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 #define FF_LAYOUTRETURN_MAXERR 20
  29
  30
  31 static struct group_info        *ff_zero_group;
  32
  33 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
  34                 struct nfs_pgio_header *hdr);
  35 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
  36                                struct nfs42_layoutstat_devinfo *devinfo,
  37                                int dev_limit);
  38 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
  39                               const struct nfs42_layoutstat_devinfo *devinfo,
  40                               struct nfs4_ff_layout_mirror *mirror);
  41
  42 static struct pnfs_layout_hdr *
  43 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  44 {
  45         struct nfs4_flexfile_layout *ffl;
  46
  47         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  48         if (ffl) {
  49                 INIT_LIST_HEAD(&ffl->error_list);
  50                 INIT_LIST_HEAD(&ffl->mirrors);
  51                 ffl->last_report_time = ktime_get();
  52                 return &ffl->generic_hdr;
  53         } else
  54                 return NULL;
  55 }
  56
  57 static void
  58 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  59 {
  60         struct nfs4_ff_layout_ds_err *err, *n;
  61
  62         list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
  63                                  list) {
  64                 list_del(&err->list);
  65                 kfree(err);
  66         }
  67         kfree(FF_LAYOUT_FROM_HDR(lo));
  68 }
  69
  70 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  71 {
  72         __be32 *p;
  73
  74         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  75         if (unlikely(p == NULL))
  76                 return -ENOBUFS;
  77         stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
  78         memcpy(stateid->data, p, NFS4_STATEID_SIZE);
  79         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  80                 p[0], p[1], p[2], p[3]);
  81         return 0;
  82 }
  83
  84 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  85 {
  86         __be32 *p;
  87
  88         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  89         if (unlikely(!p))
  90                 return -ENOBUFS;
  91         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  92         nfs4_print_deviceid(devid);
  93         return 0;
  94 }
  95
  96 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
  97 {
  98         __be32 *p;
  99
 100         p = xdr_inline_decode(xdr, 4);
 101         if (unlikely(!p))
 102                 return -ENOBUFS;
 103         fh->size = be32_to_cpup(p++);
 104         if (fh->size > NFS_MAXFHSIZE) {
 105                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
 106                        fh->size);
 107                 return -EOVERFLOW;
 108         }
 109         /* fh.data */
 110         p = xdr_inline_decode(xdr, fh->size);
 111         if (unlikely(!p))
 112                 return -ENOBUFS;
 113         memcpy(&fh->data, p, fh->size);
 114         dprintk("%s: fh len %d\n", __func__, fh->size);
 115
 116         return 0;
 117 }
 118
 119 /*
 120  * Currently only stringified uids and gids are accepted.
 121  * I.e., kerberos is not supported to the DSes, so no pricipals.
 122  *
 123  * That means that one common function will suffice, but when
 124  * principals are added, this should be split to accomodate
 125  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 126  */
 127 static int
 128 decode_name(struct xdr_stream *xdr, u32 *id)
 129 {
 130         __be32 *p;
 131         int len;
 132
 133         /* opaque_length(4)*/
 134         p = xdr_inline_decode(xdr, 4);
 135         if (unlikely(!p))
 136                 return -ENOBUFS;
 137         len = be32_to_cpup(p++);
 138         if (len < 0)
 139                 return -EINVAL;
 140
 141         dprintk("%s: len %u\n", __func__, len);
 142
 143         /* opaque body */
 144         p = xdr_inline_decode(xdr, len);
 145         if (unlikely(!p))
 146                 return -ENOBUFS;
 147
 148         if (!nfs_map_string_to_numeric((char *)p, len, id))
 149                 return -EINVAL;
 150
 151         return 0;
 152 }
 153
 154 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 155                 const struct nfs4_ff_layout_mirror *m2)
 156 {
 157         int i, j;
 158
 159         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 160                 return false;
 161         for (i = 0; i < m1->fh_versions_cnt; i++) {
 162                 bool found_fh = false;
 163                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 164                         if (nfs_compare_fh(&m1->fh_versions[i],
 165                                         &m2->fh_versions[j]) == 0) {
 166                                 found_fh = true;
 167                                 break;
 168                         }
 169                 }
 170                 if (!found_fh)
 171                         return false;
 172         }
 173         return true;
 174 }
 175
 176 static struct nfs4_ff_layout_mirror *
 177 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 178                 struct nfs4_ff_layout_mirror *mirror)
 179 {
 180         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 181         struct nfs4_ff_layout_mirror *pos;
 182         struct inode *inode = lo->plh_inode;
 183
 184         spin_lock(&inode->i_lock);
 185         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 186                 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
 187                         continue;
 188                 if (!ff_mirror_match_fh(mirror, pos))
 189                         continue;
 190                 if (atomic_inc_not_zero(&pos->ref)) {
 191                         spin_unlock(&inode->i_lock);
 192                         return pos;
 193                 }
 194         }
 195         list_add(&mirror->mirrors, &ff_layout->mirrors);
 196         mirror->layout = lo;
 197         spin_unlock(&inode->i_lock);
 198         return mirror;
 199 }
 200
 201 static void
 202 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 203 {
 204         struct inode *inode;
 205         if (mirror->layout == NULL)
 206                 return;
 207         inode = mirror->layout->plh_inode;
 208         spin_lock(&inode->i_lock);
 209         list_del(&mirror->mirrors);
 210         spin_unlock(&inode->i_lock);
 211         mirror->layout = NULL;
 212 }
 213
 214 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 215 {
 216         struct nfs4_ff_layout_mirror *mirror;
 217
 218         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 219         if (mirror != NULL) {
 220                 spin_lock_init(&mirror->lock);
 221                 atomic_set(&mirror->ref, 1);
 222                 INIT_LIST_HEAD(&mirror->mirrors);
 223         }
 224         return mirror;
 225 }
 226
 227 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 228 {
 229         struct rpc_cred *cred;
 230
 231         ff_layout_remove_mirror(mirror);
 232         kfree(mirror->fh_versions);
 233         cred = rcu_access_pointer(mirror->ro_cred);
 234         if (cred)
 235                 put_rpccred(cred);
 236         cred = rcu_access_pointer(mirror->rw_cred);
 237         if (cred)
 238                 put_rpccred(cred);
 239         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 240         kfree(mirror);
 241 }
 242
 243 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 244 {
 245         if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
 246                 ff_layout_free_mirror(mirror);
 247 }
 248
 249 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 250 {
 251         int i;
 252
 253         if (fls->mirror_array) {
 254                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 255                         /* normally mirror_ds is freed in
 256                          * .free_deviceid_node but we still do it here
 257                          * for .alloc_lseg error path */
 258                         ff_layout_put_mirror(fls->mirror_array[i]);
 259                 }
 260                 kfree(fls->mirror_array);
 261                 fls->mirror_array = NULL;
 262         }
 263 }
 264
 265 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 266 {
 267         int ret = 0;
 268
 269         dprintk("--> %s\n", __func__);
 270
 271         /* FIXME: remove this check when layout segment support is added */
 272         if (lgr->range.offset != 0 ||
 273             lgr->range.length != NFS4_MAX_UINT64) {
 274                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 275                         __func__);
 276                 ret = -EINVAL;
 277         }
 278
 279         dprintk("--> %s returns %d\n", __func__, ret);
 280         return ret;
 281 }
 282
 283 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 284 {
 285         if (fls) {
 286                 ff_layout_free_mirror_array(fls);
 287                 kfree(fls);
 288         }
 289 }
 290
 291 static bool
 292 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 293                 const struct pnfs_layout_range *l2)
 294 {
 295         u64 end1, end2;
 296
 297         if (l1->iomode != l2->iomode)
 298                 return l1->iomode != IOMODE_READ;
 299         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 300         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 301         if (end1 < l2->offset)
 302                 return false;
 303         if (end2 < l1->offset)
 304                 return true;
 305         return l2->offset <= l1->offset;
 306 }
 307
 308 static bool
 309 ff_lseg_merge(struct pnfs_layout_segment *new,
 310                 struct pnfs_layout_segment *old)
 311 {
 312         u64 new_end, old_end;
 313
 314         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 315                 return false;
 316         if (new->pls_range.iomode != old->pls_range.iomode)
 317                 return false;
 318         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 319                         old->pls_range.length);
 320         if (old_end < new->pls_range.offset)
 321                 return false;
 322         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 323                         new->pls_range.length);
 324         if (new_end < old->pls_range.offset)
 325                 return false;
 326
 327         /* Mergeable: copy info from 'old' to 'new' */
 328         if (new_end < old_end)
 329                 new_end = old_end;
 330         if (new->pls_range.offset < old->pls_range.offset)
 331                 new->pls_range.offset = old->pls_range.offset;
 332         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 333                         new_end);
 334         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 335                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 336         return true;
 337 }
 338
 339 static void
 340 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 341                 struct pnfs_layout_segment *lseg,
 342                 struct list_head *free_me)
 343 {
 344         pnfs_generic_layout_insert_lseg(lo, lseg,
 345                         ff_lseg_range_is_after,
 346                         ff_lseg_merge,
 347                         free_me);
 348 }
 349
 350 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 351 {
 352         int i, j;
 353
 354         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 355                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 356                         if (fls->mirror_array[i]->efficiency <
 357                             fls->mirror_array[j]->efficiency)
 358                                 swap(fls->mirror_array[i],
 359                                      fls->mirror_array[j]);
 360         }
 361 }
 362
 363 static struct pnfs_layout_segment *
 364 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 365                      struct nfs4_layoutget_res *lgr,
 366                      gfp_t gfp_flags)
 367 {
 368         struct pnfs_layout_segment *ret;
 369         struct nfs4_ff_layout_segment *fls = NULL;
 370         struct xdr_stream stream;
 371         struct xdr_buf buf;
 372         struct page *scratch;
 373         u64 stripe_unit;
 374         u32 mirror_array_cnt;
 375         __be32 *p;
 376         int i, rc;
 377
 378         dprintk("--> %s\n", __func__);
 379         scratch = alloc_page(gfp_flags);
 380         if (!scratch)
 381                 return ERR_PTR(-ENOMEM);
 382
 383         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 384                               lgr->layoutp->len);
 385         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 386
 387         /* stripe unit and mirror_array_cnt */
 388         rc = -EIO;
 389         p = xdr_inline_decode(&stream, 8 + 4);
 390         if (!p)
 391                 goto out_err_free;
 392
 393         p = xdr_decode_hyper(p, &stripe_unit);
 394         mirror_array_cnt = be32_to_cpup(p++);
 395         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 396                 stripe_unit, mirror_array_cnt);
 397
 398         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 399             mirror_array_cnt == 0)
 400                 goto out_err_free;
 401
 402         rc = -ENOMEM;
 403         fls = kzalloc(sizeof(*fls), gfp_flags);
 404         if (!fls)
 405                 goto out_err_free;
 406
 407         fls->mirror_array_cnt = mirror_array_cnt;
 408         fls->stripe_unit = stripe_unit;
 409         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 410                                     sizeof(fls->mirror_array[0]), gfp_flags);
 411         if (fls->mirror_array == NULL)
 412                 goto out_err_free;
 413
 414         for (i = 0; i < fls->mirror_array_cnt; i++) {
 415                 struct nfs4_ff_layout_mirror *mirror;
 416                 struct auth_cred acred = { .group_info = ff_zero_group };
 417                 struct rpc_cred __rcu *cred;
 418                 u32 ds_count, fh_count, id;
 419                 int j;
 420
 421                 rc = -EIO;
 422                 p = xdr_inline_decode(&stream, 4);
 423                 if (!p)
 424                         goto out_err_free;
 425                 ds_count = be32_to_cpup(p);
 426
 427                 /* FIXME: allow for striping? */
 428                 if (ds_count != 1)
 429                         goto out_err_free;
 430
 431                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 432                 if (fls->mirror_array[i] == NULL) {
 433                         rc = -ENOMEM;
 434                         goto out_err_free;
 435                 }
 436
 437                 fls->mirror_array[i]->ds_count = ds_count;
 438
 439                 /* deviceid */
 440                 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
 441                 if (rc)
 442                         goto out_err_free;
 443
 444                 /* efficiency */
 445                 rc = -EIO;
 446                 p = xdr_inline_decode(&stream, 4);
 447                 if (!p)
 448                         goto out_err_free;
 449                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 450
 451                 /* stateid */
 452                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 453                 if (rc)
 454                         goto out_err_free;
 455
 456                 /* fh */
 457                 rc = -EIO;
 458                 p = xdr_inline_decode(&stream, 4);
 459                 if (!p)
 460                         goto out_err_free;
 461                 fh_count = be32_to_cpup(p);
 462
 463                 fls->mirror_array[i]->fh_versions =
 464                         kzalloc(fh_count * sizeof(struct nfs_fh),
 465                                 gfp_flags);
 466                 if (fls->mirror_array[i]->fh_versions == NULL) {
 467                         rc = -ENOMEM;
 468                         goto out_err_free;
 469                 }
 470
 471                 for (j = 0; j < fh_count; j++) {
 472                         rc = decode_nfs_fh(&stream,
 473                                            &fls->mirror_array[i]->fh_versions[j]);
 474                         if (rc)
 475                                 goto out_err_free;
 476                 }
 477
 478                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 479
 480                 /* user */
 481                 rc = decode_name(&stream, &id);
 482                 if (rc)
 483                         goto out_err_free;
 484
 485                 acred.uid = make_kuid(&init_user_ns, id);
 486
 487                 /* group */
 488                 rc = decode_name(&stream, &id);
 489                 if (rc)
 490                         goto out_err_free;
 491
 492                 acred.gid = make_kgid(&init_user_ns, id);
 493
 494                 /* find the cred for it */
 495                 rcu_assign_pointer(cred, rpc_lookup_generic_cred(&acred, 0, gfp_flags));
 496                 if (IS_ERR(cred)) {
 497                         rc = PTR_ERR(cred);
 498                         goto out_err_free;
 499                 }
 500
 501                 if (lgr->range.iomode == IOMODE_READ)
 502                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 503                 else
 504                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 505
 506                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 507                 if (mirror != fls->mirror_array[i]) {
 508                         /* swap cred ptrs so free_mirror will clean up old */
 509                         if (lgr->range.iomode == IOMODE_READ) {
 510                                 cred = xchg(&mirror->ro_cred, cred);
 511                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 512                         } else {
 513                                 cred = xchg(&mirror->rw_cred, cred);
 514                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 515                         }
 516                         ff_layout_free_mirror(fls->mirror_array[i]);
 517                         fls->mirror_array[i] = mirror;
 518                 }
 519
 520                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 521                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 522                         from_kuid(&init_user_ns, acred.uid),
 523                         from_kgid(&init_user_ns, acred.gid));
 524         }
 525
 526         p = xdr_inline_decode(&stream, 4);
 527         if (!p)
 528                 goto out_sort_mirrors;
 529         fls->flags = be32_to_cpup(p);
 530
 531         p = xdr_inline_decode(&stream, 4);
 532         if (!p)
 533                 goto out_sort_mirrors;
 534         for (i=0; i < fls->mirror_array_cnt; i++)
 535                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 536
 537 out_sort_mirrors:
 538         ff_layout_sort_mirrors(fls);
 539         rc = ff_layout_check_layout(lgr);
 540         if (rc)
 541                 goto out_err_free;
 542         ret = &fls->generic_hdr;
 543         dprintk("<-- %s (success)\n", __func__);
 544 out_free_page:
 545         __free_page(scratch);
 546         return ret;
 547 out_err_free:
 548         _ff_layout_free_lseg(fls);
 549         ret = ERR_PTR(rc);
 550         dprintk("<-- %s (%d)\n", __func__, rc);
 551         goto out_free_page;
 552 }
 553
 554 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 555 {
 556         struct pnfs_layout_segment *lseg;
 557
 558         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 559                 if (lseg->pls_range.iomode == IOMODE_RW)
 560                         return true;
 561
 562         return false;
 563 }
 564
 565 static void
 566 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 567 {
 568         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 569
 570         dprintk("--> %s\n", __func__);
 571
 572         if (lseg->pls_range.iomode == IOMODE_RW) {
 573                 struct nfs4_flexfile_layout *ffl;
 574                 struct inode *inode;
 575
 576                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 577                 inode = ffl->generic_hdr.plh_inode;
 578                 spin_lock(&inode->i_lock);
 579                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 580                         ffl->commit_info.nbuckets = 0;
 581                         kfree(ffl->commit_info.buckets);
 582                         ffl->commit_info.buckets = NULL;
 583                 }
 584                 spin_unlock(&inode->i_lock);
 585         }
 586         _ff_layout_free_lseg(fls);
 587 }
 588
 589 /* Return 1 until we have multiple lsegs support */
 590 static int
 591 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 592 {
 593         return 1;
 594 }
 595
 596 static void
 597 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 598 {
 599         /* first IO request? */
 600         if (atomic_inc_return(&timer->n_ops) == 1) {
 601                 timer->start_time = now;
 602         }
 603 }
 604
 605 static ktime_t
 606 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 607 {
 608         ktime_t start;
 609
 610         if (atomic_dec_return(&timer->n_ops) < 0)
 611                 WARN_ON_ONCE(1);
 612
 613         start = timer->start_time;
 614         timer->start_time = now;
 615         return ktime_sub(now, start);
 616 }
 617
 618 static bool
 619 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 620                             struct nfs4_ff_layoutstat *layoutstat,
 621                             ktime_t now)
 622 {
 623         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 624         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
 625
 626         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 627         if (!mirror->start_time)
 628                 mirror->start_time = now;
 629         if (mirror->report_interval != 0)
 630                 report_interval = (s64)mirror->report_interval * 1000LL;
 631         else if (layoutstats_timer != 0)
 632                 report_interval = (s64)layoutstats_timer * 1000LL;
 633         if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
 634                         report_interval) {
 635                 ffl->last_report_time = now;
 636                 return true;
 637         }
 638
 639         return false;
 640 }
 641
 642 static void
 643 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 644                 __u64 requested)
 645 {
 646         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 647
 648         iostat->ops_requested++;
 649         iostat->bytes_requested += requested;
 650 }
 651
 652 static void
 653 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 654                 __u64 requested,
 655                 __u64 completed,
 656                 ktime_t time_completed,
 657                 ktime_t time_started)
 658 {
 659         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 660         ktime_t completion_time = ktime_sub(time_completed, time_started);
 661         ktime_t timer;
 662
 663         iostat->ops_completed++;
 664         iostat->bytes_completed += completed;
 665         iostat->bytes_not_delivered += requested - completed;
 666
 667         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 668         iostat->total_busy_time =
 669                         ktime_add(iostat->total_busy_time, timer);
 670         iostat->aggregate_completion_time =
 671                         ktime_add(iostat->aggregate_completion_time,
 672                                         completion_time);
 673 }
 674
 675 static void
 676 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 677                 struct nfs4_ff_layout_mirror *mirror,
 678                 __u64 requested, ktime_t now)
 679 {
 680         bool report;
 681
 682         spin_lock(&mirror->lock);
 683         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 684         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 685         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 686         spin_unlock(&mirror->lock);
 687
 688         if (report)
 689                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 690 }
 691
 692 static void
 693 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 694                 struct nfs4_ff_layout_mirror *mirror,
 695                 __u64 requested,
 696                 __u64 completed)
 697 {
 698         spin_lock(&mirror->lock);
 699         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 700                         requested, completed,
 701                         ktime_get(), task->tk_start);
 702         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 703         spin_unlock(&mirror->lock);
 704 }
 705
 706 static void
 707 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 708                 struct nfs4_ff_layout_mirror *mirror,
 709                 __u64 requested, ktime_t now)
 710 {
 711         bool report;
 712
 713         spin_lock(&mirror->lock);
 714         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 715         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 716         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 717         spin_unlock(&mirror->lock);
 718
 719         if (report)
 720                 pnfs_report_layoutstat(inode, GFP_NOIO);
 721 }
 722
 723 static void
 724 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 725                 struct nfs4_ff_layout_mirror *mirror,
 726                 __u64 requested,
 727                 __u64 completed,
 728                 enum nfs3_stable_how committed)
 729 {
 730         if (committed == NFS_UNSTABLE)
 731                 requested = completed = 0;
 732
 733         spin_lock(&mirror->lock);
 734         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 735                         requested, completed, ktime_get(), task->tk_start);
 736         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 737         spin_unlock(&mirror->lock);
 738 }
 739
 740 static int
 741 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 742                             struct nfs_commit_info *cinfo,
 743                             gfp_t gfp_flags)
 744 {
 745         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 746         struct pnfs_commit_bucket *buckets;
 747         int size;
 748
 749         if (cinfo->ds->nbuckets != 0) {
 750                 /* This assumes there is only one RW lseg per file.
 751                  * To support multiple lseg per file, we need to
 752                  * change struct pnfs_commit_bucket to allow dynamic
 753                  * increasing nbuckets.
 754                  */
 755                 return 0;
 756         }
 757
 758         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 759
 760         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 761                           gfp_flags);
 762         if (!buckets)
 763                 return -ENOMEM;
 764         else {
 765                 int i;
 766
 767                 spin_lock(&cinfo->inode->i_lock);
 768                 if (cinfo->ds->nbuckets != 0)
 769                         kfree(buckets);
 770                 else {
 771                         cinfo->ds->buckets = buckets;
 772                         cinfo->ds->nbuckets = size;
 773                         for (i = 0; i < size; i++) {
 774                                 INIT_LIST_HEAD(&buckets[i].written);
 775                                 INIT_LIST_HEAD(&buckets[i].committing);
 776                                 /* mark direct verifier as unset */
 777                                 buckets[i].direct_verf.committed =
 778                                         NFS_INVALID_STABLE_HOW;
 779                         }
 780                 }
 781                 spin_unlock(&cinfo->inode->i_lock);
 782                 return 0;
 783         }
 784 }
 785
 786 static struct nfs4_pnfs_ds *
 787 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 788                                   int start_idx,
 789                                   int *best_idx)
 790 {
 791         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 792         struct nfs4_pnfs_ds *ds;
 793         bool fail_return = false;
 794         int idx;
 795
 796         /* mirrors are sorted by efficiency */
 797         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 798                 if (idx+1 == fls->mirror_array_cnt)
 799                         fail_return = true;
 800                 ds = nfs4_ff_layout_prepare_ds(lseg, idx, fail_return);
 801                 if (ds) {
 802                         *best_idx = idx;
 803                         return ds;
 804                 }
 805         }
 806
 807         return NULL;
 808 }
 809
 810 static void
 811 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 812                       struct nfs_page *req,
 813                       bool strict_iomode)
 814 {
 815 retry_strict:
 816         pnfs_put_lseg(pgio->pg_lseg);
 817         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 818                                            req->wb_context,
 819                                            0,
 820                                            NFS4_MAX_UINT64,
 821                                            IOMODE_READ,
 822                                            strict_iomode,
 823                                            GFP_KERNEL);
 824         if (IS_ERR(pgio->pg_lseg)) {
 825                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 826                 pgio->pg_lseg = NULL;
 827         }
 828
 829         /* If we don't have checking, do get a IOMODE_RW
 830          * segment, and the server wants to avoid READs
 831          * there, then retry!
 832          */
 833         if (pgio->pg_lseg && !strict_iomode &&
 834             ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 835                 strict_iomode = true;
 836                 goto retry_strict;
 837         }
 838 }
 839
 840 static void
 841 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 842                         struct nfs_page *req)
 843 {
 844         struct nfs_pgio_mirror *pgm;
 845         struct nfs4_ff_layout_mirror *mirror;
 846         struct nfs4_pnfs_ds *ds;
 847         int ds_idx;
 848
 849 retry:
 850         pnfs_generic_pg_check_layout(pgio);
 851         /* Use full layout for now */
 852         if (!pgio->pg_lseg)
 853                 ff_layout_pg_get_read(pgio, req, false);
 854         else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg))
 855                 ff_layout_pg_get_read(pgio, req, true);
 856
 857         /* If no lseg, fall back to read through mds */
 858         if (pgio->pg_lseg == NULL)
 859                 goto out_mds;
 860
 861         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 862         if (!ds) {
 863                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 864                         goto out_mds;
 865                 pnfs_put_lseg(pgio->pg_lseg);
 866                 pgio->pg_lseg = NULL;
 867                 /* Sleep for 1 second before retrying */
 868                 ssleep(1);
 869                 goto retry;
 870         }
 871
 872         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 873
 874         pgio->pg_mirror_idx = ds_idx;
 875
 876         /* read always uses only one mirror - idx 0 for pgio layer */
 877         pgm = &pgio->pg_mirrors[0];
 878         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 879
 880         return;
 881 out_mds:
 882         pnfs_put_lseg(pgio->pg_lseg);
 883         pgio->pg_lseg = NULL;
 884         nfs_pageio_reset_read_mds(pgio);
 885 }
 886
 887 static void
 888 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 889                         struct nfs_page *req)
 890 {
 891         struct nfs4_ff_layout_mirror *mirror;
 892         struct nfs_pgio_mirror *pgm;
 893         struct nfs_commit_info cinfo;
 894         struct nfs4_pnfs_ds *ds;
 895         int i;
 896         int status;
 897
 898 retry:
 899         pnfs_generic_pg_check_layout(pgio);
 900         if (!pgio->pg_lseg) {
 901                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 902                                                    req->wb_context,
 903                                                    0,
 904                                                    NFS4_MAX_UINT64,
 905                                                    IOMODE_RW,
 906                                                    false,
 907                                                    GFP_NOFS);
 908                 if (IS_ERR(pgio->pg_lseg)) {
 909                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 910                         pgio->pg_lseg = NULL;
 911                         return;
 912                 }
 913         }
 914         /* If no lseg, fall back to write through mds */
 915         if (pgio->pg_lseg == NULL)
 916                 goto out_mds;
 917
 918         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 919         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 920         if (status < 0)
 921                 goto out_mds;
 922
 923         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 924         if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
 925                 goto out_eagain;
 926
 927         for (i = 0; i < pgio->pg_mirror_count; i++) {
 928                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 929                 if (!ds) {
 930                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 931                                 goto out_mds;
 932                         pnfs_put_lseg(pgio->pg_lseg);
 933                         pgio->pg_lseg = NULL;
 934                         /* Sleep for 1 second before retrying */
 935                         ssleep(1);
 936                         goto retry;
 937                 }
 938                 pgm = &pgio->pg_mirrors[i];
 939                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 940                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 941         }
 942
 943         return;
 944 out_eagain:
 945         pnfs_generic_pg_cleanup(pgio);
 946         pgio->pg_error = -EAGAIN;
 947         return;
 948 out_mds:
 949         pnfs_put_lseg(pgio->pg_lseg);
 950         pgio->pg_lseg = NULL;
 951         nfs_pageio_reset_write_mds(pgio);
 952         pgio->pg_error = -EAGAIN;
 953 }
 954
 955 static unsigned int
 956 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 957                                     struct nfs_page *req)
 958 {
 959         if (!pgio->pg_lseg) {
 960                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 961                                                    req->wb_context,
 962                                                    0,
 963                                                    NFS4_MAX_UINT64,
 964                                                    IOMODE_RW,
 965                                                    false,
 966                                                    GFP_NOFS);
 967                 if (IS_ERR(pgio->pg_lseg)) {
 968                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 969                         pgio->pg_lseg = NULL;
 970                         goto out;
 971                 }
 972         }
 973         if (pgio->pg_lseg)
 974                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 975
 976         /* no lseg means that pnfs is not in use, so no mirroring here */
 977         nfs_pageio_reset_write_mds(pgio);
 978 out:
 979         return 1;
 980 }
 981
 982 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 983         .pg_init = ff_layout_pg_init_read,
 984         .pg_test = pnfs_generic_pg_test,
 985         .pg_doio = pnfs_generic_pg_readpages,
 986         .pg_cleanup = pnfs_generic_pg_cleanup,
 987 };
 988
 989 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 990         .pg_init = ff_layout_pg_init_write,
 991         .pg_test = pnfs_generic_pg_test,
 992         .pg_doio = pnfs_generic_pg_writepages,
 993         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 994         .pg_cleanup = pnfs_generic_pg_cleanup,
 995 };
 996
 997 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 998 {
 999         struct rpc_task *task = &hdr->task;
1000
1001         pnfs_layoutcommit_inode(hdr->inode, false);
1002
1003         if (retry_pnfs) {
1004                 dprintk("%s Reset task %5u for i/o through pNFS "
1005                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1006                         hdr->task.tk_pid,
1007                         hdr->inode->i_sb->s_id,
1008                         (unsigned long long)NFS_FILEID(hdr->inode),
1009                         hdr->args.count,
1010                         (unsigned long long)hdr->args.offset);
1011
1012                 hdr->completion_ops->reschedule_io(hdr);
1013                 return;
1014         }
1015
1016         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1017                 dprintk("%s Reset task %5u for i/o through MDS "
1018                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1019                         hdr->task.tk_pid,
1020                         hdr->inode->i_sb->s_id,
1021                         (unsigned long long)NFS_FILEID(hdr->inode),
1022                         hdr->args.count,
1023                         (unsigned long long)hdr->args.offset);
1024
1025                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1026         }
1027 }
1028
1029 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1030 {
1031         struct rpc_task *task = &hdr->task;
1032
1033         pnfs_layoutcommit_inode(hdr->inode, false);
1034
1035         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1036                 dprintk("%s Reset task %5u for i/o through MDS "
1037                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1038                         hdr->task.tk_pid,
1039                         hdr->inode->i_sb->s_id,
1040                         (unsigned long long)NFS_FILEID(hdr->inode),
1041                         hdr->args.count,
1042                         (unsigned long long)hdr->args.offset);
1043
1044                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1045         }
1046 }
1047
1048 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1049                                            struct nfs4_state *state,
1050                                            struct nfs_client *clp,
1051                                            struct pnfs_layout_segment *lseg,
1052                                            int idx)
1053 {
1054         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1055         struct inode *inode = lo->plh_inode;
1056         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1057         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1058
1059         switch (task->tk_status) {
1060         case -NFS4ERR_BADSESSION:
1061         case -NFS4ERR_BADSLOT:
1062         case -NFS4ERR_BAD_HIGH_SLOT:
1063         case -NFS4ERR_DEADSESSION:
1064         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1065         case -NFS4ERR_SEQ_FALSE_RETRY:
1066         case -NFS4ERR_SEQ_MISORDERED:
1067                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1068                         "flags 0x%x\n", __func__, task->tk_status,
1069                         clp->cl_exchange_flags);
1070                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1071                 break;
1072         case -NFS4ERR_DELAY:
1073         case -NFS4ERR_GRACE:
1074                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1075                 break;
1076         case -NFS4ERR_RETRY_UNCACHED_REP:
1077                 break;
1078         /* Invalidate Layout errors */
1079         case -NFS4ERR_PNFS_NO_LAYOUT:
1080         case -ESTALE:           /* mapped NFS4ERR_STALE */
1081         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1082         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1083         case -NFS4ERR_FHEXPIRED:
1084         case -NFS4ERR_WRONG_TYPE:
1085                 dprintk("%s Invalid layout error %d\n", __func__,
1086                         task->tk_status);
1087                 /*
1088                  * Destroy layout so new i/o will get a new layout.
1089                  * Layout will not be destroyed until all current lseg
1090                  * references are put. Mark layout as invalid to resend failed
1091                  * i/o and all i/o waiting on the slot table to the MDS until
1092                  * layout is destroyed and a new valid layout is obtained.
1093                  */
1094                 pnfs_destroy_layout(NFS_I(inode));
1095                 rpc_wake_up(&tbl->slot_tbl_waitq);
1096                 goto reset;
1097         /* RPC connection errors */
1098         case -ECONNREFUSED:
1099         case -EHOSTDOWN:
1100         case -EHOSTUNREACH:
1101         case -ENETUNREACH:
1102         case -EIO:
1103         case -ETIMEDOUT:
1104         case -EPIPE:
1105                 dprintk("%s DS connection error %d\n", __func__,
1106                         task->tk_status);
1107                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1108                                 &devid->deviceid);
1109                 rpc_wake_up(&tbl->slot_tbl_waitq);
1110                 /* fall through */
1111         default:
1112                 if (ff_layout_avoid_mds_available_ds(lseg))
1113                         return -NFS4ERR_RESET_TO_PNFS;
1114 reset:
1115                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1116                         task->tk_status);
1117                 return -NFS4ERR_RESET_TO_MDS;
1118         }
1119         task->tk_status = 0;
1120         return -EAGAIN;
1121 }
1122
1123 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1124 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1125                                            struct pnfs_layout_segment *lseg,
1126                                            int idx)
1127 {
1128         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1129
1130         switch (task->tk_status) {
1131         /* File access problems. Don't mark the device as unavailable */
1132         case -EACCES:
1133         case -ESTALE:
1134         case -EISDIR:
1135         case -EBADHANDLE:
1136         case -ELOOP:
1137         case -ENOSPC:
1138                 break;
1139         case -EJUKEBOX:
1140                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1141                 goto out_retry;
1142         default:
1143                 dprintk("%s DS connection error %d\n", __func__,
1144                         task->tk_status);
1145                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1146                                 &devid->deviceid);
1147         }
1148         /* FIXME: Need to prevent infinite looping here. */
1149         return -NFS4ERR_RESET_TO_PNFS;
1150 out_retry:
1151         task->tk_status = 0;
1152         rpc_restart_call_prepare(task);
1153         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1154         return -EAGAIN;
1155 }
1156
1157 static int ff_layout_async_handle_error(struct rpc_task *task,
1158                                         struct nfs4_state *state,
1159                                         struct nfs_client *clp,
1160                                         struct pnfs_layout_segment *lseg,
1161                                         int idx)
1162 {
1163         int vers = clp->cl_nfs_mod->rpc_vers->number;
1164
1165         if (task->tk_status >= 0)
1166                 return 0;
1167
1168         /* Handle the case of an invalid layout segment */
1169         if (!pnfs_is_valid_lseg(lseg))
1170                 return -NFS4ERR_RESET_TO_PNFS;
1171
1172         switch (vers) {
1173         case 3:
1174                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1175         case 4:
1176                 return ff_layout_async_handle_error_v4(task, state, clp,
1177                                                        lseg, idx);
1178         default:
1179                 /* should never happen */
1180                 WARN_ON_ONCE(1);
1181                 return 0;
1182         }
1183 }
1184
1185 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1186                                         int idx, u64 offset, u64 length,
1187                                         u32 status, int opnum, int error)
1188 {
1189         struct nfs4_ff_layout_mirror *mirror;
1190         int err;
1191
1192         if (status == 0) {
1193                 switch (error) {
1194                 case -ETIMEDOUT:
1195                 case -EPFNOSUPPORT:
1196                 case -EPROTONOSUPPORT:
1197                 case -EOPNOTSUPP:
1198                 case -ECONNREFUSED:
1199                 case -ECONNRESET:
1200                 case -EHOSTDOWN:
1201                 case -EHOSTUNREACH:
1202                 case -ENETUNREACH:
1203                 case -EADDRINUSE:
1204                 case -ENOBUFS:
1205                 case -EPIPE:
1206                 case -EPERM:
1207                         status = NFS4ERR_NXIO;
1208                         break;
1209                 case -EACCES:
1210                         status = NFS4ERR_ACCESS;
1211                         break;
1212                 default:
1213                         return;
1214                 }
1215         }
1216
1217         switch (status) {
1218         case NFS4ERR_DELAY:
1219         case NFS4ERR_GRACE:
1220                 return;
1221         default:
1222                 break;
1223         }
1224
1225         mirror = FF_LAYOUT_COMP(lseg, idx);
1226         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1227                                        mirror, offset, length, status, opnum,
1228                                        GFP_NOIO);
1229         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1230         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1231 }
1232
1233 /* NFS_PROTO call done callback routines */
1234 static int ff_layout_read_done_cb(struct rpc_task *task,
1235                                 struct nfs_pgio_header *hdr)
1236 {
1237         int err;
1238
1239         trace_nfs4_pnfs_read(hdr, task->tk_status);
1240         if (task->tk_status < 0)
1241                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1242                                             hdr->args.offset, hdr->args.count,
1243                                             hdr->res.op_status, OP_READ,
1244                                             task->tk_status);
1245         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1246                                            hdr->ds_clp, hdr->lseg,
1247                                            hdr->pgio_mirror_idx);
1248
1249         switch (err) {
1250         case -NFS4ERR_RESET_TO_PNFS:
1251                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1252                                         hdr->pgio_mirror_idx + 1,
1253                                         &hdr->pgio_mirror_idx))
1254                         goto out_eagain;
1255                 ff_layout_read_record_layoutstats_done(task, hdr);
1256                 pnfs_read_resend_pnfs(hdr);
1257                 return task->tk_status;
1258         case -NFS4ERR_RESET_TO_MDS:
1259                 ff_layout_reset_read(hdr);
1260                 return task->tk_status;
1261         case -EAGAIN:
1262                 goto out_eagain;
1263         }
1264
1265         return 0;
1266 out_eagain:
1267         rpc_restart_call_prepare(task);
1268         return -EAGAIN;
1269 }
1270
1271 static bool
1272 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1273 {
1274         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1275 }
1276
1277 /*
1278  * We reference the rpc_cred of the first WRITE that triggers the need for
1279  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1280  * rfc5661 is not clear about which credential should be used.
1281  *
1282  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1283  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1284  * we always send layoutcommit after DS writes.
1285  */
1286 static void
1287 ff_layout_set_layoutcommit(struct inode *inode,
1288                 struct pnfs_layout_segment *lseg,
1289                 loff_t end_offset)
1290 {
1291         if (!ff_layout_need_layoutcommit(lseg))
1292                 return;
1293
1294         pnfs_set_layoutcommit(inode, lseg, end_offset);
1295         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1296                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1297 }
1298
1299 static bool
1300 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx)
1301 {
1302         /* No mirroring for now */
1303         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1304
1305         return ff_layout_test_devid_unavailable(node);
1306 }
1307
1308 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1309                 struct nfs_pgio_header *hdr)
1310 {
1311         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1312                 return;
1313         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1314                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1315                         hdr->args.count,
1316                         task->tk_start);
1317 }
1318
1319 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1320                 struct nfs_pgio_header *hdr)
1321 {
1322         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1323                 return;
1324         nfs4_ff_layout_stat_io_end_read(task,
1325                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1326                         hdr->args.count,
1327                         hdr->res.count);
1328 }
1329
1330 static int ff_layout_read_prepare_common(struct rpc_task *task,
1331                                          struct nfs_pgio_header *hdr)
1332 {
1333         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1334                 rpc_exit(task, -EIO);
1335                 return -EIO;
1336         }
1337         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1338                 rpc_exit(task, -EHOSTDOWN);
1339                 return -EAGAIN;
1340         }
1341
1342         ff_layout_read_record_layoutstats_start(task, hdr);
1343         return 0;
1344 }
1345
1346 /*
1347  * Call ops for the async read/write cases
1348  * In the case of dense layouts, the offset needs to be reset to its
1349  * original value.
1350  */
1351 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1352 {
1353         struct nfs_pgio_header *hdr = data;
1354
1355         if (ff_layout_read_prepare_common(task, hdr))
1356                 return;
1357
1358         rpc_call_start(task);
1359 }
1360
1361 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1362 {
1363         struct nfs_pgio_header *hdr = data;
1364
1365         if (nfs4_setup_sequence(hdr->ds_clp,
1366                                 &hdr->args.seq_args,
1367                                 &hdr->res.seq_res,
1368                                 task))
1369                 return;
1370
1371         ff_layout_read_prepare_common(task, hdr);
1372 }
1373
1374 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1375 {
1376         struct nfs_pgio_header *hdr = data;
1377
1378         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1379
1380         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1381             task->tk_status == 0) {
1382                 nfs4_sequence_done(task, &hdr->res.seq_res);
1383                 return;
1384         }
1385
1386         /* Note this may cause RPC to be resent */
1387         hdr->mds_ops->rpc_call_done(task, hdr);
1388 }
1389
1390 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1391 {
1392         struct nfs_pgio_header *hdr = data;
1393
1394         ff_layout_read_record_layoutstats_done(task, hdr);
1395         rpc_count_iostats_metrics(task,
1396             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1397 }
1398
1399 static void ff_layout_read_release(void *data)
1400 {
1401         struct nfs_pgio_header *hdr = data;
1402
1403         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1404         pnfs_generic_rw_release(data);
1405 }
1406
1407
1408 static int ff_layout_write_done_cb(struct rpc_task *task,
1409                                 struct nfs_pgio_header *hdr)
1410 {
1411         loff_t end_offs = 0;
1412         int err;
1413
1414         trace_nfs4_pnfs_write(hdr, task->tk_status);
1415         if (task->tk_status < 0)
1416                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1417                                             hdr->args.offset, hdr->args.count,
1418                                             hdr->res.op_status, OP_WRITE,
1419                                             task->tk_status);
1420         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1421                                            hdr->ds_clp, hdr->lseg,
1422                                            hdr->pgio_mirror_idx);
1423
1424         switch (err) {
1425         case -NFS4ERR_RESET_TO_PNFS:
1426                 ff_layout_reset_write(hdr, true);
1427                 return task->tk_status;
1428         case -NFS4ERR_RESET_TO_MDS:
1429                 ff_layout_reset_write(hdr, false);
1430                 return task->tk_status;
1431         case -EAGAIN:
1432                 return -EAGAIN;
1433         }
1434
1435         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1436             hdr->res.verf->committed == NFS_DATA_SYNC)
1437                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1438
1439         /* Note: if the write is unstable, don't set end_offs until commit */
1440         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1441
1442         /* zero out fattr since we don't care DS attr at all */
1443         hdr->fattr.valid = 0;
1444         if (task->tk_status >= 0)
1445                 nfs_writeback_update_inode(hdr);
1446
1447         return 0;
1448 }
1449
1450 static int ff_layout_commit_done_cb(struct rpc_task *task,
1451                                      struct nfs_commit_data *data)
1452 {
1453         int err;
1454
1455         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1456         if (task->tk_status < 0)
1457                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1458                                             data->args.offset, data->args.count,
1459                                             data->res.op_status, OP_COMMIT,
1460                                             task->tk_status);
1461         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1462                                            data->lseg, data->ds_commit_index);
1463
1464         switch (err) {
1465         case -NFS4ERR_RESET_TO_PNFS:
1466                 pnfs_generic_prepare_to_resend_writes(data);
1467                 return -EAGAIN;
1468         case -NFS4ERR_RESET_TO_MDS:
1469                 pnfs_generic_prepare_to_resend_writes(data);
1470                 return -EAGAIN;
1471         case -EAGAIN:
1472                 rpc_restart_call_prepare(task);
1473                 return -EAGAIN;
1474         }
1475
1476         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1477
1478         return 0;
1479 }
1480
1481 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1482                 struct nfs_pgio_header *hdr)
1483 {
1484         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1485                 return;
1486         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1487                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1488                         hdr->args.count,
1489                         task->tk_start);
1490 }
1491
1492 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1493                 struct nfs_pgio_header *hdr)
1494 {
1495         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1496                 return;
1497         nfs4_ff_layout_stat_io_end_write(task,
1498                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1499                         hdr->args.count, hdr->res.count,
1500                         hdr->res.verf->committed);
1501 }
1502
1503 static int ff_layout_write_prepare_common(struct rpc_task *task,
1504                                           struct nfs_pgio_header *hdr)
1505 {
1506         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1507                 rpc_exit(task, -EIO);
1508                 return -EIO;
1509         }
1510
1511         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1512                 rpc_exit(task, -EHOSTDOWN);
1513                 return -EAGAIN;
1514         }
1515
1516         ff_layout_write_record_layoutstats_start(task, hdr);
1517         return 0;
1518 }
1519
1520 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1521 {
1522         struct nfs_pgio_header *hdr = data;
1523
1524         if (ff_layout_write_prepare_common(task, hdr))
1525                 return;
1526
1527         rpc_call_start(task);
1528 }
1529
1530 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1531 {
1532         struct nfs_pgio_header *hdr = data;
1533
1534         if (nfs4_setup_sequence(hdr->ds_clp,
1535                                 &hdr->args.seq_args,
1536                                 &hdr->res.seq_res,
1537                                 task))
1538                 return;
1539
1540         ff_layout_write_prepare_common(task, hdr);
1541 }
1542
1543 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1544 {
1545         struct nfs_pgio_header *hdr = data;
1546
1547         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1548             task->tk_status == 0) {
1549                 nfs4_sequence_done(task, &hdr->res.seq_res);
1550                 return;
1551         }
1552
1553         /* Note this may cause RPC to be resent */
1554         hdr->mds_ops->rpc_call_done(task, hdr);
1555 }
1556
1557 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1558 {
1559         struct nfs_pgio_header *hdr = data;
1560
1561         ff_layout_write_record_layoutstats_done(task, hdr);
1562         rpc_count_iostats_metrics(task,
1563             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1564 }
1565
1566 static void ff_layout_write_release(void *data)
1567 {
1568         struct nfs_pgio_header *hdr = data;
1569
1570         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1571         pnfs_generic_rw_release(data);
1572 }
1573
1574 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1575                 struct nfs_commit_data *cdata)
1576 {
1577         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1578                 return;
1579         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1580                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1581                         0, task->tk_start);
1582 }
1583
1584 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1585                 struct nfs_commit_data *cdata)
1586 {
1587         struct nfs_page *req;
1588         __u64 count = 0;
1589
1590         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1591                 return;
1592
1593         if (task->tk_status == 0) {
1594                 list_for_each_entry(req, &cdata->pages, wb_list)
1595                         count += req->wb_bytes;
1596         }
1597         nfs4_ff_layout_stat_io_end_write(task,
1598                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1599                         count, count, NFS_FILE_SYNC);
1600 }
1601
1602 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1603                 struct nfs_commit_data *cdata)
1604 {
1605         ff_layout_commit_record_layoutstats_start(task, cdata);
1606 }
1607
1608 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1609 {
1610         ff_layout_commit_prepare_common(task, data);
1611         rpc_call_start(task);
1612 }
1613
1614 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1615 {
1616         struct nfs_commit_data *wdata = data;
1617
1618         if (nfs4_setup_sequence(wdata->ds_clp,
1619                                 &wdata->args.seq_args,
1620                                 &wdata->res.seq_res,
1621                                 task))
1622                 return;
1623         ff_layout_commit_prepare_common(task, data);
1624 }
1625
1626 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1627 {
1628         pnfs_generic_write_commit_done(task, data);
1629 }
1630
1631 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1632 {
1633         struct nfs_commit_data *cdata = data;
1634
1635         ff_layout_commit_record_layoutstats_done(task, cdata);
1636         rpc_count_iostats_metrics(task,
1637             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1638 }
1639
1640 static void ff_layout_commit_release(void *data)
1641 {
1642         struct nfs_commit_data *cdata = data;
1643
1644         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1645         pnfs_generic_commit_release(data);
1646 }
1647
1648 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1649         .rpc_call_prepare = ff_layout_read_prepare_v3,
1650         .rpc_call_done = ff_layout_read_call_done,
1651         .rpc_count_stats = ff_layout_read_count_stats,
1652         .rpc_release = ff_layout_read_release,
1653 };
1654
1655 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1656         .rpc_call_prepare = ff_layout_read_prepare_v4,
1657         .rpc_call_done = ff_layout_read_call_done,
1658         .rpc_count_stats = ff_layout_read_count_stats,
1659         .rpc_release = ff_layout_read_release,
1660 };
1661
1662 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1663         .rpc_call_prepare = ff_layout_write_prepare_v3,
1664         .rpc_call_done = ff_layout_write_call_done,
1665         .rpc_count_stats = ff_layout_write_count_stats,
1666         .rpc_release = ff_layout_write_release,
1667 };
1668
1669 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1670         .rpc_call_prepare = ff_layout_write_prepare_v4,
1671         .rpc_call_done = ff_layout_write_call_done,
1672         .rpc_count_stats = ff_layout_write_count_stats,
1673         .rpc_release = ff_layout_write_release,
1674 };
1675
1676 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1677         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1678         .rpc_call_done = ff_layout_commit_done,
1679         .rpc_count_stats = ff_layout_commit_count_stats,
1680         .rpc_release = ff_layout_commit_release,
1681 };
1682
1683 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1684         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1685         .rpc_call_done = ff_layout_commit_done,
1686         .rpc_count_stats = ff_layout_commit_count_stats,
1687         .rpc_release = ff_layout_commit_release,
1688 };
1689
1690 static enum pnfs_try_status
1691 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1692 {
1693         struct pnfs_layout_segment *lseg = hdr->lseg;
1694         struct nfs4_pnfs_ds *ds;
1695         struct rpc_clnt *ds_clnt;
1696         struct rpc_cred *ds_cred;
1697         loff_t offset = hdr->args.offset;
1698         u32 idx = hdr->pgio_mirror_idx;
1699         int vers;
1700         struct nfs_fh *fh;
1701
1702         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1703                 __func__, hdr->inode->i_ino,
1704                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1705
1706         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1707         if (!ds)
1708                 goto out_failed;
1709
1710         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1711                                                    hdr->inode);
1712         if (IS_ERR(ds_clnt))
1713                 goto out_failed;
1714
1715         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1716         if (!ds_cred)
1717                 goto out_failed;
1718
1719         vers = nfs4_ff_layout_ds_version(lseg, idx);
1720
1721         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1722                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1723
1724         hdr->pgio_done_cb = ff_layout_read_done_cb;
1725         atomic_inc(&ds->ds_clp->cl_count);
1726         hdr->ds_clp = ds->ds_clp;
1727         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1728         if (fh)
1729                 hdr->args.fh = fh;
1730
1731         if (vers == 4 &&
1732                 !nfs4_ff_layout_select_ds_stateid(lseg, idx, &hdr->args.stateid))
1733                 goto out_failed;
1734
1735         /*
1736          * Note that if we ever decide to split across DSes,
1737          * then we may need to handle dense-like offsets.
1738          */
1739         hdr->args.offset = offset;
1740         hdr->mds_offset = offset;
1741
1742         /* Perform an asynchronous read to ds */
1743         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1744                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1745                                       &ff_layout_read_call_ops_v4,
1746                           0, RPC_TASK_SOFTCONN);
1747         put_rpccred(ds_cred);
1748         return PNFS_ATTEMPTED;
1749
1750 out_failed:
1751         if (ff_layout_avoid_mds_available_ds(lseg))
1752                 return PNFS_TRY_AGAIN;
1753         return PNFS_NOT_ATTEMPTED;
1754 }
1755
1756 /* Perform async writes. */
1757 static enum pnfs_try_status
1758 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1759 {
1760         struct pnfs_layout_segment *lseg = hdr->lseg;
1761         struct nfs4_pnfs_ds *ds;
1762         struct rpc_clnt *ds_clnt;
1763         struct rpc_cred *ds_cred;
1764         loff_t offset = hdr->args.offset;
1765         int vers;
1766         struct nfs_fh *fh;
1767         int idx = hdr->pgio_mirror_idx;
1768
1769         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1770         if (!ds)
1771                 goto out_failed;
1772
1773         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1774                                                    hdr->inode);
1775         if (IS_ERR(ds_clnt))
1776                 goto out_failed;
1777
1778         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1779         if (!ds_cred)
1780                 goto out_failed;
1781
1782         vers = nfs4_ff_layout_ds_version(lseg, idx);
1783
1784         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1785                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1786                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1787                 vers);
1788
1789         hdr->pgio_done_cb = ff_layout_write_done_cb;
1790         atomic_inc(&ds->ds_clp->cl_count);
1791         hdr->ds_clp = ds->ds_clp;
1792         hdr->ds_commit_idx = idx;
1793         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1794         if (fh)
1795                 hdr->args.fh = fh;
1796
1797         if (vers == 4 &&
1798                 !nfs4_ff_layout_select_ds_stateid(lseg, idx, &hdr->args.stateid))
1799                 goto out_failed;
1800
1801         /*
1802          * Note that if we ever decide to split across DSes,
1803          * then we may need to handle dense-like offsets.
1804          */
1805         hdr->args.offset = offset;
1806
1807         /* Perform an asynchronous write */
1808         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1809                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1810                                       &ff_layout_write_call_ops_v4,
1811                           sync, RPC_TASK_SOFTCONN);
1812         put_rpccred(ds_cred);
1813         return PNFS_ATTEMPTED;
1814
1815 out_failed:
1816         if (ff_layout_avoid_mds_available_ds(lseg))
1817                 return PNFS_TRY_AGAIN;
1818         return PNFS_NOT_ATTEMPTED;
1819 }
1820
1821 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1822 {
1823         return i;
1824 }
1825
1826 static struct nfs_fh *
1827 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1828 {
1829         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1830
1831         /* FIXME: Assume that there is only one NFS version available
1832          * for the DS.
1833          */
1834         return &flseg->mirror_array[i]->fh_versions[0];
1835 }
1836
1837 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1838 {
1839         struct pnfs_layout_segment *lseg = data->lseg;
1840         struct nfs4_pnfs_ds *ds;
1841         struct rpc_clnt *ds_clnt;
1842         struct rpc_cred *ds_cred;
1843         u32 idx;
1844         int vers, ret;
1845         struct nfs_fh *fh;
1846
1847         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1848             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1849                 goto out_err;
1850
1851         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1852         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1853         if (!ds)
1854                 goto out_err;
1855
1856         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1857                                                    data->inode);
1858         if (IS_ERR(ds_clnt))
1859                 goto out_err;
1860
1861         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1862         if (!ds_cred)
1863                 goto out_err;
1864
1865         vers = nfs4_ff_layout_ds_version(lseg, idx);
1866
1867         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1868                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1869                 vers);
1870         data->commit_done_cb = ff_layout_commit_done_cb;
1871         data->cred = ds_cred;
1872         atomic_inc(&ds->ds_clp->cl_count);
1873         data->ds_clp = ds->ds_clp;
1874         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1875         if (fh)
1876                 data->args.fh = fh;
1877
1878         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1879                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1880                                                &ff_layout_commit_call_ops_v4,
1881                                    how, RPC_TASK_SOFTCONN);
1882         put_rpccred(ds_cred);
1883         return ret;
1884 out_err:
1885         pnfs_generic_prepare_to_resend_writes(data);
1886         pnfs_generic_commit_release(data);
1887         return -EAGAIN;
1888 }
1889
1890 static int
1891 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1892                            int how, struct nfs_commit_info *cinfo)
1893 {
1894         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1895                                             ff_layout_initiate_commit);
1896 }
1897
1898 static struct pnfs_ds_commit_info *
1899 ff_layout_get_ds_info(struct inode *inode)
1900 {
1901         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1902
1903         if (layout == NULL)
1904                 return NULL;
1905
1906         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1907 }
1908
1909 static void
1910 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1911 {
1912         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1913                                                   id_node));
1914 }
1915
1916 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1917                                   const struct nfs4_layoutreturn_args *args,
1918                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1919 {
1920         __be32 *start;
1921
1922         start = xdr_reserve_space(xdr, 4);
1923         if (unlikely(!start))
1924                 return -E2BIG;
1925
1926         *start = cpu_to_be32(ff_args->num_errors);
1927         /* This assume we always return _ALL_ layouts */
1928         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
1929 }
1930
1931 static void
1932 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1933 {
1934         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
1935 }
1936
1937 static void
1938 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
1939                             const nfs4_stateid *stateid,
1940                             const struct nfs42_layoutstat_devinfo *devinfo)
1941 {
1942         __be32 *p;
1943
1944         p = xdr_reserve_space(xdr, 8 + 8);
1945         p = xdr_encode_hyper(p, devinfo->offset);
1946         p = xdr_encode_hyper(p, devinfo->length);
1947         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
1948         p = xdr_reserve_space(xdr, 4*8);
1949         p = xdr_encode_hyper(p, devinfo->read_count);
1950         p = xdr_encode_hyper(p, devinfo->read_bytes);
1951         p = xdr_encode_hyper(p, devinfo->write_count);
1952         p = xdr_encode_hyper(p, devinfo->write_bytes);
1953         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
1954 }
1955
1956 static void
1957 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
1958                             const nfs4_stateid *stateid,
1959                             const struct nfs42_layoutstat_devinfo *devinfo)
1960 {
1961         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
1962         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
1963                         devinfo->ld_private.data);
1964 }
1965
1966 /* report nothing for now */
1967 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
1968                 const struct nfs4_layoutreturn_args *args,
1969                 struct nfs4_flexfile_layoutreturn_args *ff_args)
1970 {
1971         __be32 *p;
1972         int i;
1973
1974         p = xdr_reserve_space(xdr, 4);
1975         *p = cpu_to_be32(ff_args->num_dev);
1976         for (i = 0; i < ff_args->num_dev; i++)
1977                 ff_layout_encode_ff_iostat(xdr,
1978                                 &args->layout->plh_stateid,
1979                                 &ff_args->devinfo[i]);
1980 }
1981
1982 static void
1983 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
1984                 unsigned int num_entries)
1985 {
1986         unsigned int i;
1987
1988         for (i = 0; i < num_entries; i++) {
1989                 if (!devinfo[i].ld_private.ops)
1990                         continue;
1991                 if (!devinfo[i].ld_private.ops->free)
1992                         continue;
1993                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
1994         }
1995 }
1996
1997 static struct nfs4_deviceid_node *
1998 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1999                               struct pnfs_device *pdev, gfp_t gfp_flags)
2000 {
2001         struct nfs4_ff_layout_ds *dsaddr;
2002
2003         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2004         if (!dsaddr)
2005                 return NULL;
2006         return &dsaddr->id_node;
2007 }
2008
2009 static void
2010 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2011                 const void *voidargs,
2012                 const struct nfs4_xdr_opaque_data *ff_opaque)
2013 {
2014         const struct nfs4_layoutreturn_args *args = voidargs;
2015         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2016         struct xdr_buf tmp_buf = {
2017                 .head = {
2018                         [0] = {
2019                                 .iov_base = page_address(ff_args->pages[0]),
2020                         },
2021                 },
2022                 .buflen = PAGE_SIZE,
2023         };
2024         struct xdr_stream tmp_xdr;
2025         __be32 *start;
2026
2027         dprintk("%s: Begin\n", __func__);
2028
2029         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL);
2030
2031         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2032         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2033
2034         start = xdr_reserve_space(xdr, 4);
2035         *start = cpu_to_be32(tmp_buf.len);
2036         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2037
2038         dprintk("%s: Return\n", __func__);
2039 }
2040
2041 static void
2042 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2043 {
2044         struct nfs4_flexfile_layoutreturn_args *ff_args;
2045
2046         if (!args->data)
2047                 return;
2048         ff_args = args->data;
2049         args->data = NULL;
2050
2051         ff_layout_free_ds_ioerr(&ff_args->errors);
2052         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2053
2054         put_page(ff_args->pages[0]);
2055         kfree(ff_args);
2056 }
2057
2058 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2059         .encode = ff_layout_encode_layoutreturn,
2060         .free = ff_layout_free_layoutreturn,
2061 };
2062
2063 static int
2064 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2065 {
2066         struct nfs4_flexfile_layoutreturn_args *ff_args;
2067         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2068
2069         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2070         if (!ff_args)
2071                 goto out_nomem;
2072         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2073         if (!ff_args->pages[0])
2074                 goto out_nomem_free;
2075
2076         INIT_LIST_HEAD(&ff_args->errors);
2077         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2078                         &args->range, &ff_args->errors,
2079                         FF_LAYOUTRETURN_MAXERR);
2080
2081         spin_lock(&args->inode->i_lock);
2082         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2083                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2084         spin_unlock(&args->inode->i_lock);
2085
2086         args->ld_private->ops = &layoutreturn_ops;
2087         args->ld_private->data = ff_args;
2088         return 0;
2089 out_nomem_free:
2090         kfree(ff_args);
2091 out_nomem:
2092         return -ENOMEM;
2093 }
2094
2095 static int
2096 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2097 {
2098         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2099
2100         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2101 }
2102
2103 static size_t
2104 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2105                           const int buflen)
2106 {
2107         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2108         const struct in6_addr *addr = &sin6->sin6_addr;
2109
2110         /*
2111          * RFC 4291, Section 2.2.2
2112          *
2113          * Shorthanded ANY address
2114          */
2115         if (ipv6_addr_any(addr))
2116                 return snprintf(buf, buflen, "::");
2117
2118         /*
2119          * RFC 4291, Section 2.2.2
2120          *
2121          * Shorthanded loopback address
2122          */
2123         if (ipv6_addr_loopback(addr))
2124                 return snprintf(buf, buflen, "::1");
2125
2126         /*
2127          * RFC 4291, Section 2.2.3
2128          *
2129          * Special presentation address format for mapped v4
2130          * addresses.
2131          */
2132         if (ipv6_addr_v4mapped(addr))
2133                 return snprintf(buf, buflen, "::ffff:%pI4",
2134                                         &addr->s6_addr32[3]);
2135
2136         /*
2137          * RFC 4291, Section 2.2.1
2138          */
2139         return snprintf(buf, buflen, "%pI6c", addr);
2140 }
2141
2142 /* Derived from rpc_sockaddr2uaddr */
2143 static void
2144 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2145 {
2146         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2147         char portbuf[RPCBIND_MAXUADDRPLEN];
2148         char addrbuf[RPCBIND_MAXUADDRLEN];
2149         char *netid;
2150         unsigned short port;
2151         int len, netid_len;
2152         __be32 *p;
2153
2154         switch (sap->sa_family) {
2155         case AF_INET:
2156                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2157                         return;
2158                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2159                 netid = "tcp";
2160                 netid_len = 3;
2161                 break;
2162         case AF_INET6:
2163                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2164                         return;
2165                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2166                 netid = "tcp6";
2167                 netid_len = 4;
2168                 break;
2169         default:
2170                 /* we only support tcp and tcp6 */
2171                 WARN_ON_ONCE(1);
2172                 return;
2173         }
2174
2175         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2176         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2177
2178         p = xdr_reserve_space(xdr, 4 + netid_len);
2179         xdr_encode_opaque(p, netid, netid_len);
2180
2181         p = xdr_reserve_space(xdr, 4 + len);
2182         xdr_encode_opaque(p, addrbuf, len);
2183 }
2184
2185 static void
2186 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2187                          ktime_t t)
2188 {
2189         struct timespec64 ts;
2190         __be32 *p;
2191
2192         p = xdr_reserve_space(xdr, 12);
2193         ts = ktime_to_timespec64(t);
2194         p = xdr_encode_hyper(p, ts.tv_sec);
2195         *p++ = cpu_to_be32(ts.tv_nsec);
2196 }
2197
2198 static void
2199 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2200                             struct nfs4_ff_io_stat *stat)
2201 {
2202         __be32 *p;
2203
2204         p = xdr_reserve_space(xdr, 5 * 8);
2205         p = xdr_encode_hyper(p, stat->ops_requested);
2206         p = xdr_encode_hyper(p, stat->bytes_requested);
2207         p = xdr_encode_hyper(p, stat->ops_completed);
2208         p = xdr_encode_hyper(p, stat->bytes_completed);
2209         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2210         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2211         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2212 }
2213
2214 static void
2215 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2216                               const struct nfs42_layoutstat_devinfo *devinfo,
2217                               struct nfs4_ff_layout_mirror *mirror)
2218 {
2219         struct nfs4_pnfs_ds_addr *da;
2220         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2221         struct nfs_fh *fh = &mirror->fh_versions[0];
2222         __be32 *p;
2223
2224         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2225         dprintk("%s: DS %s: encoding address %s\n",
2226                 __func__, ds->ds_remotestr, da->da_remotestr);
2227         /* netaddr4 */
2228         ff_layout_encode_netaddr(xdr, da);
2229         /* nfs_fh4 */
2230         p = xdr_reserve_space(xdr, 4 + fh->size);
2231         xdr_encode_opaque(p, fh->data, fh->size);
2232         /* ff_io_latency4 read */
2233         spin_lock(&mirror->lock);
2234         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2235         /* ff_io_latency4 write */
2236         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2237         spin_unlock(&mirror->lock);
2238         /* nfstime4 */
2239         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2240         /* bool */
2241         p = xdr_reserve_space(xdr, 4);
2242         *p = cpu_to_be32(false);
2243 }
2244
2245 static void
2246 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2247                              const struct nfs4_xdr_opaque_data *opaque)
2248 {
2249         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2250                         struct nfs42_layoutstat_devinfo, ld_private);
2251         __be32 *start;
2252
2253         /* layoutupdate length */
2254         start = xdr_reserve_space(xdr, 4);
2255         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2256
2257         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2258 }
2259
2260 static void
2261 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2262 {
2263         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2264
2265         ff_layout_put_mirror(mirror);
2266 }
2267
2268 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2269         .encode = ff_layout_encode_layoutstats,
2270         .free   = ff_layout_free_layoutstats,
2271 };
2272
2273 static int
2274 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2275                                struct nfs42_layoutstat_devinfo *devinfo,
2276                                int dev_limit)
2277 {
2278         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2279         struct nfs4_ff_layout_mirror *mirror;
2280         struct nfs4_deviceid_node *dev;
2281         int i = 0;
2282
2283         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2284                 if (i >= dev_limit)
2285                         break;
2286                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2287                         continue;
2288                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2289                         continue;
2290                 /* mirror refcount put in cleanup_layoutstats */
2291                 if (!atomic_inc_not_zero(&mirror->ref))
2292                         continue;
2293                 dev = &mirror->mirror_ds->id_node;
2294                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2295                 devinfo->offset = 0;
2296                 devinfo->length = NFS4_MAX_UINT64;
2297                 spin_lock(&mirror->lock);
2298                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2299                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2300                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2301                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2302                 spin_unlock(&mirror->lock);
2303                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2304                 devinfo->ld_private.ops = &layoutstat_ops;
2305                 devinfo->ld_private.data = mirror;
2306
2307                 devinfo++;
2308                 i++;
2309         }
2310         return i;
2311 }
2312
2313 static int
2314 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2315 {
2316         struct nfs4_flexfile_layout *ff_layout;
2317         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2318
2319         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2320         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2321         if (!args->devinfo)
2322                 return -ENOMEM;
2323
2324         spin_lock(&args->inode->i_lock);
2325         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2326         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2327                         &args->devinfo[0], dev_count);
2328         spin_unlock(&args->inode->i_lock);
2329         if (!args->num_dev) {
2330                 kfree(args->devinfo);
2331                 args->devinfo = NULL;
2332                 return -ENOENT;
2333         }
2334
2335         return 0;
2336 }
2337
2338 static int
2339 ff_layout_set_layoutdriver(struct nfs_server *server,
2340                 const struct nfs_fh *dummy)
2341 {
2342 #if IS_ENABLED(CONFIG_NFS_V4_2)
2343         server->caps |= NFS_CAP_LAYOUTSTATS;
2344 #endif
2345         return 0;
2346 }
2347
2348 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2349         .id                     = LAYOUT_FLEX_FILES,
2350         .name                   = "LAYOUT_FLEX_FILES",
2351         .owner                  = THIS_MODULE,
2352         .set_layoutdriver       = ff_layout_set_layoutdriver,
2353         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2354         .free_layout_hdr        = ff_layout_free_layout_hdr,
2355         .alloc_lseg             = ff_layout_alloc_lseg,
2356         .free_lseg              = ff_layout_free_lseg,
2357         .add_lseg               = ff_layout_add_lseg,
2358         .pg_read_ops            = &ff_layout_pg_read_ops,
2359         .pg_write_ops           = &ff_layout_pg_write_ops,
2360         .get_ds_info            = ff_layout_get_ds_info,
2361         .free_deviceid_node     = ff_layout_free_deviceid_node,
2362         .mark_request_commit    = pnfs_layout_mark_request_commit,
2363         .clear_request_commit   = pnfs_generic_clear_request_commit,
2364         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2365         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2366         .commit_pagelist        = ff_layout_commit_pagelist,
2367         .read_pagelist          = ff_layout_read_pagelist,
2368         .write_pagelist         = ff_layout_write_pagelist,
2369         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2370         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2371         .sync                   = pnfs_nfs_generic_sync,
2372         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2373 };
2374
2375 static int __init nfs4flexfilelayout_init(void)
2376 {
2377         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2378                __func__);
2379         if (!ff_zero_group) {
2380                 ff_zero_group = groups_alloc(0);
2381                 if (!ff_zero_group)
2382                         return -ENOMEM;
2383         }
2384         return pnfs_register_layoutdriver(&flexfilelayout_type);
2385 }
2386
2387 static void __exit nfs4flexfilelayout_exit(void)
2388 {
2389         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2390                __func__);
2391         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2392         if (ff_zero_group) {
2393                 put_group_info(ff_zero_group);
2394                 ff_zero_group = NULL;
2395         }
2396 }
2397
2398 MODULE_ALIAS("nfs-layouttype4-4");
2399
2400 MODULE_LICENSE("GPL");
2401 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2402
2403 module_init(nfs4flexfilelayout_init);
2404 module_exit(nfs4flexfilelayout_exit);