list_head sfd_siblings;
        struct sg_device *parentdp;     /* owning device */
        wait_queue_head_t read_wait;    /* queue read until command done */
        rwlock_t rq_list_lock;  /* protect access to list in req_arr */
@@ -168,7 +170,7 @@ typedef struct sg_device { /* holds the state of each scsi generic device */
        struct scsi_device *device;
        int sg_tablesize;       /* adapter's max scatter-gather table size */
        u32 index;              /* device index number */
-       spinlock_t sfd_lock;    /* protect file descriptor list for device */
+       /* sfds is protected by sg_index_lock */
        struct list_head sfds;
        struct rw_semaphore o_sem;      /* exclude open should hold this rwsem */
        volatile char detached; /* 0->attached, 1->detached pending removal */
@@ -225,9 +227,9 @@ static int sfds_list_empty(Sg_device *sdp)
        unsigned long flags;
        int ret;
 
-       spin_lock_irqsave(&sdp->sfd_lock, flags);
+       read_lock_irqsave(&sg_index_lock, flags);
        ret = list_empty(&sdp->sfds);
-       spin_unlock_irqrestore(&sdp->sfd_lock, flags);
+       read_unlock_irqrestore(&sg_index_lock, flags);
        return ret;
 }
 
@@ -1391,7 +1393,6 @@ static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
        disk->first_minor = k;
        sdp->disk = disk;
        sdp->device = scsidp;
-       spin_lock_init(&sdp->sfd_lock);
        INIT_LIST_HEAD(&sdp->sfds);
        init_rwsem(&sdp->o_sem);
        sdp->sg_tablesize = queue_max_segments(q);
@@ -1526,13 +1527,11 @@ static void sg_remove(struct device *cl_dev, struct class_interface *cl_intf)
 
        /* Need a write lock to set sdp->detached. */
        write_lock_irqsave(&sg_index_lock, iflags);
-       spin_lock(&sdp->sfd_lock);
        sdp->detached = 1;
        list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
                wake_up_interruptible(&sfp->read_wait);
                kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
        }
-       spin_unlock(&sdp->sfd_lock);
        write_unlock_irqrestore(&sg_index_lock, iflags);
 
        sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
@@ -2057,13 +2056,13 @@ sg_add_sfp(Sg_device * sdp, int dev)
        sfp->cmd_q = SG_DEF_COMMAND_Q;
        sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
        sfp->parentdp = sdp;
-       spin_lock_irqsave(&sdp->sfd_lock, iflags);
+       write_lock_irqsave(&sg_index_lock, iflags);
        if (sdp->detached) {
-               spin_unlock_irqrestore(&sdp->sfd_lock, iflags);
+               write_unlock_irqrestore(&sg_index_lock, iflags);
                return ERR_PTR(-ENODEV);
        }
        list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
-       spin_unlock_irqrestore(&sdp->sfd_lock, iflags);
+       write_unlock_irqrestore(&sg_index_lock, iflags);
        SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp: sfp=0x%p\n", sfp));
        if (unlikely(sg_big_buff != def_reserved_size))
                sg_big_buff = def_reserved_size;
@@ -2110,12 +2109,11 @@ static void sg_remove_sfp_usercontext(struct work_struct *work)
 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;
        unsigned long iflags;
 
-       spin_lock_irqsave(&sdp->sfd_lock, iflags);
+       write_lock_irqsave(&sg_index_lock, iflags);
        list_del(&sfp->sfd_siblings);
-       spin_unlock_irqrestore(&sdp->sfd_lock, iflags);
+       write_unlock_irqrestore(&sg_index_lock, iflags);
 
        INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
        schedule_work(&sfp->ew.work);
@@ -2502,7 +2500,7 @@ static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
        return 0;
 }
 
-/* must be called while holding sg_index_lock and sfd_lock */
+/* 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;
@@ -2587,26 +2585,22 @@ static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
 
        read_lock_irqsave(&sg_index_lock, iflags);
        sdp = it ? sg_lookup_dev(it->index) : NULL;
-       if (sdp) {
-               spin_lock(&sdp->sfd_lock);
-               if (!list_empty(&sdp->sfds)) {
-                       struct scsi_device *scsidp = sdp->device;
+       if (sdp && !list_empty(&sdp->sfds)) {
+               struct scsi_device *scsidp = sdp->device;
 
-                       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);
-               }
-               spin_unlock(&sdp->sfd_lock);
+               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;