fs/nfs/flexfilelayout/flexfilelayout.c

   1 /*
   2  * Module for pnfs flexfile layout driver.
   3  *
   4  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   5  *
   6  * Tao Peng <bergwolf@primarydata.com>
   7  */
   8
   9 #include <linux/nfs_fs.h>
  10 #include <linux/nfs_page.h>
  11 #include <linux/module.h>
  12
  13 #include <linux/sunrpc/metrics.h>
  14
  15 #include "flexfilelayout.h"
  16 #include "../nfs4session.h"
  17 #include "../nfs4idmap.h"
  18 #include "../internal.h"
  19 #include "../delegation.h"
  20 #include "../nfs4trace.h"
  21 #include "../iostat.h"
  22 #include "../nfs.h"
  23 #include "../nfs42.h"
  24
  25 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  26
  27 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  28
  29 static struct pnfs_layout_hdr *
  30 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  31 {
  32         struct nfs4_flexfile_layout *ffl;
  33
  34         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  35         if (ffl) {
  36                 INIT_LIST_HEAD(&ffl->error_list);
  37                 return &ffl->generic_hdr;
  38         } else
  39                 return NULL;
  40 }
  41
  42 static void
  43 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  44 {
  45         struct nfs4_ff_layout_ds_err *err, *n;
  46
  47         list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
  48                                  list) {
  49                 list_del(&err->list);
  50                 kfree(err);
  51         }
  52         kfree(FF_LAYOUT_FROM_HDR(lo));
  53 }
  54
  55 static int decode_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  56 {
  57         __be32 *p;
  58
  59         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  60         if (unlikely(p == NULL))
  61                 return -ENOBUFS;
  62         memcpy(stateid, p, NFS4_STATEID_SIZE);
  63         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  64                 p[0], p[1], p[2], p[3]);
  65         return 0;
  66 }
  67
  68 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  69 {
  70         __be32 *p;
  71
  72         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  73         if (unlikely(!p))
  74                 return -ENOBUFS;
  75         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  76         nfs4_print_deviceid(devid);
  77         return 0;
  78 }
  79
  80 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
  81 {
  82         __be32 *p;
  83
  84         p = xdr_inline_decode(xdr, 4);
  85         if (unlikely(!p))
  86                 return -ENOBUFS;
  87         fh->size = be32_to_cpup(p++);
  88         if (fh->size > sizeof(struct nfs_fh)) {
  89                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
  90                        fh->size);
  91                 return -EOVERFLOW;
  92         }
  93         /* fh.data */
  94         p = xdr_inline_decode(xdr, fh->size);
  95         if (unlikely(!p))
  96                 return -ENOBUFS;
  97         memcpy(&fh->data, p, fh->size);
  98         dprintk("%s: fh len %d\n", __func__, fh->size);
  99
 100         return 0;
 101 }
 102
 103 /*
 104  * Currently only stringified uids and gids are accepted.
 105  * I.e., kerberos is not supported to the DSes, so no pricipals.
 106  *
 107  * That means that one common function will suffice, but when
 108  * principals are added, this should be split to accomodate
 109  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 110  */
 111 static int
 112 decode_name(struct xdr_stream *xdr, u32 *id)
 113 {
 114         __be32 *p;
 115         int len;
 116
 117         /* opaque_length(4)*/
 118         p = xdr_inline_decode(xdr, 4);
 119         if (unlikely(!p))
 120                 return -ENOBUFS;
 121         len = be32_to_cpup(p++);
 122         if (len < 0)
 123                 return -EINVAL;
 124
 125         dprintk("%s: len %u\n", __func__, len);
 126
 127         /* opaque body */
 128         p = xdr_inline_decode(xdr, len);
 129         if (unlikely(!p))
 130                 return -ENOBUFS;
 131
 132         if (!nfs_map_string_to_numeric((char *)p, len, id))
 133                 return -EINVAL;
 134
 135         return 0;
 136 }
 137
 138 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 139 {
 140         int i;
 141
 142         if (fls->mirror_array) {
 143                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 144                         /* normally mirror_ds is freed in
 145                          * .free_deviceid_node but we still do it here
 146                          * for .alloc_lseg error path */
 147                         if (fls->mirror_array[i]) {
 148                                 kfree(fls->mirror_array[i]->fh_versions);
 149                                 nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
 150                                 kfree(fls->mirror_array[i]);
 151                         }
 152                 }
 153                 kfree(fls->mirror_array);
 154                 fls->mirror_array = NULL;
 155         }
 156 }
 157
 158 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 159 {
 160         int ret = 0;
 161
 162         dprintk("--> %s\n", __func__);
 163
 164         /* FIXME: remove this check when layout segment support is added */
 165         if (lgr->range.offset != 0 ||
 166             lgr->range.length != NFS4_MAX_UINT64) {
 167                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 168                         __func__);
 169                 ret = -EINVAL;
 170         }
 171
 172         dprintk("--> %s returns %d\n", __func__, ret);
 173         return ret;
 174 }
 175
 176 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 177 {
 178         if (fls) {
 179                 ff_layout_free_mirror_array(fls);
 180                 kfree(fls);
 181         }
 182 }
 183
 184 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 185 {
 186         int i, j;
 187
 188         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 189                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 190                         if (fls->mirror_array[i]->efficiency <
 191                             fls->mirror_array[j]->efficiency)
 192                                 swap(fls->mirror_array[i],
 193                                      fls->mirror_array[j]);
 194         }
 195 }
 196
 197 static struct pnfs_layout_segment *
 198 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 199                      struct nfs4_layoutget_res *lgr,
 200                      gfp_t gfp_flags)
 201 {
 202         struct pnfs_layout_segment *ret;
 203         struct nfs4_ff_layout_segment *fls = NULL;
 204         struct xdr_stream stream;
 205         struct xdr_buf buf;
 206         struct page *scratch;
 207         u64 stripe_unit;
 208         u32 mirror_array_cnt;
 209         __be32 *p;
 210         int i, rc;
 211
 212         dprintk("--> %s\n", __func__);
 213         scratch = alloc_page(gfp_flags);
 214         if (!scratch)
 215                 return ERR_PTR(-ENOMEM);
 216
 217         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 218                               lgr->layoutp->len);
 219         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 220
 221         /* stripe unit and mirror_array_cnt */
 222         rc = -EIO;
 223         p = xdr_inline_decode(&stream, 8 + 4);
 224         if (!p)
 225                 goto out_err_free;
 226
 227         p = xdr_decode_hyper(p, &stripe_unit);
 228         mirror_array_cnt = be32_to_cpup(p++);
 229         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 230                 stripe_unit, mirror_array_cnt);
 231
 232         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 233             mirror_array_cnt == 0)
 234                 goto out_err_free;
 235
 236         rc = -ENOMEM;
 237         fls = kzalloc(sizeof(*fls), gfp_flags);
 238         if (!fls)
 239                 goto out_err_free;
 240
 241         fls->mirror_array_cnt = mirror_array_cnt;
 242         fls->stripe_unit = stripe_unit;
 243         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 244                                     sizeof(fls->mirror_array[0]), gfp_flags);
 245         if (fls->mirror_array == NULL)
 246                 goto out_err_free;
 247
 248         for (i = 0; i < fls->mirror_array_cnt; i++) {
 249                 struct nfs4_deviceid devid;
 250                 struct nfs4_deviceid_node *idnode;
 251                 u32 ds_count;
 252                 u32 fh_count;
 253                 int j;
 254
 255                 rc = -EIO;
 256                 p = xdr_inline_decode(&stream, 4);
 257                 if (!p)
 258                         goto out_err_free;
 259                 ds_count = be32_to_cpup(p);
 260
 261                 /* FIXME: allow for striping? */
 262                 if (ds_count != 1)
 263                         goto out_err_free;
 264
 265                 fls->mirror_array[i] =
 266                         kzalloc(sizeof(struct nfs4_ff_layout_mirror),
 267                                 gfp_flags);
 268                 if (fls->mirror_array[i] == NULL) {
 269                         rc = -ENOMEM;
 270                         goto out_err_free;
 271                 }
 272
 273                 spin_lock_init(&fls->mirror_array[i]->lock);
 274                 fls->mirror_array[i]->ds_count = ds_count;
 275                 fls->mirror_array[i]->lseg = &fls->generic_hdr;
 276
 277                 /* deviceid */
 278                 rc = decode_deviceid(&stream, &devid);
 279                 if (rc)
 280                         goto out_err_free;
 281
 282                 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 283                                                 &devid, lh->plh_lc_cred,
 284                                                 gfp_flags);
 285                 /*
 286                  * upon success, mirror_ds is allocated by previous
 287                  * getdeviceinfo, or newly by .alloc_deviceid_node
 288                  * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
 289                  */
 290                 if (idnode)
 291                         fls->mirror_array[i]->mirror_ds =
 292                                 FF_LAYOUT_MIRROR_DS(idnode);
 293                 else
 294                         goto out_err_free;
 295
 296                 /* efficiency */
 297                 rc = -EIO;
 298                 p = xdr_inline_decode(&stream, 4);
 299                 if (!p)
 300                         goto out_err_free;
 301                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 302
 303                 /* stateid */
 304                 rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
 305                 if (rc)
 306                         goto out_err_free;
 307
 308                 /* fh */
 309                 p = xdr_inline_decode(&stream, 4);
 310                 if (!p)
 311                         goto out_err_free;
 312                 fh_count = be32_to_cpup(p);
 313
 314                 fls->mirror_array[i]->fh_versions =
 315                         kzalloc(fh_count * sizeof(struct nfs_fh),
 316                                 gfp_flags);
 317                 if (fls->mirror_array[i]->fh_versions == NULL) {
 318                         rc = -ENOMEM;
 319                         goto out_err_free;
 320                 }
 321
 322                 for (j = 0; j < fh_count; j++) {
 323                         rc = decode_nfs_fh(&stream,
 324                                            &fls->mirror_array[i]->fh_versions[j]);
 325                         if (rc)
 326                                 goto out_err_free;
 327                 }
 328
 329                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 330
 331                 /* user */
 332                 rc = decode_name(&stream, &fls->mirror_array[i]->uid);
 333                 if (rc)
 334                         goto out_err_free;
 335
 336                 /* group */
 337                 rc = decode_name(&stream, &fls->mirror_array[i]->gid);
 338                 if (rc)
 339                         goto out_err_free;
 340
 341                 dprintk("%s: uid %d gid %d\n", __func__,
 342                         fls->mirror_array[i]->uid,
 343                         fls->mirror_array[i]->gid);
 344         }
 345
 346         p = xdr_inline_decode(&stream, 4);
 347         if (p)
 348                 fls->flags = be32_to_cpup(p);
 349
 350         ff_layout_sort_mirrors(fls);
 351         rc = ff_layout_check_layout(lgr);
 352         if (rc)
 353                 goto out_err_free;
 354
 355         ret = &fls->generic_hdr;
 356         dprintk("<-- %s (success)\n", __func__);
 357 out_free_page:
 358         __free_page(scratch);
 359         return ret;
 360 out_err_free:
 361         _ff_layout_free_lseg(fls);
 362         ret = ERR_PTR(rc);
 363         dprintk("<-- %s (%d)\n", __func__, rc);
 364         goto out_free_page;
 365 }
 366
 367 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 368 {
 369         struct pnfs_layout_segment *lseg;
 370
 371         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 372                 if (lseg->pls_range.iomode == IOMODE_RW)
 373                         return true;
 374
 375         return false;
 376 }
 377
 378 static void
 379 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 380 {
 381         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 382         int i;
 383
 384         dprintk("--> %s\n", __func__);
 385
 386         for (i = 0; i < fls->mirror_array_cnt; i++) {
 387                 if (fls->mirror_array[i]) {
 388                         nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
 389                         fls->mirror_array[i]->mirror_ds = NULL;
 390                         if (fls->mirror_array[i]->cred) {
 391                                 put_rpccred(fls->mirror_array[i]->cred);
 392                                 fls->mirror_array[i]->cred = NULL;
 393                         }
 394                 }
 395         }
 396
 397         if (lseg->pls_range.iomode == IOMODE_RW) {
 398                 struct nfs4_flexfile_layout *ffl;
 399                 struct inode *inode;
 400
 401                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 402                 inode = ffl->generic_hdr.plh_inode;
 403                 spin_lock(&inode->i_lock);
 404                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 405                         ffl->commit_info.nbuckets = 0;
 406                         kfree(ffl->commit_info.buckets);
 407                         ffl->commit_info.buckets = NULL;
 408                 }
 409                 spin_unlock(&inode->i_lock);
 410         }
 411         _ff_layout_free_lseg(fls);
 412 }
 413
 414 /* Return 1 until we have multiple lsegs support */
 415 static int
 416 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 417 {
 418         return 1;
 419 }
 420
 421 static void
 422 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer)
 423 {
 424         /* first IO request? */
 425         if (atomic_inc_return(&timer->n_ops) == 1) {
 426                 timer->start_time = ktime_get();
 427         }
 428 }
 429
 430 static ktime_t
 431 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer)
 432 {
 433         ktime_t start, now;
 434
 435         if (atomic_dec_return(&timer->n_ops) < 0)
 436                 WARN_ON_ONCE(1);
 437
 438         now = ktime_get();
 439         start = timer->start_time;
 440         timer->start_time = now;
 441         return ktime_sub(now, start);
 442 }
 443
 444 static ktime_t
 445 nfs4_ff_layout_calc_completion_time(struct rpc_task *task)
 446 {
 447         return ktime_sub(ktime_get(), task->tk_start);
 448 }
 449
 450 static bool
 451 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 452                             struct nfs4_ff_layoutstat *layoutstat)
 453 {
 454         static const ktime_t notime = {0};
 455         ktime_t now = ktime_get();
 456
 457         nfs4_ff_start_busy_timer(&layoutstat->busy_timer);
 458         if (ktime_equal(mirror->start_time, notime))
 459                 mirror->start_time = now;
 460         if (ktime_equal(mirror->last_report_time, notime))
 461                 mirror->last_report_time = now;
 462         if (ktime_to_ms(ktime_sub(now, mirror->last_report_time)) >=
 463                         FF_LAYOUTSTATS_REPORT_INTERVAL) {
 464                 mirror->last_report_time = now;
 465                 return true;
 466         }
 467
 468         return false;
 469 }
 470
 471 static void
 472 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 473                 __u64 requested)
 474 {
 475         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 476
 477         iostat->ops_requested++;
 478         iostat->bytes_requested += requested;
 479 }
 480
 481 static void
 482 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 483                 __u64 requested,
 484                 __u64 completed,
 485                 ktime_t time_completed)
 486 {
 487         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 488         ktime_t timer;
 489
 490         iostat->ops_completed++;
 491         iostat->bytes_completed += completed;
 492         iostat->bytes_not_delivered += requested - completed;
 493
 494         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer);
 495         iostat->total_busy_time =
 496                         ktime_add(iostat->total_busy_time, timer);
 497         iostat->aggregate_completion_time =
 498                         ktime_add(iostat->aggregate_completion_time, time_completed);
 499 }
 500
 501 static void
 502 nfs4_ff_layout_stat_io_start_read(struct nfs4_ff_layout_mirror *mirror,
 503                 __u64 requested)
 504 {
 505         bool report;
 506
 507         spin_lock(&mirror->lock);
 508         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat);
 509         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 510         spin_unlock(&mirror->lock);
 511
 512         if (report)
 513                 pnfs_report_layoutstat(mirror->lseg->pls_layout->plh_inode);
 514 }
 515
 516 static void
 517 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 518                 struct nfs4_ff_layout_mirror *mirror,
 519                 __u64 requested,
 520                 __u64 completed)
 521 {
 522         spin_lock(&mirror->lock);
 523         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 524                         requested, completed,
 525                         nfs4_ff_layout_calc_completion_time(task));
 526         spin_unlock(&mirror->lock);
 527 }
 528
 529 static void
 530 nfs4_ff_layout_stat_io_start_write(struct nfs4_ff_layout_mirror *mirror,
 531                 __u64 requested)
 532 {
 533         bool report;
 534
 535         spin_lock(&mirror->lock);
 536         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat);
 537         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 538         spin_unlock(&mirror->lock);
 539
 540         if (report)
 541                 pnfs_report_layoutstat(mirror->lseg->pls_layout->plh_inode);
 542 }
 543
 544 static void
 545 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 546                 struct nfs4_ff_layout_mirror *mirror,
 547                 __u64 requested,
 548                 __u64 completed,
 549                 enum nfs3_stable_how committed)
 550 {
 551         if (committed == NFS_UNSTABLE)
 552                 requested = completed = 0;
 553
 554         spin_lock(&mirror->lock);
 555         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 556                         requested, completed,
 557                         nfs4_ff_layout_calc_completion_time(task));
 558         spin_unlock(&mirror->lock);
 559 }
 560
 561 static int
 562 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 563                             struct nfs_commit_info *cinfo,
 564                             gfp_t gfp_flags)
 565 {
 566         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 567         struct pnfs_commit_bucket *buckets;
 568         int size;
 569
 570         if (cinfo->ds->nbuckets != 0) {
 571                 /* This assumes there is only one RW lseg per file.
 572                  * To support multiple lseg per file, we need to
 573                  * change struct pnfs_commit_bucket to allow dynamic
 574                  * increasing nbuckets.
 575                  */
 576                 return 0;
 577         }
 578
 579         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 580
 581         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 582                           gfp_flags);
 583         if (!buckets)
 584                 return -ENOMEM;
 585         else {
 586                 int i;
 587
 588                 spin_lock(cinfo->lock);
 589                 if (cinfo->ds->nbuckets != 0)
 590                         kfree(buckets);
 591                 else {
 592                         cinfo->ds->buckets = buckets;
 593                         cinfo->ds->nbuckets = size;
 594                         for (i = 0; i < size; i++) {
 595                                 INIT_LIST_HEAD(&buckets[i].written);
 596                                 INIT_LIST_HEAD(&buckets[i].committing);
 597                                 /* mark direct verifier as unset */
 598                                 buckets[i].direct_verf.committed =
 599                                         NFS_INVALID_STABLE_HOW;
 600                         }
 601                 }
 602                 spin_unlock(cinfo->lock);
 603                 return 0;
 604         }
 605 }
 606
 607 static struct nfs4_pnfs_ds *
 608 ff_layout_choose_best_ds_for_read(struct nfs_pageio_descriptor *pgio,
 609                                   int *best_idx)
 610 {
 611         struct nfs4_ff_layout_segment *fls;
 612         struct nfs4_pnfs_ds *ds;
 613         int idx;
 614
 615         fls = FF_LAYOUT_LSEG(pgio->pg_lseg);
 616         /* mirrors are sorted by efficiency */
 617         for (idx = 0; idx < fls->mirror_array_cnt; idx++) {
 618                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, idx, false);
 619                 if (ds) {
 620                         *best_idx = idx;
 621                         return ds;
 622                 }
 623         }
 624
 625         return NULL;
 626 }
 627
 628 static void
 629 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 630                         struct nfs_page *req)
 631 {
 632         struct nfs_pgio_mirror *pgm;
 633         struct nfs4_ff_layout_mirror *mirror;
 634         struct nfs4_pnfs_ds *ds;
 635         int ds_idx;
 636
 637         /* Use full layout for now */
 638         if (!pgio->pg_lseg)
 639                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 640                                                    req->wb_context,
 641                                                    0,
 642                                                    NFS4_MAX_UINT64,
 643                                                    IOMODE_READ,
 644                                                    GFP_KERNEL);
 645         /* If no lseg, fall back to read through mds */
 646         if (pgio->pg_lseg == NULL)
 647                 goto out_mds;
 648
 649         ds = ff_layout_choose_best_ds_for_read(pgio, &ds_idx);
 650         if (!ds)
 651                 goto out_mds;
 652         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 653
 654         pgio->pg_mirror_idx = ds_idx;
 655
 656         /* read always uses only one mirror - idx 0 for pgio layer */
 657         pgm = &pgio->pg_mirrors[0];
 658         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 659
 660         return;
 661 out_mds:
 662         pnfs_put_lseg(pgio->pg_lseg);
 663         pgio->pg_lseg = NULL;
 664         nfs_pageio_reset_read_mds(pgio);
 665 }
 666
 667 static void
 668 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 669                         struct nfs_page *req)
 670 {
 671         struct nfs4_ff_layout_mirror *mirror;
 672         struct nfs_pgio_mirror *pgm;
 673         struct nfs_commit_info cinfo;
 674         struct nfs4_pnfs_ds *ds;
 675         int i;
 676         int status;
 677
 678         if (!pgio->pg_lseg)
 679                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 680                                                    req->wb_context,
 681                                                    0,
 682                                                    NFS4_MAX_UINT64,
 683                                                    IOMODE_RW,
 684                                                    GFP_NOFS);
 685         /* If no lseg, fall back to write through mds */
 686         if (pgio->pg_lseg == NULL)
 687                 goto out_mds;
 688
 689         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 690         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 691         if (status < 0)
 692                 goto out_mds;
 693
 694         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 695         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 696             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 697                 goto out_mds;
 698
 699         for (i = 0; i < pgio->pg_mirror_count; i++) {
 700                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 701                 if (!ds)
 702                         goto out_mds;
 703                 pgm = &pgio->pg_mirrors[i];
 704                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 705                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 706         }
 707
 708         return;
 709
 710 out_mds:
 711         pnfs_put_lseg(pgio->pg_lseg);
 712         pgio->pg_lseg = NULL;
 713         nfs_pageio_reset_write_mds(pgio);
 714 }
 715
 716 static unsigned int
 717 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 718                                     struct nfs_page *req)
 719 {
 720         if (!pgio->pg_lseg)
 721                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 722                                                    req->wb_context,
 723                                                    0,
 724                                                    NFS4_MAX_UINT64,
 725                                                    IOMODE_RW,
 726                                                    GFP_NOFS);
 727         if (pgio->pg_lseg)
 728                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 729
 730         /* no lseg means that pnfs is not in use, so no mirroring here */
 731         pnfs_put_lseg(pgio->pg_lseg);
 732         pgio->pg_lseg = NULL;
 733         nfs_pageio_reset_write_mds(pgio);
 734         return 1;
 735 }
 736
 737 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 738         .pg_init = ff_layout_pg_init_read,
 739         .pg_test = pnfs_generic_pg_test,
 740         .pg_doio = pnfs_generic_pg_readpages,
 741         .pg_cleanup = pnfs_generic_pg_cleanup,
 742 };
 743
 744 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 745         .pg_init = ff_layout_pg_init_write,
 746         .pg_test = pnfs_generic_pg_test,
 747         .pg_doio = pnfs_generic_pg_writepages,
 748         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 749         .pg_cleanup = pnfs_generic_pg_cleanup,
 750 };
 751
 752 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 753 {
 754         struct rpc_task *task = &hdr->task;
 755
 756         pnfs_layoutcommit_inode(hdr->inode, false);
 757
 758         if (retry_pnfs) {
 759                 dprintk("%s Reset task %5u for i/o through pNFS "
 760                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 761                         hdr->task.tk_pid,
 762                         hdr->inode->i_sb->s_id,
 763                         (unsigned long long)NFS_FILEID(hdr->inode),
 764                         hdr->args.count,
 765                         (unsigned long long)hdr->args.offset);
 766
 767                 if (!hdr->dreq) {
 768                         struct nfs_open_context *ctx;
 769
 770                         ctx = nfs_list_entry(hdr->pages.next)->wb_context;
 771                         set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
 772                         hdr->completion_ops->error_cleanup(&hdr->pages);
 773                 } else {
 774                         nfs_direct_set_resched_writes(hdr->dreq);
 775                         /* fake unstable write to let common nfs resend pages */
 776                         hdr->verf.committed = NFS_UNSTABLE;
 777                         hdr->good_bytes = hdr->args.count;
 778                 }
 779                 return;
 780         }
 781
 782         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 783                 dprintk("%s Reset task %5u for i/o through MDS "
 784                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 785                         hdr->task.tk_pid,
 786                         hdr->inode->i_sb->s_id,
 787                         (unsigned long long)NFS_FILEID(hdr->inode),
 788                         hdr->args.count,
 789                         (unsigned long long)hdr->args.offset);
 790
 791                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
 792         }
 793 }
 794
 795 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
 796 {
 797         struct rpc_task *task = &hdr->task;
 798
 799         pnfs_layoutcommit_inode(hdr->inode, false);
 800
 801         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 802                 dprintk("%s Reset task %5u for i/o through MDS "
 803                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 804                         hdr->task.tk_pid,
 805                         hdr->inode->i_sb->s_id,
 806                         (unsigned long long)NFS_FILEID(hdr->inode),
 807                         hdr->args.count,
 808                         (unsigned long long)hdr->args.offset);
 809
 810                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
 811         }
 812 }
 813
 814 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
 815                                            struct nfs4_state *state,
 816                                            struct nfs_client *clp,
 817                                            struct pnfs_layout_segment *lseg,
 818                                            int idx)
 819 {
 820         struct pnfs_layout_hdr *lo = lseg->pls_layout;
 821         struct inode *inode = lo->plh_inode;
 822         struct nfs_server *mds_server = NFS_SERVER(inode);
 823
 824         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 825         struct nfs_client *mds_client = mds_server->nfs_client;
 826         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 827
 828         if (task->tk_status >= 0)
 829                 return 0;
 830
 831         switch (task->tk_status) {
 832         /* MDS state errors */
 833         case -NFS4ERR_DELEG_REVOKED:
 834         case -NFS4ERR_ADMIN_REVOKED:
 835         case -NFS4ERR_BAD_STATEID:
 836                 if (state == NULL)
 837                         break;
 838                 nfs_remove_bad_delegation(state->inode);
 839         case -NFS4ERR_OPENMODE:
 840                 if (state == NULL)
 841                         break;
 842                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
 843                         goto out_bad_stateid;
 844                 goto wait_on_recovery;
 845         case -NFS4ERR_EXPIRED:
 846                 if (state != NULL) {
 847                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
 848                                 goto out_bad_stateid;
 849                 }
 850                 nfs4_schedule_lease_recovery(mds_client);
 851                 goto wait_on_recovery;
 852         /* DS session errors */
 853         case -NFS4ERR_BADSESSION:
 854         case -NFS4ERR_BADSLOT:
 855         case -NFS4ERR_BAD_HIGH_SLOT:
 856         case -NFS4ERR_DEADSESSION:
 857         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 858         case -NFS4ERR_SEQ_FALSE_RETRY:
 859         case -NFS4ERR_SEQ_MISORDERED:
 860                 dprintk("%s ERROR %d, Reset session. Exchangeid "
 861                         "flags 0x%x\n", __func__, task->tk_status,
 862                         clp->cl_exchange_flags);
 863                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
 864                 break;
 865         case -NFS4ERR_DELAY:
 866         case -NFS4ERR_GRACE:
 867                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
 868                 break;
 869         case -NFS4ERR_RETRY_UNCACHED_REP:
 870                 break;
 871         /* Invalidate Layout errors */
 872         case -NFS4ERR_PNFS_NO_LAYOUT:
 873         case -ESTALE:           /* mapped NFS4ERR_STALE */
 874         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
 875         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
 876         case -NFS4ERR_FHEXPIRED:
 877         case -NFS4ERR_WRONG_TYPE:
 878                 dprintk("%s Invalid layout error %d\n", __func__,
 879                         task->tk_status);
 880                 /*
 881                  * Destroy layout so new i/o will get a new layout.
 882                  * Layout will not be destroyed until all current lseg
 883                  * references are put. Mark layout as invalid to resend failed
 884                  * i/o and all i/o waiting on the slot table to the MDS until
 885                  * layout is destroyed and a new valid layout is obtained.
 886                  */
 887                 pnfs_destroy_layout(NFS_I(inode));
 888                 rpc_wake_up(&tbl->slot_tbl_waitq);
 889                 goto reset;
 890         /* RPC connection errors */
 891         case -ECONNREFUSED:
 892         case -EHOSTDOWN:
 893         case -EHOSTUNREACH:
 894         case -ENETUNREACH:
 895         case -EIO:
 896         case -ETIMEDOUT:
 897         case -EPIPE:
 898                 dprintk("%s DS connection error %d\n", __func__,
 899                         task->tk_status);
 900                 nfs4_mark_deviceid_unavailable(devid);
 901                 rpc_wake_up(&tbl->slot_tbl_waitq);
 902                 /* fall through */
 903         default:
 904                 if (ff_layout_has_available_ds(lseg))
 905                         return -NFS4ERR_RESET_TO_PNFS;
 906 reset:
 907                 dprintk("%s Retry through MDS. Error %d\n", __func__,
 908                         task->tk_status);
 909                 return -NFS4ERR_RESET_TO_MDS;
 910         }
 911 out:
 912         task->tk_status = 0;
 913         return -EAGAIN;
 914 out_bad_stateid:
 915         task->tk_status = -EIO;
 916         return 0;
 917 wait_on_recovery:
 918         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
 919         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
 920                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
 921         goto out;
 922 }
 923
 924 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
 925 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
 926                                            struct pnfs_layout_segment *lseg,
 927                                            int idx)
 928 {
 929         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 930
 931         if (task->tk_status >= 0)
 932                 return 0;
 933
 934         if (task->tk_status != -EJUKEBOX) {
 935                 dprintk("%s DS connection error %d\n", __func__,
 936                         task->tk_status);
 937                 nfs4_mark_deviceid_unavailable(devid);
 938                 if (ff_layout_has_available_ds(lseg))
 939                         return -NFS4ERR_RESET_TO_PNFS;
 940                 else
 941                         return -NFS4ERR_RESET_TO_MDS;
 942         }
 943
 944         if (task->tk_status == -EJUKEBOX)
 945                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
 946         task->tk_status = 0;
 947         rpc_restart_call(task);
 948         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
 949         return -EAGAIN;
 950 }
 951
 952 static int ff_layout_async_handle_error(struct rpc_task *task,
 953                                         struct nfs4_state *state,
 954                                         struct nfs_client *clp,
 955                                         struct pnfs_layout_segment *lseg,
 956                                         int idx)
 957 {
 958         int vers = clp->cl_nfs_mod->rpc_vers->number;
 959
 960         switch (vers) {
 961         case 3:
 962                 return ff_layout_async_handle_error_v3(task, lseg, idx);
 963         case 4:
 964                 return ff_layout_async_handle_error_v4(task, state, clp,
 965                                                        lseg, idx);
 966         default:
 967                 /* should never happen */
 968                 WARN_ON_ONCE(1);
 969                 return 0;
 970         }
 971 }
 972
 973 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
 974                                         int idx, u64 offset, u64 length,
 975                                         u32 status, int opnum)
 976 {
 977         struct nfs4_ff_layout_mirror *mirror;
 978         int err;
 979
 980         mirror = FF_LAYOUT_COMP(lseg, idx);
 981         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
 982                                        mirror, offset, length, status, opnum,
 983                                        GFP_NOIO);
 984         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
 985 }
 986
 987 /* NFS_PROTO call done callback routines */
 988
 989 static int ff_layout_read_done_cb(struct rpc_task *task,
 990                                 struct nfs_pgio_header *hdr)
 991 {
 992         struct inode *inode;
 993         int err;
 994
 995         trace_nfs4_pnfs_read(hdr, task->tk_status);
 996         if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
 997                 hdr->res.op_status = NFS4ERR_NXIO;
 998         if (task->tk_status < 0 && hdr->res.op_status)
 999                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1000                                             hdr->args.offset, hdr->args.count,
1001                                             hdr->res.op_status, OP_READ);
1002         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1003                                            hdr->ds_clp, hdr->lseg,
1004                                            hdr->pgio_mirror_idx);
1005
1006         switch (err) {
1007         case -NFS4ERR_RESET_TO_PNFS:
1008                 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1009                         &hdr->lseg->pls_layout->plh_flags);
1010                 pnfs_read_resend_pnfs(hdr);
1011                 return task->tk_status;
1012         case -NFS4ERR_RESET_TO_MDS:
1013                 inode = hdr->lseg->pls_layout->plh_inode;
1014                 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1015                 ff_layout_reset_read(hdr);
1016                 return task->tk_status;
1017         case -EAGAIN:
1018                 rpc_restart_call_prepare(task);
1019                 return -EAGAIN;
1020         }
1021
1022         return 0;
1023 }
1024
1025 static bool
1026 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1027 {
1028         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1029 }
1030
1031 /*
1032  * We reference the rpc_cred of the first WRITE that triggers the need for
1033  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1034  * rfc5661 is not clear about which credential should be used.
1035  *
1036  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1037  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1038  * we always send layoutcommit after DS writes.
1039  */
1040 static void
1041 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
1042 {
1043         if (!ff_layout_need_layoutcommit(hdr->lseg))
1044                 return;
1045
1046         pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1047                         hdr->mds_offset + hdr->res.count);
1048         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
1049                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
1050 }
1051
1052 static bool
1053 ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
1054 {
1055         /* No mirroring for now */
1056         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1057
1058         return ff_layout_test_devid_unavailable(node);
1059 }
1060
1061 static int ff_layout_read_prepare_common(struct rpc_task *task,
1062                                          struct nfs_pgio_header *hdr)
1063 {
1064         nfs4_ff_layout_stat_io_start_read(
1065                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1066                         hdr->args.count);
1067
1068         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1069                 rpc_exit(task, -EIO);
1070                 return -EIO;
1071         }
1072         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1073                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
1074                 if (ff_layout_has_available_ds(hdr->lseg))
1075                         pnfs_read_resend_pnfs(hdr);
1076                 else
1077                         ff_layout_reset_read(hdr);
1078                 rpc_exit(task, 0);
1079                 return -EAGAIN;
1080         }
1081         hdr->pgio_done_cb = ff_layout_read_done_cb;
1082
1083         return 0;
1084 }
1085
1086 /*
1087  * Call ops for the async read/write cases
1088  * In the case of dense layouts, the offset needs to be reset to its
1089  * original value.
1090  */
1091 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1092 {
1093         struct nfs_pgio_header *hdr = data;
1094
1095         if (ff_layout_read_prepare_common(task, hdr))
1096                 return;
1097
1098         rpc_call_start(task);
1099 }
1100
1101 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1102                                     struct nfs4_sequence_args *args,
1103                                     struct nfs4_sequence_res *res,
1104                                     struct rpc_task *task)
1105 {
1106         if (ds_clp->cl_session)
1107                 return nfs41_setup_sequence(ds_clp->cl_session,
1108                                            args,
1109                                            res,
1110                                            task);
1111         return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1112                                    args,
1113                                    res,
1114                                    task);
1115 }
1116
1117 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1118 {
1119         struct nfs_pgio_header *hdr = data;
1120
1121         if (ff_layout_setup_sequence(hdr->ds_clp,
1122                                      &hdr->args.seq_args,
1123                                      &hdr->res.seq_res,
1124                                      task))
1125                 return;
1126
1127         if (ff_layout_read_prepare_common(task, hdr))
1128                 return;
1129
1130         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1131                         hdr->args.lock_context, FMODE_READ) == -EIO)
1132                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1133 }
1134
1135 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1136 {
1137         struct nfs_pgio_header *hdr = data;
1138
1139         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1140
1141         nfs4_ff_layout_stat_io_end_read(task,
1142                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1143                         hdr->args.count, hdr->res.count);
1144
1145         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1146             task->tk_status == 0) {
1147                 nfs4_sequence_done(task, &hdr->res.seq_res);
1148                 return;
1149         }
1150
1151         /* Note this may cause RPC to be resent */
1152         hdr->mds_ops->rpc_call_done(task, hdr);
1153 }
1154
1155 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1156 {
1157         struct nfs_pgio_header *hdr = data;
1158
1159         rpc_count_iostats_metrics(task,
1160             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1161 }
1162
1163 static int ff_layout_write_done_cb(struct rpc_task *task,
1164                                 struct nfs_pgio_header *hdr)
1165 {
1166         struct inode *inode;
1167         int err;
1168
1169         trace_nfs4_pnfs_write(hdr, task->tk_status);
1170         if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
1171                 hdr->res.op_status = NFS4ERR_NXIO;
1172         if (task->tk_status < 0 && hdr->res.op_status)
1173                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1174                                             hdr->args.offset, hdr->args.count,
1175                                             hdr->res.op_status, OP_WRITE);
1176         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1177                                            hdr->ds_clp, hdr->lseg,
1178                                            hdr->pgio_mirror_idx);
1179
1180         switch (err) {
1181         case -NFS4ERR_RESET_TO_PNFS:
1182         case -NFS4ERR_RESET_TO_MDS:
1183                 inode = hdr->lseg->pls_layout->plh_inode;
1184                 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1185                 if (err == -NFS4ERR_RESET_TO_PNFS) {
1186                         pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
1187                         ff_layout_reset_write(hdr, true);
1188                 } else {
1189                         pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
1190                         ff_layout_reset_write(hdr, false);
1191                 }
1192                 return task->tk_status;
1193         case -EAGAIN:
1194                 rpc_restart_call_prepare(task);
1195                 return -EAGAIN;
1196         }
1197
1198         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1199             hdr->res.verf->committed == NFS_DATA_SYNC)
1200                 ff_layout_set_layoutcommit(hdr);
1201
1202         return 0;
1203 }
1204
1205 static int ff_layout_commit_done_cb(struct rpc_task *task,
1206                                      struct nfs_commit_data *data)
1207 {
1208         struct inode *inode;
1209         int err;
1210
1211         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1212         if (task->tk_status == -ETIMEDOUT && !data->res.op_status)
1213                 data->res.op_status = NFS4ERR_NXIO;
1214         if (task->tk_status < 0 && data->res.op_status)
1215                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1216                                             data->args.offset, data->args.count,
1217                                             data->res.op_status, OP_COMMIT);
1218         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1219                                            data->lseg, data->ds_commit_index);
1220
1221         switch (err) {
1222         case -NFS4ERR_RESET_TO_PNFS:
1223         case -NFS4ERR_RESET_TO_MDS:
1224                 inode = data->lseg->pls_layout->plh_inode;
1225                 pnfs_error_mark_layout_for_return(inode, data->lseg);
1226                 if (err == -NFS4ERR_RESET_TO_PNFS)
1227                         pnfs_set_retry_layoutget(data->lseg->pls_layout);
1228                 else
1229                         pnfs_clear_retry_layoutget(data->lseg->pls_layout);
1230                 pnfs_generic_prepare_to_resend_writes(data);
1231                 return -EAGAIN;
1232         case -EAGAIN:
1233                 rpc_restart_call_prepare(task);
1234                 return -EAGAIN;
1235         }
1236
1237         if (data->verf.committed == NFS_UNSTABLE
1238             && ff_layout_need_layoutcommit(data->lseg))
1239                 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
1240
1241         return 0;
1242 }
1243
1244 static int ff_layout_write_prepare_common(struct rpc_task *task,
1245                                           struct nfs_pgio_header *hdr)
1246 {
1247         nfs4_ff_layout_stat_io_start_write(
1248                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1249                         hdr->args.count);
1250
1251         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1252                 rpc_exit(task, -EIO);
1253                 return -EIO;
1254         }
1255
1256         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1257                 bool retry_pnfs;
1258
1259                 retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1260                 dprintk("%s task %u reset io to %s\n", __func__,
1261                         task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1262                 ff_layout_reset_write(hdr, retry_pnfs);
1263                 rpc_exit(task, 0);
1264                 return -EAGAIN;
1265         }
1266
1267         return 0;
1268 }
1269
1270 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1271 {
1272         struct nfs_pgio_header *hdr = data;
1273
1274         if (ff_layout_write_prepare_common(task, hdr))
1275                 return;
1276
1277         rpc_call_start(task);
1278 }
1279
1280 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1281 {
1282         struct nfs_pgio_header *hdr = data;
1283
1284         if (ff_layout_setup_sequence(hdr->ds_clp,
1285                                      &hdr->args.seq_args,
1286                                      &hdr->res.seq_res,
1287                                      task))
1288                 return;
1289
1290         if (ff_layout_write_prepare_common(task, hdr))
1291                 return;
1292
1293         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1294                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1295                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1296 }
1297
1298 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1299 {
1300         struct nfs_pgio_header *hdr = data;
1301
1302         nfs4_ff_layout_stat_io_end_write(task,
1303                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1304                         hdr->args.count, hdr->res.count,
1305                         hdr->res.verf->committed);
1306
1307         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1308             task->tk_status == 0) {
1309                 nfs4_sequence_done(task, &hdr->res.seq_res);
1310                 return;
1311         }
1312
1313         /* Note this may cause RPC to be resent */
1314         hdr->mds_ops->rpc_call_done(task, hdr);
1315 }
1316
1317 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1318 {
1319         struct nfs_pgio_header *hdr = data;
1320
1321         rpc_count_iostats_metrics(task,
1322             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1323 }
1324
1325 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1326                 struct nfs_commit_data *cdata)
1327 {
1328         nfs4_ff_layout_stat_io_start_write(
1329                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1330                         0);
1331 }
1332
1333 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1334 {
1335         ff_layout_commit_prepare_common(task, data);
1336         rpc_call_start(task);
1337 }
1338
1339 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1340 {
1341         struct nfs_commit_data *wdata = data;
1342
1343         if (ff_layout_setup_sequence(wdata->ds_clp,
1344                                  &wdata->args.seq_args,
1345                                  &wdata->res.seq_res,
1346                                  task))
1347                 return;
1348         ff_layout_commit_prepare_common(task, data);
1349 }
1350
1351 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1352 {
1353         struct nfs_commit_data *cdata = data;
1354         struct nfs_page *req;
1355         __u64 count = 0;
1356
1357         if (task->tk_status == 0) {
1358                 list_for_each_entry(req, &cdata->pages, wb_list)
1359                         count += req->wb_bytes;
1360         }
1361
1362         nfs4_ff_layout_stat_io_end_write(task,
1363                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1364                         count, count, NFS_FILE_SYNC);
1365
1366         pnfs_generic_write_commit_done(task, data);
1367 }
1368
1369 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1370 {
1371         struct nfs_commit_data *cdata = data;
1372
1373         rpc_count_iostats_metrics(task,
1374             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1375 }
1376
1377 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1378         .rpc_call_prepare = ff_layout_read_prepare_v3,
1379         .rpc_call_done = ff_layout_read_call_done,
1380         .rpc_count_stats = ff_layout_read_count_stats,
1381         .rpc_release = pnfs_generic_rw_release,
1382 };
1383
1384 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1385         .rpc_call_prepare = ff_layout_read_prepare_v4,
1386         .rpc_call_done = ff_layout_read_call_done,
1387         .rpc_count_stats = ff_layout_read_count_stats,
1388         .rpc_release = pnfs_generic_rw_release,
1389 };
1390
1391 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1392         .rpc_call_prepare = ff_layout_write_prepare_v3,
1393         .rpc_call_done = ff_layout_write_call_done,
1394         .rpc_count_stats = ff_layout_write_count_stats,
1395         .rpc_release = pnfs_generic_rw_release,
1396 };
1397
1398 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1399         .rpc_call_prepare = ff_layout_write_prepare_v4,
1400         .rpc_call_done = ff_layout_write_call_done,
1401         .rpc_count_stats = ff_layout_write_count_stats,
1402         .rpc_release = pnfs_generic_rw_release,
1403 };
1404
1405 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1406         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1407         .rpc_call_done = ff_layout_commit_done,
1408         .rpc_count_stats = ff_layout_commit_count_stats,
1409         .rpc_release = pnfs_generic_commit_release,
1410 };
1411
1412 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1413         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1414         .rpc_call_done = ff_layout_commit_done,
1415         .rpc_count_stats = ff_layout_commit_count_stats,
1416         .rpc_release = pnfs_generic_commit_release,
1417 };
1418
1419 static enum pnfs_try_status
1420 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1421 {
1422         struct pnfs_layout_segment *lseg = hdr->lseg;
1423         struct nfs4_pnfs_ds *ds;
1424         struct rpc_clnt *ds_clnt;
1425         struct rpc_cred *ds_cred;
1426         loff_t offset = hdr->args.offset;
1427         u32 idx = hdr->pgio_mirror_idx;
1428         int vers;
1429         struct nfs_fh *fh;
1430
1431         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1432                 __func__, hdr->inode->i_ino,
1433                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1434
1435         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1436         if (!ds)
1437                 goto out_failed;
1438
1439         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1440                                                    hdr->inode);
1441         if (IS_ERR(ds_clnt))
1442                 goto out_failed;
1443
1444         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1445         if (IS_ERR(ds_cred))
1446                 goto out_failed;
1447
1448         vers = nfs4_ff_layout_ds_version(lseg, idx);
1449
1450         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1451                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1452
1453         atomic_inc(&ds->ds_clp->cl_count);
1454         hdr->ds_clp = ds->ds_clp;
1455         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1456         if (fh)
1457                 hdr->args.fh = fh;
1458         /*
1459          * Note that if we ever decide to split across DSes,
1460          * then we may need to handle dense-like offsets.
1461          */
1462         hdr->args.offset = offset;
1463         hdr->mds_offset = offset;
1464
1465         /* Perform an asynchronous read to ds */
1466         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1467                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1468                                       &ff_layout_read_call_ops_v4,
1469                           0, RPC_TASK_SOFTCONN);
1470
1471         return PNFS_ATTEMPTED;
1472
1473 out_failed:
1474         if (ff_layout_has_available_ds(lseg))
1475                 return PNFS_TRY_AGAIN;
1476         return PNFS_NOT_ATTEMPTED;
1477 }
1478
1479 /* Perform async writes. */
1480 static enum pnfs_try_status
1481 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1482 {
1483         struct pnfs_layout_segment *lseg = hdr->lseg;
1484         struct nfs4_pnfs_ds *ds;
1485         struct rpc_clnt *ds_clnt;
1486         struct rpc_cred *ds_cred;
1487         loff_t offset = hdr->args.offset;
1488         int vers;
1489         struct nfs_fh *fh;
1490         int idx = hdr->pgio_mirror_idx;
1491
1492         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1493         if (!ds)
1494                 return PNFS_NOT_ATTEMPTED;
1495
1496         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1497                                                    hdr->inode);
1498         if (IS_ERR(ds_clnt))
1499                 return PNFS_NOT_ATTEMPTED;
1500
1501         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1502         if (IS_ERR(ds_cred))
1503                 return PNFS_NOT_ATTEMPTED;
1504
1505         vers = nfs4_ff_layout_ds_version(lseg, idx);
1506
1507         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1508                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1509                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1510                 vers);
1511
1512         hdr->pgio_done_cb = ff_layout_write_done_cb;
1513         atomic_inc(&ds->ds_clp->cl_count);
1514         hdr->ds_clp = ds->ds_clp;
1515         hdr->ds_commit_idx = idx;
1516         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1517         if (fh)
1518                 hdr->args.fh = fh;
1519
1520         /*
1521          * Note that if we ever decide to split across DSes,
1522          * then we may need to handle dense-like offsets.
1523          */
1524         hdr->args.offset = offset;
1525
1526         /* Perform an asynchronous write */
1527         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1528                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1529                                       &ff_layout_write_call_ops_v4,
1530                           sync, RPC_TASK_SOFTCONN);
1531         return PNFS_ATTEMPTED;
1532 }
1533
1534 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1535 {
1536         return i;
1537 }
1538
1539 static struct nfs_fh *
1540 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1541 {
1542         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1543
1544         /* FIXME: Assume that there is only one NFS version available
1545          * for the DS.
1546          */
1547         return &flseg->mirror_array[i]->fh_versions[0];
1548 }
1549
1550 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1551 {
1552         struct pnfs_layout_segment *lseg = data->lseg;
1553         struct nfs4_pnfs_ds *ds;
1554         struct rpc_clnt *ds_clnt;
1555         struct rpc_cred *ds_cred;
1556         u32 idx;
1557         int vers;
1558         struct nfs_fh *fh;
1559
1560         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1561         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1562         if (!ds)
1563                 goto out_err;
1564
1565         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1566                                                    data->inode);
1567         if (IS_ERR(ds_clnt))
1568                 goto out_err;
1569
1570         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1571         if (IS_ERR(ds_cred))
1572                 goto out_err;
1573
1574         vers = nfs4_ff_layout_ds_version(lseg, idx);
1575
1576         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1577                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1578                 vers);
1579         data->commit_done_cb = ff_layout_commit_done_cb;
1580         data->cred = ds_cred;
1581         atomic_inc(&ds->ds_clp->cl_count);
1582         data->ds_clp = ds->ds_clp;
1583         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1584         if (fh)
1585                 data->args.fh = fh;
1586
1587         return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1588                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1589                                                &ff_layout_commit_call_ops_v4,
1590                                    how, RPC_TASK_SOFTCONN);
1591 out_err:
1592         pnfs_generic_prepare_to_resend_writes(data);
1593         pnfs_generic_commit_release(data);
1594         return -EAGAIN;
1595 }
1596
1597 static int
1598 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1599                            int how, struct nfs_commit_info *cinfo)
1600 {
1601         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1602                                             ff_layout_initiate_commit);
1603 }
1604
1605 static struct pnfs_ds_commit_info *
1606 ff_layout_get_ds_info(struct inode *inode)
1607 {
1608         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1609
1610         if (layout == NULL)
1611                 return NULL;
1612
1613         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1614 }
1615
1616 static void
1617 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1618 {
1619         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1620                                                   id_node));
1621 }
1622
1623 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1624                                   struct xdr_stream *xdr,
1625                                   const struct nfs4_layoutreturn_args *args)
1626 {
1627         struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1628         __be32 *start;
1629         int count = 0, ret = 0;
1630
1631         start = xdr_reserve_space(xdr, 4);
1632         if (unlikely(!start))
1633                 return -E2BIG;
1634
1635         /* This assume we always return _ALL_ layouts */
1636         spin_lock(&hdr->plh_inode->i_lock);
1637         ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1638         spin_unlock(&hdr->plh_inode->i_lock);
1639
1640         *start = cpu_to_be32(count);
1641
1642         return ret;
1643 }
1644
1645 /* report nothing for now */
1646 static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1647                                      struct xdr_stream *xdr,
1648                                      const struct nfs4_layoutreturn_args *args)
1649 {
1650         __be32 *p;
1651
1652         p = xdr_reserve_space(xdr, 4);
1653         if (likely(p))
1654                 *p = cpu_to_be32(0);
1655 }
1656
1657 static struct nfs4_deviceid_node *
1658 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1659                               struct pnfs_device *pdev, gfp_t gfp_flags)
1660 {
1661         struct nfs4_ff_layout_ds *dsaddr;
1662
1663         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1664         if (!dsaddr)
1665                 return NULL;
1666         return &dsaddr->id_node;
1667 }
1668
1669 static void
1670 ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1671                               struct xdr_stream *xdr,
1672                               const struct nfs4_layoutreturn_args *args)
1673 {
1674         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1675         __be32 *start;
1676
1677         dprintk("%s: Begin\n", __func__);
1678         start = xdr_reserve_space(xdr, 4);
1679         BUG_ON(!start);
1680
1681         if (ff_layout_encode_ioerr(flo, xdr, args))
1682                 goto out;
1683
1684         ff_layout_encode_iostats(flo, xdr, args);
1685 out:
1686         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1687         dprintk("%s: Return\n", __func__);
1688 }
1689
1690 static int
1691 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
1692 {
1693         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
1694
1695         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
1696 }
1697
1698 static size_t
1699 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
1700                           const int buflen)
1701 {
1702         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
1703         const struct in6_addr *addr = &sin6->sin6_addr;
1704
1705         /*
1706          * RFC 4291, Section 2.2.2
1707          *
1708          * Shorthanded ANY address
1709          */
1710         if (ipv6_addr_any(addr))
1711                 return snprintf(buf, buflen, "::");
1712
1713         /*
1714          * RFC 4291, Section 2.2.2
1715          *
1716          * Shorthanded loopback address
1717          */
1718         if (ipv6_addr_loopback(addr))
1719                 return snprintf(buf, buflen, "::1");
1720
1721         /*
1722          * RFC 4291, Section 2.2.3
1723          *
1724          * Special presentation address format for mapped v4
1725          * addresses.
1726          */
1727         if (ipv6_addr_v4mapped(addr))
1728                 return snprintf(buf, buflen, "::ffff:%pI4",
1729                                         &addr->s6_addr32[3]);
1730
1731         /*
1732          * RFC 4291, Section 2.2.1
1733          */
1734         return snprintf(buf, buflen, "%pI6c", addr);
1735 }
1736
1737 /* Derived from rpc_sockaddr2uaddr */
1738 static void
1739 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
1740 {
1741         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
1742         char portbuf[RPCBIND_MAXUADDRPLEN];
1743         char addrbuf[RPCBIND_MAXUADDRLEN];
1744         char *netid;
1745         unsigned short port;
1746         int len, netid_len;
1747         __be32 *p;
1748
1749         switch (sap->sa_family) {
1750         case AF_INET:
1751                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
1752                         return;
1753                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
1754                 netid = "tcp";
1755                 netid_len = 3;
1756                 break;
1757         case AF_INET6:
1758                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
1759                         return;
1760                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
1761                 netid = "tcp6";
1762                 netid_len = 4;
1763                 break;
1764         default:
1765                 /* we only support tcp and tcp6 */
1766                 WARN_ON_ONCE(1);
1767                 return;
1768         }
1769
1770         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
1771         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
1772
1773         p = xdr_reserve_space(xdr, 4 + netid_len);
1774         xdr_encode_opaque(p, netid, netid_len);
1775
1776         p = xdr_reserve_space(xdr, 4 + len);
1777         xdr_encode_opaque(p, addrbuf, len);
1778 }
1779
1780 static void
1781 ff_layout_encode_nfstime(struct xdr_stream *xdr,
1782                          ktime_t t)
1783 {
1784         struct timespec64 ts;
1785         __be32 *p;
1786
1787         p = xdr_reserve_space(xdr, 12);
1788         ts = ktime_to_timespec64(t);
1789         p = xdr_encode_hyper(p, ts.tv_sec);
1790         *p++ = cpu_to_be32(ts.tv_nsec);
1791 }
1792
1793 static void
1794 ff_layout_encode_io_latency(struct xdr_stream *xdr,
1795                             struct nfs4_ff_io_stat *stat)
1796 {
1797         __be32 *p;
1798
1799         p = xdr_reserve_space(xdr, 5 * 8);
1800         p = xdr_encode_hyper(p, stat->ops_requested);
1801         p = xdr_encode_hyper(p, stat->bytes_requested);
1802         p = xdr_encode_hyper(p, stat->ops_completed);
1803         p = xdr_encode_hyper(p, stat->bytes_completed);
1804         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
1805         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
1806         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
1807 }
1808
1809 static void
1810 ff_layout_encode_layoutstats(struct xdr_stream *xdr,
1811                              struct nfs42_layoutstat_args *args,
1812                              struct nfs42_layoutstat_devinfo *devinfo)
1813 {
1814         struct nfs4_ff_layout_mirror *mirror = devinfo->layout_private;
1815         struct nfs4_pnfs_ds_addr *da;
1816         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
1817         struct nfs_fh *fh = &mirror->fh_versions[0];
1818         __be32 *p, *start;
1819
1820         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
1821         dprintk("%s: DS %s: encoding address %s\n",
1822                 __func__, ds->ds_remotestr, da->da_remotestr);
1823         /* layoutupdate length */
1824         start = xdr_reserve_space(xdr, 4);
1825         /* netaddr4 */
1826         ff_layout_encode_netaddr(xdr, da);
1827         /* nfs_fh4 */
1828         p = xdr_reserve_space(xdr, 4 + fh->size);
1829         xdr_encode_opaque(p, fh->data, fh->size);
1830         /* ff_io_latency4 read */
1831         spin_lock(&mirror->lock);
1832         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
1833         /* ff_io_latency4 write */
1834         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
1835         spin_unlock(&mirror->lock);
1836         /* nfstime4 */
1837         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
1838         /* bool */
1839         p = xdr_reserve_space(xdr, 4);
1840         *p = cpu_to_be32(false);
1841
1842         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1843 }
1844
1845 static bool
1846 ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
1847                                struct pnfs_layout_segment *pls,
1848                                int *dev_count, int dev_limit)
1849 {
1850         struct nfs4_ff_layout_mirror *mirror;
1851         struct nfs4_deviceid_node *dev;
1852         struct nfs42_layoutstat_devinfo *devinfo;
1853         int i;
1854
1855         for (i = 0; i < FF_LAYOUT_MIRROR_COUNT(pls); i++) {
1856                 if (*dev_count >= dev_limit)
1857                         break;
1858                 mirror = FF_LAYOUT_COMP(pls, i);
1859                 if (!mirror || !mirror->mirror_ds)
1860                         continue;
1861                 dev = FF_LAYOUT_DEVID_NODE(pls, i);
1862                 devinfo = &args->devinfo[*dev_count];
1863                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
1864                 devinfo->offset = pls->pls_range.offset;
1865                 devinfo->length = pls->pls_range.length;
1866                 /* well, we don't really know if IO is continuous or not! */
1867                 devinfo->read_count = mirror->read_stat.io_stat.bytes_completed;
1868                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
1869                 devinfo->write_count = mirror->write_stat.io_stat.bytes_completed;
1870                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
1871                 devinfo->layout_type = LAYOUT_FLEX_FILES;
1872                 devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
1873                 devinfo->layout_private = mirror;
1874                 /* lseg refcount put in cleanup_layoutstats */
1875                 pnfs_get_lseg(pls);
1876
1877                 ++(*dev_count);
1878         }
1879
1880         return *dev_count < dev_limit;
1881 }
1882
1883 static int
1884 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
1885 {
1886         struct pnfs_layout_segment *pls;
1887         int dev_count = 0;
1888
1889         spin_lock(&args->inode->i_lock);
1890         list_for_each_entry(pls, &NFS_I(args->inode)->layout->plh_segs, pls_list) {
1891                 dev_count += FF_LAYOUT_MIRROR_COUNT(pls);
1892         }
1893         spin_unlock(&args->inode->i_lock);
1894         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
1895         if (dev_count > PNFS_LAYOUTSTATS_MAXDEV) {
1896                 dprintk("%s: truncating devinfo to limit (%d:%d)\n",
1897                         __func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
1898                 dev_count = PNFS_LAYOUTSTATS_MAXDEV;
1899         }
1900         args->devinfo = kmalloc(dev_count * sizeof(*args->devinfo), GFP_KERNEL);
1901         if (!args->devinfo)
1902                 return -ENOMEM;
1903
1904         dev_count = 0;
1905         spin_lock(&args->inode->i_lock);
1906         list_for_each_entry(pls, &NFS_I(args->inode)->layout->plh_segs, pls_list) {
1907                 if (!ff_layout_mirror_prepare_stats(args, pls, &dev_count,
1908                                                     PNFS_LAYOUTSTATS_MAXDEV)) {
1909                         break;
1910                 }
1911         }
1912         spin_unlock(&args->inode->i_lock);
1913         args->num_dev = dev_count;
1914
1915         return 0;
1916 }
1917
1918 static void
1919 ff_layout_cleanup_layoutstats(struct nfs42_layoutstat_data *data)
1920 {
1921         struct nfs4_ff_layout_mirror *mirror;
1922         int i;
1923
1924         for (i = 0; i < data->args.num_dev; i++) {
1925                 mirror = data->args.devinfo[i].layout_private;
1926                 data->args.devinfo[i].layout_private = NULL;
1927                 pnfs_put_lseg(mirror->lseg);
1928         }
1929 }
1930
1931 static struct pnfs_layoutdriver_type flexfilelayout_type = {
1932         .id                     = LAYOUT_FLEX_FILES,
1933         .name                   = "LAYOUT_FLEX_FILES",
1934         .owner                  = THIS_MODULE,
1935         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
1936         .free_layout_hdr        = ff_layout_free_layout_hdr,
1937         .alloc_lseg             = ff_layout_alloc_lseg,
1938         .free_lseg              = ff_layout_free_lseg,
1939         .pg_read_ops            = &ff_layout_pg_read_ops,
1940         .pg_write_ops           = &ff_layout_pg_write_ops,
1941         .get_ds_info            = ff_layout_get_ds_info,
1942         .free_deviceid_node     = ff_layout_free_deviceid_node,
1943         .mark_request_commit    = pnfs_layout_mark_request_commit,
1944         .clear_request_commit   = pnfs_generic_clear_request_commit,
1945         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
1946         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
1947         .commit_pagelist        = ff_layout_commit_pagelist,
1948         .read_pagelist          = ff_layout_read_pagelist,
1949         .write_pagelist         = ff_layout_write_pagelist,
1950         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
1951         .encode_layoutreturn    = ff_layout_encode_layoutreturn,
1952         .sync                   = pnfs_nfs_generic_sync,
1953         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
1954         .cleanup_layoutstats    = ff_layout_cleanup_layoutstats,
1955 };
1956
1957 static int __init nfs4flexfilelayout_init(void)
1958 {
1959         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
1960                __func__);
1961         return pnfs_register_layoutdriver(&flexfilelayout_type);
1962 }
1963
1964 static void __exit nfs4flexfilelayout_exit(void)
1965 {
1966         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
1967                __func__);
1968         pnfs_unregister_layoutdriver(&flexfilelayout_type);
1969 }
1970
1971 MODULE_ALIAS("nfs-layouttype4-4");
1972
1973 MODULE_LICENSE("GPL");
1974 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
1975
1976 module_init(nfs4flexfilelayout_init);
1977 module_exit(nfs4flexfilelayout_exit);