fs/nfs/nfs4filelayout.c

   1 /*
   2  *  Module for the pnfs nfs4 file layout driver.
   3  *  Defines all I/O and Policy interface operations, plus code
   4  *  to register itself with the pNFS client.
   5  *
   6  *  Copyright (c) 2002
   7  *  The Regents of the University of Michigan
   8  *  All Rights Reserved
   9  *
  10  *  Dean Hildebrand <dhildebz@umich.edu>
  11  *
  12  *  Permission is granted to use, copy, create derivative works, and
  13  *  redistribute this software and such derivative works for any purpose,
  14  *  so long as the name of the University of Michigan is not used in
  15  *  any advertising or publicity pertaining to the use or distribution
  16  *  of this software without specific, written prior authorization. If
  17  *  the above copyright notice or any other identification of the
  18  *  University of Michigan is included in any copy of any portion of
  19  *  this software, then the disclaimer below must also be included.
  20  *
  21  *  This software is provided as is, without representation or warranty
  22  *  of any kind either express or implied, including without limitation
  23  *  the implied warranties of merchantability, fitness for a particular
  24  *  purpose, or noninfringement.  The Regents of the University of
  25  *  Michigan shall not be liable for any damages, including special,
  26  *  indirect, incidental, or consequential damages, with respect to any
  27  *  claim arising out of or in connection with the use of the software,
  28  *  even if it has been or is hereafter advised of the possibility of
  29  *  such damages.
  30  */
  31
  32 #include <linux/nfs_fs.h>
  33 #include <linux/nfs_page.h>
  34 #include <linux/module.h>
  35
  36 #include <linux/sunrpc/metrics.h>
  37
  38 #include "internal.h"
  39 #include "delegation.h"
  40 #include "nfs4filelayout.h"
  41
  42 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  43
  44 MODULE_LICENSE("GPL");
  45 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
  46 MODULE_DESCRIPTION("The NFSv4 file layout driver");
  47
  48 #define FILELAYOUT_POLL_RETRY_MAX     (15*HZ)
  49
  50 static loff_t
  51 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
  52                             loff_t offset)
  53 {
  54         u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
  55         u64 stripe_no;
  56         u32 rem;
  57
  58         offset -= flseg->pattern_offset;
  59         stripe_no = div_u64(offset, stripe_width);
  60         div_u64_rem(offset, flseg->stripe_unit, &rem);
  61
  62         return stripe_no * flseg->stripe_unit + rem;
  63 }
  64
  65 /* This function is used by the layout driver to calculate the
  66  * offset of the file on the dserver based on whether the
  67  * layout type is STRIPE_DENSE or STRIPE_SPARSE
  68  */
  69 static loff_t
  70 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
  71 {
  72         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
  73
  74         switch (flseg->stripe_type) {
  75         case STRIPE_SPARSE:
  76                 return offset;
  77
  78         case STRIPE_DENSE:
  79                 return filelayout_get_dense_offset(flseg, offset);
  80         }
  81
  82         BUG();
  83 }
  84
  85 static void filelayout_reset_write(struct nfs_write_data *data)
  86 {
  87         struct nfs_pgio_header *hdr = data->header;
  88         struct rpc_task *task = &data->task;
  89
  90         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
  91                 dprintk("%s Reset task %5u for i/o through MDS "
  92                         "(req %s/%lld, %u bytes @ offset %llu)\n", __func__,
  93                         data->task.tk_pid,
  94                         hdr->inode->i_sb->s_id,
  95                         (long long)NFS_FILEID(hdr->inode),
  96                         data->args.count,
  97                         (unsigned long long)data->args.offset);
  98
  99                 task->tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
 100                                                         &hdr->pages,
 101                                                         hdr->completion_ops);
 102         }
 103 }
 104
 105 static void filelayout_reset_read(struct nfs_read_data *data)
 106 {
 107         struct nfs_pgio_header *hdr = data->header;
 108         struct rpc_task *task = &data->task;
 109
 110         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 111                 dprintk("%s Reset task %5u for i/o through MDS "
 112                         "(req %s/%lld, %u bytes @ offset %llu)\n", __func__,
 113                         data->task.tk_pid,
 114                         hdr->inode->i_sb->s_id,
 115                         (long long)NFS_FILEID(hdr->inode),
 116                         data->args.count,
 117                         (unsigned long long)data->args.offset);
 118
 119                 task->tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
 120                                                         &hdr->pages,
 121                                                         hdr->completion_ops);
 122         }
 123 }
 124
 125 static int filelayout_async_handle_error(struct rpc_task *task,
 126                                          struct nfs4_state *state,
 127                                          struct nfs_client *clp,
 128                                          struct pnfs_layout_segment *lseg)
 129 {
 130         struct inode *inode = lseg->pls_layout->plh_inode;
 131         struct nfs_server *mds_server = NFS_SERVER(inode);
 132         struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
 133         struct nfs_client *mds_client = mds_server->nfs_client;
 134         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 135
 136         if (task->tk_status >= 0)
 137                 return 0;
 138
 139         switch (task->tk_status) {
 140         /* MDS state errors */
 141         case -NFS4ERR_DELEG_REVOKED:
 142         case -NFS4ERR_ADMIN_REVOKED:
 143         case -NFS4ERR_BAD_STATEID:
 144                 if (state == NULL)
 145                         break;
 146                 nfs_remove_bad_delegation(state->inode);
 147         case -NFS4ERR_OPENMODE:
 148                 if (state == NULL)
 149                         break;
 150                 nfs4_schedule_stateid_recovery(mds_server, state);
 151                 goto wait_on_recovery;
 152         case -NFS4ERR_EXPIRED:
 153                 if (state != NULL)
 154                         nfs4_schedule_stateid_recovery(mds_server, state);
 155                 nfs4_schedule_lease_recovery(mds_client);
 156                 goto wait_on_recovery;
 157         /* DS session errors */
 158         case -NFS4ERR_BADSESSION:
 159         case -NFS4ERR_BADSLOT:
 160         case -NFS4ERR_BAD_HIGH_SLOT:
 161         case -NFS4ERR_DEADSESSION:
 162         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 163         case -NFS4ERR_SEQ_FALSE_RETRY:
 164         case -NFS4ERR_SEQ_MISORDERED:
 165                 dprintk("%s ERROR %d, Reset session. Exchangeid "
 166                         "flags 0x%x\n", __func__, task->tk_status,
 167                         clp->cl_exchange_flags);
 168                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
 169                 break;
 170         case -NFS4ERR_DELAY:
 171         case -NFS4ERR_GRACE:
 172         case -EKEYEXPIRED:
 173                 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
 174                 break;
 175         case -NFS4ERR_RETRY_UNCACHED_REP:
 176                 break;
 177         /* Invalidate Layout errors */
 178         case -NFS4ERR_PNFS_NO_LAYOUT:
 179         case -ESTALE:           /* mapped NFS4ERR_STALE */
 180         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
 181         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
 182         case -NFS4ERR_FHEXPIRED:
 183         case -NFS4ERR_WRONG_TYPE:
 184                 dprintk("%s Invalid layout error %d\n", __func__,
 185                         task->tk_status);
 186                 /*
 187                  * Destroy layout so new i/o will get a new layout.
 188                  * Layout will not be destroyed until all current lseg
 189                  * references are put. Mark layout as invalid to resend failed
 190                  * i/o and all i/o waiting on the slot table to the MDS until
 191                  * layout is destroyed and a new valid layout is obtained.
 192                  */
 193                 pnfs_destroy_layout(NFS_I(inode));
 194                 rpc_wake_up(&tbl->slot_tbl_waitq);
 195                 goto reset;
 196         /* RPC connection errors */
 197         case -ECONNREFUSED:
 198         case -EHOSTDOWN:
 199         case -EHOSTUNREACH:
 200         case -ENETUNREACH:
 201         case -EIO:
 202         case -ETIMEDOUT:
 203         case -EPIPE:
 204                 dprintk("%s DS connection error %d\n", __func__,
 205                         task->tk_status);
 206                 nfs4_mark_deviceid_unavailable(devid);
 207                 clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags);
 208                 _pnfs_return_layout(inode);
 209                 rpc_wake_up(&tbl->slot_tbl_waitq);
 210                 nfs4_ds_disconnect(clp);
 211                 /* fall through */
 212         default:
 213 reset:
 214                 dprintk("%s Retry through MDS. Error %d\n", __func__,
 215                         task->tk_status);
 216                 return -NFS4ERR_RESET_TO_MDS;
 217         }
 218 out:
 219         task->tk_status = 0;
 220         return -EAGAIN;
 221 wait_on_recovery:
 222         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
 223         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
 224                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
 225         goto out;
 226 }
 227
 228 /* NFS_PROTO call done callback routines */
 229
 230 static int filelayout_read_done_cb(struct rpc_task *task,
 231                                 struct nfs_read_data *data)
 232 {
 233         struct nfs_pgio_header *hdr = data->header;
 234         int err;
 235
 236         err = filelayout_async_handle_error(task, data->args.context->state,
 237                                             data->ds_clp, hdr->lseg);
 238
 239         switch (err) {
 240         case -NFS4ERR_RESET_TO_MDS:
 241                 filelayout_reset_read(data);
 242                 return task->tk_status;
 243         case -EAGAIN:
 244                 rpc_restart_call_prepare(task);
 245                 return -EAGAIN;
 246         }
 247
 248         return 0;
 249 }
 250
 251 /*
 252  * We reference the rpc_cred of the first WRITE that triggers the need for
 253  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
 254  * rfc5661 is not clear about which credential should be used.
 255  */
 256 static void
 257 filelayout_set_layoutcommit(struct nfs_write_data *wdata)
 258 {
 259         struct nfs_pgio_header *hdr = wdata->header;
 260
 261         if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
 262             wdata->res.verf->committed == NFS_FILE_SYNC)
 263                 return;
 264
 265         pnfs_set_layoutcommit(wdata);
 266         dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
 267                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
 268 }
 269
 270 bool
 271 filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
 272 {
 273         return filelayout_test_devid_invalid(node) ||
 274                 nfs4_test_deviceid_unavailable(node);
 275 }
 276
 277 static bool
 278 filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
 279 {
 280         struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
 281
 282         return filelayout_test_devid_unavailable(node);
 283 }
 284
 285 /*
 286  * Call ops for the async read/write cases
 287  * In the case of dense layouts, the offset needs to be reset to its
 288  * original value.
 289  */
 290 static void filelayout_read_prepare(struct rpc_task *task, void *data)
 291 {
 292         struct nfs_read_data *rdata = data;
 293
 294         if (filelayout_reset_to_mds(rdata->header->lseg)) {
 295                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
 296                 filelayout_reset_read(rdata);
 297                 rpc_exit(task, 0);
 298                 return;
 299         }
 300         rdata->read_done_cb = filelayout_read_done_cb;
 301
 302         if (nfs41_setup_sequence(rdata->ds_clp->cl_session,
 303                                 &rdata->args.seq_args, &rdata->res.seq_res,
 304                                 task))
 305                 return;
 306
 307         rpc_call_start(task);
 308 }
 309
 310 static void filelayout_read_call_done(struct rpc_task *task, void *data)
 311 {
 312         struct nfs_read_data *rdata = data;
 313
 314         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
 315
 316         if (test_bit(NFS_IOHDR_REDO, &rdata->header->flags) &&
 317             task->tk_status == 0)
 318                 return;
 319
 320         /* Note this may cause RPC to be resent */
 321         rdata->header->mds_ops->rpc_call_done(task, data);
 322 }
 323
 324 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
 325 {
 326         struct nfs_read_data *rdata = data;
 327
 328         rpc_count_iostats(task, NFS_SERVER(rdata->header->inode)->client->cl_metrics);
 329 }
 330
 331 static void filelayout_read_release(void *data)
 332 {
 333         struct nfs_read_data *rdata = data;
 334
 335         nfs_put_client(rdata->ds_clp);
 336         rdata->header->mds_ops->rpc_release(data);
 337 }
 338
 339 static int filelayout_write_done_cb(struct rpc_task *task,
 340                                 struct nfs_write_data *data)
 341 {
 342         struct nfs_pgio_header *hdr = data->header;
 343         int err;
 344
 345         err = filelayout_async_handle_error(task, data->args.context->state,
 346                                             data->ds_clp, hdr->lseg);
 347
 348         switch (err) {
 349         case -NFS4ERR_RESET_TO_MDS:
 350                 filelayout_reset_write(data);
 351                 return task->tk_status;
 352         case -EAGAIN:
 353                 rpc_restart_call_prepare(task);
 354                 return -EAGAIN;
 355         }
 356
 357         filelayout_set_layoutcommit(data);
 358         return 0;
 359 }
 360
 361 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
 362 static void prepare_to_resend_writes(struct nfs_commit_data *data)
 363 {
 364         struct nfs_page *first = nfs_list_entry(data->pages.next);
 365
 366         data->task.tk_status = 0;
 367         memcpy(&data->verf.verifier, &first->wb_verf,
 368                sizeof(data->verf.verifier));
 369         data->verf.verifier.data[0]++; /* ensure verifier mismatch */
 370 }
 371
 372 static int filelayout_commit_done_cb(struct rpc_task *task,
 373                                      struct nfs_commit_data *data)
 374 {
 375         int err;
 376
 377         err = filelayout_async_handle_error(task, NULL, data->ds_clp,
 378                                             data->lseg);
 379
 380         switch (err) {
 381         case -NFS4ERR_RESET_TO_MDS:
 382                 prepare_to_resend_writes(data);
 383                 return -EAGAIN;
 384         case -EAGAIN:
 385                 rpc_restart_call_prepare(task);
 386                 return -EAGAIN;
 387         }
 388
 389         return 0;
 390 }
 391
 392 static void filelayout_write_prepare(struct rpc_task *task, void *data)
 393 {
 394         struct nfs_write_data *wdata = data;
 395
 396         if (filelayout_reset_to_mds(wdata->header->lseg)) {
 397                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
 398                 filelayout_reset_write(wdata);
 399                 rpc_exit(task, 0);
 400                 return;
 401         }
 402         if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
 403                                 &wdata->args.seq_args, &wdata->res.seq_res,
 404                                 task))
 405                 return;
 406
 407         rpc_call_start(task);
 408 }
 409
 410 static void filelayout_write_call_done(struct rpc_task *task, void *data)
 411 {
 412         struct nfs_write_data *wdata = data;
 413
 414         if (test_bit(NFS_IOHDR_REDO, &wdata->header->flags) &&
 415             task->tk_status == 0)
 416                 return;
 417
 418         /* Note this may cause RPC to be resent */
 419         wdata->header->mds_ops->rpc_call_done(task, data);
 420 }
 421
 422 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
 423 {
 424         struct nfs_write_data *wdata = data;
 425
 426         rpc_count_iostats(task, NFS_SERVER(wdata->header->inode)->client->cl_metrics);
 427 }
 428
 429 static void filelayout_write_release(void *data)
 430 {
 431         struct nfs_write_data *wdata = data;
 432
 433         nfs_put_client(wdata->ds_clp);
 434         wdata->header->mds_ops->rpc_release(data);
 435 }
 436
 437 static void filelayout_commit_prepare(struct rpc_task *task, void *data)
 438 {
 439         struct nfs_commit_data *wdata = data;
 440
 441         if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
 442                                 &wdata->args.seq_args, &wdata->res.seq_res,
 443                                 task))
 444                 return;
 445
 446         rpc_call_start(task);
 447 }
 448
 449 static void filelayout_write_commit_done(struct rpc_task *task, void *data)
 450 {
 451         struct nfs_commit_data *wdata = data;
 452
 453         /* Note this may cause RPC to be resent */
 454         wdata->mds_ops->rpc_call_done(task, data);
 455 }
 456
 457 static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
 458 {
 459         struct nfs_commit_data *cdata = data;
 460
 461         rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
 462 }
 463
 464 static void filelayout_commit_release(void *calldata)
 465 {
 466         struct nfs_commit_data *data = calldata;
 467
 468         data->completion_ops->completion(data);
 469         pnfs_put_lseg(data->lseg);
 470         nfs_put_client(data->ds_clp);
 471         nfs_commitdata_release(data);
 472 }
 473
 474 static const struct rpc_call_ops filelayout_read_call_ops = {
 475         .rpc_call_prepare = filelayout_read_prepare,
 476         .rpc_call_done = filelayout_read_call_done,
 477         .rpc_count_stats = filelayout_read_count_stats,
 478         .rpc_release = filelayout_read_release,
 479 };
 480
 481 static const struct rpc_call_ops filelayout_write_call_ops = {
 482         .rpc_call_prepare = filelayout_write_prepare,
 483         .rpc_call_done = filelayout_write_call_done,
 484         .rpc_count_stats = filelayout_write_count_stats,
 485         .rpc_release = filelayout_write_release,
 486 };
 487
 488 static const struct rpc_call_ops filelayout_commit_call_ops = {
 489         .rpc_call_prepare = filelayout_commit_prepare,
 490         .rpc_call_done = filelayout_write_commit_done,
 491         .rpc_count_stats = filelayout_commit_count_stats,
 492         .rpc_release = filelayout_commit_release,
 493 };
 494
 495 static enum pnfs_try_status
 496 filelayout_read_pagelist(struct nfs_read_data *data)
 497 {
 498         struct nfs_pgio_header *hdr = data->header;
 499         struct pnfs_layout_segment *lseg = hdr->lseg;
 500         struct nfs4_pnfs_ds *ds;
 501         loff_t offset = data->args.offset;
 502         u32 j, idx;
 503         struct nfs_fh *fh;
 504         int status;
 505
 506         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
 507                 __func__, hdr->inode->i_ino,
 508                 data->args.pgbase, (size_t)data->args.count, offset);
 509
 510         /* Retrieve the correct rpc_client for the byte range */
 511         j = nfs4_fl_calc_j_index(lseg, offset);
 512         idx = nfs4_fl_calc_ds_index(lseg, j);
 513         ds = nfs4_fl_prepare_ds(lseg, idx);
 514         if (!ds)
 515                 return PNFS_NOT_ATTEMPTED;
 516         dprintk("%s USE DS: %s cl_count %d\n", __func__,
 517                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
 518
 519         /* No multipath support. Use first DS */
 520         atomic_inc(&ds->ds_clp->cl_count);
 521         data->ds_clp = ds->ds_clp;
 522         fh = nfs4_fl_select_ds_fh(lseg, j);
 523         if (fh)
 524                 data->args.fh = fh;
 525
 526         data->args.offset = filelayout_get_dserver_offset(lseg, offset);
 527         data->mds_offset = offset;
 528
 529         /* Perform an asynchronous read to ds */
 530         status = nfs_initiate_read(ds->ds_clp->cl_rpcclient, data,
 531                                   &filelayout_read_call_ops, RPC_TASK_SOFTCONN);
 532         BUG_ON(status != 0);
 533         return PNFS_ATTEMPTED;
 534 }
 535
 536 /* Perform async writes. */
 537 static enum pnfs_try_status
 538 filelayout_write_pagelist(struct nfs_write_data *data, int sync)
 539 {
 540         struct nfs_pgio_header *hdr = data->header;
 541         struct pnfs_layout_segment *lseg = hdr->lseg;
 542         struct nfs4_pnfs_ds *ds;
 543         loff_t offset = data->args.offset;
 544         u32 j, idx;
 545         struct nfs_fh *fh;
 546         int status;
 547
 548         /* Retrieve the correct rpc_client for the byte range */
 549         j = nfs4_fl_calc_j_index(lseg, offset);
 550         idx = nfs4_fl_calc_ds_index(lseg, j);
 551         ds = nfs4_fl_prepare_ds(lseg, idx);
 552         if (!ds)
 553                 return PNFS_NOT_ATTEMPTED;
 554         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
 555                 __func__, hdr->inode->i_ino, sync, (size_t) data->args.count,
 556                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
 557
 558         data->write_done_cb = filelayout_write_done_cb;
 559         atomic_inc(&ds->ds_clp->cl_count);
 560         data->ds_clp = ds->ds_clp;
 561         fh = nfs4_fl_select_ds_fh(lseg, j);
 562         if (fh)
 563                 data->args.fh = fh;
 564         /*
 565          * Get the file offset on the dserver. Set the write offset to
 566          * this offset and save the original offset.
 567          */
 568         data->args.offset = filelayout_get_dserver_offset(lseg, offset);
 569
 570         /* Perform an asynchronous write */
 571         status = nfs_initiate_write(ds->ds_clp->cl_rpcclient, data,
 572                                     &filelayout_write_call_ops, sync,
 573                                     RPC_TASK_SOFTCONN);
 574         BUG_ON(status != 0);
 575         return PNFS_ATTEMPTED;
 576 }
 577
 578 /*
 579  * filelayout_check_layout()
 580  *
 581  * Make sure layout segment parameters are sane WRT the device.
 582  * At this point no generic layer initialization of the lseg has occurred,
 583  * and nothing has been added to the layout_hdr cache.
 584  *
 585  */
 586 static int
 587 filelayout_check_layout(struct pnfs_layout_hdr *lo,
 588                         struct nfs4_filelayout_segment *fl,
 589                         struct nfs4_layoutget_res *lgr,
 590                         struct nfs4_deviceid *id,
 591                         gfp_t gfp_flags)
 592 {
 593         struct nfs4_deviceid_node *d;
 594         struct nfs4_file_layout_dsaddr *dsaddr;
 595         int status = -EINVAL;
 596         struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
 597
 598         dprintk("--> %s\n", __func__);
 599
 600         /* FIXME: remove this check when layout segment support is added */
 601         if (lgr->range.offset != 0 ||
 602             lgr->range.length != NFS4_MAX_UINT64) {
 603                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 604                         __func__);
 605                 goto out;
 606         }
 607
 608         if (fl->pattern_offset > lgr->range.offset) {
 609                 dprintk("%s pattern_offset %lld too large\n",
 610                                 __func__, fl->pattern_offset);
 611                 goto out;
 612         }
 613
 614         if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
 615                 dprintk("%s Invalid stripe unit (%u)\n",
 616                         __func__, fl->stripe_unit);
 617                 goto out;
 618         }
 619
 620         /* find and reference the deviceid */
 621         d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
 622                                    NFS_SERVER(lo->plh_inode)->nfs_client, id);
 623         if (d == NULL) {
 624                 dsaddr = filelayout_get_device_info(lo->plh_inode, id, gfp_flags);
 625                 if (dsaddr == NULL)
 626                         goto out;
 627         } else
 628                 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
 629         /* Found deviceid is unavailable */
 630         if (filelayout_test_devid_unavailable(&dsaddr->id_node))
 631                         goto out_put;
 632
 633         fl->dsaddr = dsaddr;
 634
 635         if (fl->first_stripe_index >= dsaddr->stripe_count) {
 636                 dprintk("%s Bad first_stripe_index %u\n",
 637                                 __func__, fl->first_stripe_index);
 638                 goto out_put;
 639         }
 640
 641         if ((fl->stripe_type == STRIPE_SPARSE &&
 642             fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
 643             (fl->stripe_type == STRIPE_DENSE &&
 644             fl->num_fh != dsaddr->stripe_count)) {
 645                 dprintk("%s num_fh %u not valid for given packing\n",
 646                         __func__, fl->num_fh);
 647                 goto out_put;
 648         }
 649
 650         if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
 651                 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
 652                         "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
 653                         nfss->wsize);
 654         }
 655
 656         status = 0;
 657 out:
 658         dprintk("--> %s returns %d\n", __func__, status);
 659         return status;
 660 out_put:
 661         nfs4_fl_put_deviceid(dsaddr);
 662         goto out;
 663 }
 664
 665 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
 666 {
 667         int i;
 668
 669         for (i = 0; i < fl->num_fh; i++) {
 670                 if (!fl->fh_array[i])
 671                         break;
 672                 kfree(fl->fh_array[i]);
 673         }
 674         kfree(fl->fh_array);
 675         fl->fh_array = NULL;
 676 }
 677
 678 static void
 679 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
 680 {
 681         filelayout_free_fh_array(fl);
 682         kfree(fl);
 683 }
 684
 685 static int
 686 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
 687                          struct nfs4_filelayout_segment *fl,
 688                          struct nfs4_layoutget_res *lgr,
 689                          struct nfs4_deviceid *id,
 690                          gfp_t gfp_flags)
 691 {
 692         struct xdr_stream stream;
 693         struct xdr_buf buf;
 694         struct page *scratch;
 695         __be32 *p;
 696         uint32_t nfl_util;
 697         int i;
 698
 699         dprintk("%s: set_layout_map Begin\n", __func__);
 700
 701         scratch = alloc_page(gfp_flags);
 702         if (!scratch)
 703                 return -ENOMEM;
 704
 705         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
 706         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 707
 708         /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
 709          * num_fh (4) */
 710         p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
 711         if (unlikely(!p))
 712                 goto out_err;
 713
 714         memcpy(id, p, sizeof(*id));
 715         p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
 716         nfs4_print_deviceid(id);
 717
 718         nfl_util = be32_to_cpup(p++);
 719         if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
 720                 fl->commit_through_mds = 1;
 721         if (nfl_util & NFL4_UFLG_DENSE)
 722                 fl->stripe_type = STRIPE_DENSE;
 723         else
 724                 fl->stripe_type = STRIPE_SPARSE;
 725         fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
 726
 727         fl->first_stripe_index = be32_to_cpup(p++);
 728         p = xdr_decode_hyper(p, &fl->pattern_offset);
 729         fl->num_fh = be32_to_cpup(p++);
 730
 731         dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
 732                 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
 733                 fl->pattern_offset);
 734
 735         /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
 736          * Futher checking is done in filelayout_check_layout */
 737         if (fl->num_fh >
 738             max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
 739                 goto out_err;
 740
 741         if (fl->num_fh > 0) {
 742                 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
 743                                        gfp_flags);
 744                 if (!fl->fh_array)
 745                         goto out_err;
 746         }
 747
 748         for (i = 0; i < fl->num_fh; i++) {
 749                 /* Do we want to use a mempool here? */
 750                 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
 751                 if (!fl->fh_array[i])
 752                         goto out_err_free;
 753
 754                 p = xdr_inline_decode(&stream, 4);
 755                 if (unlikely(!p))
 756                         goto out_err_free;
 757                 fl->fh_array[i]->size = be32_to_cpup(p++);
 758                 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
 759                         printk(KERN_ERR "NFS: Too big fh %d received %d\n",
 760                                i, fl->fh_array[i]->size);
 761                         goto out_err_free;
 762                 }
 763
 764                 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
 765                 if (unlikely(!p))
 766                         goto out_err_free;
 767                 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
 768                 dprintk("DEBUG: %s: fh len %d\n", __func__,
 769                         fl->fh_array[i]->size);
 770         }
 771
 772         __free_page(scratch);
 773         return 0;
 774
 775 out_err_free:
 776         filelayout_free_fh_array(fl);
 777 out_err:
 778         __free_page(scratch);
 779         return -EIO;
 780 }
 781
 782 static void
 783 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
 784 {
 785         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
 786
 787         dprintk("--> %s\n", __func__);
 788         nfs4_fl_put_deviceid(fl->dsaddr);
 789         /* This assumes a single RW lseg */
 790         if (lseg->pls_range.iomode == IOMODE_RW) {
 791                 struct nfs4_filelayout *flo;
 792
 793                 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
 794                 flo->commit_info.nbuckets = 0;
 795                 kfree(flo->commit_info.buckets);
 796                 flo->commit_info.buckets = NULL;
 797         }
 798         _filelayout_free_lseg(fl);
 799 }
 800
 801 static int
 802 filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 803                              struct nfs_commit_info *cinfo,
 804                              gfp_t gfp_flags)
 805 {
 806         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
 807         struct pnfs_commit_bucket *buckets;
 808         int size;
 809
 810         if (fl->commit_through_mds)
 811                 return 0;
 812         if (cinfo->ds->nbuckets != 0) {
 813                 /* This assumes there is only one IOMODE_RW lseg.  What
 814                  * we really want to do is have a layout_hdr level
 815                  * dictionary of <multipath_list4, fh> keys, each
 816                  * associated with a struct list_head, populated by calls
 817                  * to filelayout_write_pagelist().
 818                  * */
 819                 return 0;
 820         }
 821
 822         size = (fl->stripe_type == STRIPE_SPARSE) ?
 823                 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
 824
 825         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 826                           gfp_flags);
 827         if (!buckets)
 828                 return -ENOMEM;
 829         else {
 830                 int i;
 831
 832                 spin_lock(cinfo->lock);
 833                 if (cinfo->ds->nbuckets != 0)
 834                         kfree(buckets);
 835                 else {
 836                         cinfo->ds->buckets = buckets;
 837                         cinfo->ds->nbuckets = size;
 838                         for (i = 0; i < size; i++) {
 839                                 INIT_LIST_HEAD(&buckets[i].written);
 840                                 INIT_LIST_HEAD(&buckets[i].committing);
 841                         }
 842                 }
 843                 spin_unlock(cinfo->lock);
 844                 return 0;
 845         }
 846 }
 847
 848 static struct pnfs_layout_segment *
 849 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
 850                       struct nfs4_layoutget_res *lgr,
 851                       gfp_t gfp_flags)
 852 {
 853         struct nfs4_filelayout_segment *fl;
 854         int rc;
 855         struct nfs4_deviceid id;
 856
 857         dprintk("--> %s\n", __func__);
 858         fl = kzalloc(sizeof(*fl), gfp_flags);
 859         if (!fl)
 860                 return NULL;
 861
 862         rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
 863         if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
 864                 _filelayout_free_lseg(fl);
 865                 return NULL;
 866         }
 867         return &fl->generic_hdr;
 868 }
 869
 870 /*
 871  * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
 872  *
 873  * return true  : coalesce page
 874  * return false : don't coalesce page
 875  */
 876 static bool
 877 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
 878                    struct nfs_page *req)
 879 {
 880         u64 p_stripe, r_stripe;
 881         u32 stripe_unit;
 882
 883         if (!pnfs_generic_pg_test(pgio, prev, req) ||
 884             !nfs_generic_pg_test(pgio, prev, req))
 885                 return false;
 886
 887         p_stripe = (u64)req_offset(prev);
 888         r_stripe = (u64)req_offset(req);
 889         stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
 890
 891         do_div(p_stripe, stripe_unit);
 892         do_div(r_stripe, stripe_unit);
 893
 894         return (p_stripe == r_stripe);
 895 }
 896
 897 static void
 898 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 899                         struct nfs_page *req)
 900 {
 901         BUG_ON(pgio->pg_lseg != NULL);
 902
 903         if (req->wb_offset != req->wb_pgbase) {
 904                 /*
 905                  * Handling unaligned pages is difficult, because have to
 906                  * somehow split a req in two in certain cases in the
 907                  * pg.test code.  Avoid this by just not using pnfs
 908                  * in this case.
 909                  */
 910                 nfs_pageio_reset_read_mds(pgio);
 911                 return;
 912         }
 913         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 914                                            req->wb_context,
 915                                            0,
 916                                            NFS4_MAX_UINT64,
 917                                            IOMODE_READ,
 918                                            GFP_KERNEL);
 919         /* If no lseg, fall back to read through mds */
 920         if (pgio->pg_lseg == NULL)
 921                 nfs_pageio_reset_read_mds(pgio);
 922 }
 923
 924 static void
 925 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 926                          struct nfs_page *req)
 927 {
 928         struct nfs_commit_info cinfo;
 929         int status;
 930
 931         BUG_ON(pgio->pg_lseg != NULL);
 932
 933         if (req->wb_offset != req->wb_pgbase)
 934                 goto out_mds;
 935         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 936                                            req->wb_context,
 937                                            0,
 938                                            NFS4_MAX_UINT64,
 939                                            IOMODE_RW,
 940                                            GFP_NOFS);
 941         /* If no lseg, fall back to write through mds */
 942         if (pgio->pg_lseg == NULL)
 943                 goto out_mds;
 944         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 945         status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 946         if (status < 0) {
 947                 pnfs_put_lseg(pgio->pg_lseg);
 948                 pgio->pg_lseg = NULL;
 949                 goto out_mds;
 950         }
 951         return;
 952 out_mds:
 953         nfs_pageio_reset_write_mds(pgio);
 954 }
 955
 956 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
 957         .pg_init = filelayout_pg_init_read,
 958         .pg_test = filelayout_pg_test,
 959         .pg_doio = pnfs_generic_pg_readpages,
 960 };
 961
 962 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
 963         .pg_init = filelayout_pg_init_write,
 964         .pg_test = filelayout_pg_test,
 965         .pg_doio = pnfs_generic_pg_writepages,
 966 };
 967
 968 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
 969 {
 970         if (fl->stripe_type == STRIPE_SPARSE)
 971                 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
 972         else
 973                 return j;
 974 }
 975
 976 /* The generic layer is about to remove the req from the commit list.
 977  * If this will make the bucket empty, it will need to put the lseg reference.
 978  */
 979 static void
 980 filelayout_clear_request_commit(struct nfs_page *req,
 981                                 struct nfs_commit_info *cinfo)
 982 {
 983         struct pnfs_layout_segment *freeme = NULL;
 984
 985         spin_lock(cinfo->lock);
 986         if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
 987                 goto out;
 988         cinfo->ds->nwritten--;
 989         if (list_is_singular(&req->wb_list)) {
 990                 struct pnfs_commit_bucket *bucket;
 991
 992                 bucket = list_first_entry(&req->wb_list,
 993                                           struct pnfs_commit_bucket,
 994                                           written);
 995                 freeme = bucket->wlseg;
 996                 bucket->wlseg = NULL;
 997         }
 998 out:
 999         nfs_request_remove_commit_list(req, cinfo);
1000         spin_unlock(cinfo->lock);
1001         pnfs_put_lseg(freeme);
1002 }
1003
1004 static struct list_head *
1005 filelayout_choose_commit_list(struct nfs_page *req,
1006                               struct pnfs_layout_segment *lseg,
1007                               struct nfs_commit_info *cinfo)
1008 {
1009         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
1010         u32 i, j;
1011         struct list_head *list;
1012         struct pnfs_commit_bucket *buckets;
1013
1014         if (fl->commit_through_mds)
1015                 return &cinfo->mds->list;
1016
1017         /* Note that we are calling nfs4_fl_calc_j_index on each page
1018          * that ends up being committed to a data server.  An attractive
1019          * alternative is to add a field to nfs_write_data and nfs_page
1020          * to store the value calculated in filelayout_write_pagelist
1021          * and just use that here.
1022          */
1023         j = nfs4_fl_calc_j_index(lseg, req_offset(req));
1024         i = select_bucket_index(fl, j);
1025         buckets = cinfo->ds->buckets;
1026         list = &buckets[i].written;
1027         if (list_empty(list)) {
1028                 /* Non-empty buckets hold a reference on the lseg.  That ref
1029                  * is normally transferred to the COMMIT call and released
1030                  * there.  It could also be released if the last req is pulled
1031                  * off due to a rewrite, in which case it will be done in
1032                  * filelayout_clear_request_commit
1033                  */
1034                 buckets[i].wlseg = pnfs_get_lseg(lseg);
1035         }
1036         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1037         cinfo->ds->nwritten++;
1038         return list;
1039 }
1040
1041 static void
1042 filelayout_mark_request_commit(struct nfs_page *req,
1043                                struct pnfs_layout_segment *lseg,
1044                                struct nfs_commit_info *cinfo)
1045 {
1046         struct list_head *list;
1047
1048         list = filelayout_choose_commit_list(req, lseg, cinfo);
1049         nfs_request_add_commit_list(req, list, cinfo);
1050 }
1051
1052 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1053 {
1054         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1055
1056         if (flseg->stripe_type == STRIPE_SPARSE)
1057                 return i;
1058         else
1059                 return nfs4_fl_calc_ds_index(lseg, i);
1060 }
1061
1062 static struct nfs_fh *
1063 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1064 {
1065         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1066
1067         if (flseg->stripe_type == STRIPE_SPARSE) {
1068                 if (flseg->num_fh == 1)
1069                         i = 0;
1070                 else if (flseg->num_fh == 0)
1071                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1072                         return NULL;
1073         }
1074         return flseg->fh_array[i];
1075 }
1076
1077 static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1078 {
1079         struct pnfs_layout_segment *lseg = data->lseg;
1080         struct nfs4_pnfs_ds *ds;
1081         u32 idx;
1082         struct nfs_fh *fh;
1083
1084         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1085         ds = nfs4_fl_prepare_ds(lseg, idx);
1086         if (!ds) {
1087                 prepare_to_resend_writes(data);
1088                 filelayout_commit_release(data);
1089                 return -EAGAIN;
1090         }
1091         dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1092                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count));
1093         data->commit_done_cb = filelayout_commit_done_cb;
1094         atomic_inc(&ds->ds_clp->cl_count);
1095         data->ds_clp = ds->ds_clp;
1096         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1097         if (fh)
1098                 data->args.fh = fh;
1099         return nfs_initiate_commit(ds->ds_clp->cl_rpcclient, data,
1100                                    &filelayout_commit_call_ops, how,
1101                                    RPC_TASK_SOFTCONN);
1102 }
1103
1104 static int
1105 transfer_commit_list(struct list_head *src, struct list_head *dst,
1106                      struct nfs_commit_info *cinfo, int max)
1107 {
1108         struct nfs_page *req, *tmp;
1109         int ret = 0;
1110
1111         list_for_each_entry_safe(req, tmp, src, wb_list) {
1112                 if (!nfs_lock_request(req))
1113                         continue;
1114                 kref_get(&req->wb_kref);
1115                 if (cond_resched_lock(cinfo->lock))
1116                         list_safe_reset_next(req, tmp, wb_list);
1117                 nfs_request_remove_commit_list(req, cinfo);
1118                 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1119                 nfs_list_add_request(req, dst);
1120                 ret++;
1121                 if ((ret == max) && !cinfo->dreq)
1122                         break;
1123         }
1124         return ret;
1125 }
1126
1127 static int
1128 filelayout_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
1129                                struct nfs_commit_info *cinfo,
1130                                int max)
1131 {
1132         struct list_head *src = &bucket->written;
1133         struct list_head *dst = &bucket->committing;
1134         int ret;
1135
1136         ret = transfer_commit_list(src, dst, cinfo, max);
1137         if (ret) {
1138                 cinfo->ds->nwritten -= ret;
1139                 cinfo->ds->ncommitting += ret;
1140                 bucket->clseg = bucket->wlseg;
1141                 if (list_empty(src))
1142                         bucket->wlseg = NULL;
1143                 else
1144                         pnfs_get_lseg(bucket->clseg);
1145         }
1146         return ret;
1147 }
1148
1149 /* Move reqs from written to committing lists, returning count of number moved.
1150  * Note called with cinfo->lock held.
1151  */
1152 static int filelayout_scan_commit_lists(struct nfs_commit_info *cinfo,
1153                                         int max)
1154 {
1155         int i, rv = 0, cnt;
1156
1157         for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
1158                 cnt = filelayout_scan_ds_commit_list(&cinfo->ds->buckets[i],
1159                                                      cinfo, max);
1160                 max -= cnt;
1161                 rv += cnt;
1162         }
1163         return rv;
1164 }
1165
1166 /* Pull everything off the committing lists and dump into @dst */
1167 static void filelayout_recover_commit_reqs(struct list_head *dst,
1168                                            struct nfs_commit_info *cinfo)
1169 {
1170         struct pnfs_commit_bucket *b;
1171         int i;
1172
1173         /* NOTE cinfo->lock is NOT held, relying on fact that this is
1174          * only called on single thread per dreq.
1175          * Can't take the lock because need to do pnfs_put_lseg
1176          */
1177         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1178                 if (transfer_commit_list(&b->written, dst, cinfo, 0)) {
1179                         BUG_ON(!list_empty(&b->written));
1180                         pnfs_put_lseg(b->wlseg);
1181                         b->wlseg = NULL;
1182                 }
1183         }
1184         cinfo->ds->nwritten = 0;
1185 }
1186
1187 static unsigned int
1188 alloc_ds_commits(struct nfs_commit_info *cinfo, struct list_head *list)
1189 {
1190         struct pnfs_ds_commit_info *fl_cinfo;
1191         struct pnfs_commit_bucket *bucket;
1192         struct nfs_commit_data *data;
1193         int i, j;
1194         unsigned int nreq = 0;
1195
1196         fl_cinfo = cinfo->ds;
1197         bucket = fl_cinfo->buckets;
1198         for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
1199                 if (list_empty(&bucket->committing))
1200                         continue;
1201                 data = nfs_commitdata_alloc();
1202                 if (!data)
1203                         break;
1204                 data->ds_commit_index = i;
1205                 data->lseg = bucket->clseg;
1206                 bucket->clseg = NULL;
1207                 list_add(&data->pages, list);
1208                 nreq++;
1209         }
1210
1211         /* Clean up on error */
1212         for (j = i; j < fl_cinfo->nbuckets; j++, bucket++) {
1213                 if (list_empty(&bucket->committing))
1214                         continue;
1215                 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo);
1216                 pnfs_put_lseg(bucket->clseg);
1217                 bucket->clseg = NULL;
1218         }
1219         /* Caller will clean up entries put on list */
1220         return nreq;
1221 }
1222
1223 /* This follows nfs_commit_list pretty closely */
1224 static int
1225 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1226                            int how, struct nfs_commit_info *cinfo)
1227 {
1228         struct nfs_commit_data *data, *tmp;
1229         LIST_HEAD(list);
1230         unsigned int nreq = 0;
1231
1232         if (!list_empty(mds_pages)) {
1233                 data = nfs_commitdata_alloc();
1234                 if (data != NULL) {
1235                         data->lseg = NULL;
1236                         list_add(&data->pages, &list);
1237                         nreq++;
1238                 } else
1239                         nfs_retry_commit(mds_pages, NULL, cinfo);
1240         }
1241
1242         nreq += alloc_ds_commits(cinfo, &list);
1243
1244         if (nreq == 0) {
1245                 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1246                 goto out;
1247         }
1248
1249         atomic_add(nreq, &cinfo->mds->rpcs_out);
1250
1251         list_for_each_entry_safe(data, tmp, &list, pages) {
1252                 list_del_init(&data->pages);
1253                 if (!data->lseg) {
1254                         nfs_init_commit(data, mds_pages, NULL, cinfo);
1255                         nfs_initiate_commit(NFS_CLIENT(inode), data,
1256                                             data->mds_ops, how, 0);
1257                 } else {
1258                         struct pnfs_commit_bucket *buckets;
1259
1260                         buckets = cinfo->ds->buckets;
1261                         nfs_init_commit(data, &buckets[data->ds_commit_index].committing, data->lseg, cinfo);
1262                         filelayout_initiate_commit(data, how);
1263                 }
1264         }
1265 out:
1266         cinfo->ds->ncommitting = 0;
1267         return PNFS_ATTEMPTED;
1268 }
1269
1270 static void
1271 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
1272 {
1273         nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1274 }
1275
1276 static struct pnfs_layout_hdr *
1277 filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1278 {
1279         struct nfs4_filelayout *flo;
1280
1281         flo = kzalloc(sizeof(*flo), gfp_flags);
1282         return &flo->generic_hdr;
1283 }
1284
1285 static void
1286 filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1287 {
1288         kfree(FILELAYOUT_FROM_HDR(lo));
1289 }
1290
1291 static struct pnfs_ds_commit_info *
1292 filelayout_get_ds_info(struct inode *inode)
1293 {
1294         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1295
1296         if (layout == NULL)
1297                 return NULL;
1298         else
1299                 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1300 }
1301
1302 static struct pnfs_layoutdriver_type filelayout_type = {
1303         .id                     = LAYOUT_NFSV4_1_FILES,
1304         .name                   = "LAYOUT_NFSV4_1_FILES",
1305         .owner                  = THIS_MODULE,
1306         .alloc_layout_hdr       = filelayout_alloc_layout_hdr,
1307         .free_layout_hdr        = filelayout_free_layout_hdr,
1308         .alloc_lseg             = filelayout_alloc_lseg,
1309         .free_lseg              = filelayout_free_lseg,
1310         .pg_read_ops            = &filelayout_pg_read_ops,
1311         .pg_write_ops           = &filelayout_pg_write_ops,
1312         .get_ds_info            = &filelayout_get_ds_info,
1313         .mark_request_commit    = filelayout_mark_request_commit,
1314         .clear_request_commit   = filelayout_clear_request_commit,
1315         .scan_commit_lists      = filelayout_scan_commit_lists,
1316         .recover_commit_reqs    = filelayout_recover_commit_reqs,
1317         .commit_pagelist        = filelayout_commit_pagelist,
1318         .read_pagelist          = filelayout_read_pagelist,
1319         .write_pagelist         = filelayout_write_pagelist,
1320         .free_deviceid_node     = filelayout_free_deveiceid_node,
1321 };
1322
1323 static int __init nfs4filelayout_init(void)
1324 {
1325         printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1326                __func__);
1327         return pnfs_register_layoutdriver(&filelayout_type);
1328 }
1329
1330 static void __exit nfs4filelayout_exit(void)
1331 {
1332         printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1333                __func__);
1334         pnfs_unregister_layoutdriver(&filelayout_type);
1335 }
1336
1337 MODULE_ALIAS("nfs-layouttype4-1");
1338
1339 module_init(nfs4filelayout_init);
1340 module_exit(nfs4filelayout_exit);