fs/nfs/flexfilelayout/flexfilelayout.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Module for pnfs flexfile layout driver.
   4  *
   5  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   6  *
   7  * Tao Peng <bergwolf@primarydata.com>
   8  */
   9
  10 #include <linux/nfs_fs.h>
  11 #include <linux/nfs_mount.h>
  12 #include <linux/nfs_page.h>
  13 #include <linux/module.h>
  14 #include <linux/sched/mm.h>
  15
  16 #include <linux/sunrpc/metrics.h>
  17
  18 #include "flexfilelayout.h"
  19 #include "../nfs4session.h"
  20 #include "../nfs4idmap.h"
  21 #include "../internal.h"
  22 #include "../delegation.h"
  23 #include "../nfs4trace.h"
  24 #include "../iostat.h"
  25 #include "../nfs.h"
  26 #include "../nfs42.h"
  27
  28 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  29
  30 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  31 #define FF_LAYOUTRETURN_MAXERR 20
  32
  33 static unsigned short io_maxretrans;
  34
  35 static const struct pnfs_commit_ops ff_layout_commit_ops;
  36 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
  37                 struct nfs_pgio_header *hdr);
  38 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
  39                                struct nfs42_layoutstat_devinfo *devinfo,
  40                                int dev_limit);
  41 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
  42                               const struct nfs42_layoutstat_devinfo *devinfo,
  43                               struct nfs4_ff_layout_mirror *mirror);
  44
  45 static struct pnfs_layout_hdr *
  46 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  47 {
  48         struct nfs4_flexfile_layout *ffl;
  49
  50         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  51         if (ffl) {
  52                 pnfs_init_ds_commit_info(&ffl->commit_info);
  53                 INIT_LIST_HEAD(&ffl->error_list);
  54                 INIT_LIST_HEAD(&ffl->mirrors);
  55                 ffl->last_report_time = ktime_get();
  56                 ffl->commit_info.ops = &ff_layout_commit_ops;
  57                 return &ffl->generic_hdr;
  58         } else
  59                 return NULL;
  60 }
  61
  62 static void
  63 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  64 {
  65         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
  66         struct nfs4_ff_layout_ds_err *err, *n;
  67
  68         list_for_each_entry_safe(err, n, &ffl->error_list, list) {
  69                 list_del(&err->list);
  70                 kfree(err);
  71         }
  72         kfree_rcu(ffl, generic_hdr.plh_rcu);
  73 }
  74
  75 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  76 {
  77         __be32 *p;
  78
  79         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  80         if (unlikely(p == NULL))
  81                 return -ENOBUFS;
  82         stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
  83         memcpy(stateid->data, p, NFS4_STATEID_SIZE);
  84         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  85                 p[0], p[1], p[2], p[3]);
  86         return 0;
  87 }
  88
  89 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  90 {
  91         __be32 *p;
  92
  93         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  94         if (unlikely(!p))
  95                 return -ENOBUFS;
  96         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  97         nfs4_print_deviceid(devid);
  98         return 0;
  99 }
 100
 101 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
 102 {
 103         __be32 *p;
 104
 105         p = xdr_inline_decode(xdr, 4);
 106         if (unlikely(!p))
 107                 return -ENOBUFS;
 108         fh->size = be32_to_cpup(p++);
 109         if (fh->size > sizeof(struct nfs_fh)) {
 110                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
 111                        fh->size);
 112                 return -EOVERFLOW;
 113         }
 114         /* fh.data */
 115         p = xdr_inline_decode(xdr, fh->size);
 116         if (unlikely(!p))
 117                 return -ENOBUFS;
 118         memcpy(&fh->data, p, fh->size);
 119         dprintk("%s: fh len %d\n", __func__, fh->size);
 120
 121         return 0;
 122 }
 123
 124 /*
 125  * Currently only stringified uids and gids are accepted.
 126  * I.e., kerberos is not supported to the DSes, so no pricipals.
 127  *
 128  * That means that one common function will suffice, but when
 129  * principals are added, this should be split to accomodate
 130  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 131  */
 132 static int
 133 decode_name(struct xdr_stream *xdr, u32 *id)
 134 {
 135         __be32 *p;
 136         int len;
 137
 138         /* opaque_length(4)*/
 139         p = xdr_inline_decode(xdr, 4);
 140         if (unlikely(!p))
 141                 return -ENOBUFS;
 142         len = be32_to_cpup(p++);
 143         if (len < 0)
 144                 return -EINVAL;
 145
 146         dprintk("%s: len %u\n", __func__, len);
 147
 148         /* opaque body */
 149         p = xdr_inline_decode(xdr, len);
 150         if (unlikely(!p))
 151                 return -ENOBUFS;
 152
 153         if (!nfs_map_string_to_numeric((char *)p, len, id))
 154                 return -EINVAL;
 155
 156         return 0;
 157 }
 158
 159 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 160                 const struct nfs4_ff_layout_mirror *m2)
 161 {
 162         int i, j;
 163
 164         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 165                 return false;
 166         for (i = 0; i < m1->fh_versions_cnt; i++) {
 167                 bool found_fh = false;
 168                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 169                         if (nfs_compare_fh(&m1->fh_versions[i],
 170                                         &m2->fh_versions[j]) == 0) {
 171                                 found_fh = true;
 172                                 break;
 173                         }
 174                 }
 175                 if (!found_fh)
 176                         return false;
 177         }
 178         return true;
 179 }
 180
 181 static struct nfs4_ff_layout_mirror *
 182 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 183                 struct nfs4_ff_layout_mirror *mirror)
 184 {
 185         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 186         struct nfs4_ff_layout_mirror *pos;
 187         struct inode *inode = lo->plh_inode;
 188
 189         spin_lock(&inode->i_lock);
 190         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 191                 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
 192                         continue;
 193                 if (!ff_mirror_match_fh(mirror, pos))
 194                         continue;
 195                 if (refcount_inc_not_zero(&pos->ref)) {
 196                         spin_unlock(&inode->i_lock);
 197                         return pos;
 198                 }
 199         }
 200         list_add(&mirror->mirrors, &ff_layout->mirrors);
 201         mirror->layout = lo;
 202         spin_unlock(&inode->i_lock);
 203         return mirror;
 204 }
 205
 206 static void
 207 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 208 {
 209         struct inode *inode;
 210         if (mirror->layout == NULL)
 211                 return;
 212         inode = mirror->layout->plh_inode;
 213         spin_lock(&inode->i_lock);
 214         list_del(&mirror->mirrors);
 215         spin_unlock(&inode->i_lock);
 216         mirror->layout = NULL;
 217 }
 218
 219 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 220 {
 221         struct nfs4_ff_layout_mirror *mirror;
 222
 223         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 224         if (mirror != NULL) {
 225                 spin_lock_init(&mirror->lock);
 226                 refcount_set(&mirror->ref, 1);
 227                 INIT_LIST_HEAD(&mirror->mirrors);
 228         }
 229         return mirror;
 230 }
 231
 232 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 233 {
 234         const struct cred       *cred;
 235
 236         ff_layout_remove_mirror(mirror);
 237         kfree(mirror->fh_versions);
 238         cred = rcu_access_pointer(mirror->ro_cred);
 239         put_cred(cred);
 240         cred = rcu_access_pointer(mirror->rw_cred);
 241         put_cred(cred);
 242         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 243         kfree(mirror);
 244 }
 245
 246 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 247 {
 248         if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
 249                 ff_layout_free_mirror(mirror);
 250 }
 251
 252 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 253 {
 254         u32 i;
 255
 256         for (i = 0; i < fls->mirror_array_cnt; i++)
 257                 ff_layout_put_mirror(fls->mirror_array[i]);
 258 }
 259
 260 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 261 {
 262         int ret = 0;
 263
 264         dprintk("--> %s\n", __func__);
 265
 266         /* FIXME: remove this check when layout segment support is added */
 267         if (lgr->range.offset != 0 ||
 268             lgr->range.length != NFS4_MAX_UINT64) {
 269                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 270                         __func__);
 271                 ret = -EINVAL;
 272         }
 273
 274         dprintk("--> %s returns %d\n", __func__, ret);
 275         return ret;
 276 }
 277
 278 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 279 {
 280         if (fls) {
 281                 ff_layout_free_mirror_array(fls);
 282                 kfree(fls);
 283         }
 284 }
 285
 286 static bool
 287 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
 288                 struct pnfs_layout_segment *l2)
 289 {
 290         const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
 291         const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
 292         u32 i;
 293
 294         if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
 295                 return false;
 296         for (i = 0; i < fl1->mirror_array_cnt; i++) {
 297                 if (fl1->mirror_array[i] != fl2->mirror_array[i])
 298                         return false;
 299         }
 300         return true;
 301 }
 302
 303 static bool
 304 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 305                 const struct pnfs_layout_range *l2)
 306 {
 307         u64 end1, end2;
 308
 309         if (l1->iomode != l2->iomode)
 310                 return l1->iomode != IOMODE_READ;
 311         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 312         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 313         if (end1 < l2->offset)
 314                 return false;
 315         if (end2 < l1->offset)
 316                 return true;
 317         return l2->offset <= l1->offset;
 318 }
 319
 320 static bool
 321 ff_lseg_merge(struct pnfs_layout_segment *new,
 322                 struct pnfs_layout_segment *old)
 323 {
 324         u64 new_end, old_end;
 325
 326         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 327                 return false;
 328         if (new->pls_range.iomode != old->pls_range.iomode)
 329                 return false;
 330         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 331                         old->pls_range.length);
 332         if (old_end < new->pls_range.offset)
 333                 return false;
 334         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 335                         new->pls_range.length);
 336         if (new_end < old->pls_range.offset)
 337                 return false;
 338         if (!ff_lseg_match_mirrors(new, old))
 339                 return false;
 340
 341         /* Mergeable: copy info from 'old' to 'new' */
 342         if (new_end < old_end)
 343                 new_end = old_end;
 344         if (new->pls_range.offset < old->pls_range.offset)
 345                 new->pls_range.offset = old->pls_range.offset;
 346         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 347                         new_end);
 348         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 349                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 350         return true;
 351 }
 352
 353 static void
 354 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 355                 struct pnfs_layout_segment *lseg,
 356                 struct list_head *free_me)
 357 {
 358         pnfs_generic_layout_insert_lseg(lo, lseg,
 359                         ff_lseg_range_is_after,
 360                         ff_lseg_merge,
 361                         free_me);
 362 }
 363
 364 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 365 {
 366         int i, j;
 367
 368         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 369                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 370                         if (fls->mirror_array[i]->efficiency <
 371                             fls->mirror_array[j]->efficiency)
 372                                 swap(fls->mirror_array[i],
 373                                      fls->mirror_array[j]);
 374         }
 375 }
 376
 377 static struct pnfs_layout_segment *
 378 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 379                      struct nfs4_layoutget_res *lgr,
 380                      gfp_t gfp_flags)
 381 {
 382         struct pnfs_layout_segment *ret;
 383         struct nfs4_ff_layout_segment *fls = NULL;
 384         struct xdr_stream stream;
 385         struct xdr_buf buf;
 386         struct page *scratch;
 387         u64 stripe_unit;
 388         u32 mirror_array_cnt;
 389         __be32 *p;
 390         int i, rc;
 391
 392         dprintk("--> %s\n", __func__);
 393         scratch = alloc_page(gfp_flags);
 394         if (!scratch)
 395                 return ERR_PTR(-ENOMEM);
 396
 397         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 398                               lgr->layoutp->len);
 399         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 400
 401         /* stripe unit and mirror_array_cnt */
 402         rc = -EIO;
 403         p = xdr_inline_decode(&stream, 8 + 4);
 404         if (!p)
 405                 goto out_err_free;
 406
 407         p = xdr_decode_hyper(p, &stripe_unit);
 408         mirror_array_cnt = be32_to_cpup(p++);
 409         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 410                 stripe_unit, mirror_array_cnt);
 411
 412         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 413             mirror_array_cnt == 0)
 414                 goto out_err_free;
 415
 416         rc = -ENOMEM;
 417         fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
 418                         gfp_flags);
 419         if (!fls)
 420                 goto out_err_free;
 421
 422         fls->mirror_array_cnt = mirror_array_cnt;
 423         fls->stripe_unit = stripe_unit;
 424
 425         for (i = 0; i < fls->mirror_array_cnt; i++) {
 426                 struct nfs4_ff_layout_mirror *mirror;
 427                 struct cred *kcred;
 428                 const struct cred __rcu *cred;
 429                 kuid_t uid;
 430                 kgid_t gid;
 431                 u32 ds_count, fh_count, id;
 432                 int j;
 433
 434                 rc = -EIO;
 435                 p = xdr_inline_decode(&stream, 4);
 436                 if (!p)
 437                         goto out_err_free;
 438                 ds_count = be32_to_cpup(p);
 439
 440                 /* FIXME: allow for striping? */
 441                 if (ds_count != 1)
 442                         goto out_err_free;
 443
 444                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 445                 if (fls->mirror_array[i] == NULL) {
 446                         rc = -ENOMEM;
 447                         goto out_err_free;
 448                 }
 449
 450                 fls->mirror_array[i]->ds_count = ds_count;
 451
 452                 /* deviceid */
 453                 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
 454                 if (rc)
 455                         goto out_err_free;
 456
 457                 /* efficiency */
 458                 rc = -EIO;
 459                 p = xdr_inline_decode(&stream, 4);
 460                 if (!p)
 461                         goto out_err_free;
 462                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 463
 464                 /* stateid */
 465                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 466                 if (rc)
 467                         goto out_err_free;
 468
 469                 /* fh */
 470                 rc = -EIO;
 471                 p = xdr_inline_decode(&stream, 4);
 472                 if (!p)
 473                         goto out_err_free;
 474                 fh_count = be32_to_cpup(p);
 475
 476                 fls->mirror_array[i]->fh_versions =
 477                         kcalloc(fh_count, sizeof(struct nfs_fh),
 478                                 gfp_flags);
 479                 if (fls->mirror_array[i]->fh_versions == NULL) {
 480                         rc = -ENOMEM;
 481                         goto out_err_free;
 482                 }
 483
 484                 for (j = 0; j < fh_count; j++) {
 485                         rc = decode_nfs_fh(&stream,
 486                                            &fls->mirror_array[i]->fh_versions[j]);
 487                         if (rc)
 488                                 goto out_err_free;
 489                 }
 490
 491                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 492
 493                 /* user */
 494                 rc = decode_name(&stream, &id);
 495                 if (rc)
 496                         goto out_err_free;
 497
 498                 uid = make_kuid(&init_user_ns, id);
 499
 500                 /* group */
 501                 rc = decode_name(&stream, &id);
 502                 if (rc)
 503                         goto out_err_free;
 504
 505                 gid = make_kgid(&init_user_ns, id);
 506
 507                 if (gfp_flags & __GFP_FS)
 508                         kcred = prepare_kernel_cred(NULL);
 509                 else {
 510                         unsigned int nofs_flags = memalloc_nofs_save();
 511                         kcred = prepare_kernel_cred(NULL);
 512                         memalloc_nofs_restore(nofs_flags);
 513                 }
 514                 rc = -ENOMEM;
 515                 if (!kcred)
 516                         goto out_err_free;
 517                 kcred->fsuid = uid;
 518                 kcred->fsgid = gid;
 519                 cred = RCU_INITIALIZER(kcred);
 520
 521                 if (lgr->range.iomode == IOMODE_READ)
 522                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 523                 else
 524                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 525
 526                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 527                 if (mirror != fls->mirror_array[i]) {
 528                         /* swap cred ptrs so free_mirror will clean up old */
 529                         if (lgr->range.iomode == IOMODE_READ) {
 530                                 cred = xchg(&mirror->ro_cred, cred);
 531                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 532                         } else {
 533                                 cred = xchg(&mirror->rw_cred, cred);
 534                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 535                         }
 536                         ff_layout_free_mirror(fls->mirror_array[i]);
 537                         fls->mirror_array[i] = mirror;
 538                 }
 539
 540                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 541                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 542                         from_kuid(&init_user_ns, uid),
 543                         from_kgid(&init_user_ns, gid));
 544         }
 545
 546         p = xdr_inline_decode(&stream, 4);
 547         if (!p)
 548                 goto out_sort_mirrors;
 549         fls->flags = be32_to_cpup(p);
 550
 551         p = xdr_inline_decode(&stream, 4);
 552         if (!p)
 553                 goto out_sort_mirrors;
 554         for (i=0; i < fls->mirror_array_cnt; i++)
 555                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 556
 557 out_sort_mirrors:
 558         ff_layout_sort_mirrors(fls);
 559         rc = ff_layout_check_layout(lgr);
 560         if (rc)
 561                 goto out_err_free;
 562         ret = &fls->generic_hdr;
 563         dprintk("<-- %s (success)\n", __func__);
 564 out_free_page:
 565         __free_page(scratch);
 566         return ret;
 567 out_err_free:
 568         _ff_layout_free_lseg(fls);
 569         ret = ERR_PTR(rc);
 570         dprintk("<-- %s (%d)\n", __func__, rc);
 571         goto out_free_page;
 572 }
 573
 574 static void
 575 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 576 {
 577         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 578
 579         dprintk("--> %s\n", __func__);
 580
 581         if (lseg->pls_range.iomode == IOMODE_RW) {
 582                 struct nfs4_flexfile_layout *ffl;
 583                 struct inode *inode;
 584
 585                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 586                 inode = ffl->generic_hdr.plh_inode;
 587                 spin_lock(&inode->i_lock);
 588                 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
 589                 spin_unlock(&inode->i_lock);
 590         }
 591         _ff_layout_free_lseg(fls);
 592 }
 593
 594 static void
 595 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 596 {
 597         /* first IO request? */
 598         if (atomic_inc_return(&timer->n_ops) == 1) {
 599                 timer->start_time = now;
 600         }
 601 }
 602
 603 static ktime_t
 604 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 605 {
 606         ktime_t start;
 607
 608         if (atomic_dec_return(&timer->n_ops) < 0)
 609                 WARN_ON_ONCE(1);
 610
 611         start = timer->start_time;
 612         timer->start_time = now;
 613         return ktime_sub(now, start);
 614 }
 615
 616 static bool
 617 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 618                             struct nfs4_ff_layoutstat *layoutstat,
 619                             ktime_t now)
 620 {
 621         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 622         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
 623
 624         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 625         if (!mirror->start_time)
 626                 mirror->start_time = now;
 627         if (mirror->report_interval != 0)
 628                 report_interval = (s64)mirror->report_interval * 1000LL;
 629         else if (layoutstats_timer != 0)
 630                 report_interval = (s64)layoutstats_timer * 1000LL;
 631         if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
 632                         report_interval) {
 633                 ffl->last_report_time = now;
 634                 return true;
 635         }
 636
 637         return false;
 638 }
 639
 640 static void
 641 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 642                 __u64 requested)
 643 {
 644         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 645
 646         iostat->ops_requested++;
 647         iostat->bytes_requested += requested;
 648 }
 649
 650 static void
 651 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 652                 __u64 requested,
 653                 __u64 completed,
 654                 ktime_t time_completed,
 655                 ktime_t time_started)
 656 {
 657         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 658         ktime_t completion_time = ktime_sub(time_completed, time_started);
 659         ktime_t timer;
 660
 661         iostat->ops_completed++;
 662         iostat->bytes_completed += completed;
 663         iostat->bytes_not_delivered += requested - completed;
 664
 665         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 666         iostat->total_busy_time =
 667                         ktime_add(iostat->total_busy_time, timer);
 668         iostat->aggregate_completion_time =
 669                         ktime_add(iostat->aggregate_completion_time,
 670                                         completion_time);
 671 }
 672
 673 static void
 674 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 675                 struct nfs4_ff_layout_mirror *mirror,
 676                 __u64 requested, ktime_t now)
 677 {
 678         bool report;
 679
 680         spin_lock(&mirror->lock);
 681         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 682         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 683         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 684         spin_unlock(&mirror->lock);
 685
 686         if (report)
 687                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 688 }
 689
 690 static void
 691 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 692                 struct nfs4_ff_layout_mirror *mirror,
 693                 __u64 requested,
 694                 __u64 completed)
 695 {
 696         spin_lock(&mirror->lock);
 697         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 698                         requested, completed,
 699                         ktime_get(), task->tk_start);
 700         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 701         spin_unlock(&mirror->lock);
 702 }
 703
 704 static void
 705 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 706                 struct nfs4_ff_layout_mirror *mirror,
 707                 __u64 requested, ktime_t now)
 708 {
 709         bool report;
 710
 711         spin_lock(&mirror->lock);
 712         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 713         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 714         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 715         spin_unlock(&mirror->lock);
 716
 717         if (report)
 718                 pnfs_report_layoutstat(inode, GFP_NOIO);
 719 }
 720
 721 static void
 722 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 723                 struct nfs4_ff_layout_mirror *mirror,
 724                 __u64 requested,
 725                 __u64 completed,
 726                 enum nfs3_stable_how committed)
 727 {
 728         if (committed == NFS_UNSTABLE)
 729                 requested = completed = 0;
 730
 731         spin_lock(&mirror->lock);
 732         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 733                         requested, completed, ktime_get(), task->tk_start);
 734         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 735         spin_unlock(&mirror->lock);
 736 }
 737
 738 static void
 739 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
 740 {
 741         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 742
 743         if (devid)
 744                 nfs4_mark_deviceid_unavailable(devid);
 745 }
 746
 747 static void
 748 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
 749 {
 750         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 751
 752         if (devid)
 753                 nfs4_mark_deviceid_available(devid);
 754 }
 755
 756 static struct nfs4_pnfs_ds *
 757 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
 758                              int start_idx, int *best_idx,
 759                              bool check_device)
 760 {
 761         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 762         struct nfs4_ff_layout_mirror *mirror;
 763         struct nfs4_pnfs_ds *ds;
 764         bool fail_return = false;
 765         int idx;
 766
 767         /* mirrors are initially sorted by efficiency */
 768         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 769                 if (idx+1 == fls->mirror_array_cnt)
 770                         fail_return = !check_device;
 771
 772                 mirror = FF_LAYOUT_COMP(lseg, idx);
 773                 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
 774                 if (!ds)
 775                         continue;
 776
 777                 if (check_device &&
 778                     nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
 779                         continue;
 780
 781                 *best_idx = idx;
 782                 return ds;
 783         }
 784
 785         return NULL;
 786 }
 787
 788 static struct nfs4_pnfs_ds *
 789 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
 790                                  int start_idx, int *best_idx)
 791 {
 792         return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
 793 }
 794
 795 static struct nfs4_pnfs_ds *
 796 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
 797                                    int start_idx, int *best_idx)
 798 {
 799         return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
 800 }
 801
 802 static struct nfs4_pnfs_ds *
 803 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 804                                   int start_idx, int *best_idx)
 805 {
 806         struct nfs4_pnfs_ds *ds;
 807
 808         ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
 809         if (ds)
 810                 return ds;
 811         return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
 812 }
 813
 814 static void
 815 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 816                       struct nfs_page *req,
 817                       bool strict_iomode)
 818 {
 819         pnfs_put_lseg(pgio->pg_lseg);
 820         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 821                                            nfs_req_openctx(req),
 822                                            0,
 823                                            NFS4_MAX_UINT64,
 824                                            IOMODE_READ,
 825                                            strict_iomode,
 826                                            GFP_KERNEL);
 827         if (IS_ERR(pgio->pg_lseg)) {
 828                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 829                 pgio->pg_lseg = NULL;
 830         }
 831 }
 832
 833 static void
 834 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 835                         struct nfs_page *req)
 836 {
 837         struct nfs_pgio_mirror *pgm;
 838         struct nfs4_ff_layout_mirror *mirror;
 839         struct nfs4_pnfs_ds *ds;
 840         int ds_idx;
 841
 842 retry:
 843         pnfs_generic_pg_check_layout(pgio);
 844         /* Use full layout for now */
 845         if (!pgio->pg_lseg) {
 846                 ff_layout_pg_get_read(pgio, req, false);
 847                 if (!pgio->pg_lseg)
 848                         goto out_nolseg;
 849         }
 850         if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 851                 ff_layout_pg_get_read(pgio, req, true);
 852                 if (!pgio->pg_lseg)
 853                         goto out_nolseg;
 854         }
 855
 856         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 857         if (!ds) {
 858                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 859                         goto out_mds;
 860                 pnfs_put_lseg(pgio->pg_lseg);
 861                 pgio->pg_lseg = NULL;
 862                 /* Sleep for 1 second before retrying */
 863                 ssleep(1);
 864                 goto retry;
 865         }
 866
 867         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 868
 869         pgio->pg_mirror_idx = ds_idx;
 870
 871         /* read always uses only one mirror - idx 0 for pgio layer */
 872         pgm = &pgio->pg_mirrors[0];
 873         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 874
 875         if (NFS_SERVER(pgio->pg_inode)->flags &
 876                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 877                 pgio->pg_maxretrans = io_maxretrans;
 878         return;
 879 out_nolseg:
 880         if (pgio->pg_error < 0)
 881                 return;
 882 out_mds:
 883         trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
 884                         0, NFS4_MAX_UINT64, IOMODE_READ,
 885                         NFS_I(pgio->pg_inode)->layout,
 886                         pgio->pg_lseg);
 887         pnfs_put_lseg(pgio->pg_lseg);
 888         pgio->pg_lseg = NULL;
 889         pgio->pg_maxretrans = 0;
 890         nfs_pageio_reset_read_mds(pgio);
 891 }
 892
 893 static void
 894 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 895                         struct nfs_page *req)
 896 {
 897         struct nfs4_ff_layout_mirror *mirror;
 898         struct nfs_pgio_mirror *pgm;
 899         struct nfs4_pnfs_ds *ds;
 900         int i;
 901
 902 retry:
 903         pnfs_generic_pg_check_layout(pgio);
 904         if (!pgio->pg_lseg) {
 905                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 906                                                    nfs_req_openctx(req),
 907                                                    0,
 908                                                    NFS4_MAX_UINT64,
 909                                                    IOMODE_RW,
 910                                                    false,
 911                                                    GFP_NOFS);
 912                 if (IS_ERR(pgio->pg_lseg)) {
 913                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 914                         pgio->pg_lseg = NULL;
 915                         return;
 916                 }
 917         }
 918         /* If no lseg, fall back to write through mds */
 919         if (pgio->pg_lseg == NULL)
 920                 goto out_mds;
 921
 922         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 923         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 924             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 925                 goto out_mds;
 926
 927         for (i = 0; i < pgio->pg_mirror_count; i++) {
 928                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 929                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
 930                 if (!ds) {
 931                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 932                                 goto out_mds;
 933                         pnfs_put_lseg(pgio->pg_lseg);
 934                         pgio->pg_lseg = NULL;
 935                         /* Sleep for 1 second before retrying */
 936                         ssleep(1);
 937                         goto retry;
 938                 }
 939                 pgm = &pgio->pg_mirrors[i];
 940                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 941         }
 942
 943         if (NFS_SERVER(pgio->pg_inode)->flags &
 944                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 945                 pgio->pg_maxretrans = io_maxretrans;
 946         return;
 947
 948 out_mds:
 949         trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
 950                         0, NFS4_MAX_UINT64, IOMODE_RW,
 951                         NFS_I(pgio->pg_inode)->layout,
 952                         pgio->pg_lseg);
 953         pnfs_put_lseg(pgio->pg_lseg);
 954         pgio->pg_lseg = NULL;
 955         pgio->pg_maxretrans = 0;
 956         nfs_pageio_reset_write_mds(pgio);
 957 }
 958
 959 static unsigned int
 960 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 961                                     struct nfs_page *req)
 962 {
 963         if (!pgio->pg_lseg) {
 964                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 965                                                    nfs_req_openctx(req),
 966                                                    0,
 967                                                    NFS4_MAX_UINT64,
 968                                                    IOMODE_RW,
 969                                                    false,
 970                                                    GFP_NOFS);
 971                 if (IS_ERR(pgio->pg_lseg)) {
 972                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 973                         pgio->pg_lseg = NULL;
 974                         goto out;
 975                 }
 976         }
 977         if (pgio->pg_lseg)
 978                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 979
 980         trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
 981                         0, NFS4_MAX_UINT64, IOMODE_RW,
 982                         NFS_I(pgio->pg_inode)->layout,
 983                         pgio->pg_lseg);
 984         /* no lseg means that pnfs is not in use, so no mirroring here */
 985         nfs_pageio_reset_write_mds(pgio);
 986 out:
 987         return 1;
 988 }
 989
 990 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 991         .pg_init = ff_layout_pg_init_read,
 992         .pg_test = pnfs_generic_pg_test,
 993         .pg_doio = pnfs_generic_pg_readpages,
 994         .pg_cleanup = pnfs_generic_pg_cleanup,
 995 };
 996
 997 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 998         .pg_init = ff_layout_pg_init_write,
 999         .pg_test = pnfs_generic_pg_test,
1000         .pg_doio = pnfs_generic_pg_writepages,
1001         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1002         .pg_cleanup = pnfs_generic_pg_cleanup,
1003 };
1004
1005 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1006 {
1007         struct rpc_task *task = &hdr->task;
1008
1009         pnfs_layoutcommit_inode(hdr->inode, false);
1010
1011         if (retry_pnfs) {
1012                 dprintk("%s Reset task %5u for i/o through pNFS "
1013                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1014                         hdr->task.tk_pid,
1015                         hdr->inode->i_sb->s_id,
1016                         (unsigned long long)NFS_FILEID(hdr->inode),
1017                         hdr->args.count,
1018                         (unsigned long long)hdr->args.offset);
1019
1020                 hdr->completion_ops->reschedule_io(hdr);
1021                 return;
1022         }
1023
1024         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1025                 dprintk("%s Reset task %5u for i/o through MDS "
1026                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1027                         hdr->task.tk_pid,
1028                         hdr->inode->i_sb->s_id,
1029                         (unsigned long long)NFS_FILEID(hdr->inode),
1030                         hdr->args.count,
1031                         (unsigned long long)hdr->args.offset);
1032
1033                 trace_pnfs_mds_fallback_write_done(hdr->inode,
1034                                 hdr->args.offset, hdr->args.count,
1035                                 IOMODE_RW, NFS_I(hdr->inode)->layout,
1036                                 hdr->lseg);
1037                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1038         }
1039 }
1040
1041 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1042 {
1043         struct rpc_task *task = &hdr->task;
1044
1045         pnfs_layoutcommit_inode(hdr->inode, false);
1046
1047         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1048                 dprintk("%s Reset task %5u for i/o through MDS "
1049                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1050                         hdr->task.tk_pid,
1051                         hdr->inode->i_sb->s_id,
1052                         (unsigned long long)NFS_FILEID(hdr->inode),
1053                         hdr->args.count,
1054                         (unsigned long long)hdr->args.offset);
1055
1056                 trace_pnfs_mds_fallback_read_done(hdr->inode,
1057                                 hdr->args.offset, hdr->args.count,
1058                                 IOMODE_READ, NFS_I(hdr->inode)->layout,
1059                                 hdr->lseg);
1060                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1061         }
1062 }
1063
1064 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1065                                            struct nfs4_state *state,
1066                                            struct nfs_client *clp,
1067                                            struct pnfs_layout_segment *lseg,
1068                                            int idx)
1069 {
1070         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1071         struct inode *inode = lo->plh_inode;
1072         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1073         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1074
1075         switch (task->tk_status) {
1076         case -NFS4ERR_BADSESSION:
1077         case -NFS4ERR_BADSLOT:
1078         case -NFS4ERR_BAD_HIGH_SLOT:
1079         case -NFS4ERR_DEADSESSION:
1080         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1081         case -NFS4ERR_SEQ_FALSE_RETRY:
1082         case -NFS4ERR_SEQ_MISORDERED:
1083                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1084                         "flags 0x%x\n", __func__, task->tk_status,
1085                         clp->cl_exchange_flags);
1086                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1087                 break;
1088         case -NFS4ERR_DELAY:
1089         case -NFS4ERR_GRACE:
1090                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1091                 break;
1092         case -NFS4ERR_RETRY_UNCACHED_REP:
1093                 break;
1094         /* Invalidate Layout errors */
1095         case -NFS4ERR_PNFS_NO_LAYOUT:
1096         case -ESTALE:           /* mapped NFS4ERR_STALE */
1097         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1098         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1099         case -NFS4ERR_FHEXPIRED:
1100         case -NFS4ERR_WRONG_TYPE:
1101                 dprintk("%s Invalid layout error %d\n", __func__,
1102                         task->tk_status);
1103                 /*
1104                  * Destroy layout so new i/o will get a new layout.
1105                  * Layout will not be destroyed until all current lseg
1106                  * references are put. Mark layout as invalid to resend failed
1107                  * i/o and all i/o waiting on the slot table to the MDS until
1108                  * layout is destroyed and a new valid layout is obtained.
1109                  */
1110                 pnfs_destroy_layout(NFS_I(inode));
1111                 rpc_wake_up(&tbl->slot_tbl_waitq);
1112                 goto reset;
1113         /* RPC connection errors */
1114         case -ECONNREFUSED:
1115         case -EHOSTDOWN:
1116         case -EHOSTUNREACH:
1117         case -ENETUNREACH:
1118         case -EIO:
1119         case -ETIMEDOUT:
1120         case -EPIPE:
1121                 dprintk("%s DS connection error %d\n", __func__,
1122                         task->tk_status);
1123                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1124                                 &devid->deviceid);
1125                 rpc_wake_up(&tbl->slot_tbl_waitq);
1126                 /* fall through */
1127         default:
1128                 if (ff_layout_avoid_mds_available_ds(lseg))
1129                         return -NFS4ERR_RESET_TO_PNFS;
1130 reset:
1131                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1132                         task->tk_status);
1133                 return -NFS4ERR_RESET_TO_MDS;
1134         }
1135         task->tk_status = 0;
1136         return -EAGAIN;
1137 }
1138
1139 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1140 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1141                                            struct pnfs_layout_segment *lseg,
1142                                            int idx)
1143 {
1144         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1145
1146         switch (task->tk_status) {
1147         /* File access problems. Don't mark the device as unavailable */
1148         case -EACCES:
1149         case -ESTALE:
1150         case -EISDIR:
1151         case -EBADHANDLE:
1152         case -ELOOP:
1153         case -ENOSPC:
1154                 break;
1155         case -EJUKEBOX:
1156                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1157                 goto out_retry;
1158         default:
1159                 dprintk("%s DS connection error %d\n", __func__,
1160                         task->tk_status);
1161                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1162                                 &devid->deviceid);
1163         }
1164         /* FIXME: Need to prevent infinite looping here. */
1165         return -NFS4ERR_RESET_TO_PNFS;
1166 out_retry:
1167         task->tk_status = 0;
1168         rpc_restart_call_prepare(task);
1169         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1170         return -EAGAIN;
1171 }
1172
1173 static int ff_layout_async_handle_error(struct rpc_task *task,
1174                                         struct nfs4_state *state,
1175                                         struct nfs_client *clp,
1176                                         struct pnfs_layout_segment *lseg,
1177                                         int idx)
1178 {
1179         int vers = clp->cl_nfs_mod->rpc_vers->number;
1180
1181         if (task->tk_status >= 0) {
1182                 ff_layout_mark_ds_reachable(lseg, idx);
1183                 return 0;
1184         }
1185
1186         /* Handle the case of an invalid layout segment */
1187         if (!pnfs_is_valid_lseg(lseg))
1188                 return -NFS4ERR_RESET_TO_PNFS;
1189
1190         switch (vers) {
1191         case 3:
1192                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1193         case 4:
1194                 return ff_layout_async_handle_error_v4(task, state, clp,
1195                                                        lseg, idx);
1196         default:
1197                 /* should never happen */
1198                 WARN_ON_ONCE(1);
1199                 return 0;
1200         }
1201 }
1202
1203 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1204                                         int idx, u64 offset, u64 length,
1205                                         u32 *op_status, int opnum, int error)
1206 {
1207         struct nfs4_ff_layout_mirror *mirror;
1208         u32 status = *op_status;
1209         int err;
1210
1211         if (status == 0) {
1212                 switch (error) {
1213                 case -ETIMEDOUT:
1214                 case -EPFNOSUPPORT:
1215                 case -EPROTONOSUPPORT:
1216                 case -EOPNOTSUPP:
1217                 case -ECONNREFUSED:
1218                 case -ECONNRESET:
1219                 case -EHOSTDOWN:
1220                 case -EHOSTUNREACH:
1221                 case -ENETUNREACH:
1222                 case -EADDRINUSE:
1223                 case -ENOBUFS:
1224                 case -EPIPE:
1225                 case -EPERM:
1226                         *op_status = status = NFS4ERR_NXIO;
1227                         break;
1228                 case -EACCES:
1229                         *op_status = status = NFS4ERR_ACCESS;
1230                         break;
1231                 default:
1232                         return;
1233                 }
1234         }
1235
1236         mirror = FF_LAYOUT_COMP(lseg, idx);
1237         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1238                                        mirror, offset, length, status, opnum,
1239                                        GFP_NOIO);
1240
1241         switch (status) {
1242         case NFS4ERR_DELAY:
1243         case NFS4ERR_GRACE:
1244                 break;
1245         case NFS4ERR_NXIO:
1246                 ff_layout_mark_ds_unreachable(lseg, idx);
1247                 /* Fallthrough */
1248         default:
1249                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1250                                                   lseg);
1251         }
1252
1253         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1254 }
1255
1256 /* NFS_PROTO call done callback routines */
1257 static int ff_layout_read_done_cb(struct rpc_task *task,
1258                                 struct nfs_pgio_header *hdr)
1259 {
1260         int new_idx = hdr->pgio_mirror_idx;
1261         int err;
1262
1263         if (task->tk_status < 0) {
1264                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1265                                             hdr->args.offset, hdr->args.count,
1266                                             &hdr->res.op_status, OP_READ,
1267                                             task->tk_status);
1268                 trace_ff_layout_read_error(hdr);
1269         }
1270
1271         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1272                                            hdr->ds_clp, hdr->lseg,
1273                                            hdr->pgio_mirror_idx);
1274
1275         trace_nfs4_pnfs_read(hdr, err);
1276         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1277         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1278         switch (err) {
1279         case -NFS4ERR_RESET_TO_PNFS:
1280                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1281                                         hdr->pgio_mirror_idx + 1,
1282                                         &new_idx))
1283                         goto out_layouterror;
1284                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1285                 return task->tk_status;
1286         case -NFS4ERR_RESET_TO_MDS:
1287                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1288                 return task->tk_status;
1289         case -EAGAIN:
1290                 goto out_eagain;
1291         }
1292
1293         return 0;
1294 out_layouterror:
1295         ff_layout_read_record_layoutstats_done(task, hdr);
1296         ff_layout_send_layouterror(hdr->lseg);
1297         hdr->pgio_mirror_idx = new_idx;
1298 out_eagain:
1299         rpc_restart_call_prepare(task);
1300         return -EAGAIN;
1301 }
1302
1303 static bool
1304 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1305 {
1306         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1307 }
1308
1309 /*
1310  * We reference the rpc_cred of the first WRITE that triggers the need for
1311  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1312  * rfc5661 is not clear about which credential should be used.
1313  *
1314  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1315  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1316  * we always send layoutcommit after DS writes.
1317  */
1318 static void
1319 ff_layout_set_layoutcommit(struct inode *inode,
1320                 struct pnfs_layout_segment *lseg,
1321                 loff_t end_offset)
1322 {
1323         if (!ff_layout_need_layoutcommit(lseg))
1324                 return;
1325
1326         pnfs_set_layoutcommit(inode, lseg, end_offset);
1327         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1328                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1329 }
1330
1331 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1332                 struct nfs_pgio_header *hdr)
1333 {
1334         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1335                 return;
1336         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1337                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1338                         hdr->args.count,
1339                         task->tk_start);
1340 }
1341
1342 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1343                 struct nfs_pgio_header *hdr)
1344 {
1345         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1346                 return;
1347         nfs4_ff_layout_stat_io_end_read(task,
1348                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1349                         hdr->args.count,
1350                         hdr->res.count);
1351         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1352 }
1353
1354 static int ff_layout_read_prepare_common(struct rpc_task *task,
1355                                          struct nfs_pgio_header *hdr)
1356 {
1357         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1358                 rpc_exit(task, -EIO);
1359                 return -EIO;
1360         }
1361
1362         ff_layout_read_record_layoutstats_start(task, hdr);
1363         return 0;
1364 }
1365
1366 /*
1367  * Call ops for the async read/write cases
1368  * In the case of dense layouts, the offset needs to be reset to its
1369  * original value.
1370  */
1371 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1372 {
1373         struct nfs_pgio_header *hdr = data;
1374
1375         if (ff_layout_read_prepare_common(task, hdr))
1376                 return;
1377
1378         rpc_call_start(task);
1379 }
1380
1381 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1382 {
1383         struct nfs_pgio_header *hdr = data;
1384
1385         if (nfs4_setup_sequence(hdr->ds_clp,
1386                                 &hdr->args.seq_args,
1387                                 &hdr->res.seq_res,
1388                                 task))
1389                 return;
1390
1391         ff_layout_read_prepare_common(task, hdr);
1392 }
1393
1394 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1395 {
1396         struct nfs_pgio_header *hdr = data;
1397
1398         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1399
1400         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1401             task->tk_status == 0) {
1402                 nfs4_sequence_done(task, &hdr->res.seq_res);
1403                 return;
1404         }
1405
1406         /* Note this may cause RPC to be resent */
1407         hdr->mds_ops->rpc_call_done(task, hdr);
1408 }
1409
1410 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1411 {
1412         struct nfs_pgio_header *hdr = data;
1413
1414         ff_layout_read_record_layoutstats_done(task, hdr);
1415         rpc_count_iostats_metrics(task,
1416             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1417 }
1418
1419 static void ff_layout_read_release(void *data)
1420 {
1421         struct nfs_pgio_header *hdr = data;
1422
1423         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1424         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1425                 ff_layout_send_layouterror(hdr->lseg);
1426                 pnfs_read_resend_pnfs(hdr);
1427         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1428                 ff_layout_reset_read(hdr);
1429         pnfs_generic_rw_release(data);
1430 }
1431
1432
1433 static int ff_layout_write_done_cb(struct rpc_task *task,
1434                                 struct nfs_pgio_header *hdr)
1435 {
1436         loff_t end_offs = 0;
1437         int err;
1438
1439         if (task->tk_status < 0) {
1440                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1441                                             hdr->args.offset, hdr->args.count,
1442                                             &hdr->res.op_status, OP_WRITE,
1443                                             task->tk_status);
1444                 trace_ff_layout_write_error(hdr);
1445         }
1446
1447         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1448                                            hdr->ds_clp, hdr->lseg,
1449                                            hdr->pgio_mirror_idx);
1450
1451         trace_nfs4_pnfs_write(hdr, err);
1452         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1453         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1454         switch (err) {
1455         case -NFS4ERR_RESET_TO_PNFS:
1456                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1457                 return task->tk_status;
1458         case -NFS4ERR_RESET_TO_MDS:
1459                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1460                 return task->tk_status;
1461         case -EAGAIN:
1462                 return -EAGAIN;
1463         }
1464
1465         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1466             hdr->res.verf->committed == NFS_DATA_SYNC)
1467                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1468
1469         /* Note: if the write is unstable, don't set end_offs until commit */
1470         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1471
1472         /* zero out fattr since we don't care DS attr at all */
1473         hdr->fattr.valid = 0;
1474         if (task->tk_status >= 0)
1475                 nfs_writeback_update_inode(hdr);
1476
1477         return 0;
1478 }
1479
1480 static int ff_layout_commit_done_cb(struct rpc_task *task,
1481                                      struct nfs_commit_data *data)
1482 {
1483         int err;
1484
1485         if (task->tk_status < 0) {
1486                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1487                                             data->args.offset, data->args.count,
1488                                             &data->res.op_status, OP_COMMIT,
1489                                             task->tk_status);
1490                 trace_ff_layout_commit_error(data);
1491         }
1492
1493         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1494                                            data->lseg, data->ds_commit_index);
1495
1496         trace_nfs4_pnfs_commit_ds(data, err);
1497         switch (err) {
1498         case -NFS4ERR_RESET_TO_PNFS:
1499                 pnfs_generic_prepare_to_resend_writes(data);
1500                 return -EAGAIN;
1501         case -NFS4ERR_RESET_TO_MDS:
1502                 pnfs_generic_prepare_to_resend_writes(data);
1503                 return -EAGAIN;
1504         case -EAGAIN:
1505                 rpc_restart_call_prepare(task);
1506                 return -EAGAIN;
1507         }
1508
1509         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1510
1511         return 0;
1512 }
1513
1514 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1515                 struct nfs_pgio_header *hdr)
1516 {
1517         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1518                 return;
1519         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1520                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1521                         hdr->args.count,
1522                         task->tk_start);
1523 }
1524
1525 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1526                 struct nfs_pgio_header *hdr)
1527 {
1528         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1529                 return;
1530         nfs4_ff_layout_stat_io_end_write(task,
1531                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1532                         hdr->args.count, hdr->res.count,
1533                         hdr->res.verf->committed);
1534         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1535 }
1536
1537 static int ff_layout_write_prepare_common(struct rpc_task *task,
1538                                           struct nfs_pgio_header *hdr)
1539 {
1540         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1541                 rpc_exit(task, -EIO);
1542                 return -EIO;
1543         }
1544
1545         ff_layout_write_record_layoutstats_start(task, hdr);
1546         return 0;
1547 }
1548
1549 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1550 {
1551         struct nfs_pgio_header *hdr = data;
1552
1553         if (ff_layout_write_prepare_common(task, hdr))
1554                 return;
1555
1556         rpc_call_start(task);
1557 }
1558
1559 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1560 {
1561         struct nfs_pgio_header *hdr = data;
1562
1563         if (nfs4_setup_sequence(hdr->ds_clp,
1564                                 &hdr->args.seq_args,
1565                                 &hdr->res.seq_res,
1566                                 task))
1567                 return;
1568
1569         ff_layout_write_prepare_common(task, hdr);
1570 }
1571
1572 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1573 {
1574         struct nfs_pgio_header *hdr = data;
1575
1576         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1577             task->tk_status == 0) {
1578                 nfs4_sequence_done(task, &hdr->res.seq_res);
1579                 return;
1580         }
1581
1582         /* Note this may cause RPC to be resent */
1583         hdr->mds_ops->rpc_call_done(task, hdr);
1584 }
1585
1586 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1587 {
1588         struct nfs_pgio_header *hdr = data;
1589
1590         ff_layout_write_record_layoutstats_done(task, hdr);
1591         rpc_count_iostats_metrics(task,
1592             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1593 }
1594
1595 static void ff_layout_write_release(void *data)
1596 {
1597         struct nfs_pgio_header *hdr = data;
1598
1599         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1600         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1601                 ff_layout_send_layouterror(hdr->lseg);
1602                 ff_layout_reset_write(hdr, true);
1603         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1604                 ff_layout_reset_write(hdr, false);
1605         pnfs_generic_rw_release(data);
1606 }
1607
1608 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1609                 struct nfs_commit_data *cdata)
1610 {
1611         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1612                 return;
1613         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1614                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1615                         0, task->tk_start);
1616 }
1617
1618 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1619                 struct nfs_commit_data *cdata)
1620 {
1621         struct nfs_page *req;
1622         __u64 count = 0;
1623
1624         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1625                 return;
1626
1627         if (task->tk_status == 0) {
1628                 list_for_each_entry(req, &cdata->pages, wb_list)
1629                         count += req->wb_bytes;
1630         }
1631         nfs4_ff_layout_stat_io_end_write(task,
1632                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1633                         count, count, NFS_FILE_SYNC);
1634         set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1635 }
1636
1637 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1638                 struct nfs_commit_data *cdata)
1639 {
1640         ff_layout_commit_record_layoutstats_start(task, cdata);
1641 }
1642
1643 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1644 {
1645         ff_layout_commit_prepare_common(task, data);
1646         rpc_call_start(task);
1647 }
1648
1649 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1650 {
1651         struct nfs_commit_data *wdata = data;
1652
1653         if (nfs4_setup_sequence(wdata->ds_clp,
1654                                 &wdata->args.seq_args,
1655                                 &wdata->res.seq_res,
1656                                 task))
1657                 return;
1658         ff_layout_commit_prepare_common(task, data);
1659 }
1660
1661 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1662 {
1663         pnfs_generic_write_commit_done(task, data);
1664 }
1665
1666 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1667 {
1668         struct nfs_commit_data *cdata = data;
1669
1670         ff_layout_commit_record_layoutstats_done(task, cdata);
1671         rpc_count_iostats_metrics(task,
1672             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1673 }
1674
1675 static void ff_layout_commit_release(void *data)
1676 {
1677         struct nfs_commit_data *cdata = data;
1678
1679         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1680         pnfs_generic_commit_release(data);
1681 }
1682
1683 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1684         .rpc_call_prepare = ff_layout_read_prepare_v3,
1685         .rpc_call_done = ff_layout_read_call_done,
1686         .rpc_count_stats = ff_layout_read_count_stats,
1687         .rpc_release = ff_layout_read_release,
1688 };
1689
1690 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1691         .rpc_call_prepare = ff_layout_read_prepare_v4,
1692         .rpc_call_done = ff_layout_read_call_done,
1693         .rpc_count_stats = ff_layout_read_count_stats,
1694         .rpc_release = ff_layout_read_release,
1695 };
1696
1697 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1698         .rpc_call_prepare = ff_layout_write_prepare_v3,
1699         .rpc_call_done = ff_layout_write_call_done,
1700         .rpc_count_stats = ff_layout_write_count_stats,
1701         .rpc_release = ff_layout_write_release,
1702 };
1703
1704 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1705         .rpc_call_prepare = ff_layout_write_prepare_v4,
1706         .rpc_call_done = ff_layout_write_call_done,
1707         .rpc_count_stats = ff_layout_write_count_stats,
1708         .rpc_release = ff_layout_write_release,
1709 };
1710
1711 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1712         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1713         .rpc_call_done = ff_layout_commit_done,
1714         .rpc_count_stats = ff_layout_commit_count_stats,
1715         .rpc_release = ff_layout_commit_release,
1716 };
1717
1718 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1719         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1720         .rpc_call_done = ff_layout_commit_done,
1721         .rpc_count_stats = ff_layout_commit_count_stats,
1722         .rpc_release = ff_layout_commit_release,
1723 };
1724
1725 static enum pnfs_try_status
1726 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1727 {
1728         struct pnfs_layout_segment *lseg = hdr->lseg;
1729         struct nfs4_pnfs_ds *ds;
1730         struct rpc_clnt *ds_clnt;
1731         struct nfs4_ff_layout_mirror *mirror;
1732         const struct cred *ds_cred;
1733         loff_t offset = hdr->args.offset;
1734         u32 idx = hdr->pgio_mirror_idx;
1735         int vers;
1736         struct nfs_fh *fh;
1737
1738         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1739                 __func__, hdr->inode->i_ino,
1740                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1741
1742         mirror = FF_LAYOUT_COMP(lseg, idx);
1743         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1744         if (!ds)
1745                 goto out_failed;
1746
1747         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1748                                                    hdr->inode);
1749         if (IS_ERR(ds_clnt))
1750                 goto out_failed;
1751
1752         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1753         if (!ds_cred)
1754                 goto out_failed;
1755
1756         vers = nfs4_ff_layout_ds_version(mirror);
1757
1758         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1759                 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1760
1761         hdr->pgio_done_cb = ff_layout_read_done_cb;
1762         refcount_inc(&ds->ds_clp->cl_count);
1763         hdr->ds_clp = ds->ds_clp;
1764         fh = nfs4_ff_layout_select_ds_fh(mirror);
1765         if (fh)
1766                 hdr->args.fh = fh;
1767
1768         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1769
1770         /*
1771          * Note that if we ever decide to split across DSes,
1772          * then we may need to handle dense-like offsets.
1773          */
1774         hdr->args.offset = offset;
1775         hdr->mds_offset = offset;
1776
1777         /* Perform an asynchronous read to ds */
1778         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1779                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1780                                       &ff_layout_read_call_ops_v4,
1781                           0, RPC_TASK_SOFTCONN);
1782         put_cred(ds_cred);
1783         return PNFS_ATTEMPTED;
1784
1785 out_failed:
1786         if (ff_layout_avoid_mds_available_ds(lseg))
1787                 return PNFS_TRY_AGAIN;
1788         trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1789                         hdr->args.offset, hdr->args.count,
1790                         IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1791         return PNFS_NOT_ATTEMPTED;
1792 }
1793
1794 /* Perform async writes. */
1795 static enum pnfs_try_status
1796 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1797 {
1798         struct pnfs_layout_segment *lseg = hdr->lseg;
1799         struct nfs4_pnfs_ds *ds;
1800         struct rpc_clnt *ds_clnt;
1801         struct nfs4_ff_layout_mirror *mirror;
1802         const struct cred *ds_cred;
1803         loff_t offset = hdr->args.offset;
1804         int vers;
1805         struct nfs_fh *fh;
1806         int idx = hdr->pgio_mirror_idx;
1807
1808         mirror = FF_LAYOUT_COMP(lseg, idx);
1809         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1810         if (!ds)
1811                 goto out_failed;
1812
1813         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1814                                                    hdr->inode);
1815         if (IS_ERR(ds_clnt))
1816                 goto out_failed;
1817
1818         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1819         if (!ds_cred)
1820                 goto out_failed;
1821
1822         vers = nfs4_ff_layout_ds_version(mirror);
1823
1824         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1825                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1826                 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1827                 vers);
1828
1829         hdr->pgio_done_cb = ff_layout_write_done_cb;
1830         refcount_inc(&ds->ds_clp->cl_count);
1831         hdr->ds_clp = ds->ds_clp;
1832         hdr->ds_commit_idx = idx;
1833         fh = nfs4_ff_layout_select_ds_fh(mirror);
1834         if (fh)
1835                 hdr->args.fh = fh;
1836
1837         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1838
1839         /*
1840          * Note that if we ever decide to split across DSes,
1841          * then we may need to handle dense-like offsets.
1842          */
1843         hdr->args.offset = offset;
1844
1845         /* Perform an asynchronous write */
1846         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1847                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1848                                       &ff_layout_write_call_ops_v4,
1849                           sync, RPC_TASK_SOFTCONN);
1850         put_cred(ds_cred);
1851         return PNFS_ATTEMPTED;
1852
1853 out_failed:
1854         if (ff_layout_avoid_mds_available_ds(lseg))
1855                 return PNFS_TRY_AGAIN;
1856         trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1857                         hdr->args.offset, hdr->args.count,
1858                         IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1859         return PNFS_NOT_ATTEMPTED;
1860 }
1861
1862 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1863 {
1864         return i;
1865 }
1866
1867 static struct nfs_fh *
1868 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1869 {
1870         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1871
1872         /* FIXME: Assume that there is only one NFS version available
1873          * for the DS.
1874          */
1875         return &flseg->mirror_array[i]->fh_versions[0];
1876 }
1877
1878 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1879 {
1880         struct pnfs_layout_segment *lseg = data->lseg;
1881         struct nfs4_pnfs_ds *ds;
1882         struct rpc_clnt *ds_clnt;
1883         struct nfs4_ff_layout_mirror *mirror;
1884         const struct cred *ds_cred;
1885         u32 idx;
1886         int vers, ret;
1887         struct nfs_fh *fh;
1888
1889         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1890             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1891                 goto out_err;
1892
1893         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1894         mirror = FF_LAYOUT_COMP(lseg, idx);
1895         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1896         if (!ds)
1897                 goto out_err;
1898
1899         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1900                                                    data->inode);
1901         if (IS_ERR(ds_clnt))
1902                 goto out_err;
1903
1904         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1905         if (!ds_cred)
1906                 goto out_err;
1907
1908         vers = nfs4_ff_layout_ds_version(mirror);
1909
1910         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1911                 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1912                 vers);
1913         data->commit_done_cb = ff_layout_commit_done_cb;
1914         data->cred = ds_cred;
1915         refcount_inc(&ds->ds_clp->cl_count);
1916         data->ds_clp = ds->ds_clp;
1917         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1918         if (fh)
1919                 data->args.fh = fh;
1920
1921         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1922                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1923                                                &ff_layout_commit_call_ops_v4,
1924                                    how, RPC_TASK_SOFTCONN);
1925         put_cred(ds_cred);
1926         return ret;
1927 out_err:
1928         pnfs_generic_prepare_to_resend_writes(data);
1929         pnfs_generic_commit_release(data);
1930         return -EAGAIN;
1931 }
1932
1933 static int
1934 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1935                            int how, struct nfs_commit_info *cinfo)
1936 {
1937         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1938                                             ff_layout_initiate_commit);
1939 }
1940
1941 static struct pnfs_ds_commit_info *
1942 ff_layout_get_ds_info(struct inode *inode)
1943 {
1944         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1945
1946         if (layout == NULL)
1947                 return NULL;
1948
1949         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1950 }
1951
1952 static void
1953 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1954                 struct pnfs_layout_segment *lseg)
1955 {
1956         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1957         struct inode *inode = lseg->pls_layout->plh_inode;
1958         struct pnfs_commit_array *array, *new;
1959
1960         new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1961         if (new) {
1962                 spin_lock(&inode->i_lock);
1963                 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1964                 spin_unlock(&inode->i_lock);
1965                 if (array != new)
1966                         pnfs_free_commit_array(new);
1967         }
1968 }
1969
1970 static void
1971 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1972                 struct inode *inode)
1973 {
1974         spin_lock(&inode->i_lock);
1975         pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1976         spin_unlock(&inode->i_lock);
1977 }
1978
1979 static void
1980 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1981 {
1982         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1983                                                   id_node));
1984 }
1985
1986 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1987                                   const struct nfs4_layoutreturn_args *args,
1988                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1989 {
1990         __be32 *start;
1991
1992         start = xdr_reserve_space(xdr, 4);
1993         if (unlikely(!start))
1994                 return -E2BIG;
1995
1996         *start = cpu_to_be32(ff_args->num_errors);
1997         /* This assume we always return _ALL_ layouts */
1998         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
1999 }
2000
2001 static void
2002 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2003 {
2004         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2005 }
2006
2007 static void
2008 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2009                             const nfs4_stateid *stateid,
2010                             const struct nfs42_layoutstat_devinfo *devinfo)
2011 {
2012         __be32 *p;
2013
2014         p = xdr_reserve_space(xdr, 8 + 8);
2015         p = xdr_encode_hyper(p, devinfo->offset);
2016         p = xdr_encode_hyper(p, devinfo->length);
2017         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2018         p = xdr_reserve_space(xdr, 4*8);
2019         p = xdr_encode_hyper(p, devinfo->read_count);
2020         p = xdr_encode_hyper(p, devinfo->read_bytes);
2021         p = xdr_encode_hyper(p, devinfo->write_count);
2022         p = xdr_encode_hyper(p, devinfo->write_bytes);
2023         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2024 }
2025
2026 static void
2027 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2028                             const nfs4_stateid *stateid,
2029                             const struct nfs42_layoutstat_devinfo *devinfo)
2030 {
2031         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2032         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2033                         devinfo->ld_private.data);
2034 }
2035
2036 /* report nothing for now */
2037 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2038                 const struct nfs4_layoutreturn_args *args,
2039                 struct nfs4_flexfile_layoutreturn_args *ff_args)
2040 {
2041         __be32 *p;
2042         int i;
2043
2044         p = xdr_reserve_space(xdr, 4);
2045         *p = cpu_to_be32(ff_args->num_dev);
2046         for (i = 0; i < ff_args->num_dev; i++)
2047                 ff_layout_encode_ff_iostat(xdr,
2048                                 &args->layout->plh_stateid,
2049                                 &ff_args->devinfo[i]);
2050 }
2051
2052 static void
2053 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2054                 unsigned int num_entries)
2055 {
2056         unsigned int i;
2057
2058         for (i = 0; i < num_entries; i++) {
2059                 if (!devinfo[i].ld_private.ops)
2060                         continue;
2061                 if (!devinfo[i].ld_private.ops->free)
2062                         continue;
2063                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2064         }
2065 }
2066
2067 static struct nfs4_deviceid_node *
2068 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2069                               struct pnfs_device *pdev, gfp_t gfp_flags)
2070 {
2071         struct nfs4_ff_layout_ds *dsaddr;
2072
2073         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2074         if (!dsaddr)
2075                 return NULL;
2076         return &dsaddr->id_node;
2077 }
2078
2079 static void
2080 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2081                 const void *voidargs,
2082                 const struct nfs4_xdr_opaque_data *ff_opaque)
2083 {
2084         const struct nfs4_layoutreturn_args *args = voidargs;
2085         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2086         struct xdr_buf tmp_buf = {
2087                 .head = {
2088                         [0] = {
2089                                 .iov_base = page_address(ff_args->pages[0]),
2090                         },
2091                 },
2092                 .buflen = PAGE_SIZE,
2093         };
2094         struct xdr_stream tmp_xdr;
2095         __be32 *start;
2096
2097         dprintk("%s: Begin\n", __func__);
2098
2099         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2100
2101         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2102         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2103
2104         start = xdr_reserve_space(xdr, 4);
2105         *start = cpu_to_be32(tmp_buf.len);
2106         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2107
2108         dprintk("%s: Return\n", __func__);
2109 }
2110
2111 static void
2112 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2113 {
2114         struct nfs4_flexfile_layoutreturn_args *ff_args;
2115
2116         if (!args->data)
2117                 return;
2118         ff_args = args->data;
2119         args->data = NULL;
2120
2121         ff_layout_free_ds_ioerr(&ff_args->errors);
2122         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2123
2124         put_page(ff_args->pages[0]);
2125         kfree(ff_args);
2126 }
2127
2128 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2129         .encode = ff_layout_encode_layoutreturn,
2130         .free = ff_layout_free_layoutreturn,
2131 };
2132
2133 static int
2134 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2135 {
2136         struct nfs4_flexfile_layoutreturn_args *ff_args;
2137         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2138
2139         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2140         if (!ff_args)
2141                 goto out_nomem;
2142         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2143         if (!ff_args->pages[0])
2144                 goto out_nomem_free;
2145
2146         INIT_LIST_HEAD(&ff_args->errors);
2147         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2148                         &args->range, &ff_args->errors,
2149                         FF_LAYOUTRETURN_MAXERR);
2150
2151         spin_lock(&args->inode->i_lock);
2152         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2153                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2154         spin_unlock(&args->inode->i_lock);
2155
2156         args->ld_private->ops = &layoutreturn_ops;
2157         args->ld_private->data = ff_args;
2158         return 0;
2159 out_nomem_free:
2160         kfree(ff_args);
2161 out_nomem:
2162         return -ENOMEM;
2163 }
2164
2165 #ifdef CONFIG_NFS_V4_2
2166 void
2167 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2168 {
2169         struct pnfs_layout_hdr *lo = lseg->pls_layout;
2170         struct nfs42_layout_error *errors;
2171         LIST_HEAD(head);
2172
2173         if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2174                 return;
2175         ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2176         if (list_empty(&head))
2177                 return;
2178
2179         errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2180                         sizeof(*errors), GFP_NOFS);
2181         if (errors != NULL) {
2182                 const struct nfs4_ff_layout_ds_err *pos;
2183                 size_t n = 0;
2184
2185                 list_for_each_entry(pos, &head, list) {
2186                         errors[n].offset = pos->offset;
2187                         errors[n].length = pos->length;
2188                         nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2189                         errors[n].errors[0].dev_id = pos->deviceid;
2190                         errors[n].errors[0].status = pos->status;
2191                         errors[n].errors[0].opnum = pos->opnum;
2192                         n++;
2193                         if (!list_is_last(&pos->list, &head) &&
2194                             n < NFS42_LAYOUTERROR_MAX)
2195                                 continue;
2196                         if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2197                                 break;
2198                         n = 0;
2199                 }
2200                 kfree(errors);
2201         }
2202         ff_layout_free_ds_ioerr(&head);
2203 }
2204 #else
2205 void
2206 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2207 {
2208 }
2209 #endif
2210
2211 static int
2212 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2213 {
2214         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2215
2216         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2217 }
2218
2219 static size_t
2220 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2221                           const int buflen)
2222 {
2223         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2224         const struct in6_addr *addr = &sin6->sin6_addr;
2225
2226         /*
2227          * RFC 4291, Section 2.2.2
2228          *
2229          * Shorthanded ANY address
2230          */
2231         if (ipv6_addr_any(addr))
2232                 return snprintf(buf, buflen, "::");
2233
2234         /*
2235          * RFC 4291, Section 2.2.2
2236          *
2237          * Shorthanded loopback address
2238          */
2239         if (ipv6_addr_loopback(addr))
2240                 return snprintf(buf, buflen, "::1");
2241
2242         /*
2243          * RFC 4291, Section 2.2.3
2244          *
2245          * Special presentation address format for mapped v4
2246          * addresses.
2247          */
2248         if (ipv6_addr_v4mapped(addr))
2249                 return snprintf(buf, buflen, "::ffff:%pI4",
2250                                         &addr->s6_addr32[3]);
2251
2252         /*
2253          * RFC 4291, Section 2.2.1
2254          */
2255         return snprintf(buf, buflen, "%pI6c", addr);
2256 }
2257
2258 /* Derived from rpc_sockaddr2uaddr */
2259 static void
2260 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2261 {
2262         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2263         char portbuf[RPCBIND_MAXUADDRPLEN];
2264         char addrbuf[RPCBIND_MAXUADDRLEN];
2265         char *netid;
2266         unsigned short port;
2267         int len, netid_len;
2268         __be32 *p;
2269
2270         switch (sap->sa_family) {
2271         case AF_INET:
2272                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2273                         return;
2274                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2275                 netid = "tcp";
2276                 netid_len = 3;
2277                 break;
2278         case AF_INET6:
2279                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2280                         return;
2281                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2282                 netid = "tcp6";
2283                 netid_len = 4;
2284                 break;
2285         default:
2286                 /* we only support tcp and tcp6 */
2287                 WARN_ON_ONCE(1);
2288                 return;
2289         }
2290
2291         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2292         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2293
2294         p = xdr_reserve_space(xdr, 4 + netid_len);
2295         xdr_encode_opaque(p, netid, netid_len);
2296
2297         p = xdr_reserve_space(xdr, 4 + len);
2298         xdr_encode_opaque(p, addrbuf, len);
2299 }
2300
2301 static void
2302 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2303                          ktime_t t)
2304 {
2305         struct timespec64 ts;
2306         __be32 *p;
2307
2308         p = xdr_reserve_space(xdr, 12);
2309         ts = ktime_to_timespec64(t);
2310         p = xdr_encode_hyper(p, ts.tv_sec);
2311         *p++ = cpu_to_be32(ts.tv_nsec);
2312 }
2313
2314 static void
2315 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2316                             struct nfs4_ff_io_stat *stat)
2317 {
2318         __be32 *p;
2319
2320         p = xdr_reserve_space(xdr, 5 * 8);
2321         p = xdr_encode_hyper(p, stat->ops_requested);
2322         p = xdr_encode_hyper(p, stat->bytes_requested);
2323         p = xdr_encode_hyper(p, stat->ops_completed);
2324         p = xdr_encode_hyper(p, stat->bytes_completed);
2325         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2326         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2327         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2328 }
2329
2330 static void
2331 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2332                               const struct nfs42_layoutstat_devinfo *devinfo,
2333                               struct nfs4_ff_layout_mirror *mirror)
2334 {
2335         struct nfs4_pnfs_ds_addr *da;
2336         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2337         struct nfs_fh *fh = &mirror->fh_versions[0];
2338         __be32 *p;
2339
2340         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2341         dprintk("%s: DS %s: encoding address %s\n",
2342                 __func__, ds->ds_remotestr, da->da_remotestr);
2343         /* netaddr4 */
2344         ff_layout_encode_netaddr(xdr, da);
2345         /* nfs_fh4 */
2346         p = xdr_reserve_space(xdr, 4 + fh->size);
2347         xdr_encode_opaque(p, fh->data, fh->size);
2348         /* ff_io_latency4 read */
2349         spin_lock(&mirror->lock);
2350         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2351         /* ff_io_latency4 write */
2352         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2353         spin_unlock(&mirror->lock);
2354         /* nfstime4 */
2355         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2356         /* bool */
2357         p = xdr_reserve_space(xdr, 4);
2358         *p = cpu_to_be32(false);
2359 }
2360
2361 static void
2362 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2363                              const struct nfs4_xdr_opaque_data *opaque)
2364 {
2365         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2366                         struct nfs42_layoutstat_devinfo, ld_private);
2367         __be32 *start;
2368
2369         /* layoutupdate length */
2370         start = xdr_reserve_space(xdr, 4);
2371         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2372
2373         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2374 }
2375
2376 static void
2377 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2378 {
2379         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2380
2381         ff_layout_put_mirror(mirror);
2382 }
2383
2384 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2385         .encode = ff_layout_encode_layoutstats,
2386         .free   = ff_layout_free_layoutstats,
2387 };
2388
2389 static int
2390 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2391                                struct nfs42_layoutstat_devinfo *devinfo,
2392                                int dev_limit)
2393 {
2394         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2395         struct nfs4_ff_layout_mirror *mirror;
2396         struct nfs4_deviceid_node *dev;
2397         int i = 0;
2398
2399         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2400                 if (i >= dev_limit)
2401                         break;
2402                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2403                         continue;
2404                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2405                         continue;
2406                 /* mirror refcount put in cleanup_layoutstats */
2407                 if (!refcount_inc_not_zero(&mirror->ref))
2408                         continue;
2409                 dev = &mirror->mirror_ds->id_node;
2410                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2411                 devinfo->offset = 0;
2412                 devinfo->length = NFS4_MAX_UINT64;
2413                 spin_lock(&mirror->lock);
2414                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2415                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2416                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2417                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2418                 spin_unlock(&mirror->lock);
2419                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2420                 devinfo->ld_private.ops = &layoutstat_ops;
2421                 devinfo->ld_private.data = mirror;
2422
2423                 devinfo++;
2424                 i++;
2425         }
2426         return i;
2427 }
2428
2429 static int
2430 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2431 {
2432         struct nfs4_flexfile_layout *ff_layout;
2433         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2434
2435         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2436         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2437         if (!args->devinfo)
2438                 return -ENOMEM;
2439
2440         spin_lock(&args->inode->i_lock);
2441         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2442         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2443                         &args->devinfo[0], dev_count);
2444         spin_unlock(&args->inode->i_lock);
2445         if (!args->num_dev) {
2446                 kfree(args->devinfo);
2447                 args->devinfo = NULL;
2448                 return -ENOENT;
2449         }
2450
2451         return 0;
2452 }
2453
2454 static int
2455 ff_layout_set_layoutdriver(struct nfs_server *server,
2456                 const struct nfs_fh *dummy)
2457 {
2458 #if IS_ENABLED(CONFIG_NFS_V4_2)
2459         server->caps |= NFS_CAP_LAYOUTSTATS;
2460 #endif
2461         return 0;
2462 }
2463
2464 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2465         .setup_ds_info          = ff_layout_setup_ds_info,
2466         .release_ds_info        = ff_layout_release_ds_info,
2467         .mark_request_commit    = pnfs_layout_mark_request_commit,
2468         .clear_request_commit   = pnfs_generic_clear_request_commit,
2469         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2470         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2471         .commit_pagelist        = ff_layout_commit_pagelist,
2472 };
2473
2474 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2475         .id                     = LAYOUT_FLEX_FILES,
2476         .name                   = "LAYOUT_FLEX_FILES",
2477         .owner                  = THIS_MODULE,
2478         .flags                  = PNFS_LAYOUTGET_ON_OPEN,
2479         .max_layoutget_response = 4096, /* 1 page or so... */
2480         .set_layoutdriver       = ff_layout_set_layoutdriver,
2481         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2482         .free_layout_hdr        = ff_layout_free_layout_hdr,
2483         .alloc_lseg             = ff_layout_alloc_lseg,
2484         .free_lseg              = ff_layout_free_lseg,
2485         .add_lseg               = ff_layout_add_lseg,
2486         .pg_read_ops            = &ff_layout_pg_read_ops,
2487         .pg_write_ops           = &ff_layout_pg_write_ops,
2488         .get_ds_info            = ff_layout_get_ds_info,
2489         .free_deviceid_node     = ff_layout_free_deviceid_node,
2490         .read_pagelist          = ff_layout_read_pagelist,
2491         .write_pagelist         = ff_layout_write_pagelist,
2492         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2493         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2494         .sync                   = pnfs_nfs_generic_sync,
2495         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2496 };
2497
2498 static int __init nfs4flexfilelayout_init(void)
2499 {
2500         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2501                __func__);
2502         return pnfs_register_layoutdriver(&flexfilelayout_type);
2503 }
2504
2505 static void __exit nfs4flexfilelayout_exit(void)
2506 {
2507         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2508                __func__);
2509         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2510 }
2511
2512 MODULE_ALIAS("nfs-layouttype4-4");
2513
2514 MODULE_LICENSE("GPL");
2515 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2516
2517 module_init(nfs4flexfilelayout_init);
2518 module_exit(nfs4flexfilelayout_exit);
2519
2520 module_param(io_maxretrans, ushort, 0644);
2521 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2522                         "retries an I/O request before returning an error. ");