xprt_rdma.h 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676
  1. /* SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause */
  2. /*
  3. * Copyright (c) 2014-2017 Oracle. All rights reserved.
  4. * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
  5. *
  6. * This software is available to you under a choice of one of two
  7. * licenses. You may choose to be licensed under the terms of the GNU
  8. * General Public License (GPL) Version 2, available from the file
  9. * COPYING in the main directory of this source tree, or the BSD-type
  10. * license below:
  11. *
  12. * Redistribution and use in source and binary forms, with or without
  13. * modification, are permitted provided that the following conditions
  14. * are met:
  15. *
  16. * Redistributions of source code must retain the above copyright
  17. * notice, this list of conditions and the following disclaimer.
  18. *
  19. * Redistributions in binary form must reproduce the above
  20. * copyright notice, this list of conditions and the following
  21. * disclaimer in the documentation and/or other materials provided
  22. * with the distribution.
  23. *
  24. * Neither the name of the Network Appliance, Inc. nor the names of
  25. * its contributors may be used to endorse or promote products
  26. * derived from this software without specific prior written
  27. * permission.
  28. *
  29. * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  30. * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  31. * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  32. * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  33. * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  34. * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  35. * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  36. * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  37. * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  38. * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  39. * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  40. */
  41. #ifndef _LINUX_SUNRPC_XPRT_RDMA_H
  42. #define _LINUX_SUNRPC_XPRT_RDMA_H
  43. #include <linux/wait.h> /* wait_queue_head_t, etc */
  44. #include <linux/spinlock.h> /* spinlock_t, etc */
  45. #include <linux/atomic.h> /* atomic_t, etc */
  46. #include <linux/workqueue.h> /* struct work_struct */
  47. #include <rdma/rdma_cm.h> /* RDMA connection api */
  48. #include <rdma/ib_verbs.h> /* RDMA verbs api */
  49. #include <linux/sunrpc/clnt.h> /* rpc_xprt */
  50. #include <linux/sunrpc/rpc_rdma.h> /* RPC/RDMA protocol */
  51. #include <linux/sunrpc/xprtrdma.h> /* xprt parameters */
  52. #define RDMA_RESOLVE_TIMEOUT (5000) /* 5 seconds */
  53. #define RDMA_CONNECT_RETRY_MAX (2) /* retries if no listener backlog */
  54. #define RPCRDMA_BIND_TO (60U * HZ)
  55. #define RPCRDMA_INIT_REEST_TO (5U * HZ)
  56. #define RPCRDMA_MAX_REEST_TO (30U * HZ)
  57. #define RPCRDMA_IDLE_DISC_TO (5U * 60 * HZ)
  58. /*
  59. * Interface Adapter -- one per transport instance
  60. */
  61. struct rpcrdma_ia {
  62. const struct rpcrdma_memreg_ops *ri_ops;
  63. struct ib_device *ri_device;
  64. struct rdma_cm_id *ri_id;
  65. struct ib_pd *ri_pd;
  66. struct completion ri_done;
  67. struct completion ri_remove_done;
  68. int ri_async_rc;
  69. unsigned int ri_max_segs;
  70. unsigned int ri_max_frwr_depth;
  71. unsigned int ri_max_inline_write;
  72. unsigned int ri_max_inline_read;
  73. unsigned int ri_max_send_sges;
  74. bool ri_implicit_roundup;
  75. enum ib_mr_type ri_mrtype;
  76. unsigned long ri_flags;
  77. struct ib_qp_attr ri_qp_attr;
  78. struct ib_qp_init_attr ri_qp_init_attr;
  79. };
  80. enum {
  81. RPCRDMA_IAF_REMOVING = 0,
  82. };
  83. /*
  84. * RDMA Endpoint -- one per transport instance
  85. */
  86. struct rpcrdma_ep {
  87. unsigned int rep_send_count;
  88. unsigned int rep_send_batch;
  89. int rep_connected;
  90. struct ib_qp_init_attr rep_attr;
  91. wait_queue_head_t rep_connect_wait;
  92. struct rpcrdma_connect_private rep_cm_private;
  93. struct rdma_conn_param rep_remote_cma;
  94. struct delayed_work rep_connect_worker;
  95. };
  96. /* Pre-allocate extra Work Requests for handling backward receives
  97. * and sends. This is a fixed value because the Work Queues are
  98. * allocated when the forward channel is set up.
  99. */
  100. #if defined(CONFIG_SUNRPC_BACKCHANNEL)
  101. #define RPCRDMA_BACKWARD_WRS (8)
  102. #else
  103. #define RPCRDMA_BACKWARD_WRS (0)
  104. #endif
  105. /* Registered buffer -- registered kmalloc'd memory for RDMA SEND/RECV
  106. *
  107. * The below structure appears at the front of a large region of kmalloc'd
  108. * memory, which always starts on a good alignment boundary.
  109. */
  110. struct rpcrdma_regbuf {
  111. struct ib_sge rg_iov;
  112. struct ib_device *rg_device;
  113. enum dma_data_direction rg_direction;
  114. __be32 rg_base[0] __attribute__ ((aligned(256)));
  115. };
  116. static inline u64
  117. rdmab_addr(struct rpcrdma_regbuf *rb)
  118. {
  119. return rb->rg_iov.addr;
  120. }
  121. static inline u32
  122. rdmab_length(struct rpcrdma_regbuf *rb)
  123. {
  124. return rb->rg_iov.length;
  125. }
  126. static inline u32
  127. rdmab_lkey(struct rpcrdma_regbuf *rb)
  128. {
  129. return rb->rg_iov.lkey;
  130. }
  131. static inline struct ib_device *
  132. rdmab_device(struct rpcrdma_regbuf *rb)
  133. {
  134. return rb->rg_device;
  135. }
  136. #define RPCRDMA_DEF_GFP (GFP_NOIO | __GFP_NOWARN)
  137. /* To ensure a transport can always make forward progress,
  138. * the number of RDMA segments allowed in header chunk lists
  139. * is capped at 8. This prevents less-capable devices and
  140. * memory registrations from overrunning the Send buffer
  141. * while building chunk lists.
  142. *
  143. * Elements of the Read list take up more room than the
  144. * Write list or Reply chunk. 8 read segments means the Read
  145. * list (or Write list or Reply chunk) cannot consume more
  146. * than
  147. *
  148. * ((8 + 2) * read segment size) + 1 XDR words, or 244 bytes.
  149. *
  150. * And the fixed part of the header is another 24 bytes.
  151. *
  152. * The smallest inline threshold is 1024 bytes, ensuring that
  153. * at least 750 bytes are available for RPC messages.
  154. */
  155. enum {
  156. RPCRDMA_MAX_HDR_SEGS = 8,
  157. RPCRDMA_HDRBUF_SIZE = 256,
  158. };
  159. /*
  160. * struct rpcrdma_rep -- this structure encapsulates state required
  161. * to receive and complete an RPC Reply, asychronously. It needs
  162. * several pieces of state:
  163. *
  164. * o receive buffer and ib_sge (donated to provider)
  165. * o status of receive (success or not, length, inv rkey)
  166. * o bookkeeping state to get run by reply handler (XDR stream)
  167. *
  168. * These structures are allocated during transport initialization.
  169. * N of these are associated with a transport instance, managed by
  170. * struct rpcrdma_buffer. N is the max number of outstanding RPCs.
  171. */
  172. struct rpcrdma_rep {
  173. struct ib_cqe rr_cqe;
  174. __be32 rr_xid;
  175. __be32 rr_vers;
  176. __be32 rr_proc;
  177. int rr_wc_flags;
  178. u32 rr_inv_rkey;
  179. bool rr_temp;
  180. struct rpcrdma_regbuf *rr_rdmabuf;
  181. struct rpcrdma_xprt *rr_rxprt;
  182. struct work_struct rr_work;
  183. struct xdr_buf rr_hdrbuf;
  184. struct xdr_stream rr_stream;
  185. struct rpc_rqst *rr_rqst;
  186. struct list_head rr_list;
  187. struct ib_recv_wr rr_recv_wr;
  188. };
  189. /* struct rpcrdma_sendctx - DMA mapped SGEs to unmap after Send completes
  190. */
  191. struct rpcrdma_req;
  192. struct rpcrdma_xprt;
  193. struct rpcrdma_sendctx {
  194. struct ib_send_wr sc_wr;
  195. struct ib_cqe sc_cqe;
  196. struct rpcrdma_xprt *sc_xprt;
  197. struct rpcrdma_req *sc_req;
  198. unsigned int sc_unmap_count;
  199. struct ib_sge sc_sges[];
  200. };
  201. /* Limit the number of SGEs that can be unmapped during one
  202. * Send completion. This caps the amount of work a single
  203. * completion can do before returning to the provider.
  204. *
  205. * Setting this to zero disables Send completion batching.
  206. */
  207. enum {
  208. RPCRDMA_MAX_SEND_BATCH = 7,
  209. };
  210. /*
  211. * struct rpcrdma_mr - external memory region metadata
  212. *
  213. * An external memory region is any buffer or page that is registered
  214. * on the fly (ie, not pre-registered).
  215. *
  216. * Each rpcrdma_buffer has a list of free MWs anchored in rb_mrs. During
  217. * call_allocate, rpcrdma_buffer_get() assigns one to each segment in
  218. * an rpcrdma_req. Then rpcrdma_register_external() grabs these to keep
  219. * track of registration metadata while each RPC is pending.
  220. * rpcrdma_deregister_external() uses this metadata to unmap and
  221. * release these resources when an RPC is complete.
  222. */
  223. enum rpcrdma_frwr_state {
  224. FRWR_IS_INVALID, /* ready to be used */
  225. FRWR_IS_VALID, /* in use */
  226. FRWR_FLUSHED_FR, /* flushed FASTREG WR */
  227. FRWR_FLUSHED_LI, /* flushed LOCALINV WR */
  228. };
  229. struct rpcrdma_frwr {
  230. struct ib_mr *fr_mr;
  231. struct ib_cqe fr_cqe;
  232. enum rpcrdma_frwr_state fr_state;
  233. struct completion fr_linv_done;
  234. union {
  235. struct ib_reg_wr fr_regwr;
  236. struct ib_send_wr fr_invwr;
  237. };
  238. };
  239. struct rpcrdma_fmr {
  240. struct ib_fmr *fm_mr;
  241. u64 *fm_physaddrs;
  242. };
  243. struct rpcrdma_mr {
  244. struct list_head mr_list;
  245. struct scatterlist *mr_sg;
  246. int mr_nents;
  247. enum dma_data_direction mr_dir;
  248. union {
  249. struct rpcrdma_fmr fmr;
  250. struct rpcrdma_frwr frwr;
  251. };
  252. struct rpcrdma_xprt *mr_xprt;
  253. u32 mr_handle;
  254. u32 mr_length;
  255. u64 mr_offset;
  256. struct list_head mr_all;
  257. };
  258. /*
  259. * struct rpcrdma_req -- structure central to the request/reply sequence.
  260. *
  261. * N of these are associated with a transport instance, and stored in
  262. * struct rpcrdma_buffer. N is the max number of outstanding requests.
  263. *
  264. * It includes pre-registered buffer memory for send AND recv.
  265. * The recv buffer, however, is not owned by this structure, and
  266. * is "donated" to the hardware when a recv is posted. When a
  267. * reply is handled, the recv buffer used is given back to the
  268. * struct rpcrdma_req associated with the request.
  269. *
  270. * In addition to the basic memory, this structure includes an array
  271. * of iovs for send operations. The reason is that the iovs passed to
  272. * ib_post_{send,recv} must not be modified until the work request
  273. * completes.
  274. */
  275. /* Maximum number of page-sized "segments" per chunk list to be
  276. * registered or invalidated. Must handle a Reply chunk:
  277. */
  278. enum {
  279. RPCRDMA_MAX_IOV_SEGS = 3,
  280. RPCRDMA_MAX_DATA_SEGS = ((1 * 1024 * 1024) / PAGE_SIZE) + 1,
  281. RPCRDMA_MAX_SEGS = RPCRDMA_MAX_DATA_SEGS +
  282. RPCRDMA_MAX_IOV_SEGS,
  283. };
  284. struct rpcrdma_mr_seg { /* chunk descriptors */
  285. u32 mr_len; /* length of chunk or segment */
  286. struct page *mr_page; /* owning page, if any */
  287. char *mr_offset; /* kva if no page, else offset */
  288. };
  289. /* The Send SGE array is provisioned to send a maximum size
  290. * inline request:
  291. * - RPC-over-RDMA header
  292. * - xdr_buf head iovec
  293. * - RPCRDMA_MAX_INLINE bytes, in pages
  294. * - xdr_buf tail iovec
  295. *
  296. * The actual number of array elements consumed by each RPC
  297. * depends on the device's max_sge limit.
  298. */
  299. enum {
  300. RPCRDMA_MIN_SEND_SGES = 3,
  301. RPCRDMA_MAX_PAGE_SGES = RPCRDMA_MAX_INLINE >> PAGE_SHIFT,
  302. RPCRDMA_MAX_SEND_SGES = 1 + 1 + RPCRDMA_MAX_PAGE_SGES + 1,
  303. };
  304. struct rpcrdma_buffer;
  305. struct rpcrdma_req {
  306. struct list_head rl_list;
  307. struct rpc_rqst rl_slot;
  308. struct rpcrdma_buffer *rl_buffer;
  309. struct rpcrdma_rep *rl_reply;
  310. struct xdr_stream rl_stream;
  311. struct xdr_buf rl_hdrbuf;
  312. struct rpcrdma_sendctx *rl_sendctx;
  313. struct rpcrdma_regbuf *rl_rdmabuf; /* xprt header */
  314. struct rpcrdma_regbuf *rl_sendbuf; /* rq_snd_buf */
  315. struct rpcrdma_regbuf *rl_recvbuf; /* rq_rcv_buf */
  316. struct list_head rl_all;
  317. unsigned long rl_flags;
  318. struct list_head rl_registered; /* registered segments */
  319. struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];
  320. };
  321. /* rl_flags */
  322. enum {
  323. RPCRDMA_REQ_F_PENDING = 0,
  324. RPCRDMA_REQ_F_TX_RESOURCES,
  325. };
  326. static inline struct rpcrdma_req *
  327. rpcr_to_rdmar(const struct rpc_rqst *rqst)
  328. {
  329. return container_of(rqst, struct rpcrdma_req, rl_slot);
  330. }
  331. static inline void
  332. rpcrdma_mr_push(struct rpcrdma_mr *mr, struct list_head *list)
  333. {
  334. list_add_tail(&mr->mr_list, list);
  335. }
  336. static inline struct rpcrdma_mr *
  337. rpcrdma_mr_pop(struct list_head *list)
  338. {
  339. struct rpcrdma_mr *mr;
  340. mr = list_first_entry(list, struct rpcrdma_mr, mr_list);
  341. list_del_init(&mr->mr_list);
  342. return mr;
  343. }
  344. /*
  345. * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
  346. * inline requests/replies, and client/server credits.
  347. *
  348. * One of these is associated with a transport instance
  349. */
  350. struct rpcrdma_buffer {
  351. spinlock_t rb_mrlock; /* protect rb_mrs list */
  352. struct list_head rb_mrs;
  353. struct list_head rb_all;
  354. unsigned long rb_sc_head;
  355. unsigned long rb_sc_tail;
  356. unsigned long rb_sc_last;
  357. struct rpcrdma_sendctx **rb_sc_ctxs;
  358. spinlock_t rb_lock; /* protect buf lists */
  359. struct list_head rb_send_bufs;
  360. struct list_head rb_recv_bufs;
  361. unsigned long rb_flags;
  362. u32 rb_max_requests;
  363. u32 rb_credits; /* most recent credit grant */
  364. int rb_posted_receives;
  365. u32 rb_bc_srv_max_requests;
  366. spinlock_t rb_reqslock; /* protect rb_allreqs */
  367. struct list_head rb_allreqs;
  368. u32 rb_bc_max_requests;
  369. spinlock_t rb_recovery_lock; /* protect rb_stale_mrs */
  370. struct list_head rb_stale_mrs;
  371. struct delayed_work rb_recovery_worker;
  372. struct delayed_work rb_refresh_worker;
  373. };
  374. #define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
  375. /* rb_flags */
  376. enum {
  377. RPCRDMA_BUF_F_EMPTY_SCQ = 0,
  378. };
  379. /*
  380. * Internal structure for transport instance creation. This
  381. * exists primarily for modularity.
  382. *
  383. * This data should be set with mount options
  384. */
  385. struct rpcrdma_create_data_internal {
  386. unsigned int max_requests; /* max requests (slots) in flight */
  387. unsigned int rsize; /* mount rsize - max read hdr+data */
  388. unsigned int wsize; /* mount wsize - max write hdr+data */
  389. unsigned int inline_rsize; /* max non-rdma read data payload */
  390. unsigned int inline_wsize; /* max non-rdma write data payload */
  391. };
  392. /*
  393. * Statistics for RPCRDMA
  394. */
  395. struct rpcrdma_stats {
  396. /* accessed when sending a call */
  397. unsigned long read_chunk_count;
  398. unsigned long write_chunk_count;
  399. unsigned long reply_chunk_count;
  400. unsigned long long total_rdma_request;
  401. /* rarely accessed error counters */
  402. unsigned long long pullup_copy_count;
  403. unsigned long hardway_register_count;
  404. unsigned long failed_marshal_count;
  405. unsigned long bad_reply_count;
  406. unsigned long mrs_recovered;
  407. unsigned long mrs_orphaned;
  408. unsigned long mrs_allocated;
  409. unsigned long empty_sendctx_q;
  410. /* accessed when receiving a reply */
  411. unsigned long long total_rdma_reply;
  412. unsigned long long fixup_copy_count;
  413. unsigned long reply_waits_for_send;
  414. unsigned long local_inv_needed;
  415. unsigned long nomsg_call_count;
  416. unsigned long bcall_count;
  417. };
  418. /*
  419. * Per-registration mode operations
  420. */
  421. struct rpcrdma_xprt;
  422. struct rpcrdma_memreg_ops {
  423. struct rpcrdma_mr_seg *
  424. (*ro_map)(struct rpcrdma_xprt *,
  425. struct rpcrdma_mr_seg *, int, bool,
  426. struct rpcrdma_mr **);
  427. int (*ro_send)(struct rpcrdma_ia *ia,
  428. struct rpcrdma_req *req);
  429. void (*ro_reminv)(struct rpcrdma_rep *rep,
  430. struct list_head *mrs);
  431. void (*ro_unmap_sync)(struct rpcrdma_xprt *,
  432. struct list_head *);
  433. void (*ro_recover_mr)(struct rpcrdma_mr *mr);
  434. int (*ro_open)(struct rpcrdma_ia *,
  435. struct rpcrdma_ep *,
  436. struct rpcrdma_create_data_internal *);
  437. size_t (*ro_maxpages)(struct rpcrdma_xprt *);
  438. int (*ro_init_mr)(struct rpcrdma_ia *,
  439. struct rpcrdma_mr *);
  440. void (*ro_release_mr)(struct rpcrdma_mr *mr);
  441. const char *ro_displayname;
  442. const int ro_send_w_inv_ok;
  443. };
  444. extern const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops;
  445. extern const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops;
  446. /*
  447. * RPCRDMA transport -- encapsulates the structures above for
  448. * integration with RPC.
  449. *
  450. * The contained structures are embedded, not pointers,
  451. * for convenience. This structure need not be visible externally.
  452. *
  453. * It is allocated and initialized during mount, and released
  454. * during unmount.
  455. */
  456. struct rpcrdma_xprt {
  457. struct rpc_xprt rx_xprt;
  458. struct rpcrdma_ia rx_ia;
  459. struct rpcrdma_ep rx_ep;
  460. struct rpcrdma_buffer rx_buf;
  461. struct rpcrdma_create_data_internal rx_data;
  462. struct delayed_work rx_connect_worker;
  463. struct rpcrdma_stats rx_stats;
  464. };
  465. #define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, rx_xprt)
  466. #define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
  467. static inline const char *
  468. rpcrdma_addrstr(const struct rpcrdma_xprt *r_xprt)
  469. {
  470. return r_xprt->rx_xprt.address_strings[RPC_DISPLAY_ADDR];
  471. }
  472. static inline const char *
  473. rpcrdma_portstr(const struct rpcrdma_xprt *r_xprt)
  474. {
  475. return r_xprt->rx_xprt.address_strings[RPC_DISPLAY_PORT];
  476. }
  477. /* Setting this to 0 ensures interoperability with early servers.
  478. * Setting this to 1 enhances certain unaligned read/write performance.
  479. * Default is 0, see sysctl entry and rpc_rdma.c rpcrdma_convert_iovs() */
  480. extern int xprt_rdma_pad_optimize;
  481. /* This setting controls the hunt for a supported memory
  482. * registration strategy.
  483. */
  484. extern unsigned int xprt_rdma_memreg_strategy;
  485. /*
  486. * Interface Adapter calls - xprtrdma/verbs.c
  487. */
  488. int rpcrdma_ia_open(struct rpcrdma_xprt *xprt);
  489. void rpcrdma_ia_remove(struct rpcrdma_ia *ia);
  490. void rpcrdma_ia_close(struct rpcrdma_ia *);
  491. bool frwr_is_supported(struct rpcrdma_ia *);
  492. bool fmr_is_supported(struct rpcrdma_ia *);
  493. extern struct workqueue_struct *rpcrdma_receive_wq;
  494. /*
  495. * Endpoint calls - xprtrdma/verbs.c
  496. */
  497. int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
  498. struct rpcrdma_create_data_internal *);
  499. void rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
  500. int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
  501. void rpcrdma_conn_func(struct rpcrdma_ep *ep);
  502. void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
  503. int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
  504. struct rpcrdma_req *);
  505. void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp);
  506. /*
  507. * Buffer calls - xprtrdma/verbs.c
  508. */
  509. struct rpcrdma_req *rpcrdma_create_req(struct rpcrdma_xprt *);
  510. void rpcrdma_destroy_req(struct rpcrdma_req *);
  511. int rpcrdma_buffer_create(struct rpcrdma_xprt *);
  512. void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
  513. struct rpcrdma_sendctx *rpcrdma_sendctx_get_locked(struct rpcrdma_buffer *buf);
  514. struct rpcrdma_mr *rpcrdma_mr_get(struct rpcrdma_xprt *r_xprt);
  515. void rpcrdma_mr_put(struct rpcrdma_mr *mr);
  516. void rpcrdma_mr_unmap_and_put(struct rpcrdma_mr *mr);
  517. void rpcrdma_mr_defer_recovery(struct rpcrdma_mr *mr);
  518. struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
  519. void rpcrdma_buffer_put(struct rpcrdma_req *);
  520. void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
  521. struct rpcrdma_regbuf *rpcrdma_alloc_regbuf(size_t, enum dma_data_direction,
  522. gfp_t);
  523. bool __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *, struct rpcrdma_regbuf *);
  524. void rpcrdma_free_regbuf(struct rpcrdma_regbuf *);
  525. static inline bool
  526. rpcrdma_regbuf_is_mapped(struct rpcrdma_regbuf *rb)
  527. {
  528. return rb->rg_device != NULL;
  529. }
  530. static inline bool
  531. rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
  532. {
  533. if (likely(rpcrdma_regbuf_is_mapped(rb)))
  534. return true;
  535. return __rpcrdma_dma_map_regbuf(ia, rb);
  536. }
  537. int rpcrdma_alloc_wq(void);
  538. void rpcrdma_destroy_wq(void);
  539. /*
  540. * Wrappers for chunk registration, shared by read/write chunk code.
  541. */
  542. static inline enum dma_data_direction
  543. rpcrdma_data_dir(bool writing)
  544. {
  545. return writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
  546. }
  547. /*
  548. * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
  549. */
  550. enum rpcrdma_chunktype {
  551. rpcrdma_noch = 0,
  552. rpcrdma_readch,
  553. rpcrdma_areadch,
  554. rpcrdma_writech,
  555. rpcrdma_replych
  556. };
  557. int rpcrdma_prepare_send_sges(struct rpcrdma_xprt *r_xprt,
  558. struct rpcrdma_req *req, u32 hdrlen,
  559. struct xdr_buf *xdr,
  560. enum rpcrdma_chunktype rtype);
  561. void rpcrdma_unmap_sendctx(struct rpcrdma_sendctx *sc);
  562. int rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst);
  563. void rpcrdma_set_max_header_sizes(struct rpcrdma_xprt *);
  564. void rpcrdma_complete_rqst(struct rpcrdma_rep *rep);
  565. void rpcrdma_reply_handler(struct rpcrdma_rep *rep);
  566. void rpcrdma_release_rqst(struct rpcrdma_xprt *r_xprt,
  567. struct rpcrdma_req *req);
  568. void rpcrdma_deferred_completion(struct work_struct *work);
  569. static inline void rpcrdma_set_xdrlen(struct xdr_buf *xdr, size_t len)
  570. {
  571. xdr->head[0].iov_len = len;
  572. xdr->len = len;
  573. }
  574. /* RPC/RDMA module init - xprtrdma/transport.c
  575. */
  576. extern unsigned int xprt_rdma_max_inline_read;
  577. void xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap);
  578. void xprt_rdma_free_addresses(struct rpc_xprt *xprt);
  579. void rpcrdma_connect_worker(struct work_struct *work);
  580. void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq);
  581. int xprt_rdma_init(void);
  582. void xprt_rdma_cleanup(void);
  583. /* Backchannel calls - xprtrdma/backchannel.c
  584. */
  585. #if defined(CONFIG_SUNRPC_BACKCHANNEL)
  586. int xprt_rdma_bc_setup(struct rpc_xprt *, unsigned int);
  587. int xprt_rdma_bc_up(struct svc_serv *, struct net *);
  588. size_t xprt_rdma_bc_maxpayload(struct rpc_xprt *);
  589. int rpcrdma_bc_post_recv(struct rpcrdma_xprt *, unsigned int);
  590. void rpcrdma_bc_receive_call(struct rpcrdma_xprt *, struct rpcrdma_rep *);
  591. int xprt_rdma_bc_send_reply(struct rpc_rqst *rqst);
  592. void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
  593. void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
  594. #endif /* CONFIG_SUNRPC_BACKCHANNEL */
  595. extern struct xprt_class xprt_rdma_bc;
  596. #endif /* _LINUX_SUNRPC_XPRT_RDMA_H */