1 /* 2 drbd_req.h 3 4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg. 5 6 Copyright (C) 2006-2008, LINBIT Information Technologies GmbH. 7 Copyright (C) 2006-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. 8 Copyright (C) 2006-2008, Philipp Reisner <philipp.reisner@linbit.com>. 9 10 DRBD is free software; you can redistribute it and/or modify 11 it under the terms of the GNU General Public License as published by 12 the Free Software Foundation; either version 2, or (at your option) 13 any later version. 14 15 DRBD is distributed in the hope that it will be useful, 16 but WITHOUT ANY WARRANTY; without even the implied warranty of 17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 18 GNU General Public License for more details. 19 20 You should have received a copy of the GNU General Public License 21 along with drbd; see the file COPYING. If not, write to 22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 23 */ 24 25 #ifndef _DRBD_REQ_H 26 #define _DRBD_REQ_H 27 28 #include <linux/module.h> 29 30 #include <linux/slab.h> 31 #include <linux/drbd.h> 32 #include "drbd_int.h" 33 #include "drbd_wrappers.h" 34 35 /* The request callbacks will be called in irq context by the IDE drivers, 36 and in Softirqs/Tasklets/BH context by the SCSI drivers, 37 and by the receiver and worker in kernel-thread context. 38 Try to get the locking right :) */ 39 40 /* 41 * Objects of type struct drbd_request do only exist on a R_PRIMARY node, and are 42 * associated with IO requests originating from the block layer above us. 43 * 44 * There are quite a few things that may happen to a drbd request 45 * during its lifetime. 46 * 47 * It will be created. 48 * It will be marked with the intention to be 49 * submitted to local disk and/or 50 * send via the network. 51 * 52 * It has to be placed on the transfer log and other housekeeping lists, 53 * In case we have a network connection. 54 * 55 * It may be identified as a concurrent (write) request 56 * and be handled accordingly. 57 * 58 * It may me handed over to the local disk subsystem. 59 * It may be completed by the local disk subsystem, 60 * either successfully or with io-error. 61 * In case it is a READ request, and it failed locally, 62 * it may be retried remotely. 63 * 64 * It may be queued for sending. 65 * It may be handed over to the network stack, 66 * which may fail. 67 * It may be acknowledged by the "peer" according to the wire_protocol in use. 68 * this may be a negative ack. 69 * It may receive a faked ack when the network connection is lost and the 70 * transfer log is cleaned up. 71 * Sending may be canceled due to network connection loss. 72 * When it finally has outlived its time, 73 * corresponding dirty bits in the resync-bitmap may be cleared or set, 74 * it will be destroyed, 75 * and completion will be signalled to the originator, 76 * with or without "success". 77 */ 78 79 enum drbd_req_event { 80 created, 81 to_be_send, 82 to_be_submitted, 83 84 /* XXX yes, now I am inconsistent... 85 * these two are not "events" but "actions" 86 * oh, well... */ 87 queue_for_net_write, 88 queue_for_net_read, 89 90 send_canceled, 91 send_failed, 92 handed_over_to_network, 93 connection_lost_while_pending, 94 read_retry_remote_canceled, 95 recv_acked_by_peer, 96 write_acked_by_peer, 97 write_acked_by_peer_and_sis, /* and set_in_sync */ 98 conflict_discarded_by_peer, 99 neg_acked, 100 barrier_acked, /* in protocol A and B */ 101 data_received, /* (remote read) */ 102 103 read_completed_with_error, 104 read_ahead_completed_with_error, 105 write_completed_with_error, 106 completed_ok, 107 resend, 108 fail_frozen_disk_io, 109 restart_frozen_disk_io, 110 nothing, /* for tracing only */ 111 }; 112 113 /* encoding of request states for now. we don't actually need that many bits. 114 * we don't need to do atomic bit operations either, since most of the time we 115 * need to look at the connection state and/or manipulate some lists at the 116 * same time, so we should hold the request lock anyways. 117 */ 118 enum drbd_req_state_bits { 119 /* 210 120 * 000: no local possible 121 * 001: to be submitted 122 * UNUSED, we could map: 011: submitted, completion still pending 123 * 110: completed ok 124 * 010: completed with error 125 */ 126 __RQ_LOCAL_PENDING, 127 __RQ_LOCAL_COMPLETED, 128 __RQ_LOCAL_OK, 129 130 /* 76543 131 * 00000: no network possible 132 * 00001: to be send 133 * 00011: to be send, on worker queue 134 * 00101: sent, expecting recv_ack (B) or write_ack (C) 135 * 11101: sent, 136 * recv_ack (B) or implicit "ack" (A), 137 * still waiting for the barrier ack. 138 * master_bio may already be completed and invalidated. 139 * 11100: write_acked (C), 140 * data_received (for remote read, any protocol) 141 * or finally the barrier ack has arrived (B,A)... 142 * request can be freed 143 * 01100: neg-acked (write, protocol C) 144 * or neg-d-acked (read, any protocol) 145 * or killed from the transfer log 146 * during cleanup after connection loss 147 * request can be freed 148 * 01000: canceled or send failed... 149 * request can be freed 150 */ 151 152 /* if "SENT" is not set, yet, this can still fail or be canceled. 153 * if "SENT" is set already, we still wait for an Ack packet. 154 * when cleared, the master_bio may be completed. 155 * in (B,A) the request object may still linger on the transaction log 156 * until the corresponding barrier ack comes in */ 157 __RQ_NET_PENDING, 158 159 /* If it is QUEUED, and it is a WRITE, it is also registered in the 160 * transfer log. Currently we need this flag to avoid conflicts between 161 * worker canceling the request and tl_clear_barrier killing it from 162 * transfer log. We should restructure the code so this conflict does 163 * no longer occur. */ 164 __RQ_NET_QUEUED, 165 166 /* well, actually only "handed over to the network stack". 167 * 168 * TODO can potentially be dropped because of the similar meaning 169 * of RQ_NET_SENT and ~RQ_NET_QUEUED. 170 * however it is not exactly the same. before we drop it 171 * we must ensure that we can tell a request with network part 172 * from a request without, regardless of what happens to it. */ 173 __RQ_NET_SENT, 174 175 /* when set, the request may be freed (if RQ_NET_QUEUED is clear). 176 * basically this means the corresponding P_BARRIER_ACK was received */ 177 __RQ_NET_DONE, 178 179 /* whether or not we know (C) or pretend (B,A) that the write 180 * was successfully written on the peer. 181 */ 182 __RQ_NET_OK, 183 184 /* peer called drbd_set_in_sync() for this write */ 185 __RQ_NET_SIS, 186 187 /* keep this last, its for the RQ_NET_MASK */ 188 __RQ_NET_MAX, 189 190 /* Set when this is a write, clear for a read */ 191 __RQ_WRITE, 192 193 /* Should call drbd_al_complete_io() for this request... */ 194 __RQ_IN_ACT_LOG, 195 }; 196 197 #define RQ_LOCAL_PENDING (1UL << __RQ_LOCAL_PENDING) 198 #define RQ_LOCAL_COMPLETED (1UL << __RQ_LOCAL_COMPLETED) 199 #define RQ_LOCAL_OK (1UL << __RQ_LOCAL_OK) 200 201 #define RQ_LOCAL_MASK ((RQ_LOCAL_OK << 1)-1) /* 0x07 */ 202 203 #define RQ_NET_PENDING (1UL << __RQ_NET_PENDING) 204 #define RQ_NET_QUEUED (1UL << __RQ_NET_QUEUED) 205 #define RQ_NET_SENT (1UL << __RQ_NET_SENT) 206 #define RQ_NET_DONE (1UL << __RQ_NET_DONE) 207 #define RQ_NET_OK (1UL << __RQ_NET_OK) 208 #define RQ_NET_SIS (1UL << __RQ_NET_SIS) 209 210 /* 0x1f8 */ 211 #define RQ_NET_MASK (((1UL << __RQ_NET_MAX)-1) & ~RQ_LOCAL_MASK) 212 213 #define RQ_WRITE (1UL << __RQ_WRITE) 214 #define RQ_IN_ACT_LOG (1UL << __RQ_IN_ACT_LOG) 215 216 /* For waking up the frozen transfer log mod_req() has to return if the request 217 should be counted in the epoch object*/ 218 #define MR_WRITE_SHIFT 0 219 #define MR_WRITE (1 << MR_WRITE_SHIFT) 220 #define MR_READ_SHIFT 1 221 #define MR_READ (1 << MR_READ_SHIFT) 222 223 /* epoch entries */ 224 static inline 225 struct hlist_head *ee_hash_slot(struct drbd_conf *mdev, sector_t sector) 226 { 227 BUG_ON(mdev->ee_hash_s == 0); 228 return mdev->ee_hash + 229 ((unsigned int)(sector>>HT_SHIFT) % mdev->ee_hash_s); 230 } 231 232 /* transfer log (drbd_request objects) */ 233 static inline 234 struct hlist_head *tl_hash_slot(struct drbd_conf *mdev, sector_t sector) 235 { 236 BUG_ON(mdev->tl_hash_s == 0); 237 return mdev->tl_hash + 238 ((unsigned int)(sector>>HT_SHIFT) % mdev->tl_hash_s); 239 } 240 241 /* application reads (drbd_request objects) */ 242 static struct hlist_head *ar_hash_slot(struct drbd_conf *mdev, sector_t sector) 243 { 244 return mdev->app_reads_hash 245 + ((unsigned int)(sector) % APP_R_HSIZE); 246 } 247 248 /* when we receive the answer for a read request, 249 * verify that we actually know about it */ 250 static inline struct drbd_request *_ar_id_to_req(struct drbd_conf *mdev, 251 u64 id, sector_t sector) 252 { 253 struct hlist_head *slot = ar_hash_slot(mdev, sector); 254 struct hlist_node *n; 255 struct drbd_request *req; 256 257 hlist_for_each_entry(req, n, slot, colision) { 258 if ((unsigned long)req == (unsigned long)id) { 259 D_ASSERT(req->sector == sector); 260 return req; 261 } 262 } 263 return NULL; 264 } 265 266 static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bio *bio_src) 267 { 268 struct bio *bio; 269 bio = bio_clone(bio_src, GFP_NOIO); /* XXX cannot fail?? */ 270 271 req->private_bio = bio; 272 273 bio->bi_private = req; 274 bio->bi_end_io = drbd_endio_pri; 275 bio->bi_next = NULL; 276 } 277 278 static inline struct drbd_request *drbd_req_new(struct drbd_conf *mdev, 279 struct bio *bio_src) 280 { 281 struct drbd_request *req = 282 mempool_alloc(drbd_request_mempool, GFP_NOIO); 283 if (likely(req)) { 284 drbd_req_make_private_bio(req, bio_src); 285 286 req->rq_state = bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0; 287 req->mdev = mdev; 288 req->master_bio = bio_src; 289 req->epoch = 0; 290 req->sector = bio_src->bi_sector; 291 req->size = bio_src->bi_size; 292 req->start_time = jiffies; 293 INIT_HLIST_NODE(&req->colision); 294 INIT_LIST_HEAD(&req->tl_requests); 295 INIT_LIST_HEAD(&req->w.list); 296 } 297 return req; 298 } 299 300 static inline void drbd_req_free(struct drbd_request *req) 301 { 302 mempool_free(req, drbd_request_mempool); 303 } 304 305 static inline int overlaps(sector_t s1, int l1, sector_t s2, int l2) 306 { 307 return !((s1 + (l1>>9) <= s2) || (s1 >= s2 + (l2>>9))); 308 } 309 310 /* Short lived temporary struct on the stack. 311 * We could squirrel the error to be returned into 312 * bio->bi_size, or similar. But that would be too ugly. */ 313 struct bio_and_error { 314 struct bio *bio; 315 int error; 316 }; 317 318 extern void _req_may_be_done(struct drbd_request *req, 319 struct bio_and_error *m); 320 extern int __req_mod(struct drbd_request *req, enum drbd_req_event what, 321 struct bio_and_error *m); 322 extern void complete_master_bio(struct drbd_conf *mdev, 323 struct bio_and_error *m); 324 325 /* use this if you don't want to deal with calling complete_master_bio() 326 * outside the spinlock, e.g. when walking some list on cleanup. */ 327 static inline int _req_mod(struct drbd_request *req, enum drbd_req_event what) 328 { 329 struct drbd_conf *mdev = req->mdev; 330 struct bio_and_error m; 331 int rv; 332 333 /* __req_mod possibly frees req, do not touch req after that! */ 334 rv = __req_mod(req, what, &m); 335 if (m.bio) 336 complete_master_bio(mdev, &m); 337 338 return rv; 339 } 340 341 /* completion of master bio is outside of spinlock. 342 * If you need it irqsave, do it your self! 343 * Which means: don't use from bio endio callback. */ 344 static inline int req_mod(struct drbd_request *req, 345 enum drbd_req_event what) 346 { 347 struct drbd_conf *mdev = req->mdev; 348 struct bio_and_error m; 349 int rv; 350 351 spin_lock_irq(&mdev->req_lock); 352 rv = __req_mod(req, what, &m); 353 spin_unlock_irq(&mdev->req_lock); 354 355 if (m.bio) 356 complete_master_bio(mdev, &m); 357 358 return rv; 359 } 360 #endif 361