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 struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 139 {
 140         struct nfs4_ff_layout_mirror *mirror;
 141
 142         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 143         if (mirror != NULL) {
 144                 spin_lock_init(&mirror->lock);
 145                 atomic_set(&mirror->ref, 1);
 146         }
 147         return mirror;
 148 }
 149
 150 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 151 {
 152         kfree(mirror->fh_versions);
 153         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 154         kfree(mirror);
 155 }
 156
 157 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 158 {
 159         if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
 160                 ff_layout_free_mirror(mirror);
 161 }
 162
 163 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 164 {
 165         int i;
 166
 167         if (fls->mirror_array) {
 168                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 169                         /* normally mirror_ds is freed in
 170                          * .free_deviceid_node but we still do it here
 171                          * for .alloc_lseg error path */
 172                         ff_layout_put_mirror(fls->mirror_array[i]);
 173                 }
 174                 kfree(fls->mirror_array);
 175                 fls->mirror_array = NULL;
 176         }
 177 }
 178
 179 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 180 {
 181         int ret = 0;
 182
 183         dprintk("--> %s\n", __func__);
 184
 185         /* FIXME: remove this check when layout segment support is added */
 186         if (lgr->range.offset != 0 ||
 187             lgr->range.length != NFS4_MAX_UINT64) {
 188                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 189                         __func__);
 190                 ret = -EINVAL;
 191         }
 192
 193         dprintk("--> %s returns %d\n", __func__, ret);
 194         return ret;
 195 }
 196
 197 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 198 {
 199         if (fls) {
 200                 ff_layout_free_mirror_array(fls);
 201                 kfree(fls);
 202         }
 203 }
 204
 205 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 206 {
 207         int i, j;
 208
 209         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 210                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 211                         if (fls->mirror_array[i]->efficiency <
 212                             fls->mirror_array[j]->efficiency)
 213                                 swap(fls->mirror_array[i],
 214                                      fls->mirror_array[j]);
 215         }
 216 }
 217
 218 static struct pnfs_layout_segment *
 219 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 220                      struct nfs4_layoutget_res *lgr,
 221                      gfp_t gfp_flags)
 222 {
 223         struct pnfs_layout_segment *ret;
 224         struct nfs4_ff_layout_segment *fls = NULL;
 225         struct xdr_stream stream;
 226         struct xdr_buf buf;
 227         struct page *scratch;
 228         u64 stripe_unit;
 229         u32 mirror_array_cnt;
 230         __be32 *p;
 231         int i, rc;
 232
 233         dprintk("--> %s\n", __func__);
 234         scratch = alloc_page(gfp_flags);
 235         if (!scratch)
 236                 return ERR_PTR(-ENOMEM);
 237
 238         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 239                               lgr->layoutp->len);
 240         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 241
 242         /* stripe unit and mirror_array_cnt */
 243         rc = -EIO;
 244         p = xdr_inline_decode(&stream, 8 + 4);
 245         if (!p)
 246                 goto out_err_free;
 247
 248         p = xdr_decode_hyper(p, &stripe_unit);
 249         mirror_array_cnt = be32_to_cpup(p++);
 250         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 251                 stripe_unit, mirror_array_cnt);
 252
 253         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 254             mirror_array_cnt == 0)
 255                 goto out_err_free;
 256
 257         rc = -ENOMEM;
 258         fls = kzalloc(sizeof(*fls), gfp_flags);
 259         if (!fls)
 260                 goto out_err_free;
 261
 262         fls->mirror_array_cnt = mirror_array_cnt;
 263         fls->stripe_unit = stripe_unit;
 264         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 265                                     sizeof(fls->mirror_array[0]), gfp_flags);
 266         if (fls->mirror_array == NULL)
 267                 goto out_err_free;
 268
 269         for (i = 0; i < fls->mirror_array_cnt; i++) {
 270                 struct nfs4_deviceid devid;
 271                 struct nfs4_deviceid_node *idnode;
 272                 u32 ds_count;
 273                 u32 fh_count;
 274                 int j;
 275
 276                 rc = -EIO;
 277                 p = xdr_inline_decode(&stream, 4);
 278                 if (!p)
 279                         goto out_err_free;
 280                 ds_count = be32_to_cpup(p);
 281
 282                 /* FIXME: allow for striping? */
 283                 if (ds_count != 1)
 284                         goto out_err_free;
 285
 286                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 287                 if (fls->mirror_array[i] == NULL) {
 288                         rc = -ENOMEM;
 289                         goto out_err_free;
 290                 }
 291
 292                 fls->mirror_array[i]->ds_count = ds_count;
 293
 294                 /* deviceid */
 295                 rc = decode_deviceid(&stream, &devid);
 296                 if (rc)
 297                         goto out_err_free;
 298
 299                 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 300                                                 &devid, lh->plh_lc_cred,
 301                                                 gfp_flags);
 302                 /*
 303                  * upon success, mirror_ds is allocated by previous
 304                  * getdeviceinfo, or newly by .alloc_deviceid_node
 305                  * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
 306                  */
 307                 if (idnode)
 308                         fls->mirror_array[i]->mirror_ds =
 309                                 FF_LAYOUT_MIRROR_DS(idnode);
 310                 else
 311                         goto out_err_free;
 312
 313                 /* efficiency */
 314                 rc = -EIO;
 315                 p = xdr_inline_decode(&stream, 4);
 316                 if (!p)
 317                         goto out_err_free;
 318                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 319
 320                 /* stateid */
 321                 rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
 322                 if (rc)
 323                         goto out_err_free;
 324
 325                 /* fh */
 326                 p = xdr_inline_decode(&stream, 4);
 327                 if (!p)
 328                         goto out_err_free;
 329                 fh_count = be32_to_cpup(p);
 330
 331                 fls->mirror_array[i]->fh_versions =
 332                         kzalloc(fh_count * sizeof(struct nfs_fh),
 333                                 gfp_flags);
 334                 if (fls->mirror_array[i]->fh_versions == NULL) {
 335                         rc = -ENOMEM;
 336                         goto out_err_free;
 337                 }
 338
 339                 for (j = 0; j < fh_count; j++) {
 340                         rc = decode_nfs_fh(&stream,
 341                                            &fls->mirror_array[i]->fh_versions[j]);
 342                         if (rc)
 343                                 goto out_err_free;
 344                 }
 345
 346                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 347
 348                 /* user */
 349                 rc = decode_name(&stream, &fls->mirror_array[i]->uid);
 350                 if (rc)
 351                         goto out_err_free;
 352
 353                 /* group */
 354                 rc = decode_name(&stream, &fls->mirror_array[i]->gid);
 355                 if (rc)
 356                         goto out_err_free;
 357
 358                 dprintk("%s: uid %d gid %d\n", __func__,
 359                         fls->mirror_array[i]->uid,
 360                         fls->mirror_array[i]->gid);
 361         }
 362
 363         p = xdr_inline_decode(&stream, 4);
 364         if (p)
 365                 fls->flags = be32_to_cpup(p);
 366
 367         ff_layout_sort_mirrors(fls);
 368         rc = ff_layout_check_layout(lgr);
 369         if (rc)
 370                 goto out_err_free;
 371
 372         ret = &fls->generic_hdr;
 373         dprintk("<-- %s (success)\n", __func__);
 374 out_free_page:
 375         __free_page(scratch);
 376         return ret;
 377 out_err_free:
 378         _ff_layout_free_lseg(fls);
 379         ret = ERR_PTR(rc);
 380         dprintk("<-- %s (%d)\n", __func__, rc);
 381         goto out_free_page;
 382 }
 383
 384 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 385 {
 386         struct pnfs_layout_segment *lseg;
 387
 388         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 389                 if (lseg->pls_range.iomode == IOMODE_RW)
 390                         return true;
 391
 392         return false;
 393 }
 394
 395 static void
 396 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 397 {
 398         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 399         int i;
 400
 401         dprintk("--> %s\n", __func__);
 402
 403         for (i = 0; i < fls->mirror_array_cnt; i++) {
 404                 if (fls->mirror_array[i]) {
 405                         nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
 406                         fls->mirror_array[i]->mirror_ds = NULL;
 407                         if (fls->mirror_array[i]->cred) {
 408                                 put_rpccred(fls->mirror_array[i]->cred);
 409                                 fls->mirror_array[i]->cred = NULL;
 410                         }
 411                 }
 412         }
 413
 414         if (lseg->pls_range.iomode == IOMODE_RW) {
 415                 struct nfs4_flexfile_layout *ffl;
 416                 struct inode *inode;
 417
 418                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 419                 inode = ffl->generic_hdr.plh_inode;
 420                 spin_lock(&inode->i_lock);
 421                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 422                         ffl->commit_info.nbuckets = 0;
 423                         kfree(ffl->commit_info.buckets);
 424                         ffl->commit_info.buckets = NULL;
 425                 }
 426                 spin_unlock(&inode->i_lock);
 427         }
 428         _ff_layout_free_lseg(fls);
 429 }
 430
 431 /* Return 1 until we have multiple lsegs support */
 432 static int
 433 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 434 {
 435         return 1;
 436 }
 437
 438 static void
 439 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 440 {
 441         /* first IO request? */
 442         if (atomic_inc_return(&timer->n_ops) == 1) {
 443                 timer->start_time = now;
 444         }
 445 }
 446
 447 static ktime_t
 448 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 449 {
 450         ktime_t start;
 451
 452         if (atomic_dec_return(&timer->n_ops) < 0)
 453                 WARN_ON_ONCE(1);
 454
 455         start = timer->start_time;
 456         timer->start_time = now;
 457         return ktime_sub(now, start);
 458 }
 459
 460 static bool
 461 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 462                             struct nfs4_ff_layoutstat *layoutstat,
 463                             ktime_t now)
 464 {
 465         static const ktime_t notime = {0};
 466
 467         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 468         if (ktime_equal(mirror->start_time, notime))
 469                 mirror->start_time = now;
 470         if (ktime_equal(mirror->last_report_time, notime))
 471                 mirror->last_report_time = now;
 472         if (ktime_to_ms(ktime_sub(now, mirror->last_report_time)) >=
 473                         FF_LAYOUTSTATS_REPORT_INTERVAL) {
 474                 mirror->last_report_time = now;
 475                 return true;
 476         }
 477
 478         return false;
 479 }
 480
 481 static void
 482 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 483                 __u64 requested)
 484 {
 485         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 486
 487         iostat->ops_requested++;
 488         iostat->bytes_requested += requested;
 489 }
 490
 491 static void
 492 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 493                 __u64 requested,
 494                 __u64 completed,
 495                 ktime_t time_completed,
 496                 ktime_t time_started)
 497 {
 498         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 499         ktime_t completion_time = ktime_sub(time_completed, time_started);
 500         ktime_t timer;
 501
 502         iostat->ops_completed++;
 503         iostat->bytes_completed += completed;
 504         iostat->bytes_not_delivered += requested - completed;
 505
 506         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 507         iostat->total_busy_time =
 508                         ktime_add(iostat->total_busy_time, timer);
 509         iostat->aggregate_completion_time =
 510                         ktime_add(iostat->aggregate_completion_time,
 511                                         completion_time);
 512 }
 513
 514 static void
 515 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 516                 struct nfs4_ff_layout_mirror *mirror,
 517                 __u64 requested, ktime_t now)
 518 {
 519         bool report;
 520
 521         spin_lock(&mirror->lock);
 522         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 523         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 524         spin_unlock(&mirror->lock);
 525
 526         if (report)
 527                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 528 }
 529
 530 static void
 531 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 532                 struct nfs4_ff_layout_mirror *mirror,
 533                 __u64 requested,
 534                 __u64 completed)
 535 {
 536         spin_lock(&mirror->lock);
 537         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 538                         requested, completed,
 539                         ktime_get(), task->tk_start);
 540         spin_unlock(&mirror->lock);
 541 }
 542
 543 static void
 544 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 545                 struct nfs4_ff_layout_mirror *mirror,
 546                 __u64 requested, ktime_t now)
 547 {
 548         bool report;
 549
 550         spin_lock(&mirror->lock);
 551         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 552         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 553         spin_unlock(&mirror->lock);
 554
 555         if (report)
 556                 pnfs_report_layoutstat(inode, GFP_NOIO);
 557 }
 558
 559 static void
 560 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 561                 struct nfs4_ff_layout_mirror *mirror,
 562                 __u64 requested,
 563                 __u64 completed,
 564                 enum nfs3_stable_how committed)
 565 {
 566         if (committed == NFS_UNSTABLE)
 567                 requested = completed = 0;
 568
 569         spin_lock(&mirror->lock);
 570         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 571                         requested, completed, ktime_get(), task->tk_start);
 572         spin_unlock(&mirror->lock);
 573 }
 574
 575 static int
 576 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 577                             struct nfs_commit_info *cinfo,
 578                             gfp_t gfp_flags)
 579 {
 580         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 581         struct pnfs_commit_bucket *buckets;
 582         int size;
 583
 584         if (cinfo->ds->nbuckets != 0) {
 585                 /* This assumes there is only one RW lseg per file.
 586                  * To support multiple lseg per file, we need to
 587                  * change struct pnfs_commit_bucket to allow dynamic
 588                  * increasing nbuckets.
 589                  */
 590                 return 0;
 591         }
 592
 593         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 594
 595         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 596                           gfp_flags);
 597         if (!buckets)
 598                 return -ENOMEM;
 599         else {
 600                 int i;
 601
 602                 spin_lock(cinfo->lock);
 603                 if (cinfo->ds->nbuckets != 0)
 604                         kfree(buckets);
 605                 else {
 606                         cinfo->ds->buckets = buckets;
 607                         cinfo->ds->nbuckets = size;
 608                         for (i = 0; i < size; i++) {
 609                                 INIT_LIST_HEAD(&buckets[i].written);
 610                                 INIT_LIST_HEAD(&buckets[i].committing);
 611                                 /* mark direct verifier as unset */
 612                                 buckets[i].direct_verf.committed =
 613                                         NFS_INVALID_STABLE_HOW;
 614                         }
 615                 }
 616                 spin_unlock(cinfo->lock);
 617                 return 0;
 618         }
 619 }
 620
 621 static struct nfs4_pnfs_ds *
 622 ff_layout_choose_best_ds_for_read(struct nfs_pageio_descriptor *pgio,
 623                                   int *best_idx)
 624 {
 625         struct nfs4_ff_layout_segment *fls;
 626         struct nfs4_pnfs_ds *ds;
 627         int idx;
 628
 629         fls = FF_LAYOUT_LSEG(pgio->pg_lseg);
 630         /* mirrors are sorted by efficiency */
 631         for (idx = 0; idx < fls->mirror_array_cnt; idx++) {
 632                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, idx, false);
 633                 if (ds) {
 634                         *best_idx = idx;
 635                         return ds;
 636                 }
 637         }
 638
 639         return NULL;
 640 }
 641
 642 static void
 643 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 644                         struct nfs_page *req)
 645 {
 646         struct nfs_pgio_mirror *pgm;
 647         struct nfs4_ff_layout_mirror *mirror;
 648         struct nfs4_pnfs_ds *ds;
 649         int ds_idx;
 650
 651         /* Use full layout for now */
 652         if (!pgio->pg_lseg)
 653                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 654                                                    req->wb_context,
 655                                                    0,
 656                                                    NFS4_MAX_UINT64,
 657                                                    IOMODE_READ,
 658                                                    GFP_KERNEL);
 659         /* If no lseg, fall back to read through mds */
 660         if (pgio->pg_lseg == NULL)
 661                 goto out_mds;
 662
 663         ds = ff_layout_choose_best_ds_for_read(pgio, &ds_idx);
 664         if (!ds)
 665                 goto out_mds;
 666         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 667
 668         pgio->pg_mirror_idx = ds_idx;
 669
 670         /* read always uses only one mirror - idx 0 for pgio layer */
 671         pgm = &pgio->pg_mirrors[0];
 672         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 673
 674         return;
 675 out_mds:
 676         pnfs_put_lseg(pgio->pg_lseg);
 677         pgio->pg_lseg = NULL;
 678         nfs_pageio_reset_read_mds(pgio);
 679 }
 680
 681 static void
 682 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 683                         struct nfs_page *req)
 684 {
 685         struct nfs4_ff_layout_mirror *mirror;
 686         struct nfs_pgio_mirror *pgm;
 687         struct nfs_commit_info cinfo;
 688         struct nfs4_pnfs_ds *ds;
 689         int i;
 690         int status;
 691
 692         if (!pgio->pg_lseg)
 693                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 694                                                    req->wb_context,
 695                                                    0,
 696                                                    NFS4_MAX_UINT64,
 697                                                    IOMODE_RW,
 698                                                    GFP_NOFS);
 699         /* If no lseg, fall back to write through mds */
 700         if (pgio->pg_lseg == NULL)
 701                 goto out_mds;
 702
 703         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 704         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 705         if (status < 0)
 706                 goto out_mds;
 707
 708         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 709         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 710             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 711                 goto out_mds;
 712
 713         for (i = 0; i < pgio->pg_mirror_count; i++) {
 714                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 715                 if (!ds)
 716                         goto out_mds;
 717                 pgm = &pgio->pg_mirrors[i];
 718                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 719                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 720         }
 721
 722         return;
 723
 724 out_mds:
 725         pnfs_put_lseg(pgio->pg_lseg);
 726         pgio->pg_lseg = NULL;
 727         nfs_pageio_reset_write_mds(pgio);
 728 }
 729
 730 static unsigned int
 731 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 732                                     struct nfs_page *req)
 733 {
 734         if (!pgio->pg_lseg)
 735                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 736                                                    req->wb_context,
 737                                                    0,
 738                                                    NFS4_MAX_UINT64,
 739                                                    IOMODE_RW,
 740                                                    GFP_NOFS);
 741         if (pgio->pg_lseg)
 742                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 743
 744         /* no lseg means that pnfs is not in use, so no mirroring here */
 745         nfs_pageio_reset_write_mds(pgio);
 746         return 1;
 747 }
 748
 749 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 750         .pg_init = ff_layout_pg_init_read,
 751         .pg_test = pnfs_generic_pg_test,
 752         .pg_doio = pnfs_generic_pg_readpages,
 753         .pg_cleanup = pnfs_generic_pg_cleanup,
 754 };
 755
 756 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 757         .pg_init = ff_layout_pg_init_write,
 758         .pg_test = pnfs_generic_pg_test,
 759         .pg_doio = pnfs_generic_pg_writepages,
 760         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 761         .pg_cleanup = pnfs_generic_pg_cleanup,
 762 };
 763
 764 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 765 {
 766         struct rpc_task *task = &hdr->task;
 767
 768         pnfs_layoutcommit_inode(hdr->inode, false);
 769
 770         if (retry_pnfs) {
 771                 dprintk("%s Reset task %5u for i/o through pNFS "
 772                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 773                         hdr->task.tk_pid,
 774                         hdr->inode->i_sb->s_id,
 775                         (unsigned long long)NFS_FILEID(hdr->inode),
 776                         hdr->args.count,
 777                         (unsigned long long)hdr->args.offset);
 778
 779                 if (!hdr->dreq) {
 780                         struct nfs_open_context *ctx;
 781
 782                         ctx = nfs_list_entry(hdr->pages.next)->wb_context;
 783                         set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
 784                         hdr->completion_ops->error_cleanup(&hdr->pages);
 785                 } else {
 786                         nfs_direct_set_resched_writes(hdr->dreq);
 787                         /* fake unstable write to let common nfs resend pages */
 788                         hdr->verf.committed = NFS_UNSTABLE;
 789                         hdr->good_bytes = hdr->args.count;
 790                 }
 791                 return;
 792         }
 793
 794         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 795                 dprintk("%s Reset task %5u for i/o through MDS "
 796                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 797                         hdr->task.tk_pid,
 798                         hdr->inode->i_sb->s_id,
 799                         (unsigned long long)NFS_FILEID(hdr->inode),
 800                         hdr->args.count,
 801                         (unsigned long long)hdr->args.offset);
 802
 803                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
 804         }
 805 }
 806
 807 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
 808 {
 809         struct rpc_task *task = &hdr->task;
 810
 811         pnfs_layoutcommit_inode(hdr->inode, false);
 812
 813         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 814                 dprintk("%s Reset task %5u for i/o through MDS "
 815                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 816                         hdr->task.tk_pid,
 817                         hdr->inode->i_sb->s_id,
 818                         (unsigned long long)NFS_FILEID(hdr->inode),
 819                         hdr->args.count,
 820                         (unsigned long long)hdr->args.offset);
 821
 822                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
 823         }
 824 }
 825
 826 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
 827                                            struct nfs4_state *state,
 828                                            struct nfs_client *clp,
 829                                            struct pnfs_layout_segment *lseg,
 830                                            int idx)
 831 {
 832         struct pnfs_layout_hdr *lo = lseg->pls_layout;
 833         struct inode *inode = lo->plh_inode;
 834         struct nfs_server *mds_server = NFS_SERVER(inode);
 835
 836         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 837         struct nfs_client *mds_client = mds_server->nfs_client;
 838         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 839
 840         if (task->tk_status >= 0)
 841                 return 0;
 842
 843         switch (task->tk_status) {
 844         /* MDS state errors */
 845         case -NFS4ERR_DELEG_REVOKED:
 846         case -NFS4ERR_ADMIN_REVOKED:
 847         case -NFS4ERR_BAD_STATEID:
 848                 if (state == NULL)
 849                         break;
 850                 nfs_remove_bad_delegation(state->inode);
 851         case -NFS4ERR_OPENMODE:
 852                 if (state == NULL)
 853                         break;
 854                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
 855                         goto out_bad_stateid;
 856                 goto wait_on_recovery;
 857         case -NFS4ERR_EXPIRED:
 858                 if (state != NULL) {
 859                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
 860                                 goto out_bad_stateid;
 861                 }
 862                 nfs4_schedule_lease_recovery(mds_client);
 863                 goto wait_on_recovery;
 864         /* DS session errors */
 865         case -NFS4ERR_BADSESSION:
 866         case -NFS4ERR_BADSLOT:
 867         case -NFS4ERR_BAD_HIGH_SLOT:
 868         case -NFS4ERR_DEADSESSION:
 869         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 870         case -NFS4ERR_SEQ_FALSE_RETRY:
 871         case -NFS4ERR_SEQ_MISORDERED:
 872                 dprintk("%s ERROR %d, Reset session. Exchangeid "
 873                         "flags 0x%x\n", __func__, task->tk_status,
 874                         clp->cl_exchange_flags);
 875                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
 876                 break;
 877         case -NFS4ERR_DELAY:
 878         case -NFS4ERR_GRACE:
 879                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
 880                 break;
 881         case -NFS4ERR_RETRY_UNCACHED_REP:
 882                 break;
 883         /* Invalidate Layout errors */
 884         case -NFS4ERR_PNFS_NO_LAYOUT:
 885         case -ESTALE:           /* mapped NFS4ERR_STALE */
 886         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
 887         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
 888         case -NFS4ERR_FHEXPIRED:
 889         case -NFS4ERR_WRONG_TYPE:
 890                 dprintk("%s Invalid layout error %d\n", __func__,
 891                         task->tk_status);
 892                 /*
 893                  * Destroy layout so new i/o will get a new layout.
 894                  * Layout will not be destroyed until all current lseg
 895                  * references are put. Mark layout as invalid to resend failed
 896                  * i/o and all i/o waiting on the slot table to the MDS until
 897                  * layout is destroyed and a new valid layout is obtained.
 898                  */
 899                 pnfs_destroy_layout(NFS_I(inode));
 900                 rpc_wake_up(&tbl->slot_tbl_waitq);
 901                 goto reset;
 902         /* RPC connection errors */
 903         case -ECONNREFUSED:
 904         case -EHOSTDOWN:
 905         case -EHOSTUNREACH:
 906         case -ENETUNREACH:
 907         case -EIO:
 908         case -ETIMEDOUT:
 909         case -EPIPE:
 910                 dprintk("%s DS connection error %d\n", __func__,
 911                         task->tk_status);
 912                 nfs4_mark_deviceid_unavailable(devid);
 913                 rpc_wake_up(&tbl->slot_tbl_waitq);
 914                 /* fall through */
 915         default:
 916                 if (ff_layout_has_available_ds(lseg))
 917                         return -NFS4ERR_RESET_TO_PNFS;
 918 reset:
 919                 dprintk("%s Retry through MDS. Error %d\n", __func__,
 920                         task->tk_status);
 921                 return -NFS4ERR_RESET_TO_MDS;
 922         }
 923 out:
 924         task->tk_status = 0;
 925         return -EAGAIN;
 926 out_bad_stateid:
 927         task->tk_status = -EIO;
 928         return 0;
 929 wait_on_recovery:
 930         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
 931         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
 932                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
 933         goto out;
 934 }
 935
 936 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
 937 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
 938                                            struct pnfs_layout_segment *lseg,
 939                                            int idx)
 940 {
 941         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 942
 943         if (task->tk_status >= 0)
 944                 return 0;
 945
 946         if (task->tk_status != -EJUKEBOX) {
 947                 dprintk("%s DS connection error %d\n", __func__,
 948                         task->tk_status);
 949                 nfs4_mark_deviceid_unavailable(devid);
 950                 if (ff_layout_has_available_ds(lseg))
 951                         return -NFS4ERR_RESET_TO_PNFS;
 952                 else
 953                         return -NFS4ERR_RESET_TO_MDS;
 954         }
 955
 956         if (task->tk_status == -EJUKEBOX)
 957                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
 958         task->tk_status = 0;
 959         rpc_restart_call(task);
 960         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
 961         return -EAGAIN;
 962 }
 963
 964 static int ff_layout_async_handle_error(struct rpc_task *task,
 965                                         struct nfs4_state *state,
 966                                         struct nfs_client *clp,
 967                                         struct pnfs_layout_segment *lseg,
 968                                         int idx)
 969 {
 970         int vers = clp->cl_nfs_mod->rpc_vers->number;
 971
 972         switch (vers) {
 973         case 3:
 974                 return ff_layout_async_handle_error_v3(task, lseg, idx);
 975         case 4:
 976                 return ff_layout_async_handle_error_v4(task, state, clp,
 977                                                        lseg, idx);
 978         default:
 979                 /* should never happen */
 980                 WARN_ON_ONCE(1);
 981                 return 0;
 982         }
 983 }
 984
 985 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
 986                                         int idx, u64 offset, u64 length,
 987                                         u32 status, int opnum, int error)
 988 {
 989         struct nfs4_ff_layout_mirror *mirror;
 990         int err;
 991
 992         if (status == 0) {
 993                 switch (error) {
 994                 case -ETIMEDOUT:
 995                 case -EPFNOSUPPORT:
 996                 case -EPROTONOSUPPORT:
 997                 case -EOPNOTSUPP:
 998                 case -ECONNREFUSED:
 999                 case -ECONNRESET:
1000                 case -EHOSTDOWN:
1001                 case -EHOSTUNREACH:
1002                 case -ENETUNREACH:
1003                 case -EADDRINUSE:
1004                 case -ENOBUFS:
1005                 case -EPIPE:
1006                 case -EPERM:
1007                         status = NFS4ERR_NXIO;
1008                         break;
1009                 case -EACCES:
1010                         status = NFS4ERR_ACCESS;
1011                         break;
1012                 default:
1013                         return;
1014                 }
1015         }
1016
1017         mirror = FF_LAYOUT_COMP(lseg, idx);
1018         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1019                                        mirror, offset, length, status, opnum,
1020                                        GFP_NOIO);
1021         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1022 }
1023
1024 /* NFS_PROTO call done callback routines */
1025
1026 static int ff_layout_read_done_cb(struct rpc_task *task,
1027                                 struct nfs_pgio_header *hdr)
1028 {
1029         struct inode *inode;
1030         int err;
1031
1032         trace_nfs4_pnfs_read(hdr, task->tk_status);
1033         if (task->tk_status < 0)
1034                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1035                                             hdr->args.offset, hdr->args.count,
1036                                             hdr->res.op_status, OP_READ,
1037                                             task->tk_status);
1038         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1039                                            hdr->ds_clp, hdr->lseg,
1040                                            hdr->pgio_mirror_idx);
1041
1042         switch (err) {
1043         case -NFS4ERR_RESET_TO_PNFS:
1044                 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1045                         &hdr->lseg->pls_layout->plh_flags);
1046                 pnfs_read_resend_pnfs(hdr);
1047                 return task->tk_status;
1048         case -NFS4ERR_RESET_TO_MDS:
1049                 inode = hdr->lseg->pls_layout->plh_inode;
1050                 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1051                 ff_layout_reset_read(hdr);
1052                 return task->tk_status;
1053         case -EAGAIN:
1054                 rpc_restart_call_prepare(task);
1055                 return -EAGAIN;
1056         }
1057
1058         return 0;
1059 }
1060
1061 static bool
1062 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1063 {
1064         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1065 }
1066
1067 /*
1068  * We reference the rpc_cred of the first WRITE that triggers the need for
1069  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1070  * rfc5661 is not clear about which credential should be used.
1071  *
1072  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1073  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1074  * we always send layoutcommit after DS writes.
1075  */
1076 static void
1077 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
1078 {
1079         if (!ff_layout_need_layoutcommit(hdr->lseg))
1080                 return;
1081
1082         pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1083                         hdr->mds_offset + hdr->res.count);
1084         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
1085                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
1086 }
1087
1088 static bool
1089 ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
1090 {
1091         /* No mirroring for now */
1092         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1093
1094         return ff_layout_test_devid_unavailable(node);
1095 }
1096
1097 static int ff_layout_read_prepare_common(struct rpc_task *task,
1098                                          struct nfs_pgio_header *hdr)
1099 {
1100         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1101                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1102                         hdr->args.count,
1103                         task->tk_start);
1104
1105         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1106                 rpc_exit(task, -EIO);
1107                 return -EIO;
1108         }
1109         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1110                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
1111                 if (ff_layout_has_available_ds(hdr->lseg))
1112                         pnfs_read_resend_pnfs(hdr);
1113                 else
1114                         ff_layout_reset_read(hdr);
1115                 rpc_exit(task, 0);
1116                 return -EAGAIN;
1117         }
1118         hdr->pgio_done_cb = ff_layout_read_done_cb;
1119
1120         return 0;
1121 }
1122
1123 /*
1124  * Call ops for the async read/write cases
1125  * In the case of dense layouts, the offset needs to be reset to its
1126  * original value.
1127  */
1128 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1129 {
1130         struct nfs_pgio_header *hdr = data;
1131
1132         if (ff_layout_read_prepare_common(task, hdr))
1133                 return;
1134
1135         rpc_call_start(task);
1136 }
1137
1138 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1139                                     struct nfs4_sequence_args *args,
1140                                     struct nfs4_sequence_res *res,
1141                                     struct rpc_task *task)
1142 {
1143         if (ds_clp->cl_session)
1144                 return nfs41_setup_sequence(ds_clp->cl_session,
1145                                            args,
1146                                            res,
1147                                            task);
1148         return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1149                                    args,
1150                                    res,
1151                                    task);
1152 }
1153
1154 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1155 {
1156         struct nfs_pgio_header *hdr = data;
1157
1158         if (ff_layout_setup_sequence(hdr->ds_clp,
1159                                      &hdr->args.seq_args,
1160                                      &hdr->res.seq_res,
1161                                      task))
1162                 return;
1163
1164         if (ff_layout_read_prepare_common(task, hdr))
1165                 return;
1166
1167         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1168                         hdr->args.lock_context, FMODE_READ) == -EIO)
1169                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1170 }
1171
1172 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1173 {
1174         struct nfs_pgio_header *hdr = data;
1175
1176         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1177
1178         nfs4_ff_layout_stat_io_end_read(task,
1179                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1180                         hdr->args.count, hdr->res.count);
1181
1182         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1183             task->tk_status == 0) {
1184                 nfs4_sequence_done(task, &hdr->res.seq_res);
1185                 return;
1186         }
1187
1188         /* Note this may cause RPC to be resent */
1189         hdr->mds_ops->rpc_call_done(task, hdr);
1190 }
1191
1192 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1193 {
1194         struct nfs_pgio_header *hdr = data;
1195
1196         rpc_count_iostats_metrics(task,
1197             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1198 }
1199
1200 static int ff_layout_write_done_cb(struct rpc_task *task,
1201                                 struct nfs_pgio_header *hdr)
1202 {
1203         struct inode *inode;
1204         int err;
1205
1206         trace_nfs4_pnfs_write(hdr, task->tk_status);
1207         if (task->tk_status < 0)
1208                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1209                                             hdr->args.offset, hdr->args.count,
1210                                             hdr->res.op_status, OP_WRITE,
1211                                             task->tk_status);
1212         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1213                                            hdr->ds_clp, hdr->lseg,
1214                                            hdr->pgio_mirror_idx);
1215
1216         switch (err) {
1217         case -NFS4ERR_RESET_TO_PNFS:
1218         case -NFS4ERR_RESET_TO_MDS:
1219                 inode = hdr->lseg->pls_layout->plh_inode;
1220                 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1221                 if (err == -NFS4ERR_RESET_TO_PNFS) {
1222                         pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
1223                         ff_layout_reset_write(hdr, true);
1224                 } else {
1225                         pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
1226                         ff_layout_reset_write(hdr, false);
1227                 }
1228                 return task->tk_status;
1229         case -EAGAIN:
1230                 rpc_restart_call_prepare(task);
1231                 return -EAGAIN;
1232         }
1233
1234         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1235             hdr->res.verf->committed == NFS_DATA_SYNC)
1236                 ff_layout_set_layoutcommit(hdr);
1237
1238         /* zero out fattr since we don't care DS attr at all */
1239         hdr->fattr.valid = 0;
1240         if (task->tk_status >= 0)
1241                 nfs_writeback_update_inode(hdr);
1242
1243         return 0;
1244 }
1245
1246 static int ff_layout_commit_done_cb(struct rpc_task *task,
1247                                      struct nfs_commit_data *data)
1248 {
1249         struct inode *inode;
1250         int err;
1251
1252         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1253         if (task->tk_status < 0)
1254                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1255                                             data->args.offset, data->args.count,
1256                                             data->res.op_status, OP_COMMIT,
1257                                             task->tk_status);
1258         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1259                                            data->lseg, data->ds_commit_index);
1260
1261         switch (err) {
1262         case -NFS4ERR_RESET_TO_PNFS:
1263         case -NFS4ERR_RESET_TO_MDS:
1264                 inode = data->lseg->pls_layout->plh_inode;
1265                 pnfs_error_mark_layout_for_return(inode, data->lseg);
1266                 if (err == -NFS4ERR_RESET_TO_PNFS)
1267                         pnfs_set_retry_layoutget(data->lseg->pls_layout);
1268                 else
1269                         pnfs_clear_retry_layoutget(data->lseg->pls_layout);
1270                 pnfs_generic_prepare_to_resend_writes(data);
1271                 return -EAGAIN;
1272         case -EAGAIN:
1273                 rpc_restart_call_prepare(task);
1274                 return -EAGAIN;
1275         }
1276
1277         if (data->verf.committed == NFS_UNSTABLE
1278             && ff_layout_need_layoutcommit(data->lseg))
1279                 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
1280
1281         return 0;
1282 }
1283
1284 static int ff_layout_write_prepare_common(struct rpc_task *task,
1285                                           struct nfs_pgio_header *hdr)
1286 {
1287         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1288                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1289                         hdr->args.count,
1290                         task->tk_start);
1291
1292         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1293                 rpc_exit(task, -EIO);
1294                 return -EIO;
1295         }
1296
1297         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1298                 bool retry_pnfs;
1299
1300                 retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1301                 dprintk("%s task %u reset io to %s\n", __func__,
1302                         task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1303                 ff_layout_reset_write(hdr, retry_pnfs);
1304                 rpc_exit(task, 0);
1305                 return -EAGAIN;
1306         }
1307
1308         return 0;
1309 }
1310
1311 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1312 {
1313         struct nfs_pgio_header *hdr = data;
1314
1315         if (ff_layout_write_prepare_common(task, hdr))
1316                 return;
1317
1318         rpc_call_start(task);
1319 }
1320
1321 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1322 {
1323         struct nfs_pgio_header *hdr = data;
1324
1325         if (ff_layout_setup_sequence(hdr->ds_clp,
1326                                      &hdr->args.seq_args,
1327                                      &hdr->res.seq_res,
1328                                      task))
1329                 return;
1330
1331         if (ff_layout_write_prepare_common(task, hdr))
1332                 return;
1333
1334         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1335                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1336                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1337 }
1338
1339 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1340 {
1341         struct nfs_pgio_header *hdr = data;
1342
1343         nfs4_ff_layout_stat_io_end_write(task,
1344                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1345                         hdr->args.count, hdr->res.count,
1346                         hdr->res.verf->committed);
1347
1348         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1349             task->tk_status == 0) {
1350                 nfs4_sequence_done(task, &hdr->res.seq_res);
1351                 return;
1352         }
1353
1354         /* Note this may cause RPC to be resent */
1355         hdr->mds_ops->rpc_call_done(task, hdr);
1356 }
1357
1358 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1359 {
1360         struct nfs_pgio_header *hdr = data;
1361
1362         rpc_count_iostats_metrics(task,
1363             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1364 }
1365
1366 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1367                 struct nfs_commit_data *cdata)
1368 {
1369         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1370                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1371                         0, task->tk_start);
1372 }
1373
1374 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1375 {
1376         ff_layout_commit_prepare_common(task, data);
1377         rpc_call_start(task);
1378 }
1379
1380 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1381 {
1382         struct nfs_commit_data *wdata = data;
1383
1384         if (ff_layout_setup_sequence(wdata->ds_clp,
1385                                  &wdata->args.seq_args,
1386                                  &wdata->res.seq_res,
1387                                  task))
1388                 return;
1389         ff_layout_commit_prepare_common(task, data);
1390 }
1391
1392 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1393 {
1394         struct nfs_commit_data *cdata = data;
1395         struct nfs_page *req;
1396         __u64 count = 0;
1397
1398         if (task->tk_status == 0) {
1399                 list_for_each_entry(req, &cdata->pages, wb_list)
1400                         count += req->wb_bytes;
1401         }
1402
1403         nfs4_ff_layout_stat_io_end_write(task,
1404                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1405                         count, count, NFS_FILE_SYNC);
1406
1407         pnfs_generic_write_commit_done(task, data);
1408 }
1409
1410 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1411 {
1412         struct nfs_commit_data *cdata = data;
1413
1414         rpc_count_iostats_metrics(task,
1415             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1416 }
1417
1418 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1419         .rpc_call_prepare = ff_layout_read_prepare_v3,
1420         .rpc_call_done = ff_layout_read_call_done,
1421         .rpc_count_stats = ff_layout_read_count_stats,
1422         .rpc_release = pnfs_generic_rw_release,
1423 };
1424
1425 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1426         .rpc_call_prepare = ff_layout_read_prepare_v4,
1427         .rpc_call_done = ff_layout_read_call_done,
1428         .rpc_count_stats = ff_layout_read_count_stats,
1429         .rpc_release = pnfs_generic_rw_release,
1430 };
1431
1432 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1433         .rpc_call_prepare = ff_layout_write_prepare_v3,
1434         .rpc_call_done = ff_layout_write_call_done,
1435         .rpc_count_stats = ff_layout_write_count_stats,
1436         .rpc_release = pnfs_generic_rw_release,
1437 };
1438
1439 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1440         .rpc_call_prepare = ff_layout_write_prepare_v4,
1441         .rpc_call_done = ff_layout_write_call_done,
1442         .rpc_count_stats = ff_layout_write_count_stats,
1443         .rpc_release = pnfs_generic_rw_release,
1444 };
1445
1446 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1447         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1448         .rpc_call_done = ff_layout_commit_done,
1449         .rpc_count_stats = ff_layout_commit_count_stats,
1450         .rpc_release = pnfs_generic_commit_release,
1451 };
1452
1453 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1454         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1455         .rpc_call_done = ff_layout_commit_done,
1456         .rpc_count_stats = ff_layout_commit_count_stats,
1457         .rpc_release = pnfs_generic_commit_release,
1458 };
1459
1460 static enum pnfs_try_status
1461 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1462 {
1463         struct pnfs_layout_segment *lseg = hdr->lseg;
1464         struct nfs4_pnfs_ds *ds;
1465         struct rpc_clnt *ds_clnt;
1466         struct rpc_cred *ds_cred;
1467         loff_t offset = hdr->args.offset;
1468         u32 idx = hdr->pgio_mirror_idx;
1469         int vers;
1470         struct nfs_fh *fh;
1471
1472         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1473                 __func__, hdr->inode->i_ino,
1474                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1475
1476         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1477         if (!ds)
1478                 goto out_failed;
1479
1480         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1481                                                    hdr->inode);
1482         if (IS_ERR(ds_clnt))
1483                 goto out_failed;
1484
1485         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1486         if (IS_ERR(ds_cred))
1487                 goto out_failed;
1488
1489         vers = nfs4_ff_layout_ds_version(lseg, idx);
1490
1491         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1492                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1493
1494         atomic_inc(&ds->ds_clp->cl_count);
1495         hdr->ds_clp = ds->ds_clp;
1496         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1497         if (fh)
1498                 hdr->args.fh = fh;
1499         /*
1500          * Note that if we ever decide to split across DSes,
1501          * then we may need to handle dense-like offsets.
1502          */
1503         hdr->args.offset = offset;
1504         hdr->mds_offset = offset;
1505
1506         /* Perform an asynchronous read to ds */
1507         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1508                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1509                                       &ff_layout_read_call_ops_v4,
1510                           0, RPC_TASK_SOFTCONN);
1511
1512         return PNFS_ATTEMPTED;
1513
1514 out_failed:
1515         if (ff_layout_has_available_ds(lseg))
1516                 return PNFS_TRY_AGAIN;
1517         return PNFS_NOT_ATTEMPTED;
1518 }
1519
1520 /* Perform async writes. */
1521 static enum pnfs_try_status
1522 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1523 {
1524         struct pnfs_layout_segment *lseg = hdr->lseg;
1525         struct nfs4_pnfs_ds *ds;
1526         struct rpc_clnt *ds_clnt;
1527         struct rpc_cred *ds_cred;
1528         loff_t offset = hdr->args.offset;
1529         int vers;
1530         struct nfs_fh *fh;
1531         int idx = hdr->pgio_mirror_idx;
1532
1533         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1534         if (!ds)
1535                 return PNFS_NOT_ATTEMPTED;
1536
1537         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1538                                                    hdr->inode);
1539         if (IS_ERR(ds_clnt))
1540                 return PNFS_NOT_ATTEMPTED;
1541
1542         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1543         if (IS_ERR(ds_cred))
1544                 return PNFS_NOT_ATTEMPTED;
1545
1546         vers = nfs4_ff_layout_ds_version(lseg, idx);
1547
1548         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1549                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1550                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1551                 vers);
1552
1553         hdr->pgio_done_cb = ff_layout_write_done_cb;
1554         atomic_inc(&ds->ds_clp->cl_count);
1555         hdr->ds_clp = ds->ds_clp;
1556         hdr->ds_commit_idx = idx;
1557         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1558         if (fh)
1559                 hdr->args.fh = fh;
1560
1561         /*
1562          * Note that if we ever decide to split across DSes,
1563          * then we may need to handle dense-like offsets.
1564          */
1565         hdr->args.offset = offset;
1566
1567         /* Perform an asynchronous write */
1568         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1569                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1570                                       &ff_layout_write_call_ops_v4,
1571                           sync, RPC_TASK_SOFTCONN);
1572         return PNFS_ATTEMPTED;
1573 }
1574
1575 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1576 {
1577         return i;
1578 }
1579
1580 static struct nfs_fh *
1581 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1582 {
1583         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1584
1585         /* FIXME: Assume that there is only one NFS version available
1586          * for the DS.
1587          */
1588         return &flseg->mirror_array[i]->fh_versions[0];
1589 }
1590
1591 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1592 {
1593         struct pnfs_layout_segment *lseg = data->lseg;
1594         struct nfs4_pnfs_ds *ds;
1595         struct rpc_clnt *ds_clnt;
1596         struct rpc_cred *ds_cred;
1597         u32 idx;
1598         int vers;
1599         struct nfs_fh *fh;
1600
1601         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1602         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1603         if (!ds)
1604                 goto out_err;
1605
1606         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1607                                                    data->inode);
1608         if (IS_ERR(ds_clnt))
1609                 goto out_err;
1610
1611         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1612         if (IS_ERR(ds_cred))
1613                 goto out_err;
1614
1615         vers = nfs4_ff_layout_ds_version(lseg, idx);
1616
1617         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1618                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1619                 vers);
1620         data->commit_done_cb = ff_layout_commit_done_cb;
1621         data->cred = ds_cred;
1622         atomic_inc(&ds->ds_clp->cl_count);
1623         data->ds_clp = ds->ds_clp;
1624         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1625         if (fh)
1626                 data->args.fh = fh;
1627
1628         return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1629                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1630                                                &ff_layout_commit_call_ops_v4,
1631                                    how, RPC_TASK_SOFTCONN);
1632 out_err:
1633         pnfs_generic_prepare_to_resend_writes(data);
1634         pnfs_generic_commit_release(data);
1635         return -EAGAIN;
1636 }
1637
1638 static int
1639 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1640                            int how, struct nfs_commit_info *cinfo)
1641 {
1642         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1643                                             ff_layout_initiate_commit);
1644 }
1645
1646 static struct pnfs_ds_commit_info *
1647 ff_layout_get_ds_info(struct inode *inode)
1648 {
1649         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1650
1651         if (layout == NULL)
1652                 return NULL;
1653
1654         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1655 }
1656
1657 static void
1658 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1659 {
1660         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1661                                                   id_node));
1662 }
1663
1664 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1665                                   struct xdr_stream *xdr,
1666                                   const struct nfs4_layoutreturn_args *args)
1667 {
1668         struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1669         __be32 *start;
1670         int count = 0, ret = 0;
1671
1672         start = xdr_reserve_space(xdr, 4);
1673         if (unlikely(!start))
1674                 return -E2BIG;
1675
1676         /* This assume we always return _ALL_ layouts */
1677         spin_lock(&hdr->plh_inode->i_lock);
1678         ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1679         spin_unlock(&hdr->plh_inode->i_lock);
1680
1681         *start = cpu_to_be32(count);
1682
1683         return ret;
1684 }
1685
1686 /* report nothing for now */
1687 static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1688                                      struct xdr_stream *xdr,
1689                                      const struct nfs4_layoutreturn_args *args)
1690 {
1691         __be32 *p;
1692
1693         p = xdr_reserve_space(xdr, 4);
1694         if (likely(p))
1695                 *p = cpu_to_be32(0);
1696 }
1697
1698 static struct nfs4_deviceid_node *
1699 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1700                               struct pnfs_device *pdev, gfp_t gfp_flags)
1701 {
1702         struct nfs4_ff_layout_ds *dsaddr;
1703
1704         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1705         if (!dsaddr)
1706                 return NULL;
1707         return &dsaddr->id_node;
1708 }
1709
1710 static void
1711 ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1712                               struct xdr_stream *xdr,
1713                               const struct nfs4_layoutreturn_args *args)
1714 {
1715         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1716         __be32 *start;
1717
1718         dprintk("%s: Begin\n", __func__);
1719         start = xdr_reserve_space(xdr, 4);
1720         BUG_ON(!start);
1721
1722         if (ff_layout_encode_ioerr(flo, xdr, args))
1723                 goto out;
1724
1725         ff_layout_encode_iostats(flo, xdr, args);
1726 out:
1727         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1728         dprintk("%s: Return\n", __func__);
1729 }
1730
1731 static int
1732 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
1733 {
1734         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
1735
1736         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
1737 }
1738
1739 static size_t
1740 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
1741                           const int buflen)
1742 {
1743         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
1744         const struct in6_addr *addr = &sin6->sin6_addr;
1745
1746         /*
1747          * RFC 4291, Section 2.2.2
1748          *
1749          * Shorthanded ANY address
1750          */
1751         if (ipv6_addr_any(addr))
1752                 return snprintf(buf, buflen, "::");
1753
1754         /*
1755          * RFC 4291, Section 2.2.2
1756          *
1757          * Shorthanded loopback address
1758          */
1759         if (ipv6_addr_loopback(addr))
1760                 return snprintf(buf, buflen, "::1");
1761
1762         /*
1763          * RFC 4291, Section 2.2.3
1764          *
1765          * Special presentation address format for mapped v4
1766          * addresses.
1767          */
1768         if (ipv6_addr_v4mapped(addr))
1769                 return snprintf(buf, buflen, "::ffff:%pI4",
1770                                         &addr->s6_addr32[3]);
1771
1772         /*
1773          * RFC 4291, Section 2.2.1
1774          */
1775         return snprintf(buf, buflen, "%pI6c", addr);
1776 }
1777
1778 /* Derived from rpc_sockaddr2uaddr */
1779 static void
1780 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
1781 {
1782         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
1783         char portbuf[RPCBIND_MAXUADDRPLEN];
1784         char addrbuf[RPCBIND_MAXUADDRLEN];
1785         char *netid;
1786         unsigned short port;
1787         int len, netid_len;
1788         __be32 *p;
1789
1790         switch (sap->sa_family) {
1791         case AF_INET:
1792                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
1793                         return;
1794                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
1795                 netid = "tcp";
1796                 netid_len = 3;
1797                 break;
1798         case AF_INET6:
1799                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
1800                         return;
1801                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
1802                 netid = "tcp6";
1803                 netid_len = 4;
1804                 break;
1805         default:
1806                 /* we only support tcp and tcp6 */
1807                 WARN_ON_ONCE(1);
1808                 return;
1809         }
1810
1811         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
1812         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
1813
1814         p = xdr_reserve_space(xdr, 4 + netid_len);
1815         xdr_encode_opaque(p, netid, netid_len);
1816
1817         p = xdr_reserve_space(xdr, 4 + len);
1818         xdr_encode_opaque(p, addrbuf, len);
1819 }
1820
1821 static void
1822 ff_layout_encode_nfstime(struct xdr_stream *xdr,
1823                          ktime_t t)
1824 {
1825         struct timespec64 ts;
1826         __be32 *p;
1827
1828         p = xdr_reserve_space(xdr, 12);
1829         ts = ktime_to_timespec64(t);
1830         p = xdr_encode_hyper(p, ts.tv_sec);
1831         *p++ = cpu_to_be32(ts.tv_nsec);
1832 }
1833
1834 static void
1835 ff_layout_encode_io_latency(struct xdr_stream *xdr,
1836                             struct nfs4_ff_io_stat *stat)
1837 {
1838         __be32 *p;
1839
1840         p = xdr_reserve_space(xdr, 5 * 8);
1841         p = xdr_encode_hyper(p, stat->ops_requested);
1842         p = xdr_encode_hyper(p, stat->bytes_requested);
1843         p = xdr_encode_hyper(p, stat->ops_completed);
1844         p = xdr_encode_hyper(p, stat->bytes_completed);
1845         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
1846         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
1847         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
1848 }
1849
1850 static void
1851 ff_layout_encode_layoutstats(struct xdr_stream *xdr,
1852                              struct nfs42_layoutstat_args *args,
1853                              struct nfs42_layoutstat_devinfo *devinfo)
1854 {
1855         struct nfs4_ff_layout_mirror *mirror = devinfo->layout_private;
1856         struct nfs4_pnfs_ds_addr *da;
1857         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
1858         struct nfs_fh *fh = &mirror->fh_versions[0];
1859         __be32 *p, *start;
1860
1861         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
1862         dprintk("%s: DS %s: encoding address %s\n",
1863                 __func__, ds->ds_remotestr, da->da_remotestr);
1864         /* layoutupdate length */
1865         start = xdr_reserve_space(xdr, 4);
1866         /* netaddr4 */
1867         ff_layout_encode_netaddr(xdr, da);
1868         /* nfs_fh4 */
1869         p = xdr_reserve_space(xdr, 4 + fh->size);
1870         xdr_encode_opaque(p, fh->data, fh->size);
1871         /* ff_io_latency4 read */
1872         spin_lock(&mirror->lock);
1873         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
1874         /* ff_io_latency4 write */
1875         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
1876         spin_unlock(&mirror->lock);
1877         /* nfstime4 */
1878         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
1879         /* bool */
1880         p = xdr_reserve_space(xdr, 4);
1881         *p = cpu_to_be32(false);
1882
1883         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1884 }
1885
1886 static bool
1887 ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
1888                                struct pnfs_layout_segment *pls,
1889                                int *dev_count, int dev_limit)
1890 {
1891         struct nfs4_ff_layout_mirror *mirror;
1892         struct nfs4_deviceid_node *dev;
1893         struct nfs42_layoutstat_devinfo *devinfo;
1894         int i;
1895
1896         for (i = 0; i < FF_LAYOUT_MIRROR_COUNT(pls); i++) {
1897                 if (*dev_count >= dev_limit)
1898                         break;
1899                 mirror = FF_LAYOUT_COMP(pls, i);
1900                 if (!mirror || !mirror->mirror_ds)
1901                         continue;
1902                 dev = FF_LAYOUT_DEVID_NODE(pls, i);
1903                 devinfo = &args->devinfo[*dev_count];
1904                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
1905                 devinfo->offset = pls->pls_range.offset;
1906                 devinfo->length = pls->pls_range.length;
1907                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
1908                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
1909                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
1910                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
1911                 devinfo->layout_type = LAYOUT_FLEX_FILES;
1912                 devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
1913                 devinfo->layout_private = mirror;
1914                 /* mirror refcount put in cleanup_layoutstats */
1915                 atomic_inc(&mirror->ref);
1916
1917                 ++(*dev_count);
1918         }
1919
1920         return *dev_count < dev_limit;
1921 }
1922
1923 static int
1924 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
1925 {
1926         struct pnfs_layout_segment *pls;
1927         int dev_count = 0;
1928
1929         spin_lock(&args->inode->i_lock);
1930         list_for_each_entry(pls, &NFS_I(args->inode)->layout->plh_segs, pls_list) {
1931                 dev_count += FF_LAYOUT_MIRROR_COUNT(pls);
1932         }
1933         spin_unlock(&args->inode->i_lock);
1934         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
1935         if (dev_count > PNFS_LAYOUTSTATS_MAXDEV) {
1936                 dprintk("%s: truncating devinfo to limit (%d:%d)\n",
1937                         __func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
1938                 dev_count = PNFS_LAYOUTSTATS_MAXDEV;
1939         }
1940         args->devinfo = kmalloc(dev_count * sizeof(*args->devinfo), GFP_KERNEL);
1941         if (!args->devinfo)
1942                 return -ENOMEM;
1943
1944         dev_count = 0;
1945         spin_lock(&args->inode->i_lock);
1946         list_for_each_entry(pls, &NFS_I(args->inode)->layout->plh_segs, pls_list) {
1947                 if (!ff_layout_mirror_prepare_stats(args, pls, &dev_count,
1948                                                     PNFS_LAYOUTSTATS_MAXDEV)) {
1949                         break;
1950                 }
1951         }
1952         spin_unlock(&args->inode->i_lock);
1953         args->num_dev = dev_count;
1954
1955         return 0;
1956 }
1957
1958 static void
1959 ff_layout_cleanup_layoutstats(struct nfs42_layoutstat_data *data)
1960 {
1961         struct nfs4_ff_layout_mirror *mirror;
1962         int i;
1963
1964         for (i = 0; i < data->args.num_dev; i++) {
1965                 mirror = data->args.devinfo[i].layout_private;
1966                 data->args.devinfo[i].layout_private = NULL;
1967                 ff_layout_put_mirror(mirror);
1968         }
1969 }
1970
1971 static struct pnfs_layoutdriver_type flexfilelayout_type = {
1972         .id                     = LAYOUT_FLEX_FILES,
1973         .name                   = "LAYOUT_FLEX_FILES",
1974         .owner                  = THIS_MODULE,
1975         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
1976         .free_layout_hdr        = ff_layout_free_layout_hdr,
1977         .alloc_lseg             = ff_layout_alloc_lseg,
1978         .free_lseg              = ff_layout_free_lseg,
1979         .pg_read_ops            = &ff_layout_pg_read_ops,
1980         .pg_write_ops           = &ff_layout_pg_write_ops,
1981         .get_ds_info            = ff_layout_get_ds_info,
1982         .free_deviceid_node     = ff_layout_free_deviceid_node,
1983         .mark_request_commit    = pnfs_layout_mark_request_commit,
1984         .clear_request_commit   = pnfs_generic_clear_request_commit,
1985         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
1986         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
1987         .commit_pagelist        = ff_layout_commit_pagelist,
1988         .read_pagelist          = ff_layout_read_pagelist,
1989         .write_pagelist         = ff_layout_write_pagelist,
1990         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
1991         .encode_layoutreturn    = ff_layout_encode_layoutreturn,
1992         .sync                   = pnfs_nfs_generic_sync,
1993         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
1994         .cleanup_layoutstats    = ff_layout_cleanup_layoutstats,
1995 };
1996
1997 static int __init nfs4flexfilelayout_init(void)
1998 {
1999         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2000                __func__);
2001         return pnfs_register_layoutdriver(&flexfilelayout_type);
2002 }
2003
2004 static void __exit nfs4flexfilelayout_exit(void)
2005 {
2006         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2007                __func__);
2008         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2009 }
2010
2011 MODULE_ALIAS("nfs-layouttype4-4");
2012
2013 MODULE_LICENSE("GPL");
2014 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2015
2016 module_init(nfs4flexfilelayout_init);
2017 module_exit(nfs4flexfilelayout_exit);