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 "nfs4session.h"
  39 #include "internal.h"
  40 #include "delegation.h"
  41 #include "nfs4filelayout.h"
  42
  43 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  44
  45 MODULE_LICENSE("GPL");
  46 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
  47 MODULE_DESCRIPTION("The NFSv4 file layout driver");
  48
  49 #define FILELAYOUT_POLL_RETRY_MAX     (15*HZ)
  50
  51 static loff_t
  52 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
  53                             loff_t offset)
  54 {
  55         u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
  56         u64 stripe_no;
  57         u32 rem;
  58
  59         offset -= flseg->pattern_offset;
  60         stripe_no = div_u64(offset, stripe_width);
  61         div_u64_rem(offset, flseg->stripe_unit, &rem);
  62
  63         return stripe_no * flseg->stripe_unit + rem;
  64 }
  65
  66 /* This function is used by the layout driver to calculate the
  67  * offset of the file on the dserver based on whether the
  68  * layout type is STRIPE_DENSE or STRIPE_SPARSE
  69  */
  70 static loff_t
  71 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
  72 {
  73         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
  74
  75         switch (flseg->stripe_type) {
  76         case STRIPE_SPARSE:
  77                 return offset;
  78
  79         case STRIPE_DENSE:
  80                 return filelayout_get_dense_offset(flseg, offset);
  81         }
  82
  83         BUG();
  84 }
  85
  86 static void filelayout_reset_write(struct nfs_write_data *data)
  87 {
  88         struct nfs_pgio_header *hdr = data->header;
  89         struct rpc_task *task = &data->task;
  90
  91         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
  92                 dprintk("%s Reset task %5u for i/o through MDS "
  93                         "(req %s/%lld, %u bytes @ offset %llu)\n", __func__,
  94                         data->task.tk_pid,
  95                         hdr->inode->i_sb->s_id,
  96                         (long long)NFS_FILEID(hdr->inode),
  97                         data->args.count,
  98                         (unsigned long long)data->args.offset);
  99
 100                 task->tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
 101                                                         &hdr->pages,
 102                                                         hdr->completion_ops,
 103                                                         hdr->dreq);
 104         }
 105 }
 106
 107 static void filelayout_reset_read(struct nfs_read_data *data)
 108 {
 109         struct nfs_pgio_header *hdr = data->header;
 110         struct rpc_task *task = &data->task;
 111
 112         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 113                 dprintk("%s Reset task %5u for i/o through MDS "
 114                         "(req %s/%lld, %u bytes @ offset %llu)\n", __func__,
 115                         data->task.tk_pid,
 116                         hdr->inode->i_sb->s_id,
 117                         (long long)NFS_FILEID(hdr->inode),
 118                         data->args.count,
 119                         (unsigned long long)data->args.offset);
 120
 121                 task->tk_status = pnfs_read_done_resend_to_mds(hdr->inode,
 122                                                         &hdr->pages,
 123                                                         hdr->completion_ops,
 124                                                         hdr->dreq);
 125         }
 126 }
 127
 128 static void filelayout_fenceme(struct inode *inode, struct pnfs_layout_hdr *lo)
 129 {
 130         if (!test_and_clear_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
 131                 return;
 132         pnfs_return_layout(inode);
 133 }
 134
 135 static int filelayout_async_handle_error(struct rpc_task *task,
 136                                          struct nfs4_state *state,
 137                                          struct nfs_client *clp,
 138                                          struct pnfs_layout_segment *lseg)
 139 {
 140         struct pnfs_layout_hdr *lo = lseg->pls_layout;
 141         struct inode *inode = lo->plh_inode;
 142         struct nfs_server *mds_server = NFS_SERVER(inode);
 143         struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
 144         struct nfs_client *mds_client = mds_server->nfs_client;
 145         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 146
 147         if (task->tk_status >= 0)
 148                 return 0;
 149
 150         switch (task->tk_status) {
 151         /* MDS state errors */
 152         case -NFS4ERR_DELEG_REVOKED:
 153         case -NFS4ERR_ADMIN_REVOKED:
 154         case -NFS4ERR_BAD_STATEID:
 155                 if (state == NULL)
 156                         break;
 157                 nfs_remove_bad_delegation(state->inode);
 158         case -NFS4ERR_OPENMODE:
 159                 if (state == NULL)
 160                         break;
 161                 nfs4_schedule_stateid_recovery(mds_server, state);
 162                 goto wait_on_recovery;
 163         case -NFS4ERR_EXPIRED:
 164                 if (state != NULL)
 165                         nfs4_schedule_stateid_recovery(mds_server, state);
 166                 nfs4_schedule_lease_recovery(mds_client);
 167                 goto wait_on_recovery;
 168         /* DS session errors */
 169         case -NFS4ERR_BADSESSION:
 170         case -NFS4ERR_BADSLOT:
 171         case -NFS4ERR_BAD_HIGH_SLOT:
 172         case -NFS4ERR_DEADSESSION:
 173         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 174         case -NFS4ERR_SEQ_FALSE_RETRY:
 175         case -NFS4ERR_SEQ_MISORDERED:
 176                 dprintk("%s ERROR %d, Reset session. Exchangeid "
 177                         "flags 0x%x\n", __func__, task->tk_status,
 178                         clp->cl_exchange_flags);
 179                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
 180                 break;
 181         case -NFS4ERR_DELAY:
 182         case -NFS4ERR_GRACE:
 183                 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
 184                 break;
 185         case -NFS4ERR_RETRY_UNCACHED_REP:
 186                 break;
 187         /* Invalidate Layout errors */
 188         case -NFS4ERR_PNFS_NO_LAYOUT:
 189         case -ESTALE:           /* mapped NFS4ERR_STALE */
 190         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
 191         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
 192         case -NFS4ERR_FHEXPIRED:
 193         case -NFS4ERR_WRONG_TYPE:
 194                 dprintk("%s Invalid layout error %d\n", __func__,
 195                         task->tk_status);
 196                 /*
 197                  * Destroy layout so new i/o will get a new layout.
 198                  * Layout will not be destroyed until all current lseg
 199                  * references are put. Mark layout as invalid to resend failed
 200                  * i/o and all i/o waiting on the slot table to the MDS until
 201                  * layout is destroyed and a new valid layout is obtained.
 202                  */
 203                 pnfs_destroy_layout(NFS_I(inode));
 204                 rpc_wake_up(&tbl->slot_tbl_waitq);
 205                 goto reset;
 206         /* RPC connection errors */
 207         case -ECONNREFUSED:
 208         case -EHOSTDOWN:
 209         case -EHOSTUNREACH:
 210         case -ENETUNREACH:
 211         case -EIO:
 212         case -ETIMEDOUT:
 213         case -EPIPE:
 214                 dprintk("%s DS connection error %d\n", __func__,
 215                         task->tk_status);
 216                 nfs4_mark_deviceid_unavailable(devid);
 217                 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
 218                 rpc_wake_up(&tbl->slot_tbl_waitq);
 219                 /* fall through */
 220         default:
 221 reset:
 222                 dprintk("%s Retry through MDS. Error %d\n", __func__,
 223                         task->tk_status);
 224                 return -NFS4ERR_RESET_TO_MDS;
 225         }
 226 out:
 227         task->tk_status = 0;
 228         return -EAGAIN;
 229 wait_on_recovery:
 230         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
 231         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
 232                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
 233         goto out;
 234 }
 235
 236 /* NFS_PROTO call done callback routines */
 237
 238 static int filelayout_read_done_cb(struct rpc_task *task,
 239                                 struct nfs_read_data *data)
 240 {
 241         struct nfs_pgio_header *hdr = data->header;
 242         int err;
 243
 244         err = filelayout_async_handle_error(task, data->args.context->state,
 245                                             data->ds_clp, hdr->lseg);
 246
 247         switch (err) {
 248         case -NFS4ERR_RESET_TO_MDS:
 249                 filelayout_reset_read(data);
 250                 return task->tk_status;
 251         case -EAGAIN:
 252                 rpc_restart_call_prepare(task);
 253                 return -EAGAIN;
 254         }
 255
 256         return 0;
 257 }
 258
 259 /*
 260  * We reference the rpc_cred of the first WRITE that triggers the need for
 261  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
 262  * rfc5661 is not clear about which credential should be used.
 263  */
 264 static void
 265 filelayout_set_layoutcommit(struct nfs_write_data *wdata)
 266 {
 267         struct nfs_pgio_header *hdr = wdata->header;
 268
 269         if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds ||
 270             wdata->res.verf->committed == NFS_FILE_SYNC)
 271                 return;
 272
 273         pnfs_set_layoutcommit(wdata);
 274         dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
 275                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
 276 }
 277
 278 bool
 279 filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node)
 280 {
 281         return filelayout_test_devid_invalid(node) ||
 282                 nfs4_test_deviceid_unavailable(node);
 283 }
 284
 285 static bool
 286 filelayout_reset_to_mds(struct pnfs_layout_segment *lseg)
 287 {
 288         struct nfs4_deviceid_node *node = FILELAYOUT_DEVID_NODE(lseg);
 289
 290         return filelayout_test_devid_unavailable(node);
 291 }
 292
 293 /*
 294  * Call ops for the async read/write cases
 295  * In the case of dense layouts, the offset needs to be reset to its
 296  * original value.
 297  */
 298 static void filelayout_read_prepare(struct rpc_task *task, void *data)
 299 {
 300         struct nfs_read_data *rdata = data;
 301
 302         if (filelayout_reset_to_mds(rdata->header->lseg)) {
 303                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
 304                 filelayout_reset_read(rdata);
 305                 rpc_exit(task, 0);
 306                 return;
 307         }
 308         rdata->read_done_cb = filelayout_read_done_cb;
 309
 310         nfs41_setup_sequence(rdata->ds_clp->cl_session,
 311                         &rdata->args.seq_args,
 312                         &rdata->res.seq_res,
 313                         task);
 314 }
 315
 316 static void filelayout_read_call_done(struct rpc_task *task, void *data)
 317 {
 318         struct nfs_read_data *rdata = data;
 319
 320         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
 321
 322         if (test_bit(NFS_IOHDR_REDO, &rdata->header->flags) &&
 323             task->tk_status == 0)
 324                 return;
 325
 326         /* Note this may cause RPC to be resent */
 327         rdata->header->mds_ops->rpc_call_done(task, data);
 328 }
 329
 330 static void filelayout_read_count_stats(struct rpc_task *task, void *data)
 331 {
 332         struct nfs_read_data *rdata = data;
 333
 334         rpc_count_iostats(task, NFS_SERVER(rdata->header->inode)->client->cl_metrics);
 335 }
 336
 337 static void filelayout_read_release(void *data)
 338 {
 339         struct nfs_read_data *rdata = data;
 340         struct pnfs_layout_hdr *lo = rdata->header->lseg->pls_layout;
 341
 342         filelayout_fenceme(lo->plh_inode, lo);
 343         nfs_put_client(rdata->ds_clp);
 344         rdata->header->mds_ops->rpc_release(data);
 345 }
 346
 347 static int filelayout_write_done_cb(struct rpc_task *task,
 348                                 struct nfs_write_data *data)
 349 {
 350         struct nfs_pgio_header *hdr = data->header;
 351         int err;
 352
 353         err = filelayout_async_handle_error(task, data->args.context->state,
 354                                             data->ds_clp, hdr->lseg);
 355
 356         switch (err) {
 357         case -NFS4ERR_RESET_TO_MDS:
 358                 filelayout_reset_write(data);
 359                 return task->tk_status;
 360         case -EAGAIN:
 361                 rpc_restart_call_prepare(task);
 362                 return -EAGAIN;
 363         }
 364
 365         filelayout_set_layoutcommit(data);
 366         return 0;
 367 }
 368
 369 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
 370 static void prepare_to_resend_writes(struct nfs_commit_data *data)
 371 {
 372         struct nfs_page *first = nfs_list_entry(data->pages.next);
 373
 374         data->task.tk_status = 0;
 375         memcpy(&data->verf.verifier, &first->wb_verf,
 376                sizeof(data->verf.verifier));
 377         data->verf.verifier.data[0]++; /* ensure verifier mismatch */
 378 }
 379
 380 static int filelayout_commit_done_cb(struct rpc_task *task,
 381                                      struct nfs_commit_data *data)
 382 {
 383         int err;
 384
 385         err = filelayout_async_handle_error(task, NULL, data->ds_clp,
 386                                             data->lseg);
 387
 388         switch (err) {
 389         case -NFS4ERR_RESET_TO_MDS:
 390                 prepare_to_resend_writes(data);
 391                 return -EAGAIN;
 392         case -EAGAIN:
 393                 rpc_restart_call_prepare(task);
 394                 return -EAGAIN;
 395         }
 396
 397         return 0;
 398 }
 399
 400 static void filelayout_write_prepare(struct rpc_task *task, void *data)
 401 {
 402         struct nfs_write_data *wdata = data;
 403
 404         if (filelayout_reset_to_mds(wdata->header->lseg)) {
 405                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
 406                 filelayout_reset_write(wdata);
 407                 rpc_exit(task, 0);
 408                 return;
 409         }
 410         nfs41_setup_sequence(wdata->ds_clp->cl_session,
 411                         &wdata->args.seq_args,
 412                         &wdata->res.seq_res,
 413                         task);
 414 }
 415
 416 static void filelayout_write_call_done(struct rpc_task *task, void *data)
 417 {
 418         struct nfs_write_data *wdata = data;
 419
 420         if (test_bit(NFS_IOHDR_REDO, &wdata->header->flags) &&
 421             task->tk_status == 0)
 422                 return;
 423
 424         /* Note this may cause RPC to be resent */
 425         wdata->header->mds_ops->rpc_call_done(task, data);
 426 }
 427
 428 static void filelayout_write_count_stats(struct rpc_task *task, void *data)
 429 {
 430         struct nfs_write_data *wdata = data;
 431
 432         rpc_count_iostats(task, NFS_SERVER(wdata->header->inode)->client->cl_metrics);
 433 }
 434
 435 static void filelayout_write_release(void *data)
 436 {
 437         struct nfs_write_data *wdata = data;
 438         struct pnfs_layout_hdr *lo = wdata->header->lseg->pls_layout;
 439
 440         filelayout_fenceme(lo->plh_inode, lo);
 441         nfs_put_client(wdata->ds_clp);
 442         wdata->header->mds_ops->rpc_release(data);
 443 }
 444
 445 static void filelayout_commit_prepare(struct rpc_task *task, void *data)
 446 {
 447         struct nfs_commit_data *wdata = data;
 448
 449         nfs41_setup_sequence(wdata->ds_clp->cl_session,
 450                         &wdata->args.seq_args,
 451                         &wdata->res.seq_res,
 452                         task);
 453 }
 454
 455 static void filelayout_write_commit_done(struct rpc_task *task, void *data)
 456 {
 457         struct nfs_commit_data *wdata = data;
 458
 459         /* Note this may cause RPC to be resent */
 460         wdata->mds_ops->rpc_call_done(task, data);
 461 }
 462
 463 static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
 464 {
 465         struct nfs_commit_data *cdata = data;
 466
 467         rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
 468 }
 469
 470 static void filelayout_commit_release(void *calldata)
 471 {
 472         struct nfs_commit_data *data = calldata;
 473
 474         data->completion_ops->completion(data);
 475         pnfs_put_lseg(data->lseg);
 476         nfs_put_client(data->ds_clp);
 477         nfs_commitdata_release(data);
 478 }
 479
 480 static const struct rpc_call_ops filelayout_read_call_ops = {
 481         .rpc_call_prepare = filelayout_read_prepare,
 482         .rpc_call_done = filelayout_read_call_done,
 483         .rpc_count_stats = filelayout_read_count_stats,
 484         .rpc_release = filelayout_read_release,
 485 };
 486
 487 static const struct rpc_call_ops filelayout_write_call_ops = {
 488         .rpc_call_prepare = filelayout_write_prepare,
 489         .rpc_call_done = filelayout_write_call_done,
 490         .rpc_count_stats = filelayout_write_count_stats,
 491         .rpc_release = filelayout_write_release,
 492 };
 493
 494 static const struct rpc_call_ops filelayout_commit_call_ops = {
 495         .rpc_call_prepare = filelayout_commit_prepare,
 496         .rpc_call_done = filelayout_write_commit_done,
 497         .rpc_count_stats = filelayout_commit_count_stats,
 498         .rpc_release = filelayout_commit_release,
 499 };
 500
 501 static enum pnfs_try_status
 502 filelayout_read_pagelist(struct nfs_read_data *data)
 503 {
 504         struct nfs_pgio_header *hdr = data->header;
 505         struct pnfs_layout_segment *lseg = hdr->lseg;
 506         struct nfs4_pnfs_ds *ds;
 507         loff_t offset = data->args.offset;
 508         u32 j, idx;
 509         struct nfs_fh *fh;
 510
 511         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
 512                 __func__, hdr->inode->i_ino,
 513                 data->args.pgbase, (size_t)data->args.count, offset);
 514
 515         /* Retrieve the correct rpc_client for the byte range */
 516         j = nfs4_fl_calc_j_index(lseg, offset);
 517         idx = nfs4_fl_calc_ds_index(lseg, j);
 518         ds = nfs4_fl_prepare_ds(lseg, idx);
 519         if (!ds)
 520                 return PNFS_NOT_ATTEMPTED;
 521         dprintk("%s USE DS: %s cl_count %d\n", __func__,
 522                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
 523
 524         /* No multipath support. Use first DS */
 525         atomic_inc(&ds->ds_clp->cl_count);
 526         data->ds_clp = ds->ds_clp;
 527         fh = nfs4_fl_select_ds_fh(lseg, j);
 528         if (fh)
 529                 data->args.fh = fh;
 530
 531         data->args.offset = filelayout_get_dserver_offset(lseg, offset);
 532         data->mds_offset = offset;
 533
 534         /* Perform an asynchronous read to ds */
 535         nfs_initiate_read(ds->ds_clp->cl_rpcclient, data,
 536                                   &filelayout_read_call_ops, RPC_TASK_SOFTCONN);
 537         return PNFS_ATTEMPTED;
 538 }
 539
 540 /* Perform async writes. */
 541 static enum pnfs_try_status
 542 filelayout_write_pagelist(struct nfs_write_data *data, int sync)
 543 {
 544         struct nfs_pgio_header *hdr = data->header;
 545         struct pnfs_layout_segment *lseg = hdr->lseg;
 546         struct nfs4_pnfs_ds *ds;
 547         loff_t offset = data->args.offset;
 548         u32 j, idx;
 549         struct nfs_fh *fh;
 550
 551         /* Retrieve the correct rpc_client for the byte range */
 552         j = nfs4_fl_calc_j_index(lseg, offset);
 553         idx = nfs4_fl_calc_ds_index(lseg, j);
 554         ds = nfs4_fl_prepare_ds(lseg, idx);
 555         if (!ds)
 556                 return PNFS_NOT_ATTEMPTED;
 557         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d\n",
 558                 __func__, hdr->inode->i_ino, sync, (size_t) data->args.count,
 559                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count));
 560
 561         data->write_done_cb = filelayout_write_done_cb;
 562         atomic_inc(&ds->ds_clp->cl_count);
 563         data->ds_clp = ds->ds_clp;
 564         fh = nfs4_fl_select_ds_fh(lseg, j);
 565         if (fh)
 566                 data->args.fh = fh;
 567         /*
 568          * Get the file offset on the dserver. Set the write offset to
 569          * this offset and save the original offset.
 570          */
 571         data->args.offset = filelayout_get_dserver_offset(lseg, offset);
 572
 573         /* Perform an asynchronous write */
 574         nfs_initiate_write(ds->ds_clp->cl_rpcclient, data,
 575                                     &filelayout_write_call_ops, sync,
 576                                     RPC_TASK_SOFTCONN);
 577         return PNFS_ATTEMPTED;
 578 }
 579
 580 /*
 581  * filelayout_check_layout()
 582  *
 583  * Make sure layout segment parameters are sane WRT the device.
 584  * At this point no generic layer initialization of the lseg has occurred,
 585  * and nothing has been added to the layout_hdr cache.
 586  *
 587  */
 588 static int
 589 filelayout_check_layout(struct pnfs_layout_hdr *lo,
 590                         struct nfs4_filelayout_segment *fl,
 591                         struct nfs4_layoutget_res *lgr,
 592                         struct nfs4_deviceid *id,
 593                         gfp_t gfp_flags)
 594 {
 595         struct nfs4_deviceid_node *d;
 596         struct nfs4_file_layout_dsaddr *dsaddr;
 597         int status = -EINVAL;
 598         struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
 599
 600         dprintk("--> %s\n", __func__);
 601
 602         /* FIXME: remove this check when layout segment support is added */
 603         if (lgr->range.offset != 0 ||
 604             lgr->range.length != NFS4_MAX_UINT64) {
 605                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 606                         __func__);
 607                 goto out;
 608         }
 609
 610         if (fl->pattern_offset > lgr->range.offset) {
 611                 dprintk("%s pattern_offset %lld too large\n",
 612                                 __func__, fl->pattern_offset);
 613                 goto out;
 614         }
 615
 616         if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
 617                 dprintk("%s Invalid stripe unit (%u)\n",
 618                         __func__, fl->stripe_unit);
 619                 goto out;
 620         }
 621
 622         /* find and reference the deviceid */
 623         d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
 624                                    NFS_SERVER(lo->plh_inode)->nfs_client, id);
 625         if (d == NULL) {
 626                 dsaddr = filelayout_get_device_info(lo->plh_inode, id, gfp_flags);
 627                 if (dsaddr == NULL)
 628                         goto out;
 629         } else
 630                 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
 631         /* Found deviceid is unavailable */
 632         if (filelayout_test_devid_unavailable(&dsaddr->id_node))
 633                         goto out_put;
 634
 635         fl->dsaddr = dsaddr;
 636
 637         if (fl->first_stripe_index >= dsaddr->stripe_count) {
 638                 dprintk("%s Bad first_stripe_index %u\n",
 639                                 __func__, fl->first_stripe_index);
 640                 goto out_put;
 641         }
 642
 643         if ((fl->stripe_type == STRIPE_SPARSE &&
 644             fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
 645             (fl->stripe_type == STRIPE_DENSE &&
 646             fl->num_fh != dsaddr->stripe_count)) {
 647                 dprintk("%s num_fh %u not valid for given packing\n",
 648                         __func__, fl->num_fh);
 649                 goto out_put;
 650         }
 651
 652         if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
 653                 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
 654                         "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
 655                         nfss->wsize);
 656         }
 657
 658         status = 0;
 659 out:
 660         dprintk("--> %s returns %d\n", __func__, status);
 661         return status;
 662 out_put:
 663         nfs4_fl_put_deviceid(dsaddr);
 664         goto out;
 665 }
 666
 667 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
 668 {
 669         int i;
 670
 671         for (i = 0; i < fl->num_fh; i++) {
 672                 if (!fl->fh_array[i])
 673                         break;
 674                 kfree(fl->fh_array[i]);
 675         }
 676         kfree(fl->fh_array);
 677         fl->fh_array = NULL;
 678 }
 679
 680 static void
 681 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
 682 {
 683         filelayout_free_fh_array(fl);
 684         kfree(fl);
 685 }
 686
 687 static int
 688 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
 689                          struct nfs4_filelayout_segment *fl,
 690                          struct nfs4_layoutget_res *lgr,
 691                          struct nfs4_deviceid *id,
 692                          gfp_t gfp_flags)
 693 {
 694         struct xdr_stream stream;
 695         struct xdr_buf buf;
 696         struct page *scratch;
 697         __be32 *p;
 698         uint32_t nfl_util;
 699         int i;
 700
 701         dprintk("%s: set_layout_map Begin\n", __func__);
 702
 703         scratch = alloc_page(gfp_flags);
 704         if (!scratch)
 705                 return -ENOMEM;
 706
 707         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
 708         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 709
 710         /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
 711          * num_fh (4) */
 712         p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
 713         if (unlikely(!p))
 714                 goto out_err;
 715
 716         memcpy(id, p, sizeof(*id));
 717         p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
 718         nfs4_print_deviceid(id);
 719
 720         nfl_util = be32_to_cpup(p++);
 721         if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
 722                 fl->commit_through_mds = 1;
 723         if (nfl_util & NFL4_UFLG_DENSE)
 724                 fl->stripe_type = STRIPE_DENSE;
 725         else
 726                 fl->stripe_type = STRIPE_SPARSE;
 727         fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
 728
 729         fl->first_stripe_index = be32_to_cpup(p++);
 730         p = xdr_decode_hyper(p, &fl->pattern_offset);
 731         fl->num_fh = be32_to_cpup(p++);
 732
 733         dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
 734                 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
 735                 fl->pattern_offset);
 736
 737         /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
 738          * Futher checking is done in filelayout_check_layout */
 739         if (fl->num_fh >
 740             max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
 741                 goto out_err;
 742
 743         if (fl->num_fh > 0) {
 744                 fl->fh_array = kcalloc(fl->num_fh, sizeof(fl->fh_array[0]),
 745                                        gfp_flags);
 746                 if (!fl->fh_array)
 747                         goto out_err;
 748         }
 749
 750         for (i = 0; i < fl->num_fh; i++) {
 751                 /* Do we want to use a mempool here? */
 752                 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
 753                 if (!fl->fh_array[i])
 754                         goto out_err_free;
 755
 756                 p = xdr_inline_decode(&stream, 4);
 757                 if (unlikely(!p))
 758                         goto out_err_free;
 759                 fl->fh_array[i]->size = be32_to_cpup(p++);
 760                 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
 761                         printk(KERN_ERR "NFS: Too big fh %d received %d\n",
 762                                i, fl->fh_array[i]->size);
 763                         goto out_err_free;
 764                 }
 765
 766                 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
 767                 if (unlikely(!p))
 768                         goto out_err_free;
 769                 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
 770                 dprintk("DEBUG: %s: fh len %d\n", __func__,
 771                         fl->fh_array[i]->size);
 772         }
 773
 774         __free_page(scratch);
 775         return 0;
 776
 777 out_err_free:
 778         filelayout_free_fh_array(fl);
 779 out_err:
 780         __free_page(scratch);
 781         return -EIO;
 782 }
 783
 784 static void
 785 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
 786 {
 787         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
 788
 789         dprintk("--> %s\n", __func__);
 790         nfs4_fl_put_deviceid(fl->dsaddr);
 791         /* This assumes a single RW lseg */
 792         if (lseg->pls_range.iomode == IOMODE_RW) {
 793                 struct nfs4_filelayout *flo;
 794
 795                 flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
 796                 flo->commit_info.nbuckets = 0;
 797                 kfree(flo->commit_info.buckets);
 798                 flo->commit_info.buckets = NULL;
 799         }
 800         _filelayout_free_lseg(fl);
 801 }
 802
 803 static int
 804 filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 805                              struct nfs_commit_info *cinfo,
 806                              gfp_t gfp_flags)
 807 {
 808         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
 809         struct pnfs_commit_bucket *buckets;
 810         int size;
 811
 812         if (fl->commit_through_mds)
 813                 return 0;
 814         if (cinfo->ds->nbuckets != 0) {
 815                 /* This assumes there is only one IOMODE_RW lseg.  What
 816                  * we really want to do is have a layout_hdr level
 817                  * dictionary of <multipath_list4, fh> keys, each
 818                  * associated with a struct list_head, populated by calls
 819                  * to filelayout_write_pagelist().
 820                  * */
 821                 return 0;
 822         }
 823
 824         size = (fl->stripe_type == STRIPE_SPARSE) ?
 825                 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
 826
 827         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 828                           gfp_flags);
 829         if (!buckets)
 830                 return -ENOMEM;
 831         else {
 832                 int i;
 833
 834                 spin_lock(cinfo->lock);
 835                 if (cinfo->ds->nbuckets != 0)
 836                         kfree(buckets);
 837                 else {
 838                         cinfo->ds->buckets = buckets;
 839                         cinfo->ds->nbuckets = size;
 840                         for (i = 0; i < size; i++) {
 841                                 INIT_LIST_HEAD(&buckets[i].written);
 842                                 INIT_LIST_HEAD(&buckets[i].committing);
 843                         }
 844                 }
 845                 spin_unlock(cinfo->lock);
 846                 return 0;
 847         }
 848 }
 849
 850 static struct pnfs_layout_segment *
 851 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
 852                       struct nfs4_layoutget_res *lgr,
 853                       gfp_t gfp_flags)
 854 {
 855         struct nfs4_filelayout_segment *fl;
 856         int rc;
 857         struct nfs4_deviceid id;
 858
 859         dprintk("--> %s\n", __func__);
 860         fl = kzalloc(sizeof(*fl), gfp_flags);
 861         if (!fl)
 862                 return NULL;
 863
 864         rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
 865         if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
 866                 _filelayout_free_lseg(fl);
 867                 return NULL;
 868         }
 869         return &fl->generic_hdr;
 870 }
 871
 872 /*
 873  * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
 874  *
 875  * return true  : coalesce page
 876  * return false : don't coalesce page
 877  */
 878 static bool
 879 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
 880                    struct nfs_page *req)
 881 {
 882         u64 p_stripe, r_stripe;
 883         u32 stripe_unit;
 884
 885         if (!pnfs_generic_pg_test(pgio, prev, req) ||
 886             !nfs_generic_pg_test(pgio, prev, req))
 887                 return false;
 888
 889         p_stripe = (u64)req_offset(prev);
 890         r_stripe = (u64)req_offset(req);
 891         stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
 892
 893         do_div(p_stripe, stripe_unit);
 894         do_div(r_stripe, stripe_unit);
 895
 896         return (p_stripe == r_stripe);
 897 }
 898
 899 static void
 900 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 901                         struct nfs_page *req)
 902 {
 903         WARN_ON_ONCE(pgio->pg_lseg != NULL);
 904
 905         if (req->wb_offset != req->wb_pgbase) {
 906                 /*
 907                  * Handling unaligned pages is difficult, because have to
 908                  * somehow split a req in two in certain cases in the
 909                  * pg.test code.  Avoid this by just not using pnfs
 910                  * in this case.
 911                  */
 912                 nfs_pageio_reset_read_mds(pgio);
 913                 return;
 914         }
 915         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 916                                            req->wb_context,
 917                                            0,
 918                                            NFS4_MAX_UINT64,
 919                                            IOMODE_READ,
 920                                            GFP_KERNEL);
 921         /* If no lseg, fall back to read through mds */
 922         if (pgio->pg_lseg == NULL)
 923                 nfs_pageio_reset_read_mds(pgio);
 924 }
 925
 926 static void
 927 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 928                          struct nfs_page *req)
 929 {
 930         struct nfs_commit_info cinfo;
 931         int status;
 932
 933         WARN_ON_ONCE(pgio->pg_lseg != NULL);
 934
 935         if (req->wb_offset != req->wb_pgbase)
 936                 goto out_mds;
 937         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 938                                            req->wb_context,
 939                                            0,
 940                                            NFS4_MAX_UINT64,
 941                                            IOMODE_RW,
 942                                            GFP_NOFS);
 943         /* If no lseg, fall back to write through mds */
 944         if (pgio->pg_lseg == NULL)
 945                 goto out_mds;
 946         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 947         status = filelayout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 948         if (status < 0) {
 949                 pnfs_put_lseg(pgio->pg_lseg);
 950                 pgio->pg_lseg = NULL;
 951                 goto out_mds;
 952         }
 953         return;
 954 out_mds:
 955         nfs_pageio_reset_write_mds(pgio);
 956 }
 957
 958 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
 959         .pg_init = filelayout_pg_init_read,
 960         .pg_test = filelayout_pg_test,
 961         .pg_doio = pnfs_generic_pg_readpages,
 962 };
 963
 964 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
 965         .pg_init = filelayout_pg_init_write,
 966         .pg_test = filelayout_pg_test,
 967         .pg_doio = pnfs_generic_pg_writepages,
 968 };
 969
 970 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
 971 {
 972         if (fl->stripe_type == STRIPE_SPARSE)
 973                 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
 974         else
 975                 return j;
 976 }
 977
 978 /* The generic layer is about to remove the req from the commit list.
 979  * If this will make the bucket empty, it will need to put the lseg reference.
 980  */
 981 static void
 982 filelayout_clear_request_commit(struct nfs_page *req,
 983                                 struct nfs_commit_info *cinfo)
 984 {
 985         struct pnfs_layout_segment *freeme = NULL;
 986
 987         spin_lock(cinfo->lock);
 988         if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
 989                 goto out;
 990         cinfo->ds->nwritten--;
 991         if (list_is_singular(&req->wb_list)) {
 992                 struct pnfs_commit_bucket *bucket;
 993
 994                 bucket = list_first_entry(&req->wb_list,
 995                                           struct pnfs_commit_bucket,
 996                                           written);
 997                 freeme = bucket->wlseg;
 998                 bucket->wlseg = NULL;
 999         }
1000 out:
1001         nfs_request_remove_commit_list(req, cinfo);
1002         spin_unlock(cinfo->lock);
1003         pnfs_put_lseg(freeme);
1004 }
1005
1006 static struct list_head *
1007 filelayout_choose_commit_list(struct nfs_page *req,
1008                               struct pnfs_layout_segment *lseg,
1009                               struct nfs_commit_info *cinfo)
1010 {
1011         struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
1012         u32 i, j;
1013         struct list_head *list;
1014         struct pnfs_commit_bucket *buckets;
1015
1016         if (fl->commit_through_mds)
1017                 return &cinfo->mds->list;
1018
1019         /* Note that we are calling nfs4_fl_calc_j_index on each page
1020          * that ends up being committed to a data server.  An attractive
1021          * alternative is to add a field to nfs_write_data and nfs_page
1022          * to store the value calculated in filelayout_write_pagelist
1023          * and just use that here.
1024          */
1025         j = nfs4_fl_calc_j_index(lseg, req_offset(req));
1026         i = select_bucket_index(fl, j);
1027         buckets = cinfo->ds->buckets;
1028         list = &buckets[i].written;
1029         if (list_empty(list)) {
1030                 /* Non-empty buckets hold a reference on the lseg.  That ref
1031                  * is normally transferred to the COMMIT call and released
1032                  * there.  It could also be released if the last req is pulled
1033                  * off due to a rewrite, in which case it will be done in
1034                  * filelayout_clear_request_commit
1035                  */
1036                 buckets[i].wlseg = pnfs_get_lseg(lseg);
1037         }
1038         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1039         cinfo->ds->nwritten++;
1040         return list;
1041 }
1042
1043 static void
1044 filelayout_mark_request_commit(struct nfs_page *req,
1045                                struct pnfs_layout_segment *lseg,
1046                                struct nfs_commit_info *cinfo)
1047 {
1048         struct list_head *list;
1049
1050         list = filelayout_choose_commit_list(req, lseg, cinfo);
1051         nfs_request_add_commit_list(req, list, cinfo);
1052 }
1053
1054 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1055 {
1056         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1057
1058         if (flseg->stripe_type == STRIPE_SPARSE)
1059                 return i;
1060         else
1061                 return nfs4_fl_calc_ds_index(lseg, i);
1062 }
1063
1064 static struct nfs_fh *
1065 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1066 {
1067         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
1068
1069         if (flseg->stripe_type == STRIPE_SPARSE) {
1070                 if (flseg->num_fh == 1)
1071                         i = 0;
1072                 else if (flseg->num_fh == 0)
1073                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
1074                         return NULL;
1075         }
1076         return flseg->fh_array[i];
1077 }
1078
1079 static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1080 {
1081         struct pnfs_layout_segment *lseg = data->lseg;
1082         struct nfs4_pnfs_ds *ds;
1083         u32 idx;
1084         struct nfs_fh *fh;
1085
1086         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1087         ds = nfs4_fl_prepare_ds(lseg, idx);
1088         if (!ds) {
1089                 prepare_to_resend_writes(data);
1090                 filelayout_commit_release(data);
1091                 return -EAGAIN;
1092         }
1093         dprintk("%s ino %lu, how %d cl_count %d\n", __func__,
1094                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count));
1095         data->commit_done_cb = filelayout_commit_done_cb;
1096         atomic_inc(&ds->ds_clp->cl_count);
1097         data->ds_clp = ds->ds_clp;
1098         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1099         if (fh)
1100                 data->args.fh = fh;
1101         return nfs_initiate_commit(ds->ds_clp->cl_rpcclient, data,
1102                                    &filelayout_commit_call_ops, how,
1103                                    RPC_TASK_SOFTCONN);
1104 }
1105
1106 static int
1107 transfer_commit_list(struct list_head *src, struct list_head *dst,
1108                      struct nfs_commit_info *cinfo, int max)
1109 {
1110         struct nfs_page *req, *tmp;
1111         int ret = 0;
1112
1113         list_for_each_entry_safe(req, tmp, src, wb_list) {
1114                 if (!nfs_lock_request(req))
1115                         continue;
1116                 kref_get(&req->wb_kref);
1117                 if (cond_resched_lock(cinfo->lock))
1118                         list_safe_reset_next(req, tmp, wb_list);
1119                 nfs_request_remove_commit_list(req, cinfo);
1120                 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1121                 nfs_list_add_request(req, dst);
1122                 ret++;
1123                 if ((ret == max) && !cinfo->dreq)
1124                         break;
1125         }
1126         return ret;
1127 }
1128
1129 static int
1130 filelayout_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
1131                                struct nfs_commit_info *cinfo,
1132                                int max)
1133 {
1134         struct list_head *src = &bucket->written;
1135         struct list_head *dst = &bucket->committing;
1136         int ret;
1137
1138         ret = transfer_commit_list(src, dst, cinfo, max);
1139         if (ret) {
1140                 cinfo->ds->nwritten -= ret;
1141                 cinfo->ds->ncommitting += ret;
1142                 bucket->clseg = bucket->wlseg;
1143                 if (list_empty(src))
1144                         bucket->wlseg = NULL;
1145                 else
1146                         pnfs_get_lseg(bucket->clseg);
1147         }
1148         return ret;
1149 }
1150
1151 /* Move reqs from written to committing lists, returning count of number moved.
1152  * Note called with cinfo->lock held.
1153  */
1154 static int filelayout_scan_commit_lists(struct nfs_commit_info *cinfo,
1155                                         int max)
1156 {
1157         int i, rv = 0, cnt;
1158
1159         for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
1160                 cnt = filelayout_scan_ds_commit_list(&cinfo->ds->buckets[i],
1161                                                      cinfo, max);
1162                 max -= cnt;
1163                 rv += cnt;
1164         }
1165         return rv;
1166 }
1167
1168 /* Pull everything off the committing lists and dump into @dst */
1169 static void filelayout_recover_commit_reqs(struct list_head *dst,
1170                                            struct nfs_commit_info *cinfo)
1171 {
1172         struct pnfs_commit_bucket *b;
1173         int i;
1174
1175         /* NOTE cinfo->lock is NOT held, relying on fact that this is
1176          * only called on single thread per dreq.
1177          * Can't take the lock because need to do pnfs_put_lseg
1178          */
1179         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1180                 if (transfer_commit_list(&b->written, dst, cinfo, 0)) {
1181                         pnfs_put_lseg(b->wlseg);
1182                         b->wlseg = NULL;
1183                 }
1184         }
1185         cinfo->ds->nwritten = 0;
1186 }
1187
1188 static unsigned int
1189 alloc_ds_commits(struct nfs_commit_info *cinfo, struct list_head *list)
1190 {
1191         struct pnfs_ds_commit_info *fl_cinfo;
1192         struct pnfs_commit_bucket *bucket;
1193         struct nfs_commit_data *data;
1194         int i, j;
1195         unsigned int nreq = 0;
1196
1197         fl_cinfo = cinfo->ds;
1198         bucket = fl_cinfo->buckets;
1199         for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
1200                 if (list_empty(&bucket->committing))
1201                         continue;
1202                 data = nfs_commitdata_alloc();
1203                 if (!data)
1204                         break;
1205                 data->ds_commit_index = i;
1206                 data->lseg = bucket->clseg;
1207                 bucket->clseg = NULL;
1208                 list_add(&data->pages, list);
1209                 nreq++;
1210         }
1211
1212         /* Clean up on error */
1213         for (j = i; j < fl_cinfo->nbuckets; j++, bucket++) {
1214                 if (list_empty(&bucket->committing))
1215                         continue;
1216                 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo);
1217                 pnfs_put_lseg(bucket->clseg);
1218                 bucket->clseg = NULL;
1219         }
1220         /* Caller will clean up entries put on list */
1221         return nreq;
1222 }
1223
1224 /* This follows nfs_commit_list pretty closely */
1225 static int
1226 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1227                            int how, struct nfs_commit_info *cinfo)
1228 {
1229         struct nfs_commit_data *data, *tmp;
1230         LIST_HEAD(list);
1231         unsigned int nreq = 0;
1232
1233         if (!list_empty(mds_pages)) {
1234                 data = nfs_commitdata_alloc();
1235                 if (data != NULL) {
1236                         data->lseg = NULL;
1237                         list_add(&data->pages, &list);
1238                         nreq++;
1239                 } else
1240                         nfs_retry_commit(mds_pages, NULL, cinfo);
1241         }
1242
1243         nreq += alloc_ds_commits(cinfo, &list);
1244
1245         if (nreq == 0) {
1246                 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1247                 goto out;
1248         }
1249
1250         atomic_add(nreq, &cinfo->mds->rpcs_out);
1251
1252         list_for_each_entry_safe(data, tmp, &list, pages) {
1253                 list_del_init(&data->pages);
1254                 if (!data->lseg) {
1255                         nfs_init_commit(data, mds_pages, NULL, cinfo);
1256                         nfs_initiate_commit(NFS_CLIENT(inode), data,
1257                                             data->mds_ops, how, 0);
1258                 } else {
1259                         struct pnfs_commit_bucket *buckets;
1260
1261                         buckets = cinfo->ds->buckets;
1262                         nfs_init_commit(data, &buckets[data->ds_commit_index].committing, data->lseg, cinfo);
1263                         filelayout_initiate_commit(data, how);
1264                 }
1265         }
1266 out:
1267         cinfo->ds->ncommitting = 0;
1268         return PNFS_ATTEMPTED;
1269 }
1270
1271 static void
1272 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
1273 {
1274         nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
1275 }
1276
1277 static struct pnfs_layout_hdr *
1278 filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
1279 {
1280         struct nfs4_filelayout *flo;
1281
1282         flo = kzalloc(sizeof(*flo), gfp_flags);
1283         return &flo->generic_hdr;
1284 }
1285
1286 static void
1287 filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
1288 {
1289         kfree(FILELAYOUT_FROM_HDR(lo));
1290 }
1291
1292 static struct pnfs_ds_commit_info *
1293 filelayout_get_ds_info(struct inode *inode)
1294 {
1295         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1296
1297         if (layout == NULL)
1298                 return NULL;
1299         else
1300                 return &FILELAYOUT_FROM_HDR(layout)->commit_info;
1301 }
1302
1303 static struct pnfs_layoutdriver_type filelayout_type = {
1304         .id                     = LAYOUT_NFSV4_1_FILES,
1305         .name                   = "LAYOUT_NFSV4_1_FILES",
1306         .owner                  = THIS_MODULE,
1307         .alloc_layout_hdr       = filelayout_alloc_layout_hdr,
1308         .free_layout_hdr        = filelayout_free_layout_hdr,
1309         .alloc_lseg             = filelayout_alloc_lseg,
1310         .free_lseg              = filelayout_free_lseg,
1311         .pg_read_ops            = &filelayout_pg_read_ops,
1312         .pg_write_ops           = &filelayout_pg_write_ops,
1313         .get_ds_info            = &filelayout_get_ds_info,
1314         .mark_request_commit    = filelayout_mark_request_commit,
1315         .clear_request_commit   = filelayout_clear_request_commit,
1316         .scan_commit_lists      = filelayout_scan_commit_lists,
1317         .recover_commit_reqs    = filelayout_recover_commit_reqs,
1318         .commit_pagelist        = filelayout_commit_pagelist,
1319         .read_pagelist          = filelayout_read_pagelist,
1320         .write_pagelist         = filelayout_write_pagelist,
1321         .free_deviceid_node     = filelayout_free_deveiceid_node,
1322 };
1323
1324 static int __init nfs4filelayout_init(void)
1325 {
1326         printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
1327                __func__);
1328         return pnfs_register_layoutdriver(&filelayout_type);
1329 }
1330
1331 static void __exit nfs4filelayout_exit(void)
1332 {
1333         printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
1334                __func__);
1335         pnfs_unregister_layoutdriver(&filelayout_type);
1336 }
1337
1338 MODULE_ALIAS("nfs-layouttype4-1");
1339
1340 module_init(nfs4filelayout_init);
1341 module_exit(nfs4filelayout_exit);