Commit e96e72c45a1e78e9266dd70113b851395a440ef3

Authored by Boaz Harrosh
Committed by James Bottomley
1 parent c4df46c49d

[SCSI] libosd: Support for scatter gather write/read commands

This patch adds the Scatter-Gather (sg) API to libosd.
Scatter-gather enables a write/read of multiple none-contiguous
areas of an object, in a single call. The extents may overlap
and/or be in any order.

The Scatter-Gather list is sent to the target in what is called
a "cdb continuation segment". This is yet another possible segment
in the osd-out-buffer. It is unlike all other segments in that it
sits before the actual "data" segment (which until now was always
first), and that it is signed by itself and not part of the data
buffer. This is because the cdb-continuation-segment is considered
a spill-over of the CDB data, and is therefor signed under
OSD_SEC_CAPKEY and higher.

TODO: A new osd_finalize_request_ex version should be supplied so
the @caps received on the network also contains a size parameter
and can be spilled over into the "cdb continuation segment".

Thanks to John Chandy <john.chandy@uconn.edu> for the original
code, and investigations. And the implementation of SG support
in the osd-target.

Original-coded-by: John Chandy <john.chandy@uconn.edu>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>

Showing 4 changed files with 198 additions and 6 deletions Side-by-side Diff

drivers/scsi/osd/osd_initiator.c
... ... @@ -464,6 +464,7 @@
464 464 _osd_free_seg(or, &or->get_attr);
465 465 _osd_free_seg(or, &or->enc_get_attr);
466 466 _osd_free_seg(or, &or->set_attr);
  467 + _osd_free_seg(or, &or->cdb_cont);
467 468  
468 469 _osd_request_free(or);
469 470 }
... ... @@ -548,6 +549,12 @@
548 549 return 0;
549 550 }
550 551  
  552 +static int _alloc_cdb_cont(struct osd_request *or, unsigned total_bytes)
  553 +{
  554 + OSD_DEBUG("total_bytes=%d\n", total_bytes);
  555 + return _osd_realloc_seg(or, &or->cdb_cont, total_bytes);
  556 +}
  557 +
551 558 static int _alloc_set_attr_list(struct osd_request *or,
552 559 const struct osd_attr *oa, unsigned nelem, unsigned add_bytes)
553 560 {
... ... @@ -886,6 +893,128 @@
886 893 }
887 894 EXPORT_SYMBOL(osd_req_read_kern);
888 895  
  896 +static int _add_sg_continuation_descriptor(struct osd_request *or,
  897 + const struct osd_sg_entry *sglist, unsigned numentries, u64 *len)
  898 +{
  899 + struct osd_sg_continuation_descriptor *oscd;
  900 + u32 oscd_size;
  901 + unsigned i;
  902 + int ret;
  903 +
  904 + oscd_size = sizeof(*oscd) + numentries * sizeof(oscd->entries[0]);
  905 +
  906 + if (!or->cdb_cont.total_bytes) {
  907 + /* First time, jump over the header, we will write to:
  908 + * cdb_cont.buff + cdb_cont.total_bytes
  909 + */
  910 + or->cdb_cont.total_bytes =
  911 + sizeof(struct osd_continuation_segment_header);
  912 + }
  913 +
  914 + ret = _alloc_cdb_cont(or, or->cdb_cont.total_bytes + oscd_size);
  915 + if (unlikely(ret))
  916 + return ret;
  917 +
  918 + oscd = or->cdb_cont.buff + or->cdb_cont.total_bytes;
  919 + oscd->hdr.type = cpu_to_be16(SCATTER_GATHER_LIST);
  920 + oscd->hdr.pad_length = 0;
  921 + oscd->hdr.length = cpu_to_be32(oscd_size - sizeof(*oscd));
  922 +
  923 + *len = 0;
  924 + /* copy the sg entries and convert to network byte order */
  925 + for (i = 0; i < numentries; i++) {
  926 + oscd->entries[i].offset = cpu_to_be64(sglist[i].offset);
  927 + oscd->entries[i].len = cpu_to_be64(sglist[i].len);
  928 + *len += sglist[i].len;
  929 + }
  930 +
  931 + or->cdb_cont.total_bytes += oscd_size;
  932 + OSD_DEBUG("total_bytes=%d oscd_size=%d numentries=%d\n",
  933 + or->cdb_cont.total_bytes, oscd_size, numentries);
  934 + return 0;
  935 +}
  936 +
  937 +static int _osd_req_finalize_cdb_cont(struct osd_request *or, const u8 *cap_key)
  938 +{
  939 + struct request_queue *req_q = osd_request_queue(or->osd_dev);
  940 + struct bio *bio;
  941 + struct osd_cdb_head *cdbh = osd_cdb_head(&or->cdb);
  942 + struct osd_continuation_segment_header *cont_seg_hdr;
  943 +
  944 + if (!or->cdb_cont.total_bytes)
  945 + return 0;
  946 +
  947 + cont_seg_hdr = or->cdb_cont.buff;
  948 + cont_seg_hdr->format = CDB_CONTINUATION_FORMAT_V2;
  949 + cont_seg_hdr->service_action = cdbh->varlen_cdb.service_action;
  950 +
  951 + /* create a bio for continuation segment */
  952 + bio = bio_map_kern(req_q, or->cdb_cont.buff, or->cdb_cont.total_bytes,
  953 + GFP_KERNEL);
  954 + if (unlikely(!bio))
  955 + return -ENOMEM;
  956 +
  957 + bio->bi_rw |= REQ_WRITE;
  958 +
  959 + /* integrity check the continuation before the bio is linked
  960 + * with the other data segments since the continuation
  961 + * integrity is separate from the other data segments.
  962 + */
  963 + osd_sec_sign_data(cont_seg_hdr->integrity_check, bio, cap_key);
  964 +
  965 + cdbh->v2.cdb_continuation_length = cpu_to_be32(or->cdb_cont.total_bytes);
  966 +
  967 + /* we can't use _req_append_segment, because we need to link in the
  968 + * continuation bio to the head of the bio list - the
  969 + * continuation segment (if it exists) is always the first segment in
  970 + * the out data buffer.
  971 + */
  972 + bio->bi_next = or->out.bio;
  973 + or->out.bio = bio;
  974 + or->out.total_bytes += or->cdb_cont.total_bytes;
  975 +
  976 + return 0;
  977 +}
  978 +
  979 +/* osd_req_write_sg: Takes a @bio that points to the data out buffer and an
  980 + * @sglist that has the scatter gather entries. Scatter-gather enables a write
  981 + * of multiple none-contiguous areas of an object, in a single call. The extents
  982 + * may overlap and/or be in any order. The only constrain is that:
  983 + * total_bytes(sglist) >= total_bytes(bio)
  984 + */
  985 +int osd_req_write_sg(struct osd_request *or,
  986 + const struct osd_obj_id *obj, struct bio *bio,
  987 + const struct osd_sg_entry *sglist, unsigned numentries)
  988 +{
  989 + u64 len;
  990 + int ret = _add_sg_continuation_descriptor(or, sglist, numentries, &len);
  991 +
  992 + if (ret)
  993 + return ret;
  994 + osd_req_write(or, obj, 0, bio, len);
  995 +
  996 + return 0;
  997 +}
  998 +EXPORT_SYMBOL(osd_req_write_sg);
  999 +
  1000 +/* osd_req_read_sg: Read multiple extents of an object into @bio
  1001 + * See osd_req_write_sg
  1002 + */
  1003 +int osd_req_read_sg(struct osd_request *or,
  1004 + const struct osd_obj_id *obj, struct bio *bio,
  1005 + const struct osd_sg_entry *sglist, unsigned numentries)
  1006 +{
  1007 + u64 len;
  1008 + int ret = _add_sg_continuation_descriptor(or, sglist, numentries, &len);
  1009 +
  1010 + if (ret)
  1011 + return ret;
  1012 + osd_req_read(or, obj, 0, bio, len);
  1013 +
  1014 + return 0;
  1015 +}
  1016 +EXPORT_SYMBOL(osd_req_read_sg);
  1017 +
889 1018 void osd_req_get_attributes(struct osd_request *or,
890 1019 const struct osd_obj_id *obj)
891 1020 {
... ... @@ -1281,7 +1410,8 @@
1281 1410 }
1282 1411  
1283 1412 static int _osd_req_finalize_data_integrity(struct osd_request *or,
1284   - bool has_in, bool has_out, u64 out_data_bytes, const u8 *cap_key)
  1413 + bool has_in, bool has_out, struct bio *out_data_bio, u64 out_data_bytes,
  1414 + const u8 *cap_key)
1285 1415 {
1286 1416 struct osd_security_parameters *sec_parms = _osd_req_sec_params(or);
1287 1417 int ret;
... ... @@ -1312,7 +1442,7 @@
1312 1442 or->out.last_seg = NULL;
1313 1443  
1314 1444 /* they are now all chained to request sign them all together */
1315   - osd_sec_sign_data(&or->out_data_integ, or->out.req->bio,
  1445 + osd_sec_sign_data(&or->out_data_integ, out_data_bio,
1316 1446 cap_key);
1317 1447 }
1318 1448  
... ... @@ -1408,6 +1538,8 @@
1408 1538 {
1409 1539 struct osd_cdb_head *cdbh = osd_cdb_head(&or->cdb);
1410 1540 bool has_in, has_out;
  1541 + /* Save for data_integrity without the cdb_continuation */
  1542 + struct bio *out_data_bio = or->out.bio;
1411 1543 u64 out_data_bytes = or->out.total_bytes;
1412 1544 int ret;
1413 1545  
1414 1546  
... ... @@ -1423,9 +1555,14 @@
1423 1555 osd_set_caps(&or->cdb, cap);
1424 1556  
1425 1557 has_in = or->in.bio || or->get_attr.total_bytes;
1426   - has_out = or->out.bio || or->set_attr.total_bytes ||
1427   - or->enc_get_attr.total_bytes;
  1558 + has_out = or->out.bio || or->cdb_cont.total_bytes ||
  1559 + or->set_attr.total_bytes || or->enc_get_attr.total_bytes;
1428 1560  
  1561 + ret = _osd_req_finalize_cdb_cont(or, cap_key);
  1562 + if (ret) {
  1563 + OSD_DEBUG("_osd_req_finalize_cdb_cont failed\n");
  1564 + return ret;
  1565 + }
1429 1566 ret = _init_blk_request(or, has_in, has_out);
1430 1567 if (ret) {
1431 1568 OSD_DEBUG("_init_blk_request failed\n");
... ... @@ -1463,7 +1600,8 @@
1463 1600 }
1464 1601  
1465 1602 ret = _osd_req_finalize_data_integrity(or, has_in, has_out,
1466   - out_data_bytes, cap_key);
  1603 + out_data_bio, out_data_bytes,
  1604 + cap_key);
1467 1605 if (ret)
1468 1606 return ret;
1469 1607  
include/scsi/osd_initiator.h
... ... @@ -137,7 +137,7 @@
137 137 void *buff;
138 138 unsigned alloc_size; /* 0 here means: don't call kfree */
139 139 unsigned total_bytes;
140   - } set_attr, enc_get_attr, get_attr;
  140 + } cdb_cont, set_attr, enc_get_attr, get_attr;
141 141  
142 142 struct _osd_io_info {
143 143 struct bio *bio;
... ... @@ -448,6 +448,13 @@
448 448 int osd_req_read_kern(struct osd_request *or,
449 449 const struct osd_obj_id *obj, u64 offset, void *buff, u64 len);
450 450  
  451 +/* Scatter/Gather write/read commands */
  452 +int osd_req_write_sg(struct osd_request *or,
  453 + const struct osd_obj_id *obj, struct bio *bio,
  454 + const struct osd_sg_entry *sglist, unsigned numentries);
  455 +int osd_req_read_sg(struct osd_request *or,
  456 + const struct osd_obj_id *obj, struct bio *bio,
  457 + const struct osd_sg_entry *sglist, unsigned numentries);
451 458 /*
452 459 * Root/Partition/Collection/Object Attributes commands
453 460 */
include/scsi/osd_protocol.h
... ... @@ -631,5 +631,47 @@
631 631 put_unaligned_le16(bit_mask, &cap->permissions_bit_mask);
632 632 }
633 633  
  634 +/* osd2r05a sec 5.3: CDB continuation segment formats */
  635 +enum osd_continuation_segment_format {
  636 + CDB_CONTINUATION_FORMAT_V2 = 0x01,
  637 +};
  638 +
  639 +struct osd_continuation_segment_header {
  640 + u8 format;
  641 + u8 reserved1;
  642 + __be16 service_action;
  643 + __be32 reserved2;
  644 + u8 integrity_check[OSDv2_CRYPTO_KEYID_SIZE];
  645 +} __packed;
  646 +
  647 +/* osd2r05a sec 5.4.1: CDB continuation descriptors */
  648 +enum osd_continuation_descriptor_type {
  649 + NO_MORE_DESCRIPTORS = 0x0000,
  650 + SCATTER_GATHER_LIST = 0x0001,
  651 + QUERY_LIST = 0x0002,
  652 + USER_OBJECT = 0x0003,
  653 + COPY_USER_OBJECT_SOURCE = 0x0101,
  654 + EXTENSION_CAPABILITIES = 0xFFEE
  655 +};
  656 +
  657 +struct osd_continuation_descriptor_header {
  658 + __be16 type;
  659 + u8 reserved;
  660 + u8 pad_length;
  661 + __be32 length;
  662 +} __packed;
  663 +
  664 +
  665 +/* osd2r05a sec 5.4.2: Scatter/gather list */
  666 +struct osd_sg_list_entry {
  667 + __be64 offset;
  668 + __be64 len;
  669 +};
  670 +
  671 +struct osd_sg_continuation_descriptor {
  672 + struct osd_continuation_descriptor_header hdr;
  673 + struct osd_sg_list_entry entries[];
  674 +};
  675 +
634 676 #endif /* ndef __OSD_PROTOCOL_H__ */
include/scsi/osd_types.h
... ... @@ -37,5 +37,10 @@
37 37 void *val_ptr; /* in network order */
38 38 };
39 39  
  40 +struct osd_sg_entry {
  41 + u64 offset;
  42 + u64 len;
  43 +};
  44 +
40 45 #endif /* ndef __OSD_TYPES_H__ */