bsg-lib: fix kernel panic resulting from missing allocation of reply-buffer
authorBenjamin Block <bblock@linux.vnet.ibm.com>
Wed, 23 Aug 2017 23:57:56 +0000 (01:57 +0200)
committerJens Axboe <axboe@kernel.dk>
Thu, 24 Aug 2017 14:22:10 +0000 (08:22 -0600)
Since we split the scsi_request out of struct request bsg fails to
provide a reply-buffer for the drivers. This was done via the pointer
for sense-data, that is not preallocated anymore.

Failing to allocate/assign it results in illegal dereferences because
LLDs use this pointer unquestioned.

An example panic on s390x, using the zFCP driver, looks like this (I had
debugging on, otherwise NULL-pointer dereferences wouldn't even panic on
s390x):

Unable to handle kernel pointer dereference in virtual kernel address space
Failing address: 6b6b6b6b6b6b6000 TEID: 6b6b6b6b6b6b6403
Fault in home space mode while using kernel ASCE.
AS:0000000001590007 R3:0000000000000024
Oops: 0038 ilc:2 [#1] PREEMPT SMP DEBUG_PAGEALLOC
Modules linked in: <Long List>
CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.12.0-bsg-regression+ #3
Hardware name: IBM 2964 N96 702 (z/VM 6.4.0)
task: 0000000065cb0100 task.stack: 0000000065cb4000
Krnl PSW : 0704e00180000000 000003ff801e4156 (zfcp_fc_ct_els_job_handler+0x16/0x58 [zfcp])
           R:0 T:1 IO:1 EX:1 Key:0 M:1 W:0 P:0 AS:3 CC:2 PM:0 RI:0 EA:3
Krnl GPRS: 0000000000000001 000000005fa9d0d0 000000005fa9d078 0000000000e16866
           000003ff00000290 6b6b6b6b6b6b6b6b 0000000059f78f00 000000000000000f
           00000000593a0958 00000000593a0958 0000000060d88800 000000005ddd4c38
           0000000058b50100 07000000659cba08 000003ff801e8556 00000000659cb9a8
Krnl Code: 000003ff801e4146e31020500004        lg      %r1,80(%r2)
           000003ff801e414c58402040           l       %r4,64(%r2)
          #000003ff801e4150e35020200004       lg      %r5,32(%r2)
          >000003ff801e415650405004           st      %r4,4(%r5)
           000003ff801e415ae54c50080000       mvhi    8(%r5),0
           000003ff801e4160e33010280012       lt      %r3,40(%r1)
           000003ff801e4166a718fffb           lhi     %r1,-5
           000003ff801e416a: 1803               lr      %r0,%r3
Call Trace:
([<000003ff801e8556>] zfcp_fsf_req_complete+0x726/0x768 [zfcp])
 [<000003ff801ea82a>] zfcp_fsf_reqid_check+0x102/0x180 [zfcp]
 [<000003ff801eb980>] zfcp_qdio_int_resp+0x230/0x278 [zfcp]
 [<00000000009b91b6>] qdio_kick_handler+0x2ae/0x2c8
 [<00000000009b9e3e>] __tiqdio_inbound_processing+0x406/0xc10
 [<00000000001684c2>] tasklet_action+0x15a/0x1d8
 [<0000000000bd28ec>] __do_softirq+0x3ec/0x848
 [<00000000001675a4>] irq_exit+0x74/0xf8
 [<000000000010dd6a>] do_IRQ+0xba/0xf0
 [<0000000000bd19e8>] io_int_handler+0x104/0x2d4
 [<00000000001033b6>] enabled_wait+0xb6/0x188
([<000000000010339e>] enabled_wait+0x9e/0x188)
 [<000000000010396a>] arch_cpu_idle+0x32/0x50
 [<0000000000bd0112>] default_idle_call+0x52/0x68
 [<00000000001cd0fa>] do_idle+0x102/0x188
 [<00000000001cd41e>] cpu_startup_entry+0x3e/0x48
 [<0000000000118c64>] smp_start_secondary+0x11c/0x130
 [<0000000000bd2016>] restart_int_handler+0x62/0x78
 [<0000000000000000>]           (null)
INFO: lockdep is turned off.
Last Breaking-Event-Address:
 [<000003ff801e41d6>] zfcp_fc_ct_job_handler+0x3e/0x48 [zfcp]

Kernel panic - not syncing: Fatal exception in interrupt

This patch moves bsg-lib to allocate and setup struct bsg_job ahead of
time, including the allocation of a buffer for the reply-data.

This means, struct bsg_job is not allocated separately anymore, but as part
of struct request allocation - similar to struct scsi_cmd. Reflect this in
the function names that used to handle creation/destruction of struct
bsg_job.

Reported-by: Steffen Maier <maier@linux.vnet.ibm.com>
Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Benjamin Block <bblock@linux.vnet.ibm.com>
Fixes: 82ed4db499b8 ("block: split scsi_request out of struct request")
Cc: <stable@vger.kernel.org> #4.11+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/bsg-lib.c
include/linux/blkdev.h
include/linux/bsg-lib.h

index c4513b23f57a6438af6ae38367c072931edf138c..dd56d7460cb91d504aa22ae9215d00f97c844359 100644 (file)
 #include <scsi/scsi_cmnd.h>
 
 /**
- * bsg_destroy_job - routine to teardown/delete a bsg job
+ * bsg_teardown_job - routine to teardown a bsg job
  * @job: bsg_job that is to be torn down
  */
-static void bsg_destroy_job(struct kref *kref)
+static void bsg_teardown_job(struct kref *kref)
 {
        struct bsg_job *job = container_of(kref, struct bsg_job, kref);
        struct request *rq = job->req;
 
-       blk_end_request_all(rq, BLK_STS_OK);
-
        put_device(job->dev);   /* release reference for the request */
 
        kfree(job->request_payload.sg_list);
        kfree(job->reply_payload.sg_list);
-       kfree(job);
+
+       blk_end_request_all(rq, BLK_STS_OK);
 }
 
 void bsg_job_put(struct bsg_job *job)
 {
-       kref_put(&job->kref, bsg_destroy_job);
+       kref_put(&job->kref, bsg_teardown_job);
 }
 EXPORT_SYMBOL_GPL(bsg_job_put);
 
@@ -100,7 +99,7 @@ EXPORT_SYMBOL_GPL(bsg_job_done);
  */
 static void bsg_softirq_done(struct request *rq)
 {
-       struct bsg_job *job = rq->special;
+       struct bsg_job *job = blk_mq_rq_to_pdu(rq);
 
        bsg_job_put(job);
 }
@@ -122,33 +121,20 @@ static int bsg_map_buffer(struct bsg_buffer *buf, struct request *req)
 }
 
 /**
- * bsg_create_job - create the bsg_job structure for the bsg request
+ * bsg_prepare_job - create the bsg_job structure for the bsg request
  * @dev: device that is being sent the bsg request
  * @req: BSG request that needs a job structure
  */
-static int bsg_create_job(struct device *dev, struct request *req)
+static int bsg_prepare_job(struct device *dev, struct request *req)
 {
        struct request *rsp = req->next_rq;
-       struct request_queue *q = req->q;
        struct scsi_request *rq = scsi_req(req);
-       struct bsg_job *job;
+       struct bsg_job *job = blk_mq_rq_to_pdu(req);
        int ret;
 
-       BUG_ON(req->special);
-
-       job = kzalloc(sizeof(struct bsg_job) + q->bsg_job_size, GFP_KERNEL);
-       if (!job)
-               return -ENOMEM;
-
-       req->special = job;
-       job->req = req;
-       if (q->bsg_job_size)
-               job->dd_data = (void *)&job[1];
        job->request = rq->cmd;
        job->request_len = rq->cmd_len;
-       job->reply = rq->sense;
-       job->reply_len = SCSI_SENSE_BUFFERSIZE; /* Size of sense buffer
-                                                * allocated */
+
        if (req->bio) {
                ret = bsg_map_buffer(&job->request_payload, req);
                if (ret)
@@ -187,7 +173,6 @@ static void bsg_request_fn(struct request_queue *q)
 {
        struct device *dev = q->queuedata;
        struct request *req;
-       struct bsg_job *job;
        int ret;
 
        if (!get_device(dev))
@@ -199,7 +184,7 @@ static void bsg_request_fn(struct request_queue *q)
                        break;
                spin_unlock_irq(q->queue_lock);
 
-               ret = bsg_create_job(dev, req);
+               ret = bsg_prepare_job(dev, req);
                if (ret) {
                        scsi_req(req)->result = ret;
                        blk_end_request_all(req, BLK_STS_OK);
@@ -207,8 +192,7 @@ static void bsg_request_fn(struct request_queue *q)
                        continue;
                }
 
-               job = req->special;
-               ret = q->bsg_job_fn(job);
+               ret = q->bsg_job_fn(blk_mq_rq_to_pdu(req));
                spin_lock_irq(q->queue_lock);
                if (ret)
                        break;
@@ -219,6 +203,35 @@ static void bsg_request_fn(struct request_queue *q)
        spin_lock_irq(q->queue_lock);
 }
 
+static int bsg_init_rq(struct request_queue *q, struct request *req, gfp_t gfp)
+{
+       struct bsg_job *job = blk_mq_rq_to_pdu(req);
+       struct scsi_request *sreq = &job->sreq;
+
+       memset(job, 0, sizeof(*job));
+
+       scsi_req_init(sreq);
+       sreq->sense_len = SCSI_SENSE_BUFFERSIZE;
+       sreq->sense = kzalloc(sreq->sense_len, gfp);
+       if (!sreq->sense)
+               return -ENOMEM;
+
+       job->req = req;
+       job->reply = sreq->sense;
+       job->reply_len = sreq->sense_len;
+       job->dd_data = job + 1;
+
+       return 0;
+}
+
+static void bsg_exit_rq(struct request_queue *q, struct request *req)
+{
+       struct bsg_job *job = blk_mq_rq_to_pdu(req);
+       struct scsi_request *sreq = &job->sreq;
+
+       kfree(sreq->sense);
+}
+
 /**
  * bsg_setup_queue - Create and add the bsg hooks so we can receive requests
  * @dev: device to attach bsg device to
@@ -235,7 +248,9 @@ struct request_queue *bsg_setup_queue(struct device *dev, char *name,
        q = blk_alloc_queue(GFP_KERNEL);
        if (!q)
                return ERR_PTR(-ENOMEM);
-       q->cmd_size = sizeof(struct scsi_request);
+       q->cmd_size = sizeof(struct bsg_job) + dd_job_size;
+       q->init_rq_fn = bsg_init_rq;
+       q->exit_rq_fn = bsg_exit_rq;
        q->request_fn = bsg_request_fn;
 
        ret = blk_init_allocated_queue(q);
@@ -243,7 +258,6 @@ struct request_queue *bsg_setup_queue(struct device *dev, char *name,
                goto out_cleanup_queue;
 
        q->queuedata = dev;
-       q->bsg_job_size = dd_job_size;
        q->bsg_job_fn = job_fn;
        queue_flag_set_unlocked(QUEUE_FLAG_BIDI, q);
        queue_flag_set_unlocked(QUEUE_FLAG_SCSI_PASSTHROUGH, q);
index 25f6a0cb27d3e9644c1ef42e7f1ff2b5d817293d..2a5d52fa90f5da353086a0faf5c8a5cdea63128a 100644 (file)
@@ -568,7 +568,6 @@ struct request_queue {
 
 #if defined(CONFIG_BLK_DEV_BSG)
        bsg_job_fn              *bsg_job_fn;
-       int                     bsg_job_size;
        struct bsg_class_device bsg_dev;
 #endif
 
index e34dde2da0ef57c3692ce7001fa66fe37c4dc578..637a20cfb237db4ab76a480172e918e6e48f849b 100644 (file)
@@ -24,6 +24,7 @@
 #define _BLK_BSG_
 
 #include <linux/blkdev.h>
+#include <scsi/scsi_request.h>
 
 struct request;
 struct device;
@@ -37,6 +38,7 @@ struct bsg_buffer {
 };
 
 struct bsg_job {
+       struct scsi_request sreq;
        struct device *dev;
        struct request *req;