Merge branch 'cpumask-cleanups' of git://git.kernel.org/pub/scm/linux/kernel/git...
[safe/jmp/linux-2.6] / drivers / scsi / sd.c
index a89c421..255da53 100644 (file)
@@ -116,6 +116,9 @@ static DEFINE_IDA(sd_index_ida);
  * object after last put) */
 static DEFINE_MUTEX(sd_ref_mutex);
 
+struct kmem_cache *sd_cdb_cache;
+mempool_t *sd_cdb_pool;
+
 static const char *sd_cache_types[] = {
        "write through", "none", "write back",
        "write back, no read (daft)"
@@ -261,6 +264,15 @@ sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
        return snprintf(buf, 20, "%u\n", sdkp->ATO);
 }
 
+static ssize_t
+sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
+                         char *buf)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(dev);
+
+       return snprintf(buf, 20, "%u\n", sdkp->thin_provisioning);
+}
+
 static struct device_attribute sd_disk_attrs[] = {
        __ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
               sd_store_cache_type),
@@ -271,6 +283,7 @@ static struct device_attribute sd_disk_attrs[] = {
               sd_store_manage_start_stop),
        __ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
        __ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
+       __ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
        __ATTR_NULL,
 };
 
@@ -370,6 +383,82 @@ static void scsi_disk_put(struct scsi_disk *sdkp)
        mutex_unlock(&sd_ref_mutex);
 }
 
+static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
+{
+       unsigned int prot_op = SCSI_PROT_NORMAL;
+       unsigned int dix = scsi_prot_sg_count(scmd);
+
+       if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
+               if (dif && dix)
+                       prot_op = SCSI_PROT_READ_PASS;
+               else if (dif && !dix)
+                       prot_op = SCSI_PROT_READ_STRIP;
+               else if (!dif && dix)
+                       prot_op = SCSI_PROT_READ_INSERT;
+       } else {
+               if (dif && dix)
+                       prot_op = SCSI_PROT_WRITE_PASS;
+               else if (dif && !dix)
+                       prot_op = SCSI_PROT_WRITE_INSERT;
+               else if (!dif && dix)
+                       prot_op = SCSI_PROT_WRITE_STRIP;
+       }
+
+       scsi_set_prot_op(scmd, prot_op);
+       scsi_set_prot_type(scmd, dif);
+}
+
+/**
+ * sd_prepare_discard - unmap blocks on thinly provisioned device
+ * @rq: Request to prepare
+ *
+ * Will issue either UNMAP or WRITE SAME(16) depending on preference
+ * indicated by target device.
+ **/
+static int sd_prepare_discard(struct request *rq)
+{
+       struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
+       struct bio *bio = rq->bio;
+       sector_t sector = bio->bi_sector;
+       unsigned int num = bio_sectors(bio);
+
+       if (sdkp->device->sector_size == 4096) {
+               sector >>= 3;
+               num >>= 3;
+       }
+
+       rq->cmd_type = REQ_TYPE_BLOCK_PC;
+       rq->timeout = SD_TIMEOUT;
+
+       memset(rq->cmd, 0, rq->cmd_len);
+
+       if (sdkp->unmap) {
+               char *buf = kmap_atomic(bio_page(bio), KM_USER0);
+
+               rq->cmd[0] = UNMAP;
+               rq->cmd[8] = 24;
+               rq->cmd_len = 10;
+
+               /* Ensure that data length matches payload */
+               rq->__data_len = bio->bi_size = bio->bi_io_vec->bv_len = 24;
+
+               put_unaligned_be16(6 + 16, &buf[0]);
+               put_unaligned_be16(16, &buf[2]);
+               put_unaligned_be64(sector, &buf[8]);
+               put_unaligned_be32(num, &buf[16]);
+
+               kunmap_atomic(buf, KM_USER0);
+       } else {
+               rq->cmd[0] = WRITE_SAME_16;
+               rq->cmd[1] = 0x8; /* UNMAP */
+               put_unaligned_be64(sector, &rq->cmd[2]);
+               put_unaligned_be32(num, &rq->cmd[10]);
+               rq->cmd_len = 16;
+       }
+
+       return BLKPREP_OK;
+}
+
 /**
  *     sd_init_command - build a scsi (read or write) command from
  *     information in the request structure.
@@ -388,6 +477,14 @@ static int sd_prep_fn(struct request_queue *q, struct request *rq)
        sector_t threshold;
        unsigned int this_count = blk_rq_sectors(rq);
        int ret, host_dif;
+       unsigned char protect;
+
+       /*
+        * Discard request come in as REQ_TYPE_FS but we turn them into
+        * block PC requests to make life easier.
+        */
+       if (blk_discard_rq(rq))
+               ret = sd_prepare_discard(rq);
 
        if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
                ret = scsi_setup_blk_pc_cmnd(sdp, rq);
@@ -520,13 +617,49 @@ static int sd_prep_fn(struct request_queue *q, struct request *rq)
        /* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
        host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
        if (host_dif)
-               SCpnt->cmnd[1] = 1 << 5;
+               protect = 1 << 5;
        else
-               SCpnt->cmnd[1] = 0;
+               protect = 0;
+
+       if (host_dif == SD_DIF_TYPE2_PROTECTION) {
+               SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
 
-       if (block > 0xffffffff) {
+               if (unlikely(SCpnt->cmnd == NULL)) {
+                       ret = BLKPREP_DEFER;
+                       goto out;
+               }
+
+               SCpnt->cmd_len = SD_EXT_CDB_SIZE;
+               memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
+               SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
+               SCpnt->cmnd[7] = 0x18;
+               SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
+               SCpnt->cmnd[10] = protect | (blk_fua_rq(rq) ? 0x8 : 0);
+
+               /* LBA */
+               SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
+               SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
+               SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
+               SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
+               SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
+               SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
+               SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
+               SCpnt->cmnd[19] = (unsigned char) block & 0xff;
+
+               /* Expected Indirect LBA */
+               SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
+               SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
+               SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
+               SCpnt->cmnd[23] = (unsigned char) block & 0xff;
+
+               /* Transfer length */
+               SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
+               SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
+               SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
+               SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
+       } else if (block > 0xffffffff) {
                SCpnt->cmnd[0] += READ_16 - READ_6;
-               SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
+               SCpnt->cmnd[1] = protect | (blk_fua_rq(rq) ? 0x8 : 0);
                SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
                SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
                SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
@@ -547,7 +680,7 @@ static int sd_prep_fn(struct request_queue *q, struct request *rq)
                        this_count = 0xffff;
 
                SCpnt->cmnd[0] += READ_10 - READ_6;
-               SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
+               SCpnt->cmnd[1] = protect | (blk_fua_rq(rq) ? 0x8 : 0);
                SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
                SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
                SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
@@ -578,8 +711,7 @@ static int sd_prep_fn(struct request_queue *q, struct request *rq)
 
        /* If DIF or DIX is enabled, tell HBA how to handle request */
        if (host_dif || scsi_prot_sg_count(SCpnt))
-               sd_dif_op(SCpnt, host_dif, scsi_prot_sg_count(SCpnt),
-                         sdkp->protection_type);
+               sd_prot_op(SCpnt, host_dif);
 
        /*
         * We shouldn't disconnect in the middle of a sector, so with a dumb
@@ -956,7 +1088,7 @@ static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
 }
 #endif
 
-static struct block_device_operations sd_fops = {
+static const struct block_device_operations sd_fops = {
        .owner                  = THIS_MODULE,
        .open                   = sd_open,
        .release                = sd_release,
@@ -1023,6 +1155,7 @@ static int sd_done(struct scsi_cmnd *SCpnt)
        int result = SCpnt->result;
        unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
        struct scsi_sense_hdr sshdr;
+       struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
        int sense_valid = 0;
        int sense_deferred = 0;
 
@@ -1084,6 +1217,10 @@ static int sd_done(struct scsi_cmnd *SCpnt)
        if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
                sd_dif_complete(SCpnt, good_bytes);
 
+       if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
+           == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd)
+               mempool_free(SCpnt->cmnd, sd_cdb_pool);
+
        return good_bytes;
 }
 
@@ -1238,34 +1375,28 @@ void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
        u8 type;
 
        if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
-               type = 0;
-       else
-               type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
+               return;
 
-       sdkp->protection_type = type;
+       type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
 
-       switch (type) {
-       case SD_DIF_TYPE0_PROTECTION:
-       case SD_DIF_TYPE1_PROTECTION:
-       case SD_DIF_TYPE3_PROTECTION:
-               break;
+       if (type == sdkp->protection_type || !sdkp->first_scan)
+               return;
 
-       case SD_DIF_TYPE2_PROTECTION:
-               sd_printk(KERN_ERR, sdkp, "formatted with DIF Type 2 "  \
-                         "protection which is currently unsupported. " \
-                         "Disabling disk!\n");
-               goto disable;
+       sdkp->protection_type = type;
 
-       default:
-               sd_printk(KERN_ERR, sdkp, "formatted with unknown "     \
-                         "protection type %d. Disabling disk!\n", type);
-               goto disable;
+       if (type > SD_DIF_TYPE3_PROTECTION) {
+               sd_printk(KERN_ERR, sdkp, "formatted with unsupported " \
+                         "protection type %u. Disabling disk!\n", type);
+               sdkp->capacity = 0;
+               return;
        }
 
-       return;
-
-disable:
-       sdkp->capacity = 0;
+       if (scsi_host_dif_capable(sdp->host, type))
+               sd_printk(KERN_NOTICE, sdkp,
+                         "Enabling DIF Type %u protection\n", type);
+       else
+               sd_printk(KERN_NOTICE, sdkp,
+                         "Disabling DIF Type %u protection\n", type);
 }
 
 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
@@ -1369,6 +1500,19 @@ static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
                sd_printk(KERN_NOTICE, sdkp,
                          "physical block alignment offset: %u\n", alignment);
 
+       if (buffer[14] & 0x80) { /* TPE */
+               struct request_queue *q = sdp->request_queue;
+
+               sdkp->thin_provisioning = 1;
+               q->limits.discard_granularity = sdkp->hw_sector_size;
+               q->limits.max_discard_sectors = 0xffffffff;
+
+               if (buffer[14] & 0x40) /* TPRZ */
+                       q->limits.discard_zeroes_data = 1;
+
+               queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
+       }
+
        sdkp->capacity = lba + 1;
        return sector_size;
 }
@@ -1800,6 +1944,7 @@ void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
  */
 static void sd_read_block_limits(struct scsi_disk *sdkp)
 {
+       struct request_queue *q = sdkp->disk->queue;
        unsigned int sector_sz = sdkp->device->sector_size;
        char *buffer;
 
@@ -1814,6 +1959,31 @@ static void sd_read_block_limits(struct scsi_disk *sdkp)
        blk_queue_io_opt(sdkp->disk->queue,
                         get_unaligned_be32(&buffer[12]) * sector_sz);
 
+       /* Thin provisioning enabled and page length indicates TP support */
+       if (sdkp->thin_provisioning && buffer[3] == 0x3c) {
+               unsigned int lba_count, desc_count, granularity;
+
+               lba_count = get_unaligned_be32(&buffer[20]);
+               desc_count = get_unaligned_be32(&buffer[24]);
+
+               if (lba_count) {
+                       q->limits.max_discard_sectors =
+                               lba_count * sector_sz >> 9;
+
+                       if (desc_count)
+                               sdkp->unmap = 1;
+               }
+
+               granularity = get_unaligned_be32(&buffer[28]);
+
+               if (granularity)
+                       q->limits.discard_granularity = granularity * sector_sz;
+
+               if (buffer[32] & 0x80)
+                       q->limits.discard_alignment =
+                               get_unaligned_be32(&buffer[32]) & ~(1 << 31);
+       }
+
        kfree(buffer);
 }
 
@@ -2300,8 +2470,24 @@ static int __init init_sd(void)
        if (err)
                goto err_out_class;
 
+       sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
+                                        0, 0, NULL);
+       if (!sd_cdb_cache) {
+               printk(KERN_ERR "sd: can't init extended cdb cache\n");
+               goto err_out_class;
+       }
+
+       sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
+       if (!sd_cdb_pool) {
+               printk(KERN_ERR "sd: can't init extended cdb pool\n");
+               goto err_out_cache;
+       }
+
        return 0;
 
+err_out_cache:
+       kmem_cache_destroy(sd_cdb_cache);
+
 err_out_class:
        class_unregister(&sd_disk_class);
 err_out:
@@ -2321,6 +2507,9 @@ static void __exit exit_sd(void)
 
        SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
 
+       mempool_destroy(sd_cdb_pool);
+       kmem_cache_destroy(sd_cdb_cache);
+
        scsi_unregister_driver(&sd_template.gendrv);
        class_unregister(&sd_disk_class);