rbd: support for data-pool + obj_request->object_no
Subsystems (3)
Ilya Dryomov
maintainercover rbd: support for data-pool + obj_request->object_no
Hello,
There are two pieces here: support for RBD_FEATURE_DATA_POOL (patches 1-10) and a couple of cleanups / minor optimizations that build on it and allow us to get rid of rbd_segment_name_cache (patches 11 - 15).
Thanks,
Ilya
Ilya Dryomov (15):
rbd: use kstrndup() in rbd_header_from_disk()
rbd: kill rbd_image_header::{crypt_type,comp_type}
rbd: initialize rbd_dev->header_oloc early
rbd: do away with obj_request in rbd_obj_read_sync()
libceph: pass reply buffer length through ceph_osdc_call()
rbd: switch rbd_obj_method_sync() to ceph_osdc_call()
rbd: remove now unused rbd_obj_request_wait() and helpers
rbd: use rbd_obj_bytes() more
rbd: introduce rbd_init_layout()
rbd: support for data-pool feature
rbd: set offset and length outside of rbd_obj_request_create()
rbd: factor out __rbd_osd_req_create()
rbd: RBD_V{1,2}_DATA_FORMAT macros
rbd: store and use obj_request->object_no
rbd: kill obj_request->object_name and rbd_segment_name_cache
drivers/block/rbd.c | 601 +++++++++++++++++--------------------------- drivers/block/rbd_types.h | 10 +- include/linux/ceph/osdmap.h | 7 - net/ceph/cls_lock_client.c | 2 +- net/ceph/osd_client.c | 7 +- 5 files changed, 238 insertions(+), 389 deletions(-)
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 1/15 rbd: use kstrndup() in rbd_header_from_disk()
--- drivers/block/rbd.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-)
drivers/block/rbd.c +3 -7
@@ -992,15 +992,11 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,/* Allocate this now to avoid having to handle failure below */ if (first_time) { -size_t len; - - len =strnlen (ondisk->object_prefix, -sizeof (ondisk->object_prefix)); - object_prefix =kmalloc (len +1 ,GFP_KERNEL ); + object_prefix =kstrndup (ondisk->object_prefix, +sizeof (ondisk->object_prefix), +GFP_KERNEL );if (!object_prefix)return -ENOMEM ; -memcpy (object_prefix, ondisk->object_prefix, len); - object_prefix[len] ='\0' ; }/* Allocate the snapshot context and fill it in */
Signature · 1 line
2.4.3
Re: [PATCH 01/15] rbd: use kstrndup() in rbd_header_from_disk()
Quoted 1 line · Ilya Dryomov
Signed-off-by Ilya Dryomov <[email protected]>
Looks good.
Quoted 28 lines
--- drivers/block/rbd.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-)drivers/block/rbd.c +3
-7@@ -992,15 +992,11 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,/* Allocate this now to avoid having to handle failure below */ if (first_time) { -size_t len; - - len =strnlen (ondisk->object_prefix, -sizeof (ondisk->object_prefix)); - object_prefix =kmalloc (len +1 ,GFP_KERNEL ); + object_prefix =kstrndup (ondisk->object_prefix, +sizeof (ondisk->object_prefix), +GFP_KERNEL );if (!object_prefix)return -ENOMEM ; -memcpy (object_prefix, ondisk->object_prefix, len); - object_prefix[len] ='\0' ; }/* Allocate the snapshot context and fill it in */
Re: [PATCH 01/15] rbd: use kstrndup() in rbd_header_from_disk()
In reply to Ilya DryomovQuoted 1 line · Ilya Dryomov
Signed-off-by Ilya Dryomov <[email protected]>
Looks good.
Quoted 28 lines
--- drivers/block/rbd.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-)drivers/block/rbd.c +3
-7@@ -992,15 +992,11 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,/* Allocate this now to avoid having to handle failure below */ if (first_time) { -size_t len; - - len =strnlen (ondisk->object_prefix, -sizeof (ondisk->object_prefix)); - object_prefix =kmalloc (len +1 ,GFP_KERNEL ); + object_prefix =kstrndup (ondisk->object_prefix, +sizeof (ondisk->object_prefix), +GFP_KERNEL );if (!object_prefix)return -ENOMEM ; -memcpy (object_prefix, ondisk->object_prefix, len); - object_prefix[len] ='\0' ; }/* Allocate the snapshot context and fill it in */
Ilya Dryomov
maintainerPATCH 2/15 rbd: kill rbd_image_header::{crypt_type,comp_type}
In reply to Ilya DryomovImage format 1 is deprecated and format 2 doesn't have these. Also, __rbd_dev_create() takes care of zeroing (or otherwise initializing) format 2 specific fields.
--- drivers/block/rbd.c | 9 --------- drivers/block/rbd_types.h | 3 --- 2 files changed, 12 deletions(-)
drivers/block/rbd.c +0 -9
@@ -144,8 +144,6 @@ struct rbd_image_header {/* These six fields never change for a given rbd image */ char *object_prefix;__u8 obj_order; -__u8 crypt_type; -__u8 comp_type;u64 stripe_unit;u64 stripe_count;u64 features;/* Might be changeable someday? */
@@ -1047,12 +1045,6 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,if (first_time) { header->object_prefix = object_prefix; header->obj_order = ondisk->options.order; - header->crypt_type = ondisk->options.crypt_type; - header->comp_type = ondisk->options.comp_type; -/* The rest aren't used for format 1 images */ - header->stripe_unit =0 ; - header->stripe_count =0 ; - header->features =0 ; }else {ceph_put_snap_context (header->snapc);kfree (header->snap_names);
@@ -5938,7 +5930,6 @@ static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)if (ret <0 ) goto out_err; } -/* No support for crypto and compression type format 2 images */ return 0 ; out_err:
drivers/block/rbd_types.h +0 -3
@@ -57,9 +57,6 @@ enum rbd_notify_op {#define RBD_MIN_OBJ_ORDER 16#define RBD_MAX_OBJ_ORDER 30 -#define RBD_COMP_NONE 0 -#define RBD_CRYPT_NONE 0 -#define RBD_HEADER_TEXT "<<< Rados Block Device Image >>>\n"#define RBD_HEADER_SIGNATURE "RBD"#define RBD_HEADER_VERSION "001.005"
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 3/15 rbd: initialize rbd_dev->header_oloc early
In reply to Ilya DryomovNo reason to delay it until image_id is known. This will be required by some rbd_obj_method_sync() callers, after rbd_obj_method_sync() is changed to take oloc.
--- drivers/block/rbd.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/block/rbd.c +1 -3
@@ -4863,7 +4863,7 @@ static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,init_rwsem (&rbd_dev->header_rwsem);ceph_oid_init (&rbd_dev->header_oid); -ceph_oloc_init (&rbd_dev->header_oloc); + rbd_dev->header_oloc.pool = spec->pool_id;mutex_init (&rbd_dev->watch_mutex); rbd_dev->watch_state =RBD_WATCH_STATE_UNREGISTERED ;
@@ -6062,8 +6062,6 @@ static int rbd_dev_header_name(struct rbd_device *rbd_dev)/* Record the header object name for this rbd image. */ rbd_assert (rbd_image_format_valid (rbd_dev->image_format)); - - rbd_dev->header_oloc.pool = rbd_dev->layout.pool_id;if (rbd_dev->image_format ==1 ) ret =ceph_oid_aprintf (&rbd_dev->header_oid,GFP_KERNEL ,"%s%s" , spec->image_name,RBD_SUFFIX );
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 4/15 rbd: do away with obj_request in rbd_obj_read_sync()
In reply to Ilya Dryomovrbd_obj_request machinery is completely unnecessary here; all that's being done is fetching a metadata object - no striping, cloning, etc. More importantly, rbd_osd_req_create() grabs pool id from layout and that is becoming a data pool id.
Kill offset argument - all metadata objects are small and read in full.
--- drivers/block/rbd.c | 81 +++++++++++++++++++++-------------------------------- 1 file changed, 32 insertions(+), 49 deletions(-)
drivers/block/rbd.c +32 -49
@@ -4242,63 +4242,46 @@ static void rbd_free_disk(struct rbd_device *rbd_dev) }static int rbd_obj_read_sync (struct rbd_device *rbd_dev, -const char *object_name, -u64 offset,u64 length,void *buf) +struct ceph_object_id *oid, +struct ceph_object_locator *oloc, +void *buf,int buf_len) { -struct rbd_obj_request *obj_request; -struct page **pages =NULL ; -u32 page_count; -size_t size; +struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; +struct ceph_osd_request *req; +struct page **pages; +int num_pages =calc_pages_for (0 , buf_len);int ret; - page_count = (u32 )calc_pages_for (offset, length); - pages =ceph_alloc_page_vector (page_count,GFP_KERNEL ); -if (IS_ERR (pages)) -return PTR_ERR (pages); - - ret = -ENOMEM ; - obj_request =rbd_obj_request_create (object_name, offset, length, -OBJ_REQUEST_PAGES ); -if (!obj_request) - goto out; - - obj_request->pages = pages; - obj_request->page_count = page_count; - - obj_request->osd_req =rbd_osd_req_create (rbd_dev,OBJ_OP_READ ,1 , - obj_request); -if (!obj_request->osd_req) - goto out; + req =ceph_osdc_alloc_request (osdc,NULL ,1 , false,GFP_KERNEL ); +if (!req) +return -ENOMEM ; -osd_req_op_extent_init (obj_request->osd_req,0 ,CEPH_OSD_OP_READ , - offset, length,0 ,0 ); -osd_req_op_extent_osd_data_pages (obj_request->osd_req,0 , - obj_request->pages, - obj_request->length, - obj_request->offset & ~PAGE_MASK , - false, false); +ceph_oid_copy (&req->r_base_oid, oid); +ceph_oloc_copy (&req->r_base_oloc, oloc); + req->r_flags =CEPH_OSD_FLAG_READ ; -rbd_obj_request_submit (obj_request); - ret =rbd_obj_request_wait (obj_request); + ret =ceph_osdc_alloc_messages (req,GFP_KERNEL );if (ret) - goto out; + goto out_req; - ret = obj_request->result; -if (ret <0 ) - goto out; + pages =ceph_alloc_page_vector (num_pages,GFP_KERNEL ); +if (IS_ERR (pages)) { + ret =PTR_ERR (pages); + goto out_req; + } -rbd_assert (obj_request->xferred <= (u64 )SIZE_MAX ); - size = (size_t ) obj_request->xferred; -ceph_copy_from_page_vector (pages, buf,0 , size); -rbd_assert (size <= (size_t )INT_MAX ); - ret = (int )size; -out: -if (obj_request) -rbd_obj_request_put (obj_request); -else -ceph_release_page_vector (pages, page_count); +osd_req_op_extent_init (req,0 ,CEPH_OSD_OP_READ ,0 , buf_len,0 ,0 ); +osd_req_op_extent_osd_data_pages (req,0 , pages, buf_len,0 , false, + true); + +ceph_osdc_start_request (osdc, req, false); + ret =ceph_osdc_wait_request (osdc, req); +if (ret >=0 ) +ceph_copy_from_page_vector (pages, buf,0 , ret); +out_req: +ceph_osdc_put_request (req);return ret; }
@@ -4334,8 +4317,8 @@ static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)if (!ondisk)return -ENOMEM ; - ret =rbd_obj_read_sync (rbd_dev, rbd_dev->header_oid.name, -0 , size, ondisk); + ret =rbd_obj_read_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc, ondisk, size);if (ret <0 ) goto out;if ((size_t )ret < size) {
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 5/15 libceph: pass reply buffer length through ceph_osdc_call()
In reply to Ilya DryomovTo spare checking for "this reply fits into a page, but does it fit into my buffer?" in some callers, osd_req_op_cls_response_data_pages() needs to know how big it is.
--- net/ceph/cls_lock_client.c | 2 +- net/ceph/osd_client.c | 7 +++++-- 2 files changed, 6 insertions(+), 3 deletions(-)
net/ceph/cls_lock_client.c +1 -1
@@ -278,7 +278,7 @@ int ceph_cls_lock_info(struct ceph_osd_client *osdc,int get_info_op_buf_size;int name_len =strlen (lock_name);struct page *get_info_op_page, *reply_page; -size_t reply_len; +size_t reply_len =PAGE_SIZE ;void *p, *end;int ret;
net/ceph/osd_client.c +5 -2
@@ -4023,7 +4023,7 @@ EXPORT_SYMBOL(ceph_osdc_maybe_request_map);* Execute an OSD class method on an object. * * @flags: CEPH_OSD_FLAG_* -* @resp_len: out param for reply length +* @resp_len: in/out param for reply length */ int ceph_osdc_call(struct ceph_osd_client *osdc, struct ceph_object_id *oid,
@@ -4036,6 +4036,9 @@ int ceph_osdc_call(struct ceph_osd_client *osdc,struct ceph_osd_request *req;int ret; +if (req_len >PAGE_SIZE || (resp_page && *resp_len >PAGE_SIZE )) +return -E2BIG ; + req =ceph_osdc_alloc_request (osdc,NULL ,1 , false,GFP_NOIO );if (!req)return -ENOMEM ;
@@ -4054,7 +4057,7 @@ int ceph_osdc_call(struct ceph_osd_client *osdc, 0, false, false);if (resp_page)osd_req_op_cls_response_data_pages (req,0 , &resp_page, -PAGE_SIZE ,0 , false, false); + *resp_len,0 , false, false);ceph_osdc_start_request (osdc, req, false); ret =ceph_osdc_wait_request (osdc, req);
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 6/15 rbd: switch rbd_obj_method_sync() to ceph_osdc_call()
In reply to Ilya DryomovAs explained in the previous commit, rbd_obj_request machinery (and rbd_osd_req_create() in particular) shouldn't be used for working with metadata objects.
Switch to the recently added ceph_osdc_call(). It assumes single pages for outbound and inbound buffers, but that's OK - none of the callers need more than that. These pages need to be allocated (messenger is in dire need of proper iterator interface!), but we are swapping for pages[] and pagelist allocations in the existing code.
Kill class_name argument - all rbd methods are under "rbd".
--- drivers/block/rbd.c | 162 ++++++++++++++++++++++------------------------------ 1 file changed, 67 insertions(+), 95 deletions(-)
drivers/block/rbd.c +67 -95
@@ -3971,17 +3971,17 @@ static void rbd_reregister_watch(struct work_struct *work)* returned in the outbound buffer, or a negative error code. */ static int rbd_obj_method_sync (struct rbd_device *rbd_dev, -const char *object_name, -const char *class_name, +struct ceph_object_id *oid, +struct ceph_object_locator *oloc,const char *method_name,const void *outbound,size_t outbound_size,void *inbound,size_t inbound_size) { -struct rbd_obj_request *obj_request; -struct page **pages; -u32 page_count; +struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; +struct page *req_page =NULL ; +struct page *reply_page;int ret;/*
@@ -3991,61 +3991,35 @@ static int rbd_obj_method_sync(struct rbd_device *rbd_dev,* method. Currently if this is present it will be a * snapshot id. */ - page_count = (u32 )calc_pages_for (0 , inbound_size); - pages =ceph_alloc_page_vector (page_count,GFP_KERNEL ); -if (IS_ERR (pages)) -return PTR_ERR (pages); +if (outbound) { +if (outbound_size >PAGE_SIZE ) +return -E2BIG ; - ret = -ENOMEM ; - obj_request =rbd_obj_request_create (object_name,0 , inbound_size, -OBJ_REQUEST_PAGES ); -if (!obj_request) - goto out; - - obj_request->pages = pages; - obj_request->page_count = page_count; - - obj_request->osd_req =rbd_osd_req_create (rbd_dev,OBJ_OP_READ ,1 , - obj_request); -if (!obj_request->osd_req) - goto out; - -osd_req_op_cls_init (obj_request->osd_req,0 ,CEPH_OSD_OP_CALL , - class_name, method_name); -if (outbound_size) { -struct ceph_pagelist *pagelist; - - pagelist =kmalloc (sizeof (*pagelist),GFP_NOFS ); -if (!pagelist) - goto out; + req_page =alloc_page (GFP_KERNEL ); +if (!req_page) +return -ENOMEM ; -ceph_pagelist_init (pagelist); -ceph_pagelist_append (pagelist, outbound, outbound_size); -osd_req_op_cls_request_data_pagelist (obj_request->osd_req,0 , - pagelist); +memcpy (page_address (req_page), outbound, outbound_size); } -osd_req_op_cls_response_data_pages (obj_request->osd_req,0 , - obj_request->pages, inbound_size, -0 , false, false); -rbd_obj_request_submit (obj_request); - ret =rbd_obj_request_wait (obj_request); -if (ret) - goto out; - - ret = obj_request->result; -if (ret <0 ) - goto out; + reply_page =alloc_page (GFP_KERNEL ); +if (!reply_page) { +if (req_page) +__free_page (req_page); +return -ENOMEM ; + } -rbd_assert (obj_request->xferred < (u64 )INT_MAX ); - ret = (int )obj_request->xferred; -ceph_copy_from_page_vector (pages, inbound,0 , obj_request->xferred); -out: -if (obj_request) -rbd_obj_request_put (obj_request); -else -ceph_release_page_vector (pages, page_count); + ret =ceph_osdc_call (osdc, oid, oloc,RBD_DRV_NAME , method_name, +CEPH_OSD_FLAG_READ , req_page, outbound_size, + reply_page, &inbound_size); +if (!ret) { +memcpy (inbound,page_address (reply_page), inbound_size); + ret = inbound_size; + } +if (req_page) +__free_page (req_page); +__free_page (reply_page);return ret; }
@@ -4939,10 +4913,10 @@ static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,__le64 size; } __attribute__ ((packed)) size_buf = { 0 }; - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_size" , - &snapid,sizeof (snapid), - &size_buf,sizeof (size_buf)); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_size" , + &snapid,sizeof (snapid), + &size_buf,sizeof (size_buf));dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 )return ret;
@@ -4979,9 +4953,9 @@ static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)if (!reply_buf)return -ENOMEM ; - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_object_prefix" ,NULL ,0 , - reply_buf,RBD_OBJ_PREFIX_LEN_MAX ); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_object_prefix" , +NULL ,0 , reply_buf,RBD_OBJ_PREFIX_LEN_MAX );dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 ) goto out;
@@ -5014,10 +4988,10 @@ static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,u64 unsup;int ret; - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_features" , - &snapid,sizeof (snapid), - &features_buf,sizeof (features_buf)); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_features" , + &snapid,sizeof (snapid), + &features_buf,sizeof (features_buf));dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 )return ret;
@@ -5076,10 +5050,9 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev) } snapid =cpu_to_le64 (rbd_dev->spec->snap_id); - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_parent" , - &snapid,sizeof (snapid), - reply_buf, size); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_parent" , + &snapid,sizeof (snapid), reply_buf, size);dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 ) goto out_err;
@@ -5179,9 +5152,9 @@ static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)u64 stripe_count;int ret; - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_stripe_unit_count" ,NULL ,0 , - (char *)&striping_info_buf, size); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_stripe_unit_count" , +NULL ,0 , &striping_info_buf, size);dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 )return ret;
@@ -5218,6 +5191,7 @@ static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)static char *rbd_dev_image_name (struct rbd_device *rbd_dev) { +CEPH_DEFINE_OID_ONSTACK (oid);size_t image_id_size;char *image_id;void *p;
@@ -5245,10 +5219,10 @@ static char *rbd_dev_image_name(struct rbd_device *rbd_dev)if (!reply_buf) goto out; - ret =rbd_obj_method_sync (rbd_dev,RBD_DIRECTORY , -"rbd" ,"dir_get_name" , - image_id, image_id_size, - reply_buf, size); +ceph_oid_printf (&oid,"%s" ,RBD_DIRECTORY ); + ret =rbd_obj_method_sync (rbd_dev, &oid, &rbd_dev->header_oloc, +"dir_get_name" , image_id, image_id_size, + reply_buf, size);if (ret <0 ) goto out; p = reply_buf;
@@ -5427,9 +5401,9 @@ static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)if (!reply_buf)return -ENOMEM ; - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_snapcontext" ,NULL ,0 , - reply_buf, size); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_snapcontext" , +NULL ,0 , reply_buf, size);dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 ) goto out;
@@ -5492,10 +5466,9 @@ static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,return ERR_PTR (-ENOMEM ); snapid =cpu_to_le64 (snap_id); - ret =rbd_obj_method_sync (rbd_dev, rbd_dev->header_oid.name, -"rbd" ,"get_snapshot_name" , - &snapid,sizeof (snapid), - reply_buf, size); + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_snapshot_name" , + &snapid,sizeof (snapid), reply_buf, size);dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret <0 ) { snap_name =ERR_PTR (ret);
@@ -5802,7 +5775,7 @@ static int rbd_dev_image_id(struct rbd_device *rbd_dev) {int ret;size_t size; -char *object_name; +CEPH_DEFINE_OID_ONSTACK (oid);void *response;char *image_id;
@@ -5822,12 +5795,12 @@ static int rbd_dev_image_id(struct rbd_device *rbd_dev)* First, see if the format 2 image id file exists, and if * so, get the image's persistent id from it. */ - size =sizeof (RBD_ID_PREFIX ) +strlen (rbd_dev->spec->image_name); - object_name =kmalloc (size,GFP_NOIO ); -if (!object_name) -return -ENOMEM ; -sprintf (object_name,"%s%s" ,RBD_ID_PREFIX , rbd_dev->spec->image_name); -dout ("rbd id object name is %s\n" , object_name); + ret =ceph_oid_aprintf (&oid,GFP_KERNEL ,"%s%s" ,RBD_ID_PREFIX , + rbd_dev->spec->image_name); +if (ret) +return ret; + +dout ("rbd id object name is %s\n" , oid.name);/* Response will be an encoded string, which includes a length */
@@ -5840,9 +5813,9 @@ static int rbd_dev_image_id(struct rbd_device *rbd_dev)/* If it doesn't exist we'll assume it's a format 1 image */ - ret =rbd_obj_method_sync (rbd_dev, object_name, -"rbd" ,"get_id" ,NULL ,0 , - response,RBD_IMAGE_ID_LEN_MAX ); + ret =rbd_obj_method_sync (rbd_dev, &oid, &rbd_dev->header_oloc, +"get_id" ,NULL ,0 , + response,RBD_IMAGE_ID_LEN_MAX );dout ("%s: rbd_obj_method_sync returned %d\n" , __func__, ret);if (ret == -ENOENT ) { image_id =kstrdup ("" ,GFP_KERNEL );
@@ -5865,8 +5838,7 @@ static int rbd_dev_image_id(struct rbd_device *rbd_dev) } out:kfree (response); -kfree (object_name); - +ceph_oid_destroy (&oid);return ret; }
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 7/15 rbd: remove now unused rbd_obj_request_wait() and helpers
In reply to Ilya Dryomov--- drivers/block/rbd.c | 38 -------------------------------------- 1 file changed, 38 deletions(-)
drivers/block/rbd.c +0 -38
@@ -1619,44 +1619,6 @@ static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)ceph_osdc_start_request (osd_req->r_osdc, osd_req, false); } -static void rbd_obj_request_end (struct rbd_obj_request *obj_request) -{ -dout ("%s %p\n" , __func__, obj_request); -ceph_osdc_cancel_request (obj_request->osd_req); -} - -/* -* Wait for an object request to complete. If interrupted, cancel the -* underlying osd request. -* -* @timeout: in jiffies, 0 means "wait forever" -*/ -static int __rbd_obj_request_wait (struct rbd_obj_request *obj_request, -unsigned long timeout) -{ -long ret; - -dout ("%s %p\n" , __func__, obj_request); - ret =wait_for_completion_interruptible_timeout ( - &obj_request->completion, -ceph_timeout_jiffies (timeout)); -if (ret <=0 ) { -if (ret ==0 ) - ret = -ETIMEDOUT ; -rbd_obj_request_end (obj_request); - }else { - ret =0 ; - } - -dout ("%s %p ret %d\n" , __func__, obj_request, (int )ret); -return ret; -} - -static int rbd_obj_request_wait (struct rbd_obj_request *obj_request) -{ -return __rbd_obj_request_wait (obj_request,0 ); -} -static void rbd_img_request_complete (struct rbd_img_request *img_request) {
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 8/15 rbd: use rbd_obj_bytes() more
In reply to Ilya DryomovReturning u64 doesn't make sense: max header->obj_order is 25 and ceph_file_layout::object_size is u32.
--- drivers/block/rbd.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
drivers/block/rbd.c +12 -12
@@ -971,6 +971,14 @@ static bool rbd_dev_ondisk_valid(struct rbd_image_header_ondisk *ondisk) }/* +* returns the size of an object in the image +*/ +static u32 rbd_obj_bytes (struct rbd_image_header *header) +{ +return 1U << header->obj_order; +} + +/* * Fill an rbd image header with information from the given format 1 * on-disk header. */
@@ -1255,7 +1263,7 @@ static const char *rbd_segment_name(struct rbd_device *rbd_dev, u64 offset)static u64 rbd_segment_offset (struct rbd_device *rbd_dev,u64 offset) { -u64 segment_size = (u64 )1 << rbd_dev->header.obj_order; +u64 segment_size =rbd_obj_bytes (&rbd_dev->header);return offset & (segment_size -1 ); }
@@ -1263,7 +1271,7 @@ static u64 rbd_segment_offset(struct rbd_device *rbd_dev, u64 offset)static u64 rbd_segment_length (struct rbd_device *rbd_dev,u64 offset,u64 length) { -u64 segment_size = (u64 )1 << rbd_dev->header.obj_order; +u64 segment_size =rbd_obj_bytes (&rbd_dev->header); offset &= segment_size -1 ;
@@ -1275,14 +1283,6 @@ static u64 rbd_segment_length(struct rbd_device *rbd_dev, }/* -* returns the size of an object in the image -*/ -static u64 rbd_obj_bytes (struct rbd_image_header *header) -{ -return 1 << header->obj_order; -} - -/* * bio helpers */
@@ -2721,7 +2721,7 @@ static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)* child image to which the original request was to be sent. */ img_offset = obj_request->img_offset - obj_request->offset; - length = (u64 )1 << rbd_dev->header.obj_order; + length =rbd_obj_bytes (&rbd_dev->header);/* * There is no defined parent data beyond the parent
@@ -5130,7 +5130,7 @@ static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)* out, and only fail if the image has non-default values. */ ret = -EINVAL ; - obj_size = (u64 )1 << rbd_dev->header.obj_order; + obj_size =rbd_obj_bytes (&rbd_dev->header); p = &striping_info_buf; stripe_unit =ceph_decode_64 (&p);if (stripe_unit != obj_size) {
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 9/15 rbd: introduce rbd_init_layout()
In reply to Ilya DryomovRather than initializing layout fields with some made up values in __rbd_dev_create(), move the initialization into rbd_init_layout() and call it after the header is actually populated.
--- drivers/block/rbd.c | 25 ++++++++++++++++++------- 1 file changed, 18 insertions(+), 7 deletions(-)
drivers/block/rbd.c +18 -7
@@ -978,6 +978,21 @@ static u32 rbd_obj_bytes(struct rbd_image_header *header)return 1U << header->obj_order; } +static void rbd_init_layout (struct rbd_device *rbd_dev) +{ +if (rbd_dev->header.stripe_unit ==0 || + rbd_dev->header.stripe_count ==0 ) { + rbd_dev->header.stripe_unit =rbd_obj_bytes (&rbd_dev->header); + rbd_dev->header.stripe_count =1 ; + } + + rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit; + rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count; + rbd_dev->layout.object_size =rbd_obj_bytes (&rbd_dev->header); + rbd_dev->layout.pool_id = rbd_dev->spec->pool_id; +RCU_INIT_POINTER (rbd_dev->layout.pool_ns,NULL ); +} +/* * Fill an rbd image header with information from the given format 1 * on-disk header.
@@ -1053,6 +1068,7 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,if (first_time) { header->object_prefix = object_prefix; header->obj_order = ondisk->options.order; +rbd_init_layout (rbd_dev); }else {ceph_put_snap_context (header->snapc);kfree (header->snap_names);
@@ -4804,12 +4820,6 @@ static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc, rbd_dev->rbd_client = rbdc; rbd_dev->spec = spec; - rbd_dev->layout.stripe_unit =1 <<RBD_MAX_OBJ_ORDER ; - rbd_dev->layout.stripe_count =1 ; - rbd_dev->layout.object_size =1 <<RBD_MAX_OBJ_ORDER ; - rbd_dev->layout.pool_id = spec->pool_id; -RCU_INIT_POINTER (rbd_dev->layout.pool_ns,NULL ); -return rbd_dev; }
@@ -5848,12 +5858,13 @@ static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) goto out_err; } +rbd_init_layout (rbd_dev);return 0 ; + out_err: rbd_dev->header.features =0 ;kfree (rbd_dev->header.object_prefix); rbd_dev->header.object_prefix =NULL ; -return ret; }
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 10/15 rbd: support for data-pool feature
In reply to Ilya DryomovAdd support for RBD_FEATURE_DATA_POOL feature. rbd_dev->layout.pool_id now stores the data pool id.
--- drivers/block/rbd.c | 33 +++++++++++++++++++++++++++++++-- 1 file changed, 31 insertions(+), 2 deletions(-)
drivers/block/rbd.c +31 -2
@@ -123,9 +123,11 @@ static int atomic_dec_return_safe(atomic_t *v)#define RBD_FEATURE_LAYERING (1<<0)#define RBD_FEATURE_STRIPINGV2 (1<<1)#define RBD_FEATURE_EXCLUSIVE_LOCK (1<<2) +#define RBD_FEATURE_DATA_POOL (1<<7)#define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \ RBD_FEATURE_STRIPINGV2 | \ - RBD_FEATURE_EXCLUSIVE_LOCK) + RBD_FEATURE_EXCLUSIVE_LOCK | \ + RBD_FEATURE_DATA_POOL)/* Features supported by this (client software) implementation. */
@@ -146,6 +148,7 @@ struct rbd_image_header {__u8 obj_order;u64 stripe_unit;u64 stripe_count; +s64 data_pool_id;u64 features;/* Might be changeable someday? */ /* The remaining fields need to be updated occasionally */
@@ -989,7 +992,8 @@ static void rbd_init_layout(struct rbd_device *rbd_dev) rbd_dev->layout.stripe_unit = rbd_dev->header.stripe_unit; rbd_dev->layout.stripe_count = rbd_dev->header.stripe_count; rbd_dev->layout.object_size =rbd_obj_bytes (&rbd_dev->header); - rbd_dev->layout.pool_id = rbd_dev->spec->pool_id; + rbd_dev->layout.pool_id = rbd_dev->header.data_pool_id ==CEPH_NOPOOL ? + rbd_dev->spec->pool_id : rbd_dev->header.data_pool_id;RCU_INIT_POINTER (rbd_dev->layout.pool_ns,NULL ); }
@@ -4797,6 +4801,7 @@ static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,INIT_LIST_HEAD (&rbd_dev->node);init_rwsem (&rbd_dev->header_rwsem); + rbd_dev->header.data_pool_id =CEPH_NOPOOL ;ceph_oid_init (&rbd_dev->header_oid); rbd_dev->header_oloc.pool = spec->pool_id;
@@ -5161,6 +5166,24 @@ static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)return 0 ; } +static int rbd_dev_v2_data_pool (struct rbd_device *rbd_dev) +{ +__le64 data_pool_id; +int ret; + + ret =rbd_obj_method_sync (rbd_dev, &rbd_dev->header_oid, + &rbd_dev->header_oloc,"get_data_pool" , +NULL ,0 , &data_pool_id,sizeof (data_pool_id)); +if (ret <0 ) +return ret; +if (ret <sizeof (data_pool_id)) +return -EBADMSG ; + + rbd_dev->header.data_pool_id =le64_to_cpu (data_pool_id); +WARN_ON (rbd_dev->header.data_pool_id ==CEPH_NOPOOL ); +return 0 ; +} +static char *rbd_dev_image_name (struct rbd_device *rbd_dev) {CEPH_DEFINE_OID_ONSTACK (oid);
@@ -5858,6 +5881,12 @@ static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev) goto out_err; } +if (rbd_dev->header.features &RBD_FEATURE_DATA_POOL ) { + ret =rbd_dev_v2_data_pool (rbd_dev); +if (ret) + goto out_err; + } +rbd_init_layout (rbd_dev);return 0 ;
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 11/15 rbd: set offset and length outside of rbd_obj_request_create()
In reply to Ilya DryomovThe allocation doesn't depend on offset and length. Both offset and length can be changed after obj_request is allocated, too.
--- drivers/block/rbd.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)
drivers/block/rbd.c +11 -15
@@ -1631,7 +1631,9 @@ static void rbd_obj_request_submit(struct rbd_obj_request *obj_request) {struct ceph_osd_request *osd_req = obj_request->osd_req; -dout ("%s %p osd_req %p\n" , __func__, obj_request, osd_req); +dout ("%s %p \"%s\" %llu~%llu osd_req %p\n" , __func__, + obj_request, obj_request->object_name, obj_request->offset, + obj_request->length, osd_req);if (obj_request_img_data_test (obj_request)) {WARN_ON (obj_request->callback != rbd_img_obj_callback);rbd_img_request_get (obj_request->img_request);
@@ -2073,7 +2075,6 @@ static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)/* object_name is assumed to be a non-null pointer and NUL-terminated */ static struct rbd_obj_request *rbd_obj_request_create (const char *object_name, -u64 offset,u64 length,enum obj_request_type type) {struct rbd_obj_request *obj_request;
@@ -2094,18 +2095,13 @@ static struct rbd_obj_request *rbd_obj_request_create(const char *object_name, } obj_request->object_name =memcpy (name, object_name, size); - obj_request->offset = offset; - obj_request->length = length; - obj_request->flags =0 ; obj_request->which =BAD_WHICH ; obj_request->type = type;INIT_LIST_HEAD (&obj_request->links);init_completion (&obj_request->completion);kref_init (&obj_request->kref); -dout ("%s: \"%s\" %llu/%llu %d -> obj %p\n" , __func__, object_name, - offset, length, (int )type, obj_request); - +dout ("%s %p\n" , __func__, obj_request);return obj_request; }
@@ -2517,21 +2513,21 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,while (resid) {struct ceph_osd_request *osd_req;const char *object_name; -u64 offset; -u64 length; +u64 offset =rbd_segment_offset (rbd_dev, img_offset); +u64 length =rbd_segment_length (rbd_dev, img_offset, resid); object_name =rbd_segment_name (rbd_dev, img_offset);if (!object_name) goto out_unwind; - offset =rbd_segment_offset (rbd_dev, img_offset); - length =rbd_segment_length (rbd_dev, img_offset, resid); - obj_request =rbd_obj_request_create (object_name, - offset, length, type); + obj_request =rbd_obj_request_create (object_name, type);/* object request has its own copy of the object name */ rbd_segment_name_free (object_name);if (!obj_request) goto out_unwind; + obj_request->offset = offset; + obj_request->length = length; +/* * set obj_request->img_request before creating the * osd_request so that it gets the right snapc
@@ -2870,7 +2866,7 @@ static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)size_t size;int ret; - stat_request =rbd_obj_request_create (obj_request->object_name,0 ,0 , + stat_request =rbd_obj_request_create (obj_request->object_name,OBJ_REQUEST_PAGES );if (!stat_request)return -ENOMEM ;
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 12/15 rbd: factor out __rbd_osd_req_create()
In reply to Ilya DryomovFactor OSD request allocation and initialization code out into __rbd_osd_req_create().
--- drivers/block/rbd.c | 103 ++++++++++++++++++++-------------------------------- 1 file changed, 40 insertions(+), 63 deletions(-)
drivers/block/rbd.c +40 -63
@@ -1952,6 +1952,38 @@ static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request) osd_req->r_data_offset = obj_request->offset; } +static struct ceph_osd_request * +__rbd_osd_req_create (struct rbd_device *rbd_dev, +struct ceph_snap_context *snapc, +int num_ops,unsigned int flags, +struct rbd_obj_request *obj_request) +{ +struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc; +struct ceph_osd_request *req; + + req =ceph_osdc_alloc_request (osdc, snapc, num_ops, false,GFP_NOIO ); +if (!req) +return NULL ; + + req->r_flags = flags; + req->r_callback = rbd_osd_req_callback; + req->r_priv = obj_request; + + req->r_base_oloc.pool = rbd_dev->layout.pool_id; +if (ceph_oid_aprintf (&req->r_base_oid,GFP_NOIO ,"%s" , + obj_request->object_name)) + goto err_req; + +if (ceph_osdc_alloc_messages (req,GFP_NOIO )) + goto err_req; + +return req; + +err_req: +ceph_osdc_put_request (req); +return NULL ; +} +/* * Create an osd request. A read request has one osd op (read). * A write request has either one (watch) or two (hint+write) osd ops.
@@ -1965,8 +1997,6 @@ static struct ceph_osd_request *rbd_osd_req_create( struct rbd_obj_request *obj_request) {struct ceph_snap_context *snapc =NULL ; -struct ceph_osd_client *osdc; -struct ceph_osd_request *osd_req;if (obj_request_img_data_test (obj_request) && (op_type ==OBJ_OP_DISCARD || op_type ==OBJ_OP_WRITE )) {
@@ -1981,35 +2011,10 @@ static struct ceph_osd_request *rbd_osd_req_create(rbd_assert (num_ops ==1 || ((op_type ==OBJ_OP_WRITE ) && num_ops ==2 )); -/* Allocate and initialize the request, for the num_ops ops */ - - osdc = &rbd_dev->rbd_client->client->osdc; - osd_req =ceph_osdc_alloc_request (osdc, snapc, num_ops, false, -GFP_NOIO ); -if (!osd_req) - goto fail; - -if (op_type ==OBJ_OP_WRITE || op_type ==OBJ_OP_DISCARD ) - osd_req->r_flags =CEPH_OSD_FLAG_WRITE |CEPH_OSD_FLAG_ONDISK ; -else - osd_req->r_flags =CEPH_OSD_FLAG_READ ; - - osd_req->r_callback = rbd_osd_req_callback; - osd_req->r_priv = obj_request; - - osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id; -if (ceph_oid_aprintf (&osd_req->r_base_oid,GFP_NOIO ,"%s" , - obj_request->object_name)) - goto fail; - -if (ceph_osdc_alloc_messages (osd_req,GFP_NOIO )) - goto fail; - -return osd_req; - -fail: -ceph_osdc_put_request (osd_req); -return NULL ; +return __rbd_osd_req_create (rbd_dev, snapc, num_ops, + (op_type ==OBJ_OP_WRITE || op_type ==OBJ_OP_DISCARD ) ? +CEPH_OSD_FLAG_WRITE |CEPH_OSD_FLAG_ONDISK :CEPH_OSD_FLAG_READ , + obj_request); }/*
@@ -2022,10 +2027,6 @@ static struct ceph_osd_request * rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request) {struct rbd_img_request *img_request; -struct ceph_snap_context *snapc; -struct rbd_device *rbd_dev; -struct ceph_osd_client *osdc; -struct ceph_osd_request *osd_req;int num_osd_ops =3 ;rbd_assert (obj_request_img_data_test (obj_request));
@@ -2037,36 +2038,12 @@ rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)if (img_request_discard_test (img_request)) num_osd_ops =2 ; -/* Allocate and initialize the request, for all the ops */ - - snapc = img_request->snapc; - rbd_dev = img_request->rbd_dev; - osdc = &rbd_dev->rbd_client->client->osdc; - osd_req =ceph_osdc_alloc_request (osdc, snapc, num_osd_ops, - false,GFP_NOIO ); -if (!osd_req) - goto fail; - - osd_req->r_flags =CEPH_OSD_FLAG_WRITE |CEPH_OSD_FLAG_ONDISK ; - osd_req->r_callback = rbd_osd_req_callback; - osd_req->r_priv = obj_request; - - osd_req->r_base_oloc.pool = rbd_dev->layout.pool_id; -if (ceph_oid_aprintf (&osd_req->r_base_oid,GFP_NOIO ,"%s" , - obj_request->object_name)) - goto fail; - -if (ceph_osdc_alloc_messages (osd_req,GFP_NOIO )) - goto fail; - -return osd_req; - -fail: -ceph_osdc_put_request (osd_req); -return NULL ; +return __rbd_osd_req_create (img_request->rbd_dev, + img_request->snapc, num_osd_ops, +CEPH_OSD_FLAG_WRITE |CEPH_OSD_FLAG_ONDISK , + obj_request); } -static void rbd_osd_req_destroy (struct ceph_osd_request *osd_req) {ceph_osdc_put_request (osd_req);
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 13/15 rbd: RBD_V{1,2}_DATA_FORMAT macros
In reply to Ilya Dryomov... and also fix up the comment -- format 1 data objects have always been 12 hex digits long.
--- drivers/block/rbd.c | 6 ++---- drivers/block/rbd_types.h | 7 ++++--- 2 files changed, 6 insertions(+), 7 deletions(-)
drivers/block/rbd.c +2 -4
@@ -1257,18 +1257,16 @@ static void rbd_segment_name_free(const char *name)static const char *rbd_segment_name (struct rbd_device *rbd_dev,u64 offset) { +const char *name_format = rbd_dev->image_format ==1 ? +RBD_V1_DATA_FORMAT :RBD_V2_DATA_FORMAT ;char *name;u64 segment;int ret; -char *name_format; name =kmem_cache_alloc (rbd_segment_name_cache,GFP_NOIO );if (!name)return NULL ; segment = offset >> rbd_dev->header.obj_order; - name_format ="%s.%012llx" ; -if (rbd_dev->image_format ==2 ) - name_format ="%s.%016llx" ; ret =snprintf (name,CEPH_MAX_OID_NAME_LEN +1 , name_format, rbd_dev->header.object_prefix, segment);if (ret <0 || ret >CEPH_MAX_OID_NAME_LEN ) {
drivers/block/rbd_types.h +4 -3
@@ -25,8 +25,8 @@*/ #define RBD_HEADER_PREFIX "rbd_header." -#define RBD_DATA_PREFIX "rbd_data."#define RBD_ID_PREFIX "rbd_id." +#define RBD_V2_DATA_FORMAT "%s.%016llx"#define RBD_LOCK_NAME "rbd_lock"#define RBD_LOCK_TAG "internal"
@@ -42,13 +42,14 @@ enum rbd_notify_op {/* * For format version 1, rbd image 'foo' consists of objects * foo.rbd - image metadata -* rb.<idhi>.<idlo>.00000000 -* rb.<idhi>.<idlo>.00000001 +* rb.<idhi>.<idlo>.000000000000 +* rb.<idhi>.<idlo>.000000000001 * ... - data * There is no notion of a persistent image id in rbd format 1. */ #define RBD_SUFFIX ".rbd" +#define RBD_V1_DATA_FORMAT "%s.%012llx"#define RBD_DIRECTORY "rbd_directory"#define RBD_INFO "rbd_info"
Signature · 1 line
2.4.3
Re: [PATCH 13/15] rbd: RBD_V{1,2}_DATA_FORMAT macros
Nit: only since you are correcting the v1 name format, it should also include 32bits of "extra" randomness (i.e. rb.<idhi>.<idlo>.<extra>.000000000000). Looks like it's documented incorrectly in librbd as well.
Quoted 72 lines · Ilya Dryomov
... and also fix up the comment -- format 1 data objects have always been 12 hex digits long.
Signed-off-by Ilya Dryomov <[email protected]>--- drivers/block/rbd.c | 6 ++---- drivers/block/rbd_types.h | 7 ++++--- 2 files changed, 6 insertions(+), 7 deletions(-)drivers/block/rbd.c +2
-4@@ -1257,18 +1257,16 @@ static void rbd_segment_name_free(const char *name)static const char *rbd_segment_name (struct rbd_device *rbd_dev,u64 offset) { +const char *name_format = rbd_dev->image_format ==1 ? +RBD_V1_DATA_FORMAT :RBD_V2_DATA_FORMAT ;char *name;u64 segment;int ret; -char *name_format; name =kmem_cache_alloc (rbd_segment_name_cache,GFP_NOIO );if (!name)return NULL ; segment = offset >> rbd_dev->header.obj_order; - name_format ="%s.%012llx" ; -if (rbd_dev->image_format ==2 ) - name_format ="%s.%016llx" ; ret =snprintf (name,CEPH_MAX_OID_NAME_LEN +1 , name_format, rbd_dev->header.object_prefix, segment);if (ret <0 || ret >CEPH_MAX_OID_NAME_LEN ) {drivers/block/rbd_types.h +4
-4@@ -25,8 +25,8 @@*/ #define RBD_HEADER_PREFIX "rbd_header." -#define RBD_DATA_PREFIX "rbd_data."#define RBD_ID_PREFIX "rbd_id." +#define RBD_V2_DATA_FORMAT "%s.%016llx"#define RBD_LOCK_NAME "rbd_lock"#define RBD_LOCK_TAG "internal"@@ -42,13 +42,14 @@ enum rbd_notify_op {/* * For format version 1, rbd image 'foo' consists of objects * foo.rbd - image metadata -* rb.<idhi>.<idlo>.00000000 -* rb.<idhi>.<idlo>.00000001 +* rb.<idhi>.<idlo>.000000000000 +* rb.<idhi>.<idlo>.000000000001 * ... - data * There is no notion of a persistent image id in rbd format 1. */ #define RBD_SUFFIX ".rbd" +#define RBD_V1_DATA_FORMAT "%s.%012llx"#define RBD_DIRECTORY "rbd_directory"#define RBD_INFO "rbd_info" --2.4.3
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Signature · 1 line
Jason
Ilya Dryomov
maintainerPATCH 14/15 rbd: store and use obj_request->object_no
In reply to Ilya Dryomovobject_no can be trivially formatted into an object name. We already store object names in OSD requests with special care to avoid dynamic allocations for short names. Storing a name in obj_request, obtained as below (!), is a waste and will be removed in the next commit.
name = kmem_cache_alloc(rbd_segment_name_cache, ...);
snprintf(name, ...);
obj_request->object_name = kstrdup(name);
kmem_cache_free(rbd_segment_name_cache, name);
...
ceph_oid_aprintf(..., "%s", obj_request->object_name);--- drivers/block/rbd.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)
drivers/block/rbd.c +13 -6
@@ -232,6 +232,7 @@ enum obj_req_flags {struct rbd_obj_request {const char *object_name; +u64 object_no;u64 offset;/* object start byte */ u64 length;/* bytes from offset */ unsigned long flags;
@@ -1629,8 +1630,8 @@ static void rbd_obj_request_submit(struct rbd_obj_request *obj_request) {struct ceph_osd_request *osd_req = obj_request->osd_req; -dout ("%s %p \"%s\" %llu~%llu osd_req %p\n" , __func__, - obj_request, obj_request->object_name, obj_request->offset, +dout ("%s %p object_no %016llx %llu~%llu osd_req %p\n" , __func__, + obj_request, obj_request->object_no, obj_request->offset, obj_request->length, osd_req);if (obj_request_img_data_test (obj_request)) {WARN_ON (obj_request->callback != rbd_img_obj_callback);
@@ -1925,8 +1926,8 @@ static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)rbd_osd_call_callback (obj_request);break ;default : -rbd_warn (NULL ,"%s: unsupported op %hu" , - obj_request->object_name, (unsigned short ) opcode); +rbd_warn (NULL ,"unexpected OSD op: object_no %016llx opcode %d" , + obj_request->object_no, opcode);break ; }
@@ -1958,6 +1959,8 @@ __rbd_osd_req_create(struct rbd_device *rbd_dev, {struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;struct ceph_osd_request *req; +const char *name_format = rbd_dev->image_format ==1 ? +RBD_V1_DATA_FORMAT :RBD_V2_DATA_FORMAT ; req =ceph_osdc_alloc_request (osdc, snapc, num_ops, false,GFP_NOIO );if (!req)
@@ -1968,8 +1971,8 @@ __rbd_osd_req_create(struct rbd_device *rbd_dev, req->r_priv = obj_request; req->r_base_oloc.pool = rbd_dev->layout.pool_id; -if (ceph_oid_aprintf (&req->r_base_oid,GFP_NOIO ,"%s" , - obj_request->object_name)) +if (ceph_oid_aprintf (&req->r_base_oid,GFP_NOIO , name_format, + rbd_dev->header.object_prefix, obj_request->object_no)) goto err_req;if (ceph_osdc_alloc_messages (req,GFP_NOIO ))
@@ -2488,6 +2491,7 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,while (resid) {struct ceph_osd_request *osd_req;const char *object_name; +u64 object_no = img_offset >> rbd_dev->header.obj_order;u64 offset =rbd_segment_offset (rbd_dev, img_offset);u64 length =rbd_segment_length (rbd_dev, img_offset, resid);
@@ -2500,6 +2504,7 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,if (!obj_request) goto out_unwind; + obj_request->object_no = object_no; obj_request->offset = offset; obj_request->length = length;
@@ -2846,6 +2851,8 @@ static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)if (!stat_request)return -ENOMEM ; + stat_request->object_no = obj_request->object_no; + stat_request->osd_req =rbd_osd_req_create (rbd_dev,OBJ_OP_READ ,1 , stat_request);if (!stat_request->osd_req) {
Signature · 1 line
2.4.3
Ilya Dryomov
maintainerPATCH 15/15 rbd: kill obj_request->object_name and rbd_segment_name_cache
In reply to Ilya Dryomov--- drivers/block/rbd.c | 79 ++++----------------------------------------- include/linux/ceph/osdmap.h | 7 ---- 2 files changed, 7 insertions(+), 79 deletions(-)
drivers/block/rbd.c +7 -72
@@ -231,7 +231,6 @@ enum obj_req_flags { };struct rbd_obj_request { -const char *object_name;u64 object_no;u64 offset;/* object start byte */ u64 length;/* bytes from offset */
@@ -440,7 +439,6 @@ static DEFINE_SPINLOCK(rbd_client_list_lock);static struct kmem_cache *rbd_img_request_cache;static struct kmem_cache *rbd_obj_request_cache; -static struct kmem_cache *rbd_segment_name_cache;static int rbd_major;static DEFINE_IDA (rbd_dev_id_ida);
@@ -1249,37 +1247,6 @@ static void rbd_dev_mapping_clear(struct rbd_device *rbd_dev) rbd_dev->mapping.features =0 ; } -static void rbd_segment_name_free (const char *name) -{ -/* The explicit cast here is needed to drop the const qualifier */ - -kmem_cache_free (rbd_segment_name_cache, (void *)name); -} - -static const char *rbd_segment_name (struct rbd_device *rbd_dev,u64 offset) -{ -const char *name_format = rbd_dev->image_format ==1 ? -RBD_V1_DATA_FORMAT :RBD_V2_DATA_FORMAT ; -char *name; -u64 segment; -int ret; - - name =kmem_cache_alloc (rbd_segment_name_cache,GFP_NOIO ); -if (!name) -return NULL ; - segment = offset >> rbd_dev->header.obj_order; - ret =snprintf (name,CEPH_MAX_OID_NAME_LEN +1 , name_format, - rbd_dev->header.object_prefix, segment); -if (ret <0 || ret >CEPH_MAX_OID_NAME_LEN ) { -pr_err ("error formatting segment name for #%llu (%d)\n" , - segment, ret); -rbd_segment_name_free (name); - name =NULL ; - } - -return name; -} -static u64 rbd_segment_offset (struct rbd_device *rbd_dev,u64 offset) {u64 segment_size =rbd_obj_bytes (&rbd_dev->header);
@@ -2050,29 +2017,17 @@ static void rbd_osd_req_destroy(struct ceph_osd_request *osd_req)ceph_osdc_put_request (osd_req); } -/* object_name is assumed to be a non-null pointer and NUL-terminated */ - -static struct rbd_obj_request *rbd_obj_request_create (const char *object_name, -enum obj_request_type type) +static struct rbd_obj_request * +rbd_obj_request_create (enum obj_request_type type) {struct rbd_obj_request *obj_request; -size_t size; -char *name;rbd_assert (obj_request_type_valid (type)); - size =strlen (object_name) +1 ; - name =kmalloc (size,GFP_NOIO ); -if (!name) -return NULL ; - obj_request =kmem_cache_zalloc (rbd_obj_request_cache,GFP_NOIO ); -if (!obj_request) { -kfree (name); +if (!obj_request)return NULL ; - } - obj_request->object_name =memcpy (name, object_name, size); obj_request->which =BAD_WHICH ; obj_request->type = type;INIT_LIST_HEAD (&obj_request->links);
@@ -2114,8 +2069,6 @@ static void rbd_obj_request_destroy(struct kref *kref)break ; } -kfree (obj_request->object_name); - obj_request->object_name =NULL ;kmem_cache_free (rbd_obj_request_cache, obj_request); }
@@ -2490,17 +2443,11 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,while (resid) {struct ceph_osd_request *osd_req; -const char *object_name;u64 object_no = img_offset >> rbd_dev->header.obj_order;u64 offset =rbd_segment_offset (rbd_dev, img_offset);u64 length =rbd_segment_length (rbd_dev, img_offset, resid); - object_name =rbd_segment_name (rbd_dev, img_offset); -if (!object_name) - goto out_unwind; - obj_request =rbd_obj_request_create (object_name, type); -/* object request has its own copy of the object name */ -rbd_segment_name_free (object_name); + obj_request =rbd_obj_request_create (type);if (!obj_request) goto out_unwind;
@@ -2846,8 +2793,7 @@ static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)size_t size;int ret; - stat_request =rbd_obj_request_create (obj_request->object_name, -OBJ_REQUEST_PAGES ); + stat_request =rbd_obj_request_create (OBJ_REQUEST_PAGES );if (!stat_request)return -ENOMEM ;
@@ -6389,27 +6335,16 @@ static int rbd_slab_init(void)if (!rbd_obj_request_cache) goto out_err; -rbd_assert (!rbd_segment_name_cache); - rbd_segment_name_cache =kmem_cache_create ("rbd_segment_name" , -CEPH_MAX_OID_NAME_LEN +1 ,1 ,0 ,NULL ); -if (rbd_segment_name_cache) -return 0 ; -out_err: -kmem_cache_destroy (rbd_obj_request_cache); - rbd_obj_request_cache =NULL ; +return 0 ; +out_err:kmem_cache_destroy (rbd_img_request_cache); rbd_img_request_cache =NULL ; -return -ENOMEM ; }static void rbd_slab_exit (void ) { -rbd_assert (rbd_segment_name_cache); -kmem_cache_destroy (rbd_segment_name_cache); - rbd_segment_name_cache =NULL ; -rbd_assert (rbd_obj_request_cache);kmem_cache_destroy (rbd_obj_request_cache); rbd_obj_request_cache =NULL ;
include/linux/ceph/osdmap.h +0 -7
@@ -82,13 +82,6 @@ void ceph_oloc_copy(struct ceph_object_locator *dest,void ceph_oloc_destroy (struct ceph_object_locator *oloc);/* -* Maximum supported by kernel client object name length -* -* (probably outdated: must be >= RBD_MAX_MD_NAME_LEN -- currently 100) -*/ -#define CEPH_MAX_OID_NAME_LEN 100 - -/* * 51-char inline_name is long enough for all cephfs and all but one * rbd requests: <imgname> in "<imgname>.rbd"/"rbd_id.<imgname>" can be * arbitrarily long (~PAGE_SIZE). It's done once during rbd map; all
Signature · 1 line
2.4.3
Quoted 42 lines · Ilya Dryomov
Hello,
There are two pieces here: support for
RBD_FEATURE_DATA_POOL(patches 1-10) and a couple of cleanups / minor optimizations that build on it and allow us to get rid ofrbd_segment_name_cache(patches 11 - 15).Thanks,
Ilya
Ilya Dryomov (15):
rbd: usekstrndup()inrbd_header_from_disk()
rbd: kill rbd_image_header::{crypt_type,comp_type}
rbd: initializerbd_dev->header_olocearly
rbd: do away withobj_requestinrbd_obj_read_sync()
libceph: pass reply buffer length throughceph_osdc_call()
rbd: switchrbd_obj_method_sync()toceph_osdc_call()
rbd: remove now unusedrbd_obj_request_wait()and helpers
rbd: userbd_obj_bytes()more
rbd: introducerbd_init_layout()
rbd: support for data-pool feature
rbd: set offset and length outside ofrbd_obj_request_create()
rbd: factor out__rbd_osd_req_create()
rbd: RBD_V{1,2}_DATA_FORMAT macros
rbd: store and useobj_request->object_no
rbd: killobj_request->object_nameandrbd_segment_name_cachedrivers/block/rbd.c | 601 +++++++++++++++++--------------------------- drivers/block/rbd_types.h | 10 +- include/linux/ceph/osdmap.h | 7 - net/ceph/cls_lock_client.c | 2 +- net/ceph/osd_client.c | 7 +- 5 files changed, 238 insertions(+), 389 deletions(-)--
2.4.3--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Signature · 1 line
Jason