X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Finclude%2Flustre_net.h;h=3c1f5950346a2b2c96b0d354d3aecb1095e27c96;hb=d2c403363f65337166dc745b58d0a4529a534b84;hp=f39d8d7140b6380c1ca89cdb88bb52c4db5aa1e4;hpb=2ec741f46ac31cb566ea17d8ffda122398896102;p=fs%2Flustre-release.git diff --git a/lustre/include/lustre_net.h b/lustre/include/lustre_net.h index f39d8d7..3c1f595 100644 --- a/lustre/include/lustre_net.h +++ b/lustre/include/lustre_net.h @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2010, 2013, Intel Corporation. + * Copyright (c) 2010, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -55,8 +51,10 @@ * @{ */ +#include #include -#include +#include +#include #include #include #include @@ -71,12 +69,16 @@ #define PTLRPC_MD_OPTIONS 0 /** - * Max # of bulk operations in one request. + * log2 max # of bulk operations in one request: 2=4MB/RPC, 5=32MB/RPC, ... * In order for the client and server to properly negotiate the maximum * possible transfer size, PTLRPC_BULK_OPS_COUNT must be a power-of-two * value. The client is free to limit the actual RPC size for any bulk - * transfer via cl_max_pages_per_rpc to some non-power-of-two value. */ -#define PTLRPC_BULK_OPS_BITS 2 + * transfer via cl_max_pages_per_rpc to some non-power-of-two value. + * NOTE: This is limited to 16 (=64GB RPCs) by IOOBJ_MAX_BRW_BITS. */ +#define PTLRPC_BULK_OPS_BITS 4 +#if PTLRPC_BULK_OPS_BITS > 16 +#error "More than 65536 BRW RPCs not allowed by IOOBJ_MAX_BRW_BITS." +#endif #define PTLRPC_BULK_OPS_COUNT (1U << PTLRPC_BULK_OPS_BITS) /** * PTLRPC_BULK_OPS_MASK is for the convenience of the client only, and @@ -96,21 +98,21 @@ */ #define PTLRPC_MAX_BRW_BITS (LNET_MTU_BITS + PTLRPC_BULK_OPS_BITS) #define PTLRPC_MAX_BRW_SIZE (1 << PTLRPC_MAX_BRW_BITS) -#define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT) +#define PTLRPC_MAX_BRW_PAGES (PTLRPC_MAX_BRW_SIZE >> PAGE_SHIFT) #define ONE_MB_BRW_SIZE (1 << LNET_MTU_BITS) #define MD_MAX_BRW_SIZE (1 << LNET_MTU_BITS) -#define MD_MAX_BRW_PAGES (MD_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT) +#define MD_MAX_BRW_PAGES (MD_MAX_BRW_SIZE >> PAGE_SHIFT) #define DT_MAX_BRW_SIZE PTLRPC_MAX_BRW_SIZE -#define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_CACHE_SHIFT) +#define DT_MAX_BRW_PAGES (DT_MAX_BRW_SIZE >> PAGE_SHIFT) #define OFD_MAX_BRW_SIZE (1 << LNET_MTU_BITS) /* When PAGE_SIZE is a constant, we can check our arithmetic here with cpp! */ #if ((PTLRPC_MAX_BRW_PAGES & (PTLRPC_MAX_BRW_PAGES - 1)) != 0) # error "PTLRPC_MAX_BRW_PAGES isn't a power of two" #endif -#if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_CACHE_SIZE)) -# error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_CACHE_SIZE" +#if (PTLRPC_MAX_BRW_SIZE != (PTLRPC_MAX_BRW_PAGES * PAGE_SIZE)) +# error "PTLRPC_MAX_BRW_SIZE isn't PTLRPC_MAX_BRW_PAGES * PAGE_SIZE" #endif #if (PTLRPC_MAX_BRW_SIZE > LNET_MTU * PTLRPC_BULK_OPS_COUNT) # error "PTLRPC_MAX_BRW_SIZE too big" @@ -363,10 +365,10 @@ /** * The update request includes all of updates from the create, which might - * include linkea (4K maxim), together with other updates, we set it to 9K: - * lustre_msg + ptlrpc_body + UPDATE_BUF_SIZE (8K) + * include linkea (4K maxim), together with other updates, we set it to 1000K: + * lustre_msg + ptlrpc_body + OUT_UPDATE_BUFFER_SIZE_MAX */ -#define OUT_MAXREQSIZE (9 * 1024) +#define OUT_MAXREQSIZE (1000 * 1024) #define OUT_MAXREPSIZE MDS_MAXREPSIZE /** MDS_BUFSIZE = max_reqsize (w/o LOV EA) + max sptlrpc payload size */ @@ -451,10 +453,9 @@ */ /* depress threads factor for VM with small memory size */ #define OSS_THR_FACTOR min_t(int, 8, \ - NUM_CACHEPAGES >> (28 - PAGE_CACHE_SHIFT)) + NUM_CACHEPAGES >> (28 - PAGE_SHIFT)) #define OSS_NTHRS_INIT (PTLRPC_NTHRS_INIT + 1) #define OSS_NTHRS_BASE 64 -#define OSS_NTHRS_MAX 512 /* threads for handling "create" request */ #define OSS_CR_THR_FACTOR 1 @@ -603,6 +604,7 @@ struct ptlrpc_request_set { set_producer_func set_producer; /** opaq argument passed to the producer callback */ void *set_producer_arg; + unsigned int set_allow_intr:1; }; /** @@ -666,6 +668,10 @@ struct ptlrpc_reply_state { unsigned long rs_committed:1;/* the transaction was committed and the rs was dispatched by ptlrpc_commit_replies */ + atomic_t rs_refcount; /* number of users */ + /** Number of locks awaiting client ACK */ + int rs_nlocks; + /** Size of the state */ int rs_size; /** opcode */ @@ -677,43 +683,41 @@ struct ptlrpc_reply_state { struct obd_export *rs_export; struct ptlrpc_service_part *rs_svcpt; /** Lnet metadata handle for the reply */ - lnet_handle_md_t rs_md_h; - atomic_t rs_refcount; + lnet_handle_md_t rs_md_h; /** Context for the sevice thread */ - struct ptlrpc_svc_ctx *rs_svc_ctx; + struct ptlrpc_svc_ctx *rs_svc_ctx; /** Reply buffer (actually sent to the client), encoded if needed */ - struct lustre_msg *rs_repbuf; /* wrapper */ - /** Size of the reply buffer */ - int rs_repbuf_len; /* wrapper buf length */ - /** Size of the reply message */ - int rs_repdata_len; /* wrapper msg length */ - /** - * Actual reply message. Its content is encrupted (if needed) to - * produce reply buffer for actual sending. In simple case - * of no network encryption we jus set \a rs_repbuf to \a rs_msg - */ - struct lustre_msg *rs_msg; /* reply message */ - - /** Number of locks awaiting client ACK */ - int rs_nlocks; - /** Handles of locks awaiting client reply ACK */ - struct lustre_handle rs_locks[RS_MAX_LOCKS]; - /** Lock modes of locks in \a rs_locks */ - ldlm_mode_t rs_modes[RS_MAX_LOCKS]; + struct lustre_msg *rs_repbuf; /* wrapper */ + /** Size of the reply buffer */ + int rs_repbuf_len; /* wrapper buf length */ + /** Size of the reply message */ + int rs_repdata_len; /* wrapper msg length */ + /** + * Actual reply message. Its content is encrupted (if needed) to + * produce reply buffer for actual sending. In simple case + * of no network encryption we jus set \a rs_repbuf to \a rs_msg + */ + struct lustre_msg *rs_msg; /* reply message */ + + /** Handles of locks awaiting client reply ACK */ + struct lustre_handle rs_locks[RS_MAX_LOCKS]; + /** Lock modes of locks in \a rs_locks */ + enum ldlm_mode rs_modes[RS_MAX_LOCKS]; }; struct ptlrpc_thread; /** RPC stages */ enum rq_phase { - RQ_PHASE_NEW = 0xebc0de00, - RQ_PHASE_RPC = 0xebc0de01, - RQ_PHASE_BULK = 0xebc0de02, - RQ_PHASE_INTERPRET = 0xebc0de03, - RQ_PHASE_COMPLETE = 0xebc0de04, - RQ_PHASE_UNREGISTERING = 0xebc0de05, - RQ_PHASE_UNDEFINED = 0xebc0de06 + RQ_PHASE_NEW = 0xebc0de00, + RQ_PHASE_RPC = 0xebc0de01, + RQ_PHASE_BULK = 0xebc0de02, + RQ_PHASE_INTERPRET = 0xebc0de03, + RQ_PHASE_COMPLETE = 0xebc0de04, + RQ_PHASE_UNREG_RPC = 0xebc0de05, + RQ_PHASE_UNREG_BULK = 0xebc0de06, + RQ_PHASE_UNDEFINED = 0xebc0de07 }; /** Type of request interpreter call-back */ @@ -738,7 +742,7 @@ struct ptlrpc_request_pool { /** Maximum message size that would fit into a rquest from this pool */ int prp_rq_size; /** Function to allocate more requests for this pool */ - void (*prp_populate)(struct ptlrpc_request_pool *, int); + int (*prp_populate)(struct ptlrpc_request_pool *, int); }; struct lu_context; @@ -746,1010 +750,7 @@ struct lu_env; struct ldlm_lock; -/** - * \defgroup nrs Network Request Scheduler - * @{ - */ -struct ptlrpc_nrs_policy; -struct ptlrpc_nrs_resource; -struct ptlrpc_nrs_request; - -/** - * NRS control operations. - * - * These are common for all policies. - */ -enum ptlrpc_nrs_ctl { - /** - * Not a valid opcode. - */ - PTLRPC_NRS_CTL_INVALID, - /** - * Activate the policy. - */ - PTLRPC_NRS_CTL_START, - /** - * Reserved for multiple primary policies, which may be a possibility - * in the future. - */ - PTLRPC_NRS_CTL_STOP, - /** - * Policies can start using opcodes from this value and onwards for - * their own purposes; the assigned value itself is arbitrary. - */ - PTLRPC_NRS_CTL_1ST_POL_SPEC = 0x20, -}; - -/** - * ORR policy operations - */ -enum nrs_ctl_orr { - NRS_CTL_ORR_RD_QUANTUM = PTLRPC_NRS_CTL_1ST_POL_SPEC, - NRS_CTL_ORR_WR_QUANTUM, - NRS_CTL_ORR_RD_OFF_TYPE, - NRS_CTL_ORR_WR_OFF_TYPE, - NRS_CTL_ORR_RD_SUPP_REQ, - NRS_CTL_ORR_WR_SUPP_REQ, -}; - -/** - * NRS policy operations. - * - * These determine the behaviour of a policy, and are called in response to - * NRS core events. - */ -struct ptlrpc_nrs_pol_ops { - /** - * Called during policy registration; this operation is optional. - * - * \param[in,out] policy The policy being initialized - */ - int (*op_policy_init) (struct ptlrpc_nrs_policy *policy); - /** - * Called during policy unregistration; this operation is optional. - * - * \param[in,out] policy The policy being unregistered/finalized - */ - void (*op_policy_fini) (struct ptlrpc_nrs_policy *policy); - /** - * Called when activating a policy via lprocfs; policies allocate and - * initialize their resources here; this operation is optional. - * - * \param[in,out] policy The policy being started - * \param[in,out] arg A generic char buffer - * - * \see nrs_policy_start_locked() - */ - int (*op_policy_start) (struct ptlrpc_nrs_policy *policy, - char *arg); - /** - * Called when deactivating a policy via lprocfs; policies deallocate - * their resources here; this operation is optional - * - * \param[in,out] policy The policy being stopped - * - * \see nrs_policy_stop0() - */ - void (*op_policy_stop) (struct ptlrpc_nrs_policy *policy); - /** - * Used for policy-specific operations; i.e. not generic ones like - * \e PTLRPC_NRS_CTL_START and \e PTLRPC_NRS_CTL_GET_INFO; analogous - * to an ioctl; this operation is optional. - * - * \param[in,out] policy The policy carrying out operation \a opc - * \param[in] opc The command operation being carried out - * \param[in,out] arg An generic buffer for communication between the - * user and the control operation - * - * \retval -ve error - * \retval 0 success - * - * \see ptlrpc_nrs_policy_control() - */ - int (*op_policy_ctl) (struct ptlrpc_nrs_policy *policy, - enum ptlrpc_nrs_ctl opc, void *arg); - - /** - * Called when obtaining references to the resources of the resource - * hierarchy for a request that has arrived for handling at the PTLRPC - * service. Policies should return -ve for requests they do not wish - * to handle. This operation is mandatory. - * - * \param[in,out] policy The policy we're getting resources for. - * \param[in,out] nrq The request we are getting resources for. - * \param[in] parent The parent resource of the resource being - * requested; set to NULL if none. - * \param[out] resp The resource is to be returned here; the - * fallback policy in an NRS head should - * \e always return a non-NULL pointer value. - * \param[in] moving_req When set, signifies that this is an attempt - * to obtain resources for a request being moved - * to the high-priority NRS head by - * ldlm_lock_reorder_req(). - * This implies two things: - * 1. We are under obd_export::exp_rpc_lock and - * so should not sleep. - * 2. We should not perform non-idempotent or can - * skip performing idempotent operations that - * were carried out when resources were first - * taken for the request when it was initialized - * in ptlrpc_nrs_req_initialize(). - * - * \retval 0, +ve The level of the returned resource in the resource - * hierarchy; currently only 0 (for a non-leaf resource) - * and 1 (for a leaf resource) are supported by the - * framework. - * \retval -ve error - * - * \see ptlrpc_nrs_req_initialize() - * \see ptlrpc_nrs_hpreq_add_nolock() - * \see ptlrpc_nrs_req_hp_move() - */ - int (*op_res_get) (struct ptlrpc_nrs_policy *policy, - struct ptlrpc_nrs_request *nrq, - const struct ptlrpc_nrs_resource *parent, - struct ptlrpc_nrs_resource **resp, - bool moving_req); - /** - * Called when releasing references taken for resources in the resource - * hierarchy for the request; this operation is optional. - * - * \param[in,out] policy The policy the resource belongs to - * \param[in] res The resource to be freed - * - * \see ptlrpc_nrs_req_finalize() - * \see ptlrpc_nrs_hpreq_add_nolock() - * \see ptlrpc_nrs_req_hp_move() - */ - void (*op_res_put) (struct ptlrpc_nrs_policy *policy, - const struct ptlrpc_nrs_resource *res); - - /** - * Obtains a request for handling from the policy, and optionally - * removes the request from the policy; this operation is mandatory. - * - * \param[in,out] policy The policy to poll - * \param[in] peek When set, signifies that we just want to - * examine the request, and not handle it, so the - * request is not removed from the policy. - * \param[in] force When set, it will force a policy to return a - * request if it has one queued. - * - * \retval NULL No request available for handling - * \retval valid-pointer The request polled for handling - * - * \see ptlrpc_nrs_req_get_nolock() - */ - struct ptlrpc_nrs_request * - (*op_req_get) (struct ptlrpc_nrs_policy *policy, bool peek, - bool force); - /** - * Called when attempting to add a request to a policy for later - * handling; this operation is mandatory. - * - * \param[in,out] policy The policy on which to enqueue \a nrq - * \param[in,out] nrq The request to enqueue - * - * \retval 0 success - * \retval != 0 error - * - * \see ptlrpc_nrs_req_add_nolock() - */ - int (*op_req_enqueue) (struct ptlrpc_nrs_policy *policy, - struct ptlrpc_nrs_request *nrq); - /** - * Removes a request from the policy's set of pending requests. Normally - * called after a request has been polled successfully from the policy - * for handling; this operation is mandatory. - * - * \param[in,out] policy The policy the request \a nrq belongs to - * \param[in,out] nrq The request to dequeue - * - * \see ptlrpc_nrs_req_del_nolock() - */ - void (*op_req_dequeue) (struct ptlrpc_nrs_policy *policy, - struct ptlrpc_nrs_request *nrq); - /** - * Called after the request being carried out. Could be used for - * job/resource control; this operation is optional. - * - * \param[in,out] policy The policy which is stopping to handle request - * \a nrq - * \param[in,out] nrq The request - * - * \pre assert_spin_locked(&svcpt->scp_req_lock) - * - * \see ptlrpc_nrs_req_stop_nolock() - */ - void (*op_req_stop) (struct ptlrpc_nrs_policy *policy, - struct ptlrpc_nrs_request *nrq); - /** - * Registers the policy's lprocfs interface with a PTLRPC service. - * - * \param[in] svc The service - * - * \retval 0 success - * \retval != 0 error - */ - int (*op_lprocfs_init) (struct ptlrpc_service *svc); - /** - * Unegisters the policy's lprocfs interface with a PTLRPC service. - * - * In cases of failed policy registration in - * \e ptlrpc_nrs_policy_register(), this function may be called for a - * service which has not registered the policy successfully, so - * implementations of this method should make sure their operations are - * safe in such cases. - * - * \param[in] svc The service - */ - void (*op_lprocfs_fini) (struct ptlrpc_service *svc); -}; - -/** - * Policy flags - */ -enum nrs_policy_flags { - /** - * Fallback policy, use this flag only on a single supported policy per - * service. The flag cannot be used on policies that use - * \e PTLRPC_NRS_FL_REG_EXTERN - */ - PTLRPC_NRS_FL_FALLBACK = (1 << 0), - /** - * Start policy immediately after registering. - */ - PTLRPC_NRS_FL_REG_START = (1 << 1), - /** - * This is a policy registering from a module different to the one NRS - * core ships in (currently ptlrpc). - */ - PTLRPC_NRS_FL_REG_EXTERN = (1 << 2), -}; - -/** - * NRS queue type. - * - * Denotes whether an NRS instance is for handling normal or high-priority - * RPCs, or whether an operation pertains to one or both of the NRS instances - * in a service. - */ -enum ptlrpc_nrs_queue_type { - PTLRPC_NRS_QUEUE_REG = (1 << 0), - PTLRPC_NRS_QUEUE_HP = (1 << 1), - PTLRPC_NRS_QUEUE_BOTH = (PTLRPC_NRS_QUEUE_REG | PTLRPC_NRS_QUEUE_HP) -}; - -/** - * NRS head - * - * A PTLRPC service has at least one NRS head instance for handling normal - * priority RPCs, and may optionally have a second NRS head instance for - * handling high-priority RPCs. Each NRS head maintains a list of available - * policies, of which one and only one policy is acting as the fallback policy, - * and optionally a different policy may be acting as the primary policy. For - * all RPCs handled by this NRS head instance, NRS core will first attempt to - * enqueue the RPC using the primary policy (if any). The fallback policy is - * used in the following cases: - * - when there was no primary policy in the - * ptlrpc_nrs_pol_state::NRS_POL_STATE_STARTED state at the time the request - * was initialized. - * - when the primary policy that was at the - * ptlrpc_nrs_pol_state::PTLRPC_NRS_POL_STATE_STARTED state at the time the - * RPC was initialized, denoted it did not wish, or for some other reason was - * not able to handle the request, by returning a non-valid NRS resource - * reference. - * - when the primary policy that was at the - * ptlrpc_nrs_pol_state::PTLRPC_NRS_POL_STATE_STARTED state at the time the - * RPC was initialized, fails later during the request enqueueing stage. - * - * \see nrs_resource_get_safe() - * \see nrs_request_enqueue() - */ -struct ptlrpc_nrs { - spinlock_t nrs_lock; - /** XXX Possibly replace svcpt->scp_req_lock with another lock here. */ - /** - * List of registered policies - */ - struct list_head nrs_policy_list; - /** - * List of policies with queued requests. Policies that have any - * outstanding requests are queued here, and this list is queried - * in a round-robin manner from NRS core when obtaining a request - * for handling. This ensures that requests from policies that at some - * point transition away from the - * ptlrpc_nrs_pol_state::NRS_POL_STATE_STARTED state are drained. - */ - struct list_head nrs_policy_queued; - /** - * Service partition for this NRS head - */ - struct ptlrpc_service_part *nrs_svcpt; - /** - * Primary policy, which is the preferred policy for handling RPCs - */ - struct ptlrpc_nrs_policy *nrs_policy_primary; - /** - * Fallback policy, which is the backup policy for handling RPCs - */ - struct ptlrpc_nrs_policy *nrs_policy_fallback; - /** - * This NRS head handles either HP or regular requests - */ - enum ptlrpc_nrs_queue_type nrs_queue_type; - /** - * # queued requests from all policies in this NRS head - */ - unsigned long nrs_req_queued; - /** - * # scheduled requests from all policies in this NRS head - */ - unsigned long nrs_req_started; - /** - * # policies on this NRS - */ - unsigned nrs_num_pols; - /** - * This NRS head is in progress of starting a policy - */ - unsigned nrs_policy_starting:1; - /** - * In progress of shutting down the whole NRS head; used during - * unregistration - */ - unsigned nrs_stopping:1; - /** - * NRS policy is throttling reqeust - */ - unsigned nrs_throttling:1; -}; - -#define NRS_POL_NAME_MAX 16 - -struct ptlrpc_nrs_pol_desc; - -/** - * Service compatibility predicate; this determines whether a policy is adequate - * for handling RPCs of a particular PTLRPC service. - * - * XXX:This should give the same result during policy registration and - * unregistration, and for all partitions of a service; so the result should not - * depend on temporal service or other properties, that may influence the - * result. - */ -typedef bool (*nrs_pol_desc_compat_t) (const struct ptlrpc_service *svc, - const struct ptlrpc_nrs_pol_desc *desc); - -struct ptlrpc_nrs_pol_conf { - /** - * Human-readable policy name - */ - char nc_name[NRS_POL_NAME_MAX]; - /** - * NRS operations for this policy - */ - const struct ptlrpc_nrs_pol_ops *nc_ops; - /** - * Service compatibility predicate - */ - nrs_pol_desc_compat_t nc_compat; - /** - * Set for policies that support a single ptlrpc service, i.e. ones that - * have \a pd_compat set to nrs_policy_compat_one(). The variable value - * depicts the name of the single service that such policies are - * compatible with. - */ - const char *nc_compat_svc_name; - /** - * Owner module for this policy descriptor; policies registering from a - * different module to the one the NRS framework is held within - * (currently ptlrpc), should set this field to THIS_MODULE. - */ - struct module *nc_owner; - /** - * Policy registration flags; a bitmast of \e nrs_policy_flags - */ - unsigned nc_flags; -}; - -/** - * NRS policy registering descriptor - * - * Is used to hold a description of a policy that can be passed to NRS core in - * order to register the policy with NRS heads in different PTLRPC services. - */ -struct ptlrpc_nrs_pol_desc { - /** - * Human-readable policy name - */ - char pd_name[NRS_POL_NAME_MAX]; - /** - * Link into nrs_core::nrs_policies - */ - struct list_head pd_list; - /** - * NRS operations for this policy - */ - const struct ptlrpc_nrs_pol_ops *pd_ops; - /** - * Service compatibility predicate - */ - nrs_pol_desc_compat_t pd_compat; - /** - * Set for policies that are compatible with only one PTLRPC service. - * - * \see ptlrpc_nrs_pol_conf::nc_compat_svc_name - */ - const char *pd_compat_svc_name; - /** - * Owner module for this policy descriptor. - * - * We need to hold a reference to the module whenever we might make use - * of any of the module's contents, i.e. - * - If one or more instances of the policy are at a state where they - * might be handling a request, i.e. - * ptlrpc_nrs_pol_state::NRS_POL_STATE_STARTED or - * ptlrpc_nrs_pol_state::NRS_POL_STATE_STOPPING as we will have to - * call into the policy's ptlrpc_nrs_pol_ops() handlers. A reference - * is taken on the module when - * \e ptlrpc_nrs_pol_desc::pd_refs becomes 1, and released when it - * becomes 0, so that we hold only one reference to the module maximum - * at any time. - * - * We do not need to hold a reference to the module, even though we - * might use code and data from the module, in the following cases: - * - During external policy registration, because this should happen in - * the module's init() function, in which case the module is safe from - * removal because a reference is being held on the module by the - * kernel, and iirc kmod (and I guess module-init-tools also) will - * serialize any racing processes properly anyway. - * - During external policy unregistration, because this should happen - * in a module's exit() function, and any attempts to start a policy - * instance would need to take a reference on the module, and this is - * not possible once we have reached the point where the exit() - * handler is called. - * - During service registration and unregistration, as service setup - * and cleanup, and policy registration, unregistration and policy - * instance starting, are serialized by \e nrs_core::nrs_mutex, so - * as long as users adhere to the convention of registering policies - * in init() and unregistering them in module exit() functions, there - * should not be a race between these operations. - * - During any policy-specific lprocfs operations, because a reference - * is held by the kernel on a proc entry that has been entered by a - * syscall, so as long as proc entries are removed during unregistration time, - * then unregistration and lprocfs operations will be properly - * serialized. - */ - struct module *pd_owner; - /** - * Bitmask of \e nrs_policy_flags - */ - unsigned pd_flags; - /** - * # of references on this descriptor - */ - atomic_t pd_refs; -}; - -/** - * NRS policy state - * - * Policies transition from one state to the other during their lifetime - */ -enum ptlrpc_nrs_pol_state { - /** - * Not a valid policy state. - */ - NRS_POL_STATE_INVALID, - /** - * Policies are at this state either at the start of their life, or - * transition here when the user selects a different policy to act - * as the primary one. - */ - NRS_POL_STATE_STOPPED, - /** - * Policy is progress of stopping - */ - NRS_POL_STATE_STOPPING, - /** - * Policy is in progress of starting - */ - NRS_POL_STATE_STARTING, - /** - * A policy is in this state in two cases: - * - it is the fallback policy, which is always in this state. - * - it has been activated by the user; i.e. it is the primary policy, - */ - NRS_POL_STATE_STARTED, -}; - -/** - * NRS policy information - * - * Used for obtaining information for the status of a policy via lprocfs - */ -struct ptlrpc_nrs_pol_info { - /** - * Policy name - */ - char pi_name[NRS_POL_NAME_MAX]; - /** - * Current policy state - */ - enum ptlrpc_nrs_pol_state pi_state; - /** - * # RPCs enqueued for later dispatching by the policy - */ - long pi_req_queued; - /** - * # RPCs started for dispatch by the policy - */ - long pi_req_started; - /** - * Is this a fallback policy? - */ - unsigned pi_fallback:1; -}; - -/** - * NRS policy - * - * There is one instance of this for each policy in each NRS head of each - * PTLRPC service partition. - */ -struct ptlrpc_nrs_policy { - /** - * Linkage into the NRS head's list of policies, - * ptlrpc_nrs:nrs_policy_list - */ - struct list_head pol_list; - /** - * Linkage into the NRS head's list of policies with enqueued - * requests ptlrpc_nrs:nrs_policy_queued - */ - struct list_head pol_list_queued; - /** - * Current state of this policy - */ - enum ptlrpc_nrs_pol_state pol_state; - /** - * Bitmask of nrs_policy_flags - */ - unsigned pol_flags; - /** - * # RPCs enqueued for later dispatching by the policy - */ - long pol_req_queued; - /** - * # RPCs started for dispatch by the policy - */ - long pol_req_started; - /** - * Usage Reference count taken on the policy instance - */ - long pol_ref; - /** - * The NRS head this policy has been created at - */ - struct ptlrpc_nrs *pol_nrs; - /** - * Private policy data; varies by policy type - */ - void *pol_private; - /** - * Policy descriptor for this policy instance. - */ - struct ptlrpc_nrs_pol_desc *pol_desc; -}; - -/** - * NRS resource - * - * Resources are embedded into two types of NRS entities: - * - Inside NRS policies, in the policy's private data in - * ptlrpc_nrs_policy::pol_private - * - In objects that act as prime-level scheduling entities in different NRS - * policies; e.g. on a policy that performs round robin or similar order - * scheduling across client NIDs, there would be one NRS resource per unique - * client NID. On a policy which performs round robin scheduling across - * backend filesystem objects, there would be one resource associated with - * each of the backend filesystem objects partaking in the scheduling - * performed by the policy. - * - * NRS resources share a parent-child relationship, in which resources embedded - * in policy instances are the parent entities, with all scheduling entities - * a policy schedules across being the children, thus forming a simple resource - * hierarchy. This hierarchy may be extended with one or more levels in the - * future if the ability to have more than one primary policy is added. - * - * Upon request initialization, references to the then active NRS policies are - * taken and used to later handle the dispatching of the request with one of - * these policies. - * - * \see nrs_resource_get_safe() - * \see ptlrpc_nrs_req_add() - */ -struct ptlrpc_nrs_resource { - /** - * This NRS resource's parent; is NULL for resources embedded in NRS - * policy instances; i.e. those are top-level ones. - */ - struct ptlrpc_nrs_resource *res_parent; - /** - * The policy associated with this resource. - */ - struct ptlrpc_nrs_policy *res_policy; -}; - -enum { - NRS_RES_FALLBACK, - NRS_RES_PRIMARY, - NRS_RES_MAX -}; - -/* \name fifo - * - * FIFO policy - * - * This policy is a logical wrapper around previous, non-NRS functionality. - * It dispatches RPCs in the same order as they arrive from the network. This - * policy is currently used as the fallback policy, and the only enabled policy - * on all NRS heads of all PTLRPC service partitions. - * @{ - */ - -/** - * Private data structure for the FIFO policy - */ -struct nrs_fifo_head { - /** - * Resource object for policy instance. - */ - struct ptlrpc_nrs_resource fh_res; - /** - * List of queued requests. - */ - struct list_head fh_list; - /** - * For debugging purposes. - */ - __u64 fh_sequence; -}; - -struct nrs_fifo_req { - struct list_head fr_list; - __u64 fr_sequence; -}; - -/** @} fifo */ - -/** - * \name CRR-N - * - * CRR-N, Client Round Robin over NIDs - * @{ - */ - -/** - * private data structure for CRR-N NRS - */ -struct nrs_crrn_net { - struct ptlrpc_nrs_resource cn_res; - cfs_binheap_t *cn_binheap; - cfs_hash_t *cn_cli_hash; - /** - * Used when a new scheduling round commences, in order to synchronize - * all clients with the new round number. - */ - __u64 cn_round; - /** - * Determines the relevant ordering amongst request batches within a - * scheduling round. - */ - __u64 cn_sequence; - /** - * Round Robin quantum; the maximum number of RPCs that each request - * batch for each client can have in a scheduling round. - */ - __u16 cn_quantum; -}; - -/** - * Object representing a client in CRR-N, as identified by its NID - */ -struct nrs_crrn_client { - struct ptlrpc_nrs_resource cc_res; - struct hlist_node cc_hnode; - lnet_nid_t cc_nid; - /** - * The round number against which this client is currently scheduling - * requests. - */ - __u64 cc_round; - /** - * The sequence number used for requests scheduled by this client during - * the current round number. - */ - __u64 cc_sequence; - atomic_t cc_ref; - /** - * Round Robin quantum; the maximum number of RPCs the client is allowed - * to schedule in a single batch of each round. - */ - __u16 cc_quantum; - /** - * # of pending requests for this client, on all existing rounds - */ - __u16 cc_active; -}; - -/** - * CRR-N NRS request definition - */ -struct nrs_crrn_req { - /** - * Round number for this request; shared with all other requests in the - * same batch. - */ - __u64 cr_round; - /** - * Sequence number for this request; shared with all other requests in - * the same batch. - */ - __u64 cr_sequence; -}; - -/** - * CRR-N policy operations. - */ -enum nrs_ctl_crr { - /** - * Read the RR quantum size of a CRR-N policy. - */ - NRS_CTL_CRRN_RD_QUANTUM = PTLRPC_NRS_CTL_1ST_POL_SPEC, - /** - * Write the RR quantum size of a CRR-N policy. - */ - NRS_CTL_CRRN_WR_QUANTUM, -}; - -/** @} CRR-N */ - -/** - * \name ORR/TRR - * - * ORR/TRR (Object-based Round Robin/Target-based Round Robin) NRS policies - * @{ - */ - -/** - * Lower and upper byte offsets of a brw RPC - */ -struct nrs_orr_req_range { - __u64 or_start; - __u64 or_end; -}; - -/** - * RPC types supported by the ORR/TRR policies - */ -enum nrs_orr_supp { - NOS_OST_READ = (1 << 0), - NOS_OST_WRITE = (1 << 1), - NOS_OST_RW = (NOS_OST_READ | NOS_OST_WRITE), - /** - * Default value for policies. - */ - NOS_DFLT = NOS_OST_READ -}; - -/** - * As unique keys for grouping RPCs together, we use the object's OST FID for - * the ORR policy, and the OST index for the TRR policy. - * - * XXX: We waste some space for TRR policy instances by using a union, but it - * allows to consolidate some of the code between ORR and TRR, and these - * policies will probably eventually merge into one anyway. - */ -struct nrs_orr_key { - union { - /** object FID for ORR */ - struct lu_fid ok_fid; - /** OST index for TRR */ - __u32 ok_idx; - }; -}; - -/** - * The largest base string for unique hash/slab object names is - * "nrs_orr_reg_", so 13 characters. We add 3 to this to be used for the CPT - * id number, so this _should_ be more than enough for the maximum number of - * CPTs on any system. If it does happen that this statement is incorrect, - * nrs_orr_genobjname() will inevitably yield a non-unique name and cause - * kmem_cache_create() to complain (on Linux), so the erroneous situation - * will hopefully not go unnoticed. - */ -#define NRS_ORR_OBJ_NAME_MAX (sizeof("nrs_orr_reg_") + 3) - -/** - * private data structure for ORR and TRR NRS - */ -struct nrs_orr_data { - struct ptlrpc_nrs_resource od_res; - cfs_binheap_t *od_binheap; - cfs_hash_t *od_obj_hash; - struct kmem_cache *od_cache; - /** - * Used when a new scheduling round commences, in order to synchronize - * all object or OST batches with the new round number. - */ - __u64 od_round; - /** - * Determines the relevant ordering amongst request batches within a - * scheduling round. - */ - __u64 od_sequence; - /** - * RPC types that are currently supported. - */ - enum nrs_orr_supp od_supp; - /** - * Round Robin quantum; the maxium number of RPCs that each request - * batch for each object or OST can have in a scheduling round. - */ - __u16 od_quantum; - /** - * Whether to use physical disk offsets or logical file offsets. - */ - bool od_physical; - /** - * XXX: We need to provide a persistently allocated string to hold - * unique object names for this policy, since in currently supported - * versions of Linux by Lustre, kmem_cache_create() just sets a pointer - * to the name string provided. kstrdup() is used in the version of - * kmeme_cache_create() in current Linux mainline, so we may be able to - * remove this in the future. - */ - char od_objname[NRS_ORR_OBJ_NAME_MAX]; -}; - -/** - * Represents a backend-fs object or OST in the ORR and TRR policies - * respectively - */ -struct nrs_orr_object { - struct ptlrpc_nrs_resource oo_res; - struct hlist_node oo_hnode; - /** - * The round number against which requests are being scheduled for this - * object or OST - */ - __u64 oo_round; - /** - * The sequence number used for requests scheduled for this object or - * OST during the current round number. - */ - __u64 oo_sequence; - /** - * The key of the object or OST for which this structure instance is - * scheduling RPCs - */ - struct nrs_orr_key oo_key; - long oo_ref; - /** - * Round Robin quantum; the maximum number of RPCs that are allowed to - * be scheduled for the object or OST in a single batch of each round. - */ - __u16 oo_quantum; - /** - * # of pending requests for this object or OST, on all existing rounds - */ - __u16 oo_active; -}; - -/** - * ORR/TRR NRS request definition - */ -struct nrs_orr_req { - /** - * The offset range this request covers - */ - struct nrs_orr_req_range or_range; - /** - * Round number for this request; shared with all other requests in the - * same batch. - */ - __u64 or_round; - /** - * Sequence number for this request; shared with all other requests in - * the same batch. - */ - __u64 or_sequence; - /** - * For debugging purposes. - */ - struct nrs_orr_key or_key; - /** - * An ORR policy instance has filled in request information while - * enqueueing the request on the service partition's regular NRS head. - */ - unsigned int or_orr_set:1; - /** - * A TRR policy instance has filled in request information while - * enqueueing the request on the service partition's regular NRS head. - */ - unsigned int or_trr_set:1; - /** - * Request offset ranges have been filled in with logical offset - * values. - */ - unsigned int or_logical_set:1; - /** - * Request offset ranges have been filled in with physical offset - * values. - */ - unsigned int or_physical_set:1; -}; - -/** @} ORR/TRR */ - -#include - -/** - * NRS request - * - * Instances of this object exist embedded within ptlrpc_request; the main - * purpose of this object is to hold references to the request's resources - * for the lifetime of the request, and to hold properties that policies use - * use for determining the request's scheduling priority. - * */ -struct ptlrpc_nrs_request { - /** - * The request's resource hierarchy. - */ - struct ptlrpc_nrs_resource *nr_res_ptrs[NRS_RES_MAX]; - /** - * Index into ptlrpc_nrs_request::nr_res_ptrs of the resource of the - * policy that was used to enqueue the request. - * - * \see nrs_request_enqueue() - */ - unsigned nr_res_idx; - unsigned nr_initialized:1; - unsigned nr_enqueued:1; - unsigned nr_started:1; - unsigned nr_finalized:1; - cfs_binheap_node_t nr_node; - - /** - * Policy-specific fields, used for determining a request's scheduling - * priority, and other supporting functionality. - */ - union { - /** - * Fields for the FIFO policy - */ - struct nrs_fifo_req fifo; - /** - * CRR-N request defintion - */ - struct nrs_crrn_req crr; - /** ORR and TRR share the same request definition */ - struct nrs_orr_req orr; - /** - * TBF request definition - */ - struct nrs_tbf_req tbf; - } nr_u; - /** - * Externally-registering policies may want to use this to allocate - * their own request properties. - */ - void *ext; -}; - -/** @} nrs */ +#include /** * Basic request prioritization operations structure. @@ -1791,6 +792,8 @@ struct ptlrpc_cli_req { time_t cr_reply_deadline; /** when req bulk unlink must finish. */ time_t cr_bulk_deadline; + /** when req unlink must finish. */ + time_t cr_req_deadline; /** Portal to which this request would be sent */ short cr_req_ptl; /** Portal where to wait for reply and where reply would be sent */ @@ -1828,6 +831,8 @@ struct ptlrpc_cli_req { union ptlrpc_async_args cr_async_args; /** Opaq data for replay and commit callbacks. */ void *cr_cb_data; + /** Link to the imp->imp_unreplied_list */ + struct list_head cr_unreplied_list; /** * Commit callback, called when request is committed and about to be * freed. @@ -1847,6 +852,7 @@ struct ptlrpc_cli_req { #define rq_real_sent rq_cli.cr_sent_out #define rq_reply_deadline rq_cli.cr_reply_deadline #define rq_bulk_deadline rq_cli.cr_bulk_deadline +#define rq_req_deadline rq_cli.cr_req_deadline #define rq_nr_resend rq_cli.cr_resend_nr #define rq_request_portal rq_cli.cr_req_ptl #define rq_reply_portal rq_cli.cr_rep_ptl @@ -1866,6 +872,7 @@ struct ptlrpc_cli_req { #define rq_resend_cb rq_cli.cr_resend_cb #define rq_async_args rq_cli.cr_async_args #define rq_cb_data rq_cli.cr_cb_data +#define rq_unreplied_list rq_cli.cr_unreplied_list #define rq_commit_cb rq_cli.cr_commit_cb #define rq_replay_cb rq_cli.cr_replay_cb @@ -1887,7 +894,7 @@ struct ptlrpc_srv_req { /** history sequence # */ __u64 sr_hist_seq; /** the index of service's srv_at_array into which request is linked */ - time_t sr_at_index; + __u32 sr_at_index; /** authed uid */ uid_t sr_auth_uid; /** authed uid mapped to */ @@ -1908,7 +915,7 @@ struct ptlrpc_srv_req { struct ptlrpc_svc_ctx *sr_svc_ctx; /** (server side), pointed directly into req buffer */ struct ptlrpc_user_desc *sr_user_desc; - /** separated reply state */ + /** separated reply state, may be vmalloc'd */ struct ptlrpc_reply_state *sr_reply_state; /** server-side hp handlers */ struct ptlrpc_hpreq_ops *sr_ops; @@ -1957,7 +964,7 @@ struct ptlrpc_request { * rq_list */ spinlock_t rq_lock; - /** client-side flags are serialized by rq_lock */ + /** client-side flags are serialized by rq_lock @{ */ unsigned int rq_intr:1, rq_replied:1, rq_err:1, rq_timedout:1, rq_resend:1, rq_restart:1, /** @@ -1973,24 +980,30 @@ struct ptlrpc_request { rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1, rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1, rq_early:1, - rq_req_unlink:1, rq_reply_unlink:1, - rq_memalloc:1, /* req originated from "kswapd" */ - /* server-side flags */ - rq_packed_final:1, /* packed final reply */ - rq_hp:1, /* high priority RPC */ - rq_at_linked:1, /* link into service's srv_at_array */ - rq_reply_truncate:1, - rq_committed:1, - /* whether the "rq_set" is a valid one */ - rq_invalid_rqset:1, + rq_req_unlinked:1, /* unlinked request buffer from lnet */ + rq_reply_unlinked:1, /* unlinked reply buffer from lnet */ + rq_memalloc:1, /* req originated from "kswapd" */ + rq_committed:1, + rq_reply_truncated:1, + /** whether the "rq_set" is a valid one */ + rq_invalid_rqset:1, rq_generation_set:1, - /* do not resend request on -EINPROGRESS */ + /** do not resend request on -EINPROGRESS */ rq_no_retry_einprogress:1, /* allow the req to be sent if the import is in recovery * status */ rq_allow_replay:1, /* bulk request, sent to server, but uncommitted */ - rq_unstable:1; + rq_unstable:1, + rq_allow_intr:1; + /** @} */ + + /** server-side flags @{ */ + unsigned int + rq_hp:1, /**< high priority RPC */ + rq_at_linked:1, /**< link into service's srv_at_array */ + rq_packed_final:1; /**< packed final reply */ + /** @} */ /** one of RQ_PHASE_* */ enum rq_phase rq_phase; @@ -2020,9 +1033,11 @@ struct ptlrpc_request { /** Transaction number */ __u64 rq_transno; /** xid */ - __u64 rq_xid; + __u64 rq_xid; + /** bulk match bits */ + __u64 rq_mbits; /** - * List item to for replay list. Not yet commited requests get linked + * List item to for replay list. Not yet committed requests get linked * there. * Also see \a rq_replay comment above. * It's also link chain on obd_export::exp_req_replay_queue @@ -2047,7 +1062,6 @@ struct ptlrpc_request { rq_bulk_write:1, /* request bulk write */ /* server authentication flags */ rq_auth_gss:1, /* authenticated by gss */ - rq_auth_remote:1, /* authed as remote user */ rq_auth_usr_root:1, /* authed as root */ rq_auth_usr_mdt:1, /* authed as mdt */ rq_auth_usr_ost:1, /* authed as ost */ @@ -2061,9 +1075,9 @@ struct ptlrpc_request { /** various buffer pointers */ - struct lustre_msg *rq_reqbuf; /**< req wrapper */ - char *rq_repbuf; /**< rep buffer */ - struct lustre_msg *rq_repdata; /**< rep wrapper msg */ + struct lustre_msg *rq_reqbuf; /**< req wrapper, vmalloc*/ + char *rq_repbuf; /**< rep buffer, vmalloc */ + struct lustre_msg *rq_repdata; /**< rep wrapper msg */ /** only in priv mode */ struct lustre_msg *rq_clrbuf; int rq_reqbuf_len; /* req wrapper buf len */ @@ -2075,7 +1089,6 @@ struct ptlrpc_request { /** early replies go to offset 0, regular replies go after that */ unsigned int rq_reply_off; - /** @} */ /** Fields that help to see if request and reply were swabbed or not */ @@ -2092,6 +1105,8 @@ struct ptlrpc_request { lnet_nid_t rq_self; /** Peer description (the other side) */ lnet_process_id_t rq_peer; + /** Descriptor for the NID from which the peer sent the request. */ + lnet_process_id_t rq_source; /** * service time estimate (secs) * If the request is not served by this time, it is marked as timed out. @@ -2214,22 +1229,24 @@ static inline void lustre_set_rep_swabbed(struct ptlrpc_request *req, static inline const char * ptlrpc_phase2str(enum rq_phase phase) { - switch (phase) { - case RQ_PHASE_NEW: - return "New"; - case RQ_PHASE_RPC: - return "Rpc"; - case RQ_PHASE_BULK: - return "Bulk"; - case RQ_PHASE_INTERPRET: - return "Interpret"; - case RQ_PHASE_COMPLETE: - return "Complete"; - case RQ_PHASE_UNREGISTERING: - return "Unregistering"; - default: - return "?Phase?"; - } + switch (phase) { + case RQ_PHASE_NEW: + return "New"; + case RQ_PHASE_RPC: + return "Rpc"; + case RQ_PHASE_BULK: + return "Bulk"; + case RQ_PHASE_INTERPRET: + return "Interpret"; + case RQ_PHASE_COMPLETE: + return "Complete"; + case RQ_PHASE_UNREG_RPC: + return "UnregRPC"; + case RQ_PHASE_UNREG_BULK: + return "UnregBULK"; + default: + return "?Phase?"; + } } /** @@ -2245,23 +1262,23 @@ ptlrpc_rqphase2str(struct ptlrpc_request *req) /** * Debugging functions and helpers to print request structure into debug log * @{ - */ + */ /* Spare the preprocessor, spoil the bugs. */ #define FLAG(field, str) (field ? str : "") /** Convert bit flags into a string */ -#define DEBUG_REQ_FLAGS(req) \ - ptlrpc_rqphase2str(req), \ - FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \ - FLAG(req->rq_err, "E"), \ - FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \ - FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \ - FLAG(req->rq_no_resend, "N"), \ - FLAG(req->rq_waiting, "W"), \ - FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \ - FLAG(req->rq_committed, "M") - -#define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s" +#define DEBUG_REQ_FLAGS(req) \ + ptlrpc_rqphase2str(req), \ + FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \ + FLAG(req->rq_err, "E"), FLAG(req->rq_net_err, "e"), \ + FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \ + FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \ + FLAG(req->rq_no_resend, "N"), \ + FLAG(req->rq_waiting, "W"), \ + FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H"), \ + FLAG(req->rq_committed, "M") + +#define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s%s" void _debug_req(struct ptlrpc_request *req, struct libcfs_debug_msg_data *data, const char *fmt, ...) @@ -2288,7 +1305,7 @@ do { \ #define DEBUG_REQ(level, req, fmt, args...) \ do { \ if ((level) & (D_ERROR | D_WARNING)) { \ - static cfs_debug_limit_state_t cdls; \ + static struct cfs_debug_limit_state cdls; \ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \ debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\ } else { \ @@ -2314,12 +1331,98 @@ struct ptlrpc_bulk_page { struct page *bp_page; }; -#define BULK_GET_SOURCE 0 -#define BULK_PUT_SINK 1 -#define BULK_GET_SINK 2 -#define BULK_PUT_SOURCE 3 +enum ptlrpc_bulk_op_type { + PTLRPC_BULK_OP_ACTIVE = 0x00000001, + PTLRPC_BULK_OP_PASSIVE = 0x00000002, + PTLRPC_BULK_OP_PUT = 0x00000004, + PTLRPC_BULK_OP_GET = 0x00000008, + PTLRPC_BULK_BUF_KVEC = 0x00000010, + PTLRPC_BULK_BUF_KIOV = 0x00000020, + PTLRPC_BULK_GET_SOURCE = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_GET, + PTLRPC_BULK_PUT_SINK = PTLRPC_BULK_OP_PASSIVE | PTLRPC_BULK_OP_PUT, + PTLRPC_BULK_GET_SINK = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_GET, + PTLRPC_BULK_PUT_SOURCE = PTLRPC_BULK_OP_ACTIVE | PTLRPC_BULK_OP_PUT, +}; -/** +static inline bool ptlrpc_is_bulk_op_get(enum ptlrpc_bulk_op_type type) +{ + return (type & PTLRPC_BULK_OP_GET) == PTLRPC_BULK_OP_GET; +} + +static inline bool ptlrpc_is_bulk_get_source(enum ptlrpc_bulk_op_type type) +{ + return (type & PTLRPC_BULK_GET_SOURCE) == PTLRPC_BULK_GET_SOURCE; +} + +static inline bool ptlrpc_is_bulk_put_sink(enum ptlrpc_bulk_op_type type) +{ + return (type & PTLRPC_BULK_PUT_SINK) == PTLRPC_BULK_PUT_SINK; +} + +static inline bool ptlrpc_is_bulk_get_sink(enum ptlrpc_bulk_op_type type) +{ + return (type & PTLRPC_BULK_GET_SINK) == PTLRPC_BULK_GET_SINK; +} + +static inline bool ptlrpc_is_bulk_put_source(enum ptlrpc_bulk_op_type type) +{ + return (type & PTLRPC_BULK_PUT_SOURCE) == PTLRPC_BULK_PUT_SOURCE; +} + +static inline bool ptlrpc_is_bulk_desc_kvec(enum ptlrpc_bulk_op_type type) +{ + return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV)) + == PTLRPC_BULK_BUF_KVEC; +} + +static inline bool ptlrpc_is_bulk_desc_kiov(enum ptlrpc_bulk_op_type type) +{ + return ((type & PTLRPC_BULK_BUF_KVEC) | (type & PTLRPC_BULK_BUF_KIOV)) + == PTLRPC_BULK_BUF_KIOV; +} + +static inline bool ptlrpc_is_bulk_op_active(enum ptlrpc_bulk_op_type type) +{ + return ((type & PTLRPC_BULK_OP_ACTIVE) | + (type & PTLRPC_BULK_OP_PASSIVE)) + == PTLRPC_BULK_OP_ACTIVE; +} + +static inline bool ptlrpc_is_bulk_op_passive(enum ptlrpc_bulk_op_type type) +{ + return ((type & PTLRPC_BULK_OP_ACTIVE) | + (type & PTLRPC_BULK_OP_PASSIVE)) + == PTLRPC_BULK_OP_PASSIVE; +} + +struct ptlrpc_bulk_frag_ops { + /** + * Add a page \a page to the bulk descriptor \a desc + * Data to transfer in the page starts at offset \a pageoffset and + * amount of data to transfer from the page is \a len + */ + void (*add_kiov_frag)(struct ptlrpc_bulk_desc *desc, + struct page *page, int pageoffset, int len); + + /* + * Add a \a fragment to the bulk descriptor \a desc. + * Data to transfer in the fragment is pointed to by \a frag + * The size of the fragment is \a len + */ + int (*add_iov_frag)(struct ptlrpc_bulk_desc *desc, void *frag, int len); + + /** + * Uninitialize and free bulk descriptor \a desc. + * Works on bulk descriptors both from server and client side. + */ + void (*release_frags)(struct ptlrpc_bulk_desc *desc); +}; + +extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops; +extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops; +extern const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kvec_ops; + +/* * Definition of bulk descriptor. * Bulks are special "Two phase" RPCs where initial request message * is sent first and it is followed bt a transfer (o receiving) of a large @@ -2332,14 +1435,14 @@ struct ptlrpc_bulk_page { struct ptlrpc_bulk_desc { /** completed with failure */ unsigned long bd_failure:1; - /** {put,get}{source,sink} */ - unsigned long bd_type:2; /** client side */ unsigned long bd_registered:1; /** For serialization with callback */ spinlock_t bd_lock; /** Import generation when request for this bulk was sent */ int bd_import_generation; + /** {put,get}{source,sink}{kvec,kiov} */ + enum ptlrpc_bulk_op_type bd_type; /** LNet portal for this bulk */ __u32 bd_portal; /** Server side - export this bulk created for */ @@ -2348,13 +1451,14 @@ struct ptlrpc_bulk_desc { struct obd_import *bd_import; /** Back pointer to the request */ struct ptlrpc_request *bd_req; + struct ptlrpc_bulk_frag_ops *bd_frag_ops; wait_queue_head_t bd_waitq; /* server side only WQ */ int bd_iov_count; /* # entries in bd_iov */ int bd_max_iov; /* allocated size of bd_iov */ int bd_nob; /* # bytes covered */ int bd_nob_transferred; /* # bytes GOT/PUT */ - __u64 bd_last_xid; + __u64 bd_last_mbits; struct ptlrpc_cb_id bd_cbid; /* network callback info */ lnet_nid_t bd_sender; /* stash event::sender */ @@ -2363,14 +1467,32 @@ struct ptlrpc_bulk_desc { /** array of associated MDs */ lnet_handle_md_t bd_mds[PTLRPC_BULK_OPS_COUNT]; - /* - * encrypt iov, size is either 0 or bd_iov_count. - */ - lnet_kiov_t *bd_enc_iov; + union { + struct { + /* + * encrypt iov, size is either 0 or bd_iov_count. + */ + lnet_kiov_t *bd_enc_vec; + lnet_kiov_t *bd_vec; + } bd_kiov; + + struct { + struct kvec *bd_enc_kvec; + struct kvec *bd_kvec; + } bd_kvec; + } bd_u; - lnet_kiov_t bd_iov[0]; }; +#define GET_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_vec) +#define BD_GET_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_vec[i]) +#define GET_ENC_KIOV(desc) ((desc)->bd_u.bd_kiov.bd_enc_vec) +#define BD_GET_ENC_KIOV(desc, i) ((desc)->bd_u.bd_kiov.bd_enc_vec[i]) +#define GET_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_kvec) +#define BD_GET_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_kvec[i]) +#define GET_ENC_KVEC(desc) ((desc)->bd_u.bd_kvec.bd_enc_kvec) +#define BD_GET_ENC_KVEC(desc, i) ((desc)->bd_u.bd_kvec.bd_enc_kvec[i]) + enum { SVC_STOPPED = 1 << 0, SVC_STOPPING = 1 << 1, @@ -2385,30 +1507,30 @@ enum { * Definition of server service thread structure */ struct ptlrpc_thread { - /** - * List of active threads in svc->srv_threads - */ + /** + * List of active threads in svc->srv_threads + */ struct list_head t_link; - /** - * thread-private data (preallocated memory) - */ - void *t_data; - __u32 t_flags; - /** - * service thread index, from ptlrpc_start_threads - */ - unsigned int t_id; - /** - * service thread pid - */ - pid_t t_pid; - /** - * put watchdog in the structure per thread b=14840 - */ - struct lc_watchdog *t_watchdog; - /** - * the svc this thread belonged to b=18582 - */ + /** + * thread-private data (preallocated vmalloc'd memory) + */ + void *t_data; + __u32 t_flags; + /** + * service thread index, from ptlrpc_start_threads + */ + unsigned int t_id; + /** + * service thread pid + */ + pid_t t_pid; + /** + * put watchdog in the structure per thread b=14840 + */ + struct lc_watchdog *t_watchdog; + /** + * the svc this thread belonged to b=18582 + */ struct ptlrpc_service_part *t_svcpt; wait_queue_head_t t_ctl_waitq; struct lu_env *t_env; @@ -2761,34 +1883,38 @@ struct ptlrpcd_ctl { * Stop completion. */ struct completion pc_finishing; - /** - * Thread requests set. - */ - struct ptlrpc_request_set *pc_set; - /** + /** + * Thread requests set. + */ + struct ptlrpc_request_set *pc_set; + /** * Thread name used in kthread_run() - */ - char pc_name[16]; - /** - * Environment for request interpreters to run in. - */ - struct lu_env pc_env; + */ + char pc_name[16]; + /** + * CPT the thread is bound on. + */ + int pc_cpt; /** * Index of ptlrpcd thread in the array. */ - int pc_index; - /** - * Number of the ptlrpcd's partners. - */ - int pc_npartners; - /** - * Pointer to the array of partners' ptlrpcd_ctl structure. - */ - struct ptlrpcd_ctl **pc_partners; - /** - * Record the partner index to be processed next. - */ - int pc_cursor; + int pc_index; + /** + * Pointer to the array of partners' ptlrpcd_ctl structure. + */ + struct ptlrpcd_ctl **pc_partners; + /** + * Number of the ptlrpcd's partners. + */ + int pc_npartners; + /** + * Record the partner index to be processed next. + */ + int pc_cursor; + /** + * Error code if the thread failed to fully start. + */ + int pc_error; }; /* Bits for pc_flags */ @@ -2811,10 +1937,6 @@ enum ptlrpcd_ctl_flags { * This is a recovery ptlrpc thread. */ LIOD_RECOVERY = 1 << 3, - /** - * The ptlrpcd is bound to some CPU core. - */ - LIOD_BIND = 1 << 4, }; /** @@ -2894,8 +2016,11 @@ extern lnet_pid_t ptl_get_pid(void); */ #ifdef HAVE_SERVER_SUPPORT struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req, - unsigned npages, unsigned max_brw, - unsigned type, unsigned portal); + unsigned nfrags, unsigned max_brw, + unsigned int type, + unsigned portal, + const struct ptlrpc_bulk_frag_ops + *ops); int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc); void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc); @@ -2918,17 +2043,16 @@ int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async); static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req) { struct ptlrpc_bulk_desc *desc; - int rc; + int rc; - LASSERT(req != NULL); + LASSERT(req != NULL); desc = req->rq_bulk; - if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) && - req->rq_bulk_deadline > cfs_time_current_sec()) - return 1; + if (req->rq_bulk_deadline > cfs_time_current_sec()) + return 1; - if (!desc) - return 0; + if (!desc) + return 0; spin_lock(&desc->bd_lock); rc = desc->bd_md_count; @@ -2942,7 +2066,6 @@ int ptlrpc_send_reply(struct ptlrpc_request *req, int flags); int ptlrpc_reply(struct ptlrpc_request *req); int ptlrpc_send_error(struct ptlrpc_request *req, int difficult); int ptlrpc_error(struct ptlrpc_request *req); -void ptlrpc_resend_req(struct ptlrpc_request *request); int ptlrpc_at_get_net_latency(struct ptlrpc_request *req); int ptl_send_rpc(struct ptlrpc_request *request, int noreply); int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd); @@ -2959,11 +2082,11 @@ void ptlrpc_request_committed(struct ptlrpc_request *req, int force); void ptlrpc_init_client(int req_portal, int rep_portal, char *name, struct ptlrpc_client *); void ptlrpc_cleanup_client(struct obd_import *imp); -struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid); +struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid, + lnet_nid_t nid4refnet); int ptlrpc_queue_wait(struct ptlrpc_request *req); int ptlrpc_replay_req(struct ptlrpc_request *req); -int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async); void ptlrpc_restart_req(struct ptlrpc_request *req); void ptlrpc_abort_inflight(struct obd_import *imp); void ptlrpc_cleanup_imp(struct obd_import *imp); @@ -2974,23 +2097,18 @@ struct ptlrpc_request_set *ptlrpc_prep_fcset(int max, set_producer_func func, void *arg); int ptlrpc_set_add_cb(struct ptlrpc_request_set *set, set_interpreter_func fn, void *data); -int ptlrpc_set_next_timeout(struct ptlrpc_request_set *); int ptlrpc_check_set(const struct lu_env *env, struct ptlrpc_request_set *set); int ptlrpc_set_wait(struct ptlrpc_request_set *); -int ptlrpc_expired_set(void *data); -void ptlrpc_interrupted_set(void *data); void ptlrpc_mark_interrupted(struct ptlrpc_request *req); void ptlrpc_set_destroy(struct ptlrpc_request_set *); void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *); -void ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc, - struct ptlrpc_request *req); void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool); -void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq); +int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq); struct ptlrpc_request_pool * ptlrpc_init_rq_pool(int, int, - void (*populate_pool)(struct ptlrpc_request_pool *, int)); + int (*populate_pool)(struct ptlrpc_request_pool *, int)); void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req); struct ptlrpc_request *ptlrpc_request_alloc(struct obd_import *imp, @@ -3007,30 +2125,21 @@ struct ptlrpc_request *ptlrpc_request_alloc_pack(struct obd_import *imp, int ptlrpc_request_bufs_pack(struct ptlrpc_request *request, __u32 version, int opcode, char **bufs, struct ptlrpc_cli_ctx *ctx); -struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version, - int opcode, int count, __u32 *lengths, - char **bufs); -struct ptlrpc_request *ptlrpc_prep_req_pool(struct obd_import *imp, - __u32 version, int opcode, - int count, __u32 *lengths, char **bufs, - struct ptlrpc_request_pool *pool); void ptlrpc_req_finished(struct ptlrpc_request *request); void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request); struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req); struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req, - unsigned npages, unsigned max_brw, - unsigned type, unsigned portal); -void __ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk, int pin); -static inline void ptlrpc_free_bulk_pin(struct ptlrpc_bulk_desc *bulk) -{ - __ptlrpc_free_bulk(bulk, 1); -} -static inline void ptlrpc_free_bulk_nopin(struct ptlrpc_bulk_desc *bulk) -{ - __ptlrpc_free_bulk(bulk, 0); -} + unsigned nfrags, unsigned max_brw, + unsigned int type, + unsigned portal, + const struct ptlrpc_bulk_frag_ops + *ops); + +int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc, + void *frag, int len); void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc, - struct page *page, int pageoffset, int len, int); + struct page *page, int pageoffset, int len, + int pin); static inline void ptlrpc_prep_bulk_page_pin(struct ptlrpc_bulk_desc *desc, struct page *page, int pageoffset, int len) @@ -3045,6 +2154,20 @@ static inline void ptlrpc_prep_bulk_page_nopin(struct ptlrpc_bulk_desc *desc, __ptlrpc_prep_bulk_page(desc, page, pageoffset, len, 0); } +void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk); + +static inline void ptlrpc_release_bulk_page_pin(struct ptlrpc_bulk_desc *desc) +{ + int i; + + for (i = 0; i < desc->bd_iov_count ; i++) + put_page(BD_GET_KIOV(desc, i).kiov_page); +} + +static inline void ptlrpc_release_bulk_noop(struct ptlrpc_bulk_desc *desc) +{ +} + void ptlrpc_retain_replayable_request(struct ptlrpc_request *req, struct obd_import *imp); __u64 ptlrpc_next_xid(void); @@ -3162,21 +2285,21 @@ int ptlrpc_init_import(struct obd_import *imp); int ptlrpc_disconnect_import(struct obd_import *imp, int noclose); int ptlrpc_import_recovery_state_machine(struct obd_import *imp); void deuuidify(char *uuid, const char *prefix, char **uuid_start, - int *uuid_len); - + int *uuid_len); +void ptlrpc_import_enter_resend(struct obd_import *imp); /* ptlrpc/pack_generic.c */ int ptlrpc_reconnect_import(struct obd_import *imp); /** @} */ /** - * ptlrpc msg buffer and swab interface + * ptlrpc msg buffer and swab interface * * @{ */ int ptlrpc_buf_need_swab(struct ptlrpc_request *req, const int inout, - int index); + __u32 index); void ptlrpc_buf_set_swabbed(struct ptlrpc_request *req, const int inout, - int index); + __u32 index); int ptlrpc_unpack_rep_msg(struct ptlrpc_request *req, int len); int ptlrpc_unpack_req_msg(struct ptlrpc_request *req, int len); @@ -3196,32 +2319,32 @@ int lustre_shrink_msg(struct lustre_msg *msg, int segment, unsigned int newlen, int move_data); void lustre_free_reply_state(struct ptlrpc_reply_state *rs); int __lustre_unpack_msg(struct lustre_msg *m, int len); -int lustre_msg_hdr_size(__u32 magic, int count); -int lustre_msg_size(__u32 magic, int count, __u32 *lengths); -int lustre_msg_size_v2(int count, __u32 *lengths); -int lustre_packed_msg_size(struct lustre_msg *msg); -int lustre_msg_early_size(void); -void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, int n, int min_size); -void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen); -int lustre_msg_buflen(struct lustre_msg *m, int n); -void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len); -int lustre_msg_bufcount(struct lustre_msg *m); -char *lustre_msg_string(struct lustre_msg *m, int n, int max_len); +__u32 lustre_msg_hdr_size(__u32 magic, __u32 count); +__u32 lustre_msg_size(__u32 magic, int count, __u32 *lengths); +__u32 lustre_msg_size_v2(int count, __u32 *lengths); +__u32 lustre_packed_msg_size(struct lustre_msg *msg); +__u32 lustre_msg_early_size(void); +void *lustre_msg_buf_v2(struct lustre_msg_v2 *m, __u32 n, __u32 min_size); +void *lustre_msg_buf(struct lustre_msg *m, __u32 n, __u32 minlen); +__u32 lustre_msg_buflen(struct lustre_msg *m, __u32 n); +void lustre_msg_set_buflen(struct lustre_msg *m, __u32 n, __u32 len); +__u32 lustre_msg_bufcount(struct lustre_msg *m); +char *lustre_msg_string(struct lustre_msg *m, __u32 n, __u32 max_len); __u32 lustre_msghdr_get_flags(struct lustre_msg *msg); void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags); __u32 lustre_msg_get_flags(struct lustre_msg *msg); -void lustre_msg_add_flags(struct lustre_msg *msg, int flags); -void lustre_msg_set_flags(struct lustre_msg *msg, int flags); -void lustre_msg_clear_flags(struct lustre_msg *msg, int flags); +void lustre_msg_add_flags(struct lustre_msg *msg, __u32 flags); +void lustre_msg_set_flags(struct lustre_msg *msg, __u32 flags); +void lustre_msg_clear_flags(struct lustre_msg *msg, __u32 flags); __u32 lustre_msg_get_op_flags(struct lustre_msg *msg); -void lustre_msg_add_op_flags(struct lustre_msg *msg, int flags); -void lustre_msg_set_op_flags(struct lustre_msg *msg, int flags); +void lustre_msg_add_op_flags(struct lustre_msg *msg, __u32 flags); struct lustre_handle *lustre_msg_get_handle(struct lustre_msg *msg); __u32 lustre_msg_get_type(struct lustre_msg *msg); __u32 lustre_msg_get_version(struct lustre_msg *msg); -void lustre_msg_add_version(struct lustre_msg *msg, int version); +void lustre_msg_add_version(struct lustre_msg *msg, __u32 version); __u32 lustre_msg_get_opc(struct lustre_msg *msg); __u64 lustre_msg_get_last_xid(struct lustre_msg *msg); +__u16 lustre_msg_get_tag(struct lustre_msg *msg); __u64 lustre_msg_get_last_committed(struct lustre_msg *msg); __u64 *lustre_msg_get_versions(struct lustre_msg *msg); __u64 lustre_msg_get_transno(struct lustre_msg *msg); @@ -3231,21 +2354,18 @@ void lustre_msg_set_slv(struct lustre_msg *msg, __u64 slv); void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit); int lustre_msg_get_status(struct lustre_msg *msg); __u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg); -int lustre_msg_is_v1(struct lustre_msg *msg); __u32 lustre_msg_get_magic(struct lustre_msg *msg); __u32 lustre_msg_get_timeout(struct lustre_msg *msg); __u32 lustre_msg_get_service_time(struct lustre_msg *msg); char *lustre_msg_get_jobid(struct lustre_msg *msg); __u32 lustre_msg_get_cksum(struct lustre_msg *msg); -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) -__u32 lustre_msg_calc_cksum(struct lustre_msg *msg, int compat18); -#else +__u64 lustre_msg_get_mbits(struct lustre_msg *msg); __u32 lustre_msg_calc_cksum(struct lustre_msg *msg); -#endif void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle); void lustre_msg_set_type(struct lustre_msg *msg, __u32 type); void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc); void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid); +void lustre_msg_set_tag(struct lustre_msg *msg, __u16 tag); void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed); void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions); void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno); @@ -3257,6 +2377,7 @@ void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout); void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time); void lustre_msg_set_jobid(struct lustre_msg *msg, char *jobid); void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum); +void lustre_msg_set_mbits(struct lustre_msg *msg, __u64 mbits); static inline void lustre_shrink_reply(struct ptlrpc_request *req, int segment, @@ -3308,13 +2429,20 @@ ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase) if (req->rq_phase == new_phase) return; - if (new_phase == RQ_PHASE_UNREGISTERING) { + if (new_phase == RQ_PHASE_UNREG_RPC || + new_phase == RQ_PHASE_UNREG_BULK) { + /* No embedded unregistering phases */ + if (req->rq_phase == RQ_PHASE_UNREG_RPC || + req->rq_phase == RQ_PHASE_UNREG_BULK) + return; + req->rq_next_phase = req->rq_phase; if (req->rq_import) atomic_inc(&req->rq_import->imp_unregistering); } - if (req->rq_phase == RQ_PHASE_UNREGISTERING) { + if (req->rq_phase == RQ_PHASE_UNREG_RPC || + req->rq_phase == RQ_PHASE_UNREG_BULK) { if (req->rq_import) atomic_dec(&req->rq_import->imp_unregistering); } @@ -3326,14 +2454,11 @@ ptlrpc_rqphase_move(struct ptlrpc_request *req, enum rq_phase new_phase) } /** - * Returns true if request \a req got early reply and hard deadline is not met + * Returns true if request \a req got early reply and hard deadline is not met */ static inline int ptlrpc_client_early(struct ptlrpc_request *req) { - if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) && - req->rq_reply_deadline > cfs_time_current_sec()) - return 0; return req->rq_early; } @@ -3343,20 +2468,18 @@ ptlrpc_client_early(struct ptlrpc_request *req) static inline int ptlrpc_client_replied(struct ptlrpc_request *req) { - if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) && - req->rq_reply_deadline > cfs_time_current_sec()) - return 0; - return req->rq_replied; + if (req->rq_reply_deadline > cfs_time_current_sec()) + return 0; + return req->rq_replied; } /** Returns true if request \a req is in process of receiving server reply */ static inline int ptlrpc_client_recv(struct ptlrpc_request *req) { - if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) && - req->rq_reply_deadline > cfs_time_current_sec()) - return 1; - return req->rq_receiving_reply; + if (req->rq_reply_deadline > cfs_time_current_sec()) + return 1; + return req->rq_receiving_reply; } static inline int @@ -3365,13 +2488,17 @@ ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req) int rc; spin_lock(&req->rq_lock); - if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) && - req->rq_reply_deadline > cfs_time_current_sec()) { + if (req->rq_reply_deadline > cfs_time_current_sec()) { + spin_unlock(&req->rq_lock); + return 1; + } + if (req->rq_req_deadline > cfs_time_current_sec()) { spin_unlock(&req->rq_lock); return 1; } - rc = req->rq_receiving_reply ; - rc = rc || req->rq_req_unlink || req->rq_reply_unlink; + + rc = !req->rq_req_unlinked || !req->rq_reply_unlinked || + req->rq_receiving_reply; spin_unlock(&req->rq_lock); return rc; } @@ -3515,43 +2642,11 @@ void ptlrpc_pinger_ir_down(void); /** @} */ int ptlrpc_pinger_suppress_pings(void); -/* ptlrpc daemon bind policy */ -typedef enum { - /* all ptlrpcd threads are free mode */ - PDB_POLICY_NONE = 1, - /* all ptlrpcd threads are bound mode */ - PDB_POLICY_FULL = 2, - /* ... */ - PDB_POLICY_PAIR = 3, - /* ... , - * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1]. - * If kernel supports NUMA, pthrpcd threads are binded and - * grouped by NUMA node */ - PDB_POLICY_NEIGHBOR = 4, -} pdb_policy_t; - -/* ptlrpc daemon load policy - * It is caller's duty to specify how to push the async RPC into some ptlrpcd - * queue, but it is not enforced, affected by "ptlrpcd_bind_policy". If it is - * "PDB_POLICY_FULL", then the RPC will be processed by the selected ptlrpcd, - * Otherwise, the RPC may be processed by the selected ptlrpcd or its partner, - * depends on which is scheduled firstly, to accelerate the RPC processing. */ -typedef enum { - /* on the same CPU core as the caller */ - PDL_POLICY_SAME = 1, - /* within the same CPU partition, but not the same core as the caller */ - PDL_POLICY_LOCAL = 2, - /* round-robin on all CPU cores, but not the same core as the caller */ - PDL_POLICY_ROUND = 3, - /* the specified CPU core is preferred, but not enforced */ - PDL_POLICY_PREFERRED = 4, -} pdl_policy_t; - /* ptlrpc/ptlrpcd.c */ void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force); void ptlrpcd_free(struct ptlrpcd_ctl *pc); void ptlrpcd_wake(struct ptlrpc_request *req); -void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx); +void ptlrpcd_add_req(struct ptlrpc_request *req); void ptlrpcd_add_rqset(struct ptlrpc_request_set *set); int ptlrpcd_addref(void); void ptlrpcd_decref(void); @@ -3562,7 +2657,8 @@ void ptlrpcd_decref(void); * @{ */ const char* ll_opcode2str(__u32 opcode); -#ifdef LPROCFS +const int ll_str2opcode(const char *ops); +#ifdef CONFIG_PROC_FS void ptlrpc_lprocfs_register_obd(struct obd_device *obd); void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd); void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes); @@ -3583,7 +2679,6 @@ int llog_origin_handle_close(struct ptlrpc_request *req); /* ptlrpc/llog_client.c */ extern struct llog_operations llog_client_ops; - /** @} net */ #endif