diff options
Diffstat (limited to 'net')
| -rw-r--r-- | net/sunrpc/auth_gss/Makefile | 4 | ||||
| -rw-r--r-- | net/sunrpc/auth_gss/gss_krb5_crypto.c | 10 | ||||
| -rw-r--r-- | net/sunrpc/auth_gss/gss_krb5_seal.c | 26 | ||||
| -rw-r--r-- | net/sunrpc/auth_gss/gss_krb5_unseal.c | 16 | ||||
| -rw-r--r-- | net/sunrpc/auth_gss/gss_krb5_wrap.c | 72 | ||||
| -rw-r--r-- | net/sunrpc/svc.c | 110 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma.c | 35 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 84 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_sendto.c | 166 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_transport.c | 195 |
10 files changed, 331 insertions, 387 deletions
diff --git a/net/sunrpc/auth_gss/Makefile b/net/sunrpc/auth_gss/Makefile index f3431a7..4de8bcf 100644 --- a/net/sunrpc/auth_gss/Makefile +++ b/net/sunrpc/auth_gss/Makefile @@ -5,12 +5,12 @@ obj-$(CONFIG_SUNRPC_GSS) += auth_rpcgss.o auth_rpcgss-objs := auth_gss.o gss_generic_token.o \ - gss_mech_switch.o svcauth_gss.o gss_krb5_crypto.o + gss_mech_switch.o svcauth_gss.o obj-$(CONFIG_RPCSEC_GSS_KRB5) += rpcsec_gss_krb5.o rpcsec_gss_krb5-objs := gss_krb5_mech.o gss_krb5_seal.o gss_krb5_unseal.o \ - gss_krb5_seqnum.o gss_krb5_wrap.o + gss_krb5_seqnum.o gss_krb5_wrap.o gss_krb5_crypto.o obj-$(CONFIG_RPCSEC_GSS_SPKM3) += rpcsec_gss_spkm3.o diff --git a/net/sunrpc/auth_gss/gss_krb5_crypto.c b/net/sunrpc/auth_gss/gss_krb5_crypto.c index 1d52308..c93fca2 100644 --- a/net/sunrpc/auth_gss/gss_krb5_crypto.c +++ b/net/sunrpc/auth_gss/gss_krb5_crypto.c @@ -83,8 +83,6 @@ out: return ret; } -EXPORT_SYMBOL(krb5_encrypt); - u32 krb5_decrypt( struct crypto_blkcipher *tfm, @@ -118,8 +116,6 @@ out: return ret; } -EXPORT_SYMBOL(krb5_decrypt); - static int checksummer(struct scatterlist *sg, void *data) { @@ -161,8 +157,6 @@ out: return err ? GSS_S_FAILURE : 0; } -EXPORT_SYMBOL(make_checksum); - struct encryptor_desc { u8 iv[8]; /* XXX hard-coded blocksize */ struct blkcipher_desc desc; @@ -262,8 +256,6 @@ gss_encrypt_xdr_buf(struct crypto_blkcipher *tfm, struct xdr_buf *buf, return ret; } -EXPORT_SYMBOL(gss_encrypt_xdr_buf); - struct decryptor_desc { u8 iv[8]; /* XXX hard-coded blocksize */ struct blkcipher_desc desc; @@ -334,5 +326,3 @@ gss_decrypt_xdr_buf(struct crypto_blkcipher *tfm, struct xdr_buf *buf, return xdr_process_buf(buf, offset, buf->len - offset, decryptor, &desc); } - -EXPORT_SYMBOL(gss_decrypt_xdr_buf); diff --git a/net/sunrpc/auth_gss/gss_krb5_seal.c b/net/sunrpc/auth_gss/gss_krb5_seal.c index 5f1d36d..b8f42ef 100644 --- a/net/sunrpc/auth_gss/gss_krb5_seal.c +++ b/net/sunrpc/auth_gss/gss_krb5_seal.c @@ -78,7 +78,7 @@ gss_get_mic_kerberos(struct gss_ctx *gss_ctx, struct xdr_buf *text, struct krb5_ctx *ctx = gss_ctx->internal_ctx_id; char cksumdata[16]; struct xdr_netobj md5cksum = {.len = 0, .data = cksumdata}; - unsigned char *ptr, *krb5_hdr, *msg_start; + unsigned char *ptr, *msg_start; s32 now; u32 seq_send; @@ -87,36 +87,36 @@ gss_get_mic_kerberos(struct gss_ctx *gss_ctx, struct xdr_buf *text, now = get_seconds(); - token->len = g_token_size(&ctx->mech_used, 24); + token->len = g_token_size(&ctx->mech_used, GSS_KRB5_TOK_HDR_LEN + 8); ptr = token->data; - g_make_token_header(&ctx->mech_used, 24, &ptr); + g_make_token_header(&ctx->mech_used, GSS_KRB5_TOK_HDR_LEN + 8, &ptr); - *ptr++ = (unsigned char) ((KG_TOK_MIC_MSG>>8)&0xff); - *ptr++ = (unsigned char) (KG_TOK_MIC_MSG&0xff); + /* ptr now at header described in rfc 1964, section 1.2.1: */ + ptr[0] = (unsigned char) ((KG_TOK_MIC_MSG >> 8) & 0xff); + ptr[1] = (unsigned char) (KG_TOK_MIC_MSG & 0xff); - /* ptr now at byte 2 of header described in rfc 1964, section 1.2.1: */ - krb5_hdr = ptr - 2; - msg_start = krb5_hdr + 24; + msg_start = ptr + GSS_KRB5_TOK_HDR_LEN + 8; - *(__be16 *)(krb5_hdr + 2) = htons(SGN_ALG_DES_MAC_MD5); - memset(krb5_hdr + 4, 0xff, 4); + *(__be16 *)(ptr + 2) = htons(SGN_ALG_DES_MAC_MD5); + memset(ptr + 4, 0xff, 4); - if (make_checksum("md5", krb5_hdr, 8, text, 0, &md5cksum)) + if (make_checksum("md5", ptr, 8, text, 0, &md5cksum)) return GSS_S_FAILURE; if (krb5_encrypt(ctx->seq, NULL, md5cksum.data, md5cksum.data, md5cksum.len)) return GSS_S_FAILURE; - memcpy(krb5_hdr + 16, md5cksum.data + md5cksum.len - 8, 8); + memcpy(ptr + GSS_KRB5_TOK_HDR_LEN, md5cksum.data + md5cksum.len - 8, 8); spin_lock(&krb5_seq_lock); seq_send = ctx->seq_send++; spin_unlock(&krb5_seq_lock); if (krb5_make_seq_num(ctx->seq, ctx->initiate ? 0 : 0xff, - seq_send, krb5_hdr + 16, krb5_hdr + 8)) + seq_send, ptr + GSS_KRB5_TOK_HDR_LEN, + ptr + 8)) return GSS_S_FAILURE; return (ctx->endtime < now) ? GSS_S_CONTEXT_EXPIRED : GSS_S_COMPLETE; diff --git a/net/sunrpc/auth_gss/gss_krb5_unseal.c b/net/sunrpc/auth_gss/gss_krb5_unseal.c index d91a5d0..066ec73 100644 --- a/net/sunrpc/auth_gss/gss_krb5_unseal.c +++ b/net/sunrpc/auth_gss/gss_krb5_unseal.c @@ -92,30 +92,30 @@ gss_verify_mic_kerberos(struct gss_ctx *gss_ctx, read_token->len)) return GSS_S_DEFECTIVE_TOKEN; - if ((*ptr++ != ((KG_TOK_MIC_MSG>>8)&0xff)) || - (*ptr++ != ( KG_TOK_MIC_MSG &0xff)) ) + if ((ptr[0] != ((KG_TOK_MIC_MSG >> 8) & 0xff)) || + (ptr[1] != (KG_TOK_MIC_MSG & 0xff))) return GSS_S_DEFECTIVE_TOKEN; /* XXX sanity-check bodysize?? */ - signalg = ptr[0] + (ptr[1] << 8); + signalg = ptr[2] + (ptr[3] << 8); if (signalg != SGN_ALG_DES_MAC_MD5) return GSS_S_DEFECTIVE_TOKEN; - sealalg = ptr[2] + (ptr[3] << 8); + sealalg = ptr[4] + (ptr[5] << 8); if (sealalg != SEAL_ALG_NONE) return GSS_S_DEFECTIVE_TOKEN; - if ((ptr[4] != 0xff) || (ptr[5] != 0xff)) + if ((ptr[6] != 0xff) || (ptr[7] != 0xff)) return GSS_S_DEFECTIVE_TOKEN; - if (make_checksum("md5", ptr - 2, 8, message_buffer, 0, &md5cksum)) + if (make_checksum("md5", ptr, 8, message_buffer, 0, &md5cksum)) return GSS_S_FAILURE; if (krb5_encrypt(ctx->seq, NULL, md5cksum.data, md5cksum.data, 16)) return GSS_S_FAILURE; - if (memcmp(md5cksum.data + 8, ptr + 14, 8)) + if (memcmp(md5cksum.data + 8, ptr + GSS_KRB5_TOK_HDR_LEN, 8)) return GSS_S_BAD_SIG; /* it got through unscathed. Make sure the context is unexpired */ @@ -127,7 +127,7 @@ gss_verify_mic_kerberos(struct gss_ctx *gss_ctx, /* do sequencing checks */ - if (krb5_get_seq_num(ctx->seq, ptr + 14, ptr + 6, &direction, &seqnum)) + if (krb5_get_seq_num(ctx->seq, ptr + GSS_KRB5_TOK_HDR_LEN, ptr + 8, &direction, &seqnum)) return GSS_S_FAILURE; if ((ctx->initiate && direction != 0xff) || diff --git a/net/sunrpc/auth_gss/gss_krb5_wrap.c b/net/sunrpc/auth_gss/gss_krb5_wrap.c index b00b1b4..ae8e69b 100644 --- a/net/sunrpc/auth_gss/gss_krb5_wrap.c +++ b/net/sunrpc/auth_gss/gss_krb5_wrap.c @@ -87,8 +87,8 @@ out: return 0; } -static inline void -make_confounder(char *p, int blocksize) +static void +make_confounder(char *p, u32 conflen) { static u64 i = 0; u64 *q = (u64 *)p; @@ -102,8 +102,22 @@ make_confounder(char *p, int blocksize) * uniqueness would mean worrying about atomicity and rollover, and I * don't care enough. */ - BUG_ON(blocksize != 8); - *q = i++; + /* initialize to random value */ + if (i == 0) { + i = random32(); + i = (i << 32) | random32(); + } + + switch (conflen) { + case 16: + *q++ = i++; + /* fall through */ + case 8: + *q++ = i++; + break; + default: + BUG(); + } } /* Assumptions: the head and tail of inbuf are ours to play with. @@ -122,7 +136,7 @@ gss_wrap_kerberos(struct gss_ctx *ctx, int offset, char cksumdata[16]; struct xdr_netobj md5cksum = {.len = 0, .data = cksumdata}; int blocksize = 0, plainlen; - unsigned char *ptr, *krb5_hdr, *msg_start; + unsigned char *ptr, *msg_start; s32 now; int headlen; struct page **tmp_pages; @@ -149,26 +163,26 @@ gss_wrap_kerberos(struct gss_ctx *ctx, int offset, buf->len += headlen; BUG_ON((buf->len - offset - headlen) % blocksize); - g_make_token_header(&kctx->mech_used, 24 + plainlen, &ptr); + g_make_token_header(&kctx->mech_used, + GSS_KRB5_TOK_HDR_LEN + 8 + plainlen, &ptr); - *ptr++ = (unsigned char) ((KG_TOK_WRAP_MSG>>8)&0xff); - *ptr++ = (unsigned char) (KG_TOK_WRAP_MSG&0xff); + /* ptr now at header described in rfc 1964, section 1.2.1: */ + ptr[0] = (unsigned char) ((KG_TOK_WRAP_MSG >> 8) & 0xff); + ptr[1] = (unsigned char) (KG_TOK_WRAP_MSG & 0xff); - /* ptr now at byte 2 of header described in rfc 1964, section 1.2.1: */ - krb5_hdr = ptr - 2; - msg_start = krb5_hdr + 24; + msg_start = ptr + 24; - *(__be16 *)(krb5_hdr + 2) = htons(SGN_ALG_DES_MAC_MD5); - memset(krb5_hdr + 4, 0xff, 4); - *(__be16 *)(krb5_hdr + 4) = htons(SEAL_ALG_DES); + *(__be16 *)(ptr + 2) = htons(SGN_ALG_DES_MAC_MD5); + memset(ptr + 4, 0xff, 4); + *(__be16 *)(ptr + 4) = htons(SEAL_ALG_DES); make_confounder(msg_start, blocksize); /* XXXJBF: UGH!: */ tmp_pages = buf->pages; buf->pages = pages; - if (make_checksum("md5", krb5_hdr, 8, buf, + if (make_checksum("md5", ptr, 8, buf, offset + headlen - blocksize, &md5cksum)) return GSS_S_FAILURE; buf->pages = tmp_pages; @@ -176,7 +190,7 @@ gss_wrap_kerberos(struct gss_ctx *ctx, int offset, if (krb5_encrypt(kctx->seq, NULL, md5cksum.data, md5cksum.data, md5cksum.len)) return GSS_S_FAILURE; - memcpy(krb5_hdr + 16, md5cksum.data + md5cksum.len - 8, 8); + memcpy(ptr + GSS_KRB5_TOK_HDR_LEN, md5cksum.data + md5cksum.len - 8, 8); spin_lock(&krb5_seq_lock); seq_send = kctx->seq_send++; @@ -185,7 +199,7 @@ gss_wrap_kerberos(struct gss_ctx *ctx, int offset, /* XXX would probably be more efficient to compute checksum * and encrypt at the same time: */ if ((krb5_make_seq_num(kctx->seq, kctx->initiate ? 0 : 0xff, - seq_send, krb5_hdr + 16, krb5_hdr + 8))) + seq_send, ptr + GSS_KRB5_TOK_HDR_LEN, ptr + 8))) return GSS_S_FAILURE; if (gss_encrypt_xdr_buf(kctx->enc, buf, offset + headlen - blocksize, @@ -219,38 +233,38 @@ gss_unwrap_kerberos(struct gss_ctx *ctx, int offset, struct xdr_buf *buf) buf->len - offset)) return GSS_S_DEFECTIVE_TOKEN; - if ((*ptr++ != ((KG_TOK_WRAP_MSG>>8)&0xff)) || - (*ptr++ != (KG_TOK_WRAP_MSG &0xff)) ) + if ((ptr[0] != ((KG_TOK_WRAP_MSG >> 8) & 0xff)) || + (ptr[1] != (KG_TOK_WRAP_MSG & 0xff))) return GSS_S_DEFECTIVE_TOKEN; /* XXX sanity-check bodysize?? */ /* get the sign and seal algorithms */ - signalg = ptr[0] + (ptr[1] << 8); + signalg = ptr[2] + (ptr[3] << 8); if (signalg != SGN_ALG_DES_MAC_MD5) return GSS_S_DEFECTIVE_TOKEN; - sealalg = ptr[2] + (ptr[3] << 8); + sealalg = ptr[4] + (ptr[5] << 8); if (sealalg != SEAL_ALG_DES) return GSS_S_DEFECTIVE_TOKEN; - if ((ptr[4] != 0xff) || (ptr[5] != 0xff)) + if ((ptr[6] != 0xff) || (ptr[7] != 0xff)) return GSS_S_DEFECTIVE_TOKEN; if (gss_decrypt_xdr_buf(kctx->enc, buf, - ptr + 22 - (unsigned char *)buf->head[0].iov_base)) + ptr + GSS_KRB5_TOK_HDR_LEN + 8 - (unsigned char *)buf->head[0].iov_base)) return GSS_S_DEFECTIVE_TOKEN; - if (make_checksum("md5", ptr - 2, 8, buf, - ptr + 22 - (unsigned char *)buf->head[0].iov_base, &md5cksum)) + if (make_checksum("md5", ptr, 8, buf, + ptr + GSS_KRB5_TOK_HDR_LEN + 8 - (unsigned char *)buf->head[0].iov_base, &md5cksum)) return GSS_S_FAILURE; if (krb5_encrypt(kctx->seq, NULL, md5cksum.data, md5cksum.data, md5cksum.len)) return GSS_S_FAILURE; - if (memcmp(md5cksum.data + 8, ptr + 14, 8)) + if (memcmp(md5cksum.data + 8, ptr + GSS_KRB5_TOK_HDR_LEN, 8)) return GSS_S_BAD_SIG; /* it got through unscathed. Make sure the context is unexpired */ @@ -262,8 +276,8 @@ gss_unwrap_kerberos(struct gss_ctx *ctx, int offset, struct xdr_buf *buf) /* do sequencing checks */ - if (krb5_get_seq_num(kctx->seq, ptr + 14, ptr + 6, &direction, - &seqnum)) + if (krb5_get_seq_num(kctx->seq, ptr + GSS_KRB5_TOK_HDR_LEN, ptr + 8, + &direction, &seqnum)) return GSS_S_BAD_SIG; if ((kctx->initiate && direction != 0xff) || @@ -274,7 +288,7 @@ gss_unwrap_kerberos(struct gss_ctx *ctx, int offset, struct xdr_buf *buf) * better to copy and encrypt at the same time. */ blocksize = crypto_blkcipher_blocksize(kctx->enc); - data_start = ptr + 22 + blocksize; + data_start = ptr + GSS_KRB5_TOK_HDR_LEN + 8 + blocksize; orig_start = buf->head[0].iov_base + offset; data_len = (buf->head[0].iov_base + buf->head[0].iov_len) - data_start; memmove(orig_start, data_start, data_len); diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c index 01c7e31..5a32cb7 100644 --- a/net/sunrpc/svc.c +++ b/net/sunrpc/svc.c @@ -18,6 +18,7 @@ #include <linux/mm.h> #include <linux/interrupt.h> #include <linux/module.h> +#include <linux/kthread.h> #include <linux/sunrpc/types.h> #include <linux/sunrpc/xdr.h> @@ -291,15 +292,14 @@ svc_pool_map_put(void) /* - * Set the current thread's cpus_allowed mask so that it + * Set the given thread's cpus_allowed mask so that it * will only run on cpus in the given pool. - * - * Returns 1 and fills in oldmask iff a cpumask was applied. */ -static inline int -svc_pool_map_set_cpumask(unsigned int pidx, cpumask_t *oldmask) +static inline void +svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) { struct svc_pool_map *m = &svc_pool_map; + unsigned int node = m->pool_to[pidx]; /* * The caller checks for sv_nrpools > 1, which @@ -307,26 +307,17 @@ svc_pool_map_set_cpumask(unsigned int pidx, cpumask_t *oldmask) */ BUG_ON(m->count == 0); - switch (m->mode) - { - default: - return 0; + switch (m->mode) { case SVC_POOL_PERCPU: { - unsigned int cpu = m->pool_to[pidx]; - - *oldmask = current->cpus_allowed; - set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); - return 1; + set_cpus_allowed_ptr(task, &cpumask_of_cpu(node)); + break; } case SVC_POOL_PERNODE: { - unsigned int node = m->pool_to[pidx]; node_to_cpumask_ptr(nodecpumask, node); - - *oldmask = current->cpus_allowed; - set_cpus_allowed_ptr(current, nodecpumask); - return 1; + set_cpus_allowed_ptr(task, nodecpumask); + break; } } } @@ -443,7 +434,7 @@ EXPORT_SYMBOL(svc_create); struct svc_serv * svc_create_pooled(struct svc_program *prog, unsigned int bufsize, void (*shutdown)(struct svc_serv *serv), - svc_thread_fn func, int sig, struct module *mod) + svc_thread_fn func, struct module *mod) { struct svc_serv *serv; unsigned int npools = svc_pool_map_get(); @@ -452,7 +443,6 @@ svc_create_pooled(struct svc_program *prog, unsigned int bufsize, if (serv != NULL) { serv->sv_function = func; - serv->sv_kill_signal = sig; serv->sv_module = mod; } @@ -461,7 +451,8 @@ svc_create_pooled(struct svc_program *prog, unsigned int bufsize, EXPORT_SYMBOL(svc_create_pooled); /* - * Destroy an RPC service. Should be called with the BKL held + * Destroy an RPC service. Should be called with appropriate locking to + * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. */ void svc_destroy(struct svc_serv *serv) @@ -578,46 +569,6 @@ out_enomem: EXPORT_SYMBOL(svc_prepare_thread); /* - * Create a thread in the given pool. Caller must hold BKL. - * On a NUMA or SMP machine, with a multi-pool serv, the thread - * will be restricted to run on the cpus belonging to the pool. - */ -static int -__svc_create_thread(svc_thread_fn func, struct svc_serv *serv, - struct svc_pool *pool) -{ - struct svc_rqst *rqstp; - int error = -ENOMEM; - int have_oldmask = 0; - cpumask_t uninitialized_var(oldmask); - - rqstp = svc_prepare_thread(serv, pool); - if (IS_ERR(rqstp)) { - error = PTR_ERR(rqstp); - goto out; - } - - if (serv->sv_nrpools > 1) - have_oldmask = svc_pool_map_set_cpumask(pool->sp_id, &oldmask); - - error = kernel_thread((int (*)(void *)) func, rqstp, 0); - - if (have_oldmask) - set_cpus_allowed(current, oldmask); - - if (error < 0) - goto out_thread; - svc_sock_update_bufs(serv); - error = 0; -out: - return error; - -out_thread: - svc_exit_thread(rqstp); - goto out; -} - -/* * Choose a pool in which to create a new thread, for svc_set_num_threads */ static inline struct svc_pool * @@ -674,7 +625,7 @@ found_pool: * of threads the given number. If `pool' is non-NULL, applies * only to threads in that pool, otherwise round-robins between * all pools. Must be called with a svc_get() reference and - * the BKL held. + * the BKL or another lock to protect access to svc_serv fields. * * Destroying threads relies on the service threads filling in * rqstp->rq_task, which only the nfs ones do. Assumes the serv @@ -686,7 +637,9 @@ found_pool: int svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) { - struct task_struct *victim; + struct svc_rqst *rqstp; + struct task_struct *task; + struct svc_pool *chosen_pool; int error = 0; unsigned int state = serv->sv_nrthreads-1; @@ -702,18 +655,34 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) /* create new threads */ while (nrservs > 0) { nrservs--; + chosen_pool = choose_pool(serv, pool, &state); + + rqstp = svc_prepare_thread(serv, chosen_pool); + if (IS_ERR(rqstp)) { + error = PTR_ERR(rqstp); + break; + } + __module_get(serv->sv_module); - error = __svc_create_thread(serv->sv_function, serv, - choose_pool(serv, pool, &state)); - if (error < 0) { + task = kthread_create(serv->sv_function, rqstp, serv->sv_name); + if (IS_ERR(task)) { + error = PTR_ERR(task); module_put(serv->sv_module); + svc_exit_thread(rqstp); break; } + + rqstp->rq_task = task; + if (serv->sv_nrpools > 1) + svc_pool_map_set_cpumask(task, chosen_pool->sp_id); + + svc_sock_update_bufs(serv); + wake_up_process(task); } /* destroy old threads */ while (nrservs < 0 && - (victim = choose_victim(serv, pool, &state)) != NULL) { - send_sig(serv->sv_kill_signal, victim, 1); + (task = choose_victim(serv, pool, &state)) != NULL) { + send_sig(SIGINT, task, 1); nrservs++; } @@ -722,7 +691,8 @@ svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) EXPORT_SYMBOL(svc_set_num_threads); /* - * Called from a server thread as it's exiting. Caller must hold BKL. + * Called from a server thread as it's exiting. Caller must hold the BKL or + * the "service mutex", whichever is appropriate for the service. */ void svc_exit_thread(struct svc_rqst *rqstp) diff --git a/net/sunrpc/xprtrdma/svc_rdma.c b/net/sunrpc/xprtrdma/svc_rdma.c index 88c0ca2..8710117 100644 --- a/net/sunrpc/xprtrdma/svc_rdma.c +++ b/net/sunrpc/xprtrdma/svc_rdma.c @@ -69,6 +69,10 @@ atomic_t rdma_stat_rq_prod; atomic_t rdma_stat_sq_poll; atomic_t rdma_stat_sq_prod; +/* Temporary NFS request map and context caches */ +struct kmem_cache *svc_rdma_map_cachep; +struct kmem_cache *svc_rdma_ctxt_cachep; + /* * This function implements reading and resetting an atomic_t stat * variable through read/write to a proc file. Any write to the file @@ -236,11 +240,14 @@ static ctl_table svcrdma_root_table[] = { void svc_rdma_cleanup(void) { dprintk("SVCRDMA Module Removed, deregister RPC RDMA transport\n"); + flush_scheduled_work(); if (svcrdma_table_header) { unregister_sysctl_table(svcrdma_table_header); svcrdma_table_header = NULL; } svc_unreg_xprt_class(&svc_rdma_class); + kmem_cache_destroy(svc_rdma_map_cachep); + kmem_cache_destroy(svc_rdma_ctxt_cachep); } int svc_rdma_init(void) @@ -255,9 +262,37 @@ int svc_rdma_init(void) svcrdma_table_header = register_sysctl_table(svcrdma_root_table); + /* Create the temporary map cache */ + svc_rdma_map_cachep = kmem_cache_create("svc_rdma_map_cache", + sizeof(struct svc_rdma_req_map), + 0, + SLAB_HWCACHE_ALIGN, + NULL); + if (!svc_rdma_map_cachep) { + printk(KERN_INFO "Could not allocate map cache.\n"); + goto err0; + } + + /* Create the temporary context cache */ + svc_rdma_ctxt_cachep = + kmem_cache_create("svc_rdma_ctxt_cache", + sizeof(struct svc_rdma_op_ctxt), + 0, + SLAB_HWCACHE_ALIGN, + NULL); + if (!svc_rdma_ctxt_cachep) { + printk(KERN_INFO "Could not allocate WR ctxt cache.\n"); + goto err1; + } + /* Register RDMA with the SVC transport switch */ svc_reg_xprt_class(&svc_rdma_class); return 0; + err1: + kmem_cache_destroy(svc_rdma_map_cachep); + err0: + unregister_sysctl_table(svcrdma_table_header); + return -ENOMEM; } MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>"); MODULE_DESCRIPTION("SVC RDMA Transport"); diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c index 06ab484..b4b17f4 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c @@ -112,11 +112,6 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp, rqstp->rq_arg.tail[0].iov_len = 0; } -struct chunk_sge { - int start; /* sge no for this chunk */ - int count; /* sge count for this chunk */ -}; - /* Encode a read-chunk-list as an array of IB SGE * * Assumptions: @@ -134,8 +129,8 @@ static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, struct svc_rdma_op_ctxt *head, struct rpcrdma_msg *rmsgp, - struct ib_sge *sge, - struct chunk_sge *ch_sge_ary, + struct svc_rdma_req_map *rpl_map, + struct svc_rdma_req_map *chl_map, int ch_count, int byte_count) { @@ -156,22 +151,18 @@ static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt, head->arg.head[0] = rqstp->rq_arg.head[0]; head->arg.tail[0] = rqstp->rq_arg.tail[0]; head->arg.pages = &head->pages[head->count]; - head->sge[0].length = head->count; /* save count of hdr pages */ + head->hdr_count = head->count; /* save count of hdr pages */ head->arg.page_base = 0; head->arg.page_len = ch_bytes; head->arg.len = rqstp->rq_arg.len + ch_bytes; head->arg.buflen = rqstp->rq_arg.buflen + ch_bytes; head->count++; - ch_sge_ary[0].start = 0; + chl_map->ch[0].start = 0; while (byte_count) { + rpl_map->sge[sge_no].iov_base = + page_address(rqstp->rq_arg.pages[page_no]) + page_off; sge_bytes = min_t(int, PAGE_SIZE-page_off, ch_bytes); - sge[sge_no].addr = - ib_dma_map_page(xprt->sc_cm_id->device, - rqstp->rq_arg.pages[page_no], - page_off, sge_bytes, - DMA_FROM_DEVICE); - sge[sge_no].length = sge_bytes; - sge[sge_no].lkey = xprt->sc_phys_mr->lkey; + rpl_map->sge[sge_no].iov_len = sge_bytes; /* * Don't bump head->count here because the same page * may be used by multiple SGE. @@ -187,11 +178,11 @@ static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt, * SGE, move to the next SGE */ if (ch_bytes == 0) { - ch_sge_ary[ch_no].count = - sge_no - ch_sge_ary[ch_no].start; + chl_map->ch[ch_no].count = + sge_no - chl_map->ch[ch_no].start; ch_no++; ch++; - ch_sge_ary[ch_no].start = sge_no; + chl_map->ch[ch_no].start = sge_no; ch_bytes = ch->rc_target.rs_length; /* If bytes remaining account for next chunk */ if (byte_count) { @@ -220,18 +211,25 @@ static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt, return sge_no; } -static void rdma_set_ctxt_sge(struct svc_rdma_op_ctxt *ctxt, - struct ib_sge *sge, +static void rdma_set_ctxt_sge(struct svcxprt_rdma *xprt, + struct svc_rdma_op_ctxt *ctxt, + struct kvec *vec, u64 *sgl_offset, int count) { int i; ctxt->count = count; + ctxt->direction = DMA_FROM_DEVICE; for (i = 0; i < count; i++) { - ctxt->sge[i].addr = sge[i].addr; - ctxt->sge[i].length = sge[i].length; - *sgl_offset = *sgl_offset + sge[i].length; + atomic_inc(&xprt->sc_dma_used); + ctxt->sge[i].addr = + ib_dma_map_single(xprt->sc_cm_id->device, + vec[i].iov_base, vec[i].iov_len, + DMA_FROM_DEVICE); + ctxt->sge[i].length = vec[i].iov_len; + ctxt->sge[i].lkey = xprt->sc_phys_mr->lkey; + *sgl_offset = *sgl_offset + vec[i].iov_len; } } @@ -282,34 +280,29 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt, struct ib_send_wr read_wr; int err = 0; int ch_no; - struct ib_sge *sge; int ch_count; int byte_count; int sge_count; u64 sgl_offset; struct rpcrdma_read_chunk *ch; struct svc_rdma_op_ctxt *ctxt = NULL; - struct svc_rdma_op_ctxt *tmp_sge_ctxt; - struct svc_rdma_op_ctxt *tmp_ch_ctxt; - struct chunk_sge *ch_sge_ary; + struct svc_rdma_req_map *rpl_map; + struct svc_rdma_req_map *chl_map; /* If no read list is present, return 0 */ ch = svc_rdma_get_read_chunk(rmsgp); if (!ch) return 0; - /* Allocate temporary contexts to keep SGE */ - BUG_ON(sizeof(struct ib_sge) < sizeof(struct chunk_sge)); - tmp_sge_ctxt = svc_rdma_get_context(xprt); - sge = tmp_sge_ctxt->sge; - tmp_ch_ctxt = svc_rdma_get_context(xprt); - ch_sge_ary = (struct chunk_sge *)tmp_ch_ctxt->sge; + /* Allocate temporary reply and chunk maps */ + rpl_map = svc_rdma_get_req_map(); + chl_map = svc_rdma_get_req_map(); svc_rdma_rcl_chunk_counts(ch, &ch_count, &byte_count); if (ch_count > RPCSVC_MAXPAGES) return -EINVAL; sge_count = rdma_rcl_to_sge(xprt, rqstp, hdr_ctxt, rmsgp, - sge, ch_sge_ary, + rpl_map, chl_map, ch_count, byte_count); sgl_offset = 0; ch_no = 0; @@ -331,14 +324,15 @@ next_sge: read_wr.wr.rdma.remote_addr = get_unaligned(&(ch->rc_target.rs_offset)) + sgl_offset; - read_wr.sg_list = &sge[ch_sge_ary[ch_no].start]; + read_wr.sg_list = ctxt->sge; read_wr.num_sge = - rdma_read_max_sge(xprt, ch_sge_ary[ch_no].count); - rdma_set_ctxt_sge(ctxt, &sge[ch_sge_ary[ch_no].start], + rdma_read_max_sge(xprt, chl_map->ch[ch_no].count); + rdma_set_ctxt_sge(xprt, ctxt, + &rpl_map->sge[chl_map->ch[ch_no].start], &sgl_offset, read_wr.num_sge); if (((ch+1)->rc_discrim == 0) && - (read_wr.num_sge == ch_sge_ary[ch_no].count)) { + (read_wr.num_sge == chl_map->ch[ch_no].count)) { /* * Mark the last RDMA_READ with a bit to * indicate all RPC data has been fetched from @@ -358,9 +352,9 @@ next_sge: } atomic_inc(&rdma_stat_read); - if (read_wr.num_sge < ch_sge_ary[ch_no].count) { - ch_sge_ary[ch_no].count -= read_wr.num_sge; - ch_sge_ary[ch_no].start += read_wr.num_sge; + if (read_wr.num_sge < chl_map->ch[ch_no].count) { + chl_map->ch[ch_no].count -= read_wr.num_sge; + chl_map->ch[ch_no].start += read_wr.num_sge; goto next_sge; } sgl_offset = 0; @@ -368,8 +362,8 @@ next_sge: } out: - svc_rdma_put_context(tmp_sge_ctxt, 0); - svc_rdma_put_context(tmp_ch_ctxt, 0); + svc_rdma_put_req_map(rpl_map); + svc_rdma_put_req_map(chl_map); /* Detach arg pages. svc_recv will replenish them */ for (ch_no = 0; &rqstp->rq_pages[ch_no] < rqstp->rq_respages; ch_no++) @@ -399,7 +393,7 @@ static int rdma_read_complete(struct svc_rqst *rqstp, rqstp->rq_pages[page_no] = head->pages[page_no]; } /* Point rq_arg.pages past header */ - rqstp->rq_arg.pages = &rqstp->rq_pages[head->sge[0].length]; + rqstp->rq_arg.pages = &rqstp->rq_pages[head->hdr_count]; rqstp->rq_arg.page_len = head->arg.page_len; rqstp->rq_arg.page_base = head->arg.page_base; diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c index fb82b1b..a19b22b 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c @@ -63,52 +63,44 @@ * SGE[2..sge_count-2] data from xdr->pages[] * SGE[sge_count-1] data from xdr->tail. * + * The max SGE we need is the length of the XDR / pagesize + one for + * head + one for tail + one for RPCRDMA header. Since RPCSVC_MAXPAGES + * reserves a page for both the request and the reply header, and this + * array is only concerned with the reply we are assured that we have + * on extra page for the RPCRMDA header. */ -static struct ib_sge *xdr_to_sge(struct svcxprt_rdma *xprt, - struct xdr_buf *xdr, - struct ib_sge *sge, - int *sge_count) +static void xdr_to_sge(struct svcxprt_rdma *xprt, + struct xdr_buf *xdr, + struct svc_rdma_req_map *vec) { - /* Max we need is the length of the XDR / pagesize + one for - * head + one for tail + one for RPCRDMA header - */ int sge_max = (xdr->len+PAGE_SIZE-1) / PAGE_SIZE + 3; int sge_no; - u32 byte_count = xdr->len; u32 sge_bytes; u32 page_bytes; - int page_off; + u32 page_off; int page_no; + BUG_ON(xdr->len != + (xdr->head[0].iov_len + xdr->page_len + xdr->tail[0].iov_len)); + /* Skip the first sge, this is for the RPCRDMA header */ sge_no = 1; /* Head SGE */ - sge[sge_no].addr = ib_dma_map_single(xprt->sc_cm_id->device, - xdr->head[0].iov_base, - xdr->head[0].iov_len, - DMA_TO_DEVICE); - sge_bytes = min_t(u32, byte_count, xdr->head[0].iov_len); - byte_count -= sge_bytes; - sge[sge_no].length = sge_bytes; - sge[sge_no].lkey = xprt->sc_phys_mr->lkey; + vec->sge[sge_no].iov_base = xdr->head[0].iov_base; + vec->sge[sge_no].iov_len = xdr->head[0].iov_len; sge_no++; /* pages SGE */ page_no = 0; page_bytes = xdr->page_len; page_off = xdr->page_base; - while (byte_count && page_bytes) { - sge_bytes = min_t(u32, byte_count, (PAGE_SIZE-page_off)); - sge[sge_no].addr = - ib_dma_map_page(xprt->sc_cm_id->device, - xdr->pages[page_no], page_off, - sge_bytes, DMA_TO_DEVICE); - sge_bytes = min(sge_bytes, page_bytes); - byte_count -= sge_bytes; + while (page_bytes) { + vec->sge[sge_no].iov_base = + page_address(xdr->pages[page_no]) + page_off; + sge_bytes = min_t(u32, page_bytes, (PAGE_SIZE - page_off)); page_bytes -= sge_bytes; - sge[sge_no].length = sge_bytes; - sge[sge_no].lkey = xprt->sc_phys_mr->lkey; + vec->sge[sge_no].iov_len = sge_bytes; sge_no++; page_no++; @@ -116,36 +108,24 @@ static struct ib_sge *xdr_to_sge(struct svcxprt_rdma *xprt, } /* Tail SGE */ - if (byte_count && xdr->tail[0].iov_len) { - sge[sge_no].addr = - ib_dma_map_single(xprt->sc_cm_id->device, - xdr->tail[0].iov_base, - xdr->tail[0].iov_len, - DMA_TO_DEVICE); - sge_bytes = min_t(u32, byte_count, xdr->tail[0].iov_len); - byte_count -= sge_bytes; - sge[sge_no].length = sge_bytes; - sge[sge_no].lkey = xprt->sc_phys_mr->lkey; + if (xdr->tail[0].iov_len) { + vec->sge[sge_no].iov_base = xdr->tail[0].iov_base; + vec->sge[sge_no].iov_len = xdr->tail[0].iov_len; sge_no++; } BUG_ON(sge_no > sge_max); - BUG_ON(byte_count != 0); - - *sge_count = sge_no; - return sge; + vec->count = sge_no; } - /* Assumptions: * - The specified write_len can be represented in sc_max_sge * PAGE_SIZE */ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, u32 rmr, u64 to, u32 xdr_off, int write_len, - struct ib_sge *xdr_sge, int sge_count) + struct svc_rdma_req_map *vec) { - struct svc_rdma_op_ctxt *tmp_sge_ctxt; struct ib_send_wr write_wr; struct ib_sge *sge; int xdr_sge_no; @@ -154,25 +134,23 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, int sge_off; int bc; struct svc_rdma_op_ctxt *ctxt; - int ret = 0; - BUG_ON(sge_count > RPCSVC_MAXPAGES); + BUG_ON(vec->count > RPCSVC_MAXPAGES); dprintk("svcrdma: RDMA_WRITE rmr=%x, to=%llx, xdr_off=%d, " - "write_len=%d, xdr_sge=%p, sge_count=%d\n", + "write_len=%d, vec->sge=%p, vec->count=%lu\n", rmr, (unsigned long long)to, xdr_off, - write_len, xdr_sge, sge_count); + write_len, vec->sge, vec->count); ctxt = svc_rdma_get_context(xprt); - ctxt->count = 0; - tmp_sge_ctxt = svc_rdma_get_context(xprt); - sge = tmp_sge_ctxt->sge; + ctxt->direction = DMA_TO_DEVICE; + sge = ctxt->sge; /* Find the SGE associated with xdr_off */ - for (bc = xdr_off, xdr_sge_no = 1; bc && xdr_sge_no < sge_count; + for (bc = xdr_off, xdr_sge_no = 1; bc && xdr_sge_no < vec->count; xdr_sge_no++) { - if (xdr_sge[xdr_sge_no].length > bc) + if (vec->sge[xdr_sge_no].iov_len > bc) break; - bc -= xdr_sge[xdr_sge_no].length; + bc -= vec->sge[xdr_sge_no].iov_len; } sge_off = bc; @@ -180,21 +158,28 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, sge_no = 0; /* Copy the remaining SGE */ - while (bc != 0 && xdr_sge_no < sge_count) { - sge[sge_no].addr = xdr_sge[xdr_sge_no].addr + sge_off; - sge[sge_no].lkey = xdr_sge[xdr_sge_no].lkey; + while (bc != 0 && xdr_sge_no < vec->count) { + sge[sge_no].lkey = xprt->sc_phys_mr->lkey; sge_bytes = min((size_t)bc, - (size_t)(xdr_sge[xdr_sge_no].length-sge_off)); + (size_t)(vec->sge[xdr_sge_no].iov_len-sge_off)); sge[sge_no].length = sge_bytes; - + atomic_inc(&xprt->sc_dma_used); + sge[sge_no].addr = + ib_dma_map_single(xprt->sc_cm_id->device, + (void *) + vec->sge[xdr_sge_no].iov_base + sge_off, + sge_bytes, DMA_TO_DEVICE); + if (dma_mapping_error(sge[sge_no].addr)) + goto err; sge_off = 0; sge_no++; + ctxt->count++; xdr_sge_no++; bc -= sge_bytes; } BUG_ON(bc != 0); - BUG_ON(xdr_sge_no > sge_count); + BUG_ON(xdr_sge_no > vec->count); /* Prepare WRITE WR */ memset(&write_wr, 0, sizeof write_wr); @@ -209,21 +194,20 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, /* Post It */ atomic_inc(&rdma_stat_write); - if (svc_rdma_send(xprt, &write_wr)) { - svc_rdma_put_context(ctxt, 1); - /* Fatal error, close transport */ - ret = -EIO; - } - svc_rdma_put_context(tmp_sge_ctxt, 0); - return ret; + if (svc_rdma_send(xprt, &write_wr)) + goto err; + return 0; + err: + svc_rdma_put_context(ctxt, 0); + /* Fatal error, close transport */ + return -EIO; } static int send_write_chunks(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rdma_argp, struct rpcrdma_msg *rdma_resp, struct svc_rqst *rqstp, - struct ib_sge *sge, - int sge_count) + struct svc_rdma_req_map *vec) { u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len; int write_len; @@ -269,8 +253,7 @@ static int send_write_chunks(struct svcxprt_rdma *xprt, rs_offset + chunk_off, xdr_off, this_write, - sge, - sge_count); + vec); if (ret) { dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n", ret); @@ -292,8 +275,7 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rdma_argp, struct rpcrdma_msg *rdma_resp, struct svc_rqst *rqstp, - struct ib_sge *sge, - int sge_count) + struct svc_rdma_req_map *vec) { u32 xfer_len = rqstp->rq_res.len; int write_len; @@ -341,8 +323,7 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt, rs_offset + chunk_off, xdr_off, this_write, - sge, - sge_count); + vec); if (ret) { dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n", ret); @@ -380,7 +361,7 @@ static int send_reply(struct svcxprt_rdma *rdma, struct page *page, struct rpcrdma_msg *rdma_resp, struct svc_rdma_op_ctxt *ctxt, - int sge_count, + struct svc_rdma_req_map *vec, int byte_count) { struct ib_send_wr send_wr; @@ -405,6 +386,7 @@ static int send_reply(struct svcxprt_rdma *rdma, ctxt->count = 1; /* Prepare the SGE for the RPCRDMA Header */ + atomic_inc(&rdma->sc_dma_used); ctxt->sge[0].addr = ib_dma_map_page(rdma->sc_cm_id->device, page, 0, PAGE_SIZE, DMA_TO_DEVICE); @@ -413,10 +395,16 @@ static int send_reply(struct svcxprt_rdma *rdma, ctxt->sge[0].lkey = rdma->sc_phys_mr->lkey; /* Determine how many of our SGE are to be transmitted */ - for (sge_no = 1; byte_count && sge_no < sge_count; sge_no++) { - sge_bytes = min((size_t)ctxt->sge[sge_no].length, - (size_t)byte_count); + for (sge_no = 1; byte_count && sge_no < vec->count; sge_no++) { + sge_bytes = min_t(size_t, vec->sge[sge_no].iov_len, byte_count); byte_count -= sge_bytes; + atomic_inc(&rdma->sc_dma_used); + ctxt->sge[sge_no].addr = + ib_dma_map_single(rdma->sc_cm_id->device, + vec->sge[sge_no].iov_base, + sge_bytes, DMA_TO_DEVICE); + ctxt->sge[sge_no].length = sge_bytes; + ctxt->sge[sge_no].lkey = rdma->sc_phys_mr->lkey; } BUG_ON(byte_count != 0); @@ -428,8 +416,10 @@ static int send_reply(struct svcxprt_rdma *rdma, ctxt->pages[page_no+1] = rqstp->rq_respages[page_no]; ctxt->count++; rqstp->rq_respages[page_no] = NULL; + /* If there are more pages than SGE, terminate SGE list */ + if (page_no+1 >= sge_no) + ctxt->sge[page_no+1].length = 0; } - BUG_ON(sge_no > rdma->sc_max_sge); memset(&send_wr, 0, sizeof send_wr); ctxt->wr_op = IB_WR_SEND; @@ -473,20 +463,20 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) enum rpcrdma_proc reply_type; int ret; int inline_bytes; - struct ib_sge *sge; - int sge_count = 0; struct page *res_page; struct svc_rdma_op_ctxt *ctxt; + struct svc_rdma_req_map *vec; dprintk("svcrdma: sending response for rqstp=%p\n", rqstp); /* Get the RDMA request header. */ rdma_argp = xdr_start(&rqstp->rq_arg); - /* Build an SGE for the XDR */ + /* Build an req vec for the XDR */ ctxt = svc_rdma_get_context(rdma); ctxt->direction = DMA_TO_DEVICE; - sge = xdr_to_sge(rdma, &rqstp->rq_res, ctxt->sge, &sge_count); + vec = svc_rdma_get_req_map(); + xdr_to_sge(rdma, &rqstp->rq_res, vec); inline_bytes = rqstp->rq_res.len; @@ -503,7 +493,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) /* Send any write-chunk data and build resp write-list */ ret = send_write_chunks(rdma, rdma_argp, rdma_resp, - rqstp, sge, sge_count); + rqstp, vec); if (ret < 0) { printk(KERN_ERR "svcrdma: failed to send write chunks, rc=%d\n", ret); @@ -513,7 +503,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) /* Send any reply-list data and update resp reply-list */ ret = send_reply_chunks(rdma, rdma_argp, rdma_resp, - rqstp, sge, sge_count); + rqstp, vec); if (ret < 0) { printk(KERN_ERR "svcrdma: failed to send reply chunks, rc=%d\n", ret); @@ -521,11 +511,13 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) } inline_bytes -= ret; - ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, sge_count, + ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, vec, inline_bytes); + svc_rdma_put_req_map(vec); dprintk("svcrdma: send_reply returns %d\n", ret); return ret; error: + svc_rdma_put_req_map(vec); svc_rdma_put_context(ctxt, 0); put_page(res_page); return ret; diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c index e132509..19ddc38 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_transport.c +++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c @@ -84,70 +84,37 @@ struct svc_xprt_class svc_rdma_class = { .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, }; -static int rdma_bump_context_cache(struct svcxprt_rdma *xprt) +/* WR context cache. Created in svc_rdma.c */ +extern struct kmem_cache *svc_rdma_ctxt_cachep; + +struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt) { - int target; - int at_least_one = 0; struct svc_rdma_op_ctxt *ctxt; - target = min(xprt->sc_ctxt_cnt + xprt->sc_ctxt_bump, - xprt->sc_ctxt_max); - - spin_lock_bh(&xprt->sc_ctxt_lock); - while (xprt->sc_ctxt_cnt < target) { - xprt->sc_ctxt_cnt++; - spin_unlock_bh(&xprt->sc_ctxt_lock); - - ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL); - - spin_lock_bh(&xprt->sc_ctxt_lock); - if (ctxt) { - at_least_one = 1; - INIT_LIST_HEAD(&ctxt->free_list); - list_add(&ctxt->free_list, &xprt->sc_ctxt_free); - } else { - /* kmalloc failed...give up for now */ - xprt->sc_ctxt_cnt--; + while (1) { + ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep, GFP_KERNEL); + if (ctxt) break; - } + schedule_timeout_uninterruptible(msecs_to_jiffies(500)); } - spin_unlock_bh(&xprt->sc_ctxt_lock); - dprintk("svcrdma: sc_ctxt_max=%d, sc_ctxt_cnt=%d\n", - xprt->sc_ctxt_max, xprt->sc_ctxt_cnt); - return at_least_one; + ctxt->xprt = xprt; + INIT_LIST_HEAD(&ctxt->dto_q); + ctxt->count = 0; + atomic_inc(&xprt->sc_ctxt_used); + return ctxt; } -struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt) +static void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt) { - struct svc_rdma_op_ctxt *ctxt; - - while (1) { - spin_lock_bh(&xprt->sc_ctxt_lock); - if (unlikely(list_empty(&xprt->sc_ctxt_free))) { - /* Try to bump my cache. */ - spin_unlock_bh(&xprt->sc_ctxt_lock); - - if (rdma_bump_context_cache(xprt)) - continue; - - printk(KERN_INFO "svcrdma: sleeping waiting for " - "context memory on xprt=%p\n", - xprt); - schedule_timeout_uninterruptible(msecs_to_jiffies(500)); - continue; - } - ctxt = list_entry(xprt->sc_ctxt_free.next, - struct svc_rdma_op_ctxt, - free_list); - list_del_init(&ctxt->free_list); - spin_unlock_bh(&xprt->sc_ctxt_lock); - ctxt->xprt = xprt; - INIT_LIST_HEAD(&ctxt->dto_q); - ctxt->count = 0; - atomic_inc(&xprt->sc_ctxt_used); - break; + struct svcxprt_rdma *xprt = ctxt->xprt; + int i; + for (i = 0; i < ctxt->count && ctxt->sge[i].length; i++) { + atomic_dec(&xprt->sc_dma_used); + ib_dma_unmap_single(xprt->sc_cm_id->device, + ctxt->sge[i].addr, + ctxt->sge[i].length, + ctxt->direction); } - return ctxt; } void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages) @@ -161,18 +128,36 @@ void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages) for (i = 0; i < ctxt->count; i++) put_page(ctxt->pages[i]); - for (i = 0; i < ctxt->count; i++) - ib_dma_unmap_single(xprt->sc_cm_id->device, - ctxt->sge[i].addr, - ctxt->sge[i].length, - ctxt->direction); - - spin_lock_bh(&xprt->sc_ctxt_lock); - list_add(&ctxt->free_list, &xprt->sc_ctxt_free); - spin_unlock_bh(&xprt->sc_ctxt_lock); + kmem_cache_free(svc_rdma_ctxt_cachep, ctxt); atomic_dec(&xprt->sc_ctxt_used); } +/* Temporary NFS request map cache. Created in svc_rdma.c */ +extern struct kmem_cache *svc_rdma_map_cachep; + +/* + * Temporary NFS req mappings are shared across all transport + * instances. These are short lived and should be bounded by the number + * of concurrent server threads * depth of the SQ. + */ +struct svc_rdma_req_map *svc_rdma_get_req_map(void) +{ + struct svc_rdma_req_map *map; + while (1) { + map = kmem_cache_alloc(svc_rdma_map_cachep, GFP_KERNEL); + if (map) + break; + schedule_timeout_uninterruptible(msecs_to_jiffies(500)); + } + map->count = 0; + return map; +} + +void svc_rdma_put_req_map(struct svc_rdma_req_map *map) +{ + kmem_cache_free(svc_rdma_map_cachep, map); +} + /* ib_cq event handler */ static void cq_event_handler(struct ib_event *event, void *context) { @@ -302,6 +287,7 @@ static void rq_cq_reap(struct svcxprt_rdma *xprt) ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id; ctxt->wc_status = wc.status; ctxt->byte_len = wc.byte_len; + svc_rdma_unmap_dma(ctxt); if (wc.status != IB_WC_SUCCESS) { /* Close the transport */ dprintk("svcrdma: transport closing putting ctxt %p\n", ctxt); @@ -351,6 +337,7 @@ static void sq_cq_reap(struct svcxprt_rdma *xprt) ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id; xprt = ctxt->xprt; + svc_rdma_unmap_dma(ctxt); if (wc.status != IB_WC_SUCCESS) /* Close the transport */ set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); @@ -361,10 +348,13 @@ static void sq_cq_reap(struct svcxprt_rdma *xprt) switch (ctxt->wr_op) { case IB_WR_SEND: - case IB_WR_RDMA_WRITE: svc_rdma_put_context(ctxt, 1); break; + case IB_WR_RDMA_WRITE: + svc_rdma_put_context(ctxt, 0); + break; + case IB_WR_RDMA_READ: if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) { struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr; @@ -423,40 +413,6 @@ static void sq_comp_handler(struct ib_cq *cq, void *cq_context) tasklet_schedule(&dto_tasklet); } -static void create_context_cache(struct svcxprt_rdma *xprt, - int ctxt_count, int ctxt_bump, int ctxt_max) -{ - struct svc_rdma_op_ctxt *ctxt; - int i; - - xprt->sc_ctxt_max = ctxt_max; - xprt->sc_ctxt_bump = ctxt_bump; - xprt->sc_ctxt_cnt = 0; - atomic_set(&xprt->sc_ctxt_used, 0); - - INIT_LIST_HEAD(&xprt->sc_ctxt_free); - for (i = 0; i < ctxt_count; i++) { - ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL); - if (ctxt) { - INIT_LIST_HEAD(&ctxt->free_list); - list_add(&ctxt->free_list, &xprt->sc_ctxt_free); - xprt->sc_ctxt_cnt++; - } - } -} - -static void destroy_context_cache(struct svcxprt_rdma *xprt) -{ - while (!list_empty(&xprt->sc_ctxt_free)) { - struct svc_rdma_op_ctxt *ctxt; - ctxt = list_entry(xprt->sc_ctxt_free.next, - struct svc_rdma_op_ctxt, - free_list); - list_del_init(&ctxt->free_list); - kfree(ctxt); - } -} - static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv, int listener) { @@ -473,7 +429,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv, spin_lock_init(&cma_xprt->sc_lock); spin_lock_init(&cma_xprt->sc_read_complete_lock); - spin_lock_init(&cma_xprt->sc_ctxt_lock); spin_lock_init(&cma_xprt->sc_rq_dto_lock); cma_xprt->sc_ord = svcrdma_ord; @@ -482,21 +437,9 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv, cma_xprt->sc_max_requests = svcrdma_max_requests; cma_xprt->sc_sq_depth = svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT; atomic_set(&cma_xprt->sc_sq_count, 0); + atomic_set(&cma_xprt->sc_ctxt_used, 0); - if (!listener) { - int reqs = cma_xprt->sc_max_requests; - create_context_cache(cma_xprt, - reqs << 1, /* starting size */ - reqs, /* bump amount */ - reqs + - cma_xprt->sc_sq_depth + - RPCRDMA_MAX_THREADS + 1); /* max */ - if (list_empty(&cma_xprt->sc_ctxt_free)) { - kfree(cma_xprt); - return NULL; - } - clear_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags); - } else + if (listener) set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags); return cma_xprt; @@ -532,6 +475,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt) BUG_ON(sge_no >= xprt->sc_max_sge); page = svc_rdma_get_page(); ctxt->pages[sge_no] = page; + atomic_inc(&xprt->sc_dma_used); pa = ib_dma_map_page(xprt->sc_cm_id->device, page, 0, PAGE_SIZE, DMA_FROM_DEVICE); @@ -566,7 +510,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt) * will call the recvfrom method on the listen xprt which will accept the new * connection. */ -static void handle_connect_req(struct rdma_cm_id *new_cma_id) +static void handle_connect_req(struct rdma_cm_id *new_cma_id, size_t client_ird) { struct svcxprt_rdma *listen_xprt = new_cma_id->context; struct svcxprt_rdma *newxprt; @@ -583,6 +527,9 @@ static void handle_connect_req(struct rdma_cm_id *new_cma_id) dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n", newxprt, newxprt->sc_cm_id, listen_xprt); + /* Save client advertised inbound read limit for use later in accept. */ + newxprt->sc_ord = client_ird; + /* Set the local and remote addresses in the transport */ sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr; svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa)); @@ -619,7 +566,8 @@ static int rdma_listen_handler(struct rdma_cm_id *cma_id, case RDMA_CM_EVENT_CONNECT_REQUEST: dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, " "event=%d\n", cma_id, cma_id->context, event->event); - handle_connect_req(cma_id); + handle_connect_req(cma_id, + event->param.conn.responder_resources); break; case RDMA_CM_EVENT_ESTABLISHED: @@ -793,8 +741,12 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt) (size_t)svcrdma_max_requests); newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests; - newxprt->sc_ord = min((size_t)devattr.max_qp_rd_atom, - (size_t)svcrdma_ord); + /* + * Limit ORD based on client limit, local device limit, and + * configured svcrdma limit. + */ + newxprt->sc_ord = min_t(size_t, devattr.max_qp_rd_atom, newxprt->sc_ord); + newxprt->sc_ord = min_t(size_t, svcrdma_ord, newxprt->sc_ord); newxprt->sc_pd = ib_alloc_pd(newxprt->sc_cm_id->device); if (IS_ERR(newxprt->sc_pd)) { @@ -987,7 +939,6 @@ static void __svc_rdma_free(struct work_struct *work) * cm_id because the device ptr is needed to unmap the dma in * svc_rdma_put_context. */ - spin_lock_bh(&rdma->sc_read_complete_lock); while (!list_empty(&rdma->sc_read_complete_q)) { struct svc_rdma_op_ctxt *ctxt; ctxt = list_entry(rdma->sc_read_complete_q.next, @@ -996,10 +947,8 @@ static void __svc_rdma_free(struct work_struct *work) list_del_init(&ctxt->dto_q); svc_rdma_put_context(ctxt, 1); } - spin_unlock_bh(&rdma->sc_read_complete_lock); /* Destroy queued, but not processed recv completions */ - spin_lock_bh(&rdma->sc_rq_dto_lock); while (!list_empty(&rdma->sc_rq_dto_q)) { struct svc_rdma_op_ctxt *ctxt; ctxt = list_entry(rdma->sc_rq_dto_q.next, @@ -1008,10 +957,10 @@ static void __svc_rdma_free(struct work_struct *work) list_del_init(&ctxt->dto_q); svc_rdma_put_context(ctxt, 1); } - spin_unlock_bh(&rdma->sc_rq_dto_lock); /* Warn if we leaked a resource or under-referenced */ WARN_ON(atomic_read(&rdma->sc_ctxt_used) != 0); + WARN_ON(atomic_read(&rdma->sc_dma_used) != 0); /* Destroy the QP if present (not a listener) */ if (rdma->sc_qp && !IS_ERR(rdma->sc_qp)) @@ -1032,7 +981,6 @@ static void __svc_rdma_free(struct work_struct *work) /* Destroy the CM ID */ rdma_destroy_id(rdma->sc_cm_id); - destroy_context_cache(rdma); kfree(rdma); } @@ -1132,6 +1080,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp, length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va); /* Prepare SGE for local address */ + atomic_inc(&xprt->sc_dma_used); sge.addr = ib_dma_map_page(xprt->sc_cm_id->device, p, 0, PAGE_SIZE, DMA_FROM_DEVICE); sge.lkey = xprt->sc_phys_mr->lkey; |
