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 void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
  36                 struct nfs_pgio_header *hdr);
  37 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
  38                                struct nfs42_layoutstat_devinfo *devinfo,
  39                                int dev_limit);
  40 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
  41                               const struct nfs42_layoutstat_devinfo *devinfo,
  42                               struct nfs4_ff_layout_mirror *mirror);
  43
  44 static struct pnfs_layout_hdr *
  45 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  46 {
  47         struct nfs4_flexfile_layout *ffl;
  48
  49         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  50         if (ffl) {
  51                 pnfs_init_ds_commit_info(&ffl->commit_info);
  52                 INIT_LIST_HEAD(&ffl->error_list);
  53                 INIT_LIST_HEAD(&ffl->mirrors);
  54                 ffl->last_report_time = ktime_get();
  55                 return &ffl->generic_hdr;
  56         } else
  57                 return NULL;
  58 }
  59
  60 static void
  61 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  62 {
  63         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
  64         struct nfs4_ff_layout_ds_err *err, *n;
  65
  66         list_for_each_entry_safe(err, n, &ffl->error_list, list) {
  67                 list_del(&err->list);
  68                 kfree(err);
  69         }
  70         kfree_rcu(ffl, generic_hdr.plh_rcu);
  71 }
  72
  73 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  74 {
  75         __be32 *p;
  76
  77         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  78         if (unlikely(p == NULL))
  79                 return -ENOBUFS;
  80         stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
  81         memcpy(stateid->data, p, NFS4_STATEID_SIZE);
  82         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  83                 p[0], p[1], p[2], p[3]);
  84         return 0;
  85 }
  86
  87 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  88 {
  89         __be32 *p;
  90
  91         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  92         if (unlikely(!p))
  93                 return -ENOBUFS;
  94         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  95         nfs4_print_deviceid(devid);
  96         return 0;
  97 }
  98
  99 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
 100 {
 101         __be32 *p;
 102
 103         p = xdr_inline_decode(xdr, 4);
 104         if (unlikely(!p))
 105                 return -ENOBUFS;
 106         fh->size = be32_to_cpup(p++);
 107         if (fh->size > sizeof(struct nfs_fh)) {
 108                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
 109                        fh->size);
 110                 return -EOVERFLOW;
 111         }
 112         /* fh.data */
 113         p = xdr_inline_decode(xdr, fh->size);
 114         if (unlikely(!p))
 115                 return -ENOBUFS;
 116         memcpy(&fh->data, p, fh->size);
 117         dprintk("%s: fh len %d\n", __func__, fh->size);
 118
 119         return 0;
 120 }
 121
 122 /*
 123  * Currently only stringified uids and gids are accepted.
 124  * I.e., kerberos is not supported to the DSes, so no pricipals.
 125  *
 126  * That means that one common function will suffice, but when
 127  * principals are added, this should be split to accomodate
 128  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 129  */
 130 static int
 131 decode_name(struct xdr_stream *xdr, u32 *id)
 132 {
 133         __be32 *p;
 134         int len;
 135
 136         /* opaque_length(4)*/
 137         p = xdr_inline_decode(xdr, 4);
 138         if (unlikely(!p))
 139                 return -ENOBUFS;
 140         len = be32_to_cpup(p++);
 141         if (len < 0)
 142                 return -EINVAL;
 143
 144         dprintk("%s: len %u\n", __func__, len);
 145
 146         /* opaque body */
 147         p = xdr_inline_decode(xdr, len);
 148         if (unlikely(!p))
 149                 return -ENOBUFS;
 150
 151         if (!nfs_map_string_to_numeric((char *)p, len, id))
 152                 return -EINVAL;
 153
 154         return 0;
 155 }
 156
 157 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 158                 const struct nfs4_ff_layout_mirror *m2)
 159 {
 160         int i, j;
 161
 162         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 163                 return false;
 164         for (i = 0; i < m1->fh_versions_cnt; i++) {
 165                 bool found_fh = false;
 166                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 167                         if (nfs_compare_fh(&m1->fh_versions[i],
 168                                         &m2->fh_versions[j]) == 0) {
 169                                 found_fh = true;
 170                                 break;
 171                         }
 172                 }
 173                 if (!found_fh)
 174                         return false;
 175         }
 176         return true;
 177 }
 178
 179 static struct nfs4_ff_layout_mirror *
 180 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 181                 struct nfs4_ff_layout_mirror *mirror)
 182 {
 183         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 184         struct nfs4_ff_layout_mirror *pos;
 185         struct inode *inode = lo->plh_inode;
 186
 187         spin_lock(&inode->i_lock);
 188         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 189                 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
 190                         continue;
 191                 if (!ff_mirror_match_fh(mirror, pos))
 192                         continue;
 193                 if (refcount_inc_not_zero(&pos->ref)) {
 194                         spin_unlock(&inode->i_lock);
 195                         return pos;
 196                 }
 197         }
 198         list_add(&mirror->mirrors, &ff_layout->mirrors);
 199         mirror->layout = lo;
 200         spin_unlock(&inode->i_lock);
 201         return mirror;
 202 }
 203
 204 static void
 205 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 206 {
 207         struct inode *inode;
 208         if (mirror->layout == NULL)
 209                 return;
 210         inode = mirror->layout->plh_inode;
 211         spin_lock(&inode->i_lock);
 212         list_del(&mirror->mirrors);
 213         spin_unlock(&inode->i_lock);
 214         mirror->layout = NULL;
 215 }
 216
 217 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 218 {
 219         struct nfs4_ff_layout_mirror *mirror;
 220
 221         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 222         if (mirror != NULL) {
 223                 spin_lock_init(&mirror->lock);
 224                 refcount_set(&mirror->ref, 1);
 225                 INIT_LIST_HEAD(&mirror->mirrors);
 226         }
 227         return mirror;
 228 }
 229
 230 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 231 {
 232         const struct cred       *cred;
 233
 234         ff_layout_remove_mirror(mirror);
 235         kfree(mirror->fh_versions);
 236         cred = rcu_access_pointer(mirror->ro_cred);
 237         put_cred(cred);
 238         cred = rcu_access_pointer(mirror->rw_cred);
 239         put_cred(cred);
 240         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 241         kfree(mirror);
 242 }
 243
 244 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 245 {
 246         if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
 247                 ff_layout_free_mirror(mirror);
 248 }
 249
 250 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 251 {
 252         u32 i;
 253
 254         for (i = 0; i < fls->mirror_array_cnt; i++)
 255                 ff_layout_put_mirror(fls->mirror_array[i]);
 256 }
 257
 258 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 259 {
 260         int ret = 0;
 261
 262         dprintk("--> %s\n", __func__);
 263
 264         /* FIXME: remove this check when layout segment support is added */
 265         if (lgr->range.offset != 0 ||
 266             lgr->range.length != NFS4_MAX_UINT64) {
 267                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 268                         __func__);
 269                 ret = -EINVAL;
 270         }
 271
 272         dprintk("--> %s returns %d\n", __func__, ret);
 273         return ret;
 274 }
 275
 276 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 277 {
 278         if (fls) {
 279                 ff_layout_free_mirror_array(fls);
 280                 kfree(fls);
 281         }
 282 }
 283
 284 static bool
 285 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 286                 const struct pnfs_layout_range *l2)
 287 {
 288         u64 end1, end2;
 289
 290         if (l1->iomode != l2->iomode)
 291                 return l1->iomode != IOMODE_READ;
 292         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 293         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 294         if (end1 < l2->offset)
 295                 return false;
 296         if (end2 < l1->offset)
 297                 return true;
 298         return l2->offset <= l1->offset;
 299 }
 300
 301 static bool
 302 ff_lseg_merge(struct pnfs_layout_segment *new,
 303                 struct pnfs_layout_segment *old)
 304 {
 305         u64 new_end, old_end;
 306
 307         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 308                 return false;
 309         if (new->pls_range.iomode != old->pls_range.iomode)
 310                 return false;
 311         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 312                         old->pls_range.length);
 313         if (old_end < new->pls_range.offset)
 314                 return false;
 315         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 316                         new->pls_range.length);
 317         if (new_end < old->pls_range.offset)
 318                 return false;
 319
 320         /* Mergeable: copy info from 'old' to 'new' */
 321         if (new_end < old_end)
 322                 new_end = old_end;
 323         if (new->pls_range.offset < old->pls_range.offset)
 324                 new->pls_range.offset = old->pls_range.offset;
 325         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 326                         new_end);
 327         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 328                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 329         return true;
 330 }
 331
 332 static void
 333 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 334                 struct pnfs_layout_segment *lseg,
 335                 struct list_head *free_me)
 336 {
 337         pnfs_generic_layout_insert_lseg(lo, lseg,
 338                         ff_lseg_range_is_after,
 339                         ff_lseg_merge,
 340                         free_me);
 341 }
 342
 343 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 344 {
 345         int i, j;
 346
 347         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 348                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 349                         if (fls->mirror_array[i]->efficiency <
 350                             fls->mirror_array[j]->efficiency)
 351                                 swap(fls->mirror_array[i],
 352                                      fls->mirror_array[j]);
 353         }
 354 }
 355
 356 static struct pnfs_layout_segment *
 357 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 358                      struct nfs4_layoutget_res *lgr,
 359                      gfp_t gfp_flags)
 360 {
 361         struct pnfs_layout_segment *ret;
 362         struct nfs4_ff_layout_segment *fls = NULL;
 363         struct xdr_stream stream;
 364         struct xdr_buf buf;
 365         struct page *scratch;
 366         u64 stripe_unit;
 367         u32 mirror_array_cnt;
 368         __be32 *p;
 369         int i, rc;
 370
 371         dprintk("--> %s\n", __func__);
 372         scratch = alloc_page(gfp_flags);
 373         if (!scratch)
 374                 return ERR_PTR(-ENOMEM);
 375
 376         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 377                               lgr->layoutp->len);
 378         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 379
 380         /* stripe unit and mirror_array_cnt */
 381         rc = -EIO;
 382         p = xdr_inline_decode(&stream, 8 + 4);
 383         if (!p)
 384                 goto out_err_free;
 385
 386         p = xdr_decode_hyper(p, &stripe_unit);
 387         mirror_array_cnt = be32_to_cpup(p++);
 388         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 389                 stripe_unit, mirror_array_cnt);
 390
 391         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 392             mirror_array_cnt == 0)
 393                 goto out_err_free;
 394
 395         rc = -ENOMEM;
 396         fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
 397                         gfp_flags);
 398         if (!fls)
 399                 goto out_err_free;
 400
 401         fls->mirror_array_cnt = mirror_array_cnt;
 402         fls->stripe_unit = stripe_unit;
 403
 404         for (i = 0; i < fls->mirror_array_cnt; i++) {
 405                 struct nfs4_ff_layout_mirror *mirror;
 406                 struct cred *kcred;
 407                 const struct cred __rcu *cred;
 408                 kuid_t uid;
 409                 kgid_t gid;
 410                 u32 ds_count, fh_count, id;
 411                 int j;
 412
 413                 rc = -EIO;
 414                 p = xdr_inline_decode(&stream, 4);
 415                 if (!p)
 416                         goto out_err_free;
 417                 ds_count = be32_to_cpup(p);
 418
 419                 /* FIXME: allow for striping? */
 420                 if (ds_count != 1)
 421                         goto out_err_free;
 422
 423                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 424                 if (fls->mirror_array[i] == NULL) {
 425                         rc = -ENOMEM;
 426                         goto out_err_free;
 427                 }
 428
 429                 fls->mirror_array[i]->ds_count = ds_count;
 430
 431                 /* deviceid */
 432                 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
 433                 if (rc)
 434                         goto out_err_free;
 435
 436                 /* efficiency */
 437                 rc = -EIO;
 438                 p = xdr_inline_decode(&stream, 4);
 439                 if (!p)
 440                         goto out_err_free;
 441                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 442
 443                 /* stateid */
 444                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 445                 if (rc)
 446                         goto out_err_free;
 447
 448                 /* fh */
 449                 rc = -EIO;
 450                 p = xdr_inline_decode(&stream, 4);
 451                 if (!p)
 452                         goto out_err_free;
 453                 fh_count = be32_to_cpup(p);
 454
 455                 fls->mirror_array[i]->fh_versions =
 456                         kcalloc(fh_count, sizeof(struct nfs_fh),
 457                                 gfp_flags);
 458                 if (fls->mirror_array[i]->fh_versions == NULL) {
 459                         rc = -ENOMEM;
 460                         goto out_err_free;
 461                 }
 462
 463                 for (j = 0; j < fh_count; j++) {
 464                         rc = decode_nfs_fh(&stream,
 465                                            &fls->mirror_array[i]->fh_versions[j]);
 466                         if (rc)
 467                                 goto out_err_free;
 468                 }
 469
 470                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 471
 472                 /* user */
 473                 rc = decode_name(&stream, &id);
 474                 if (rc)
 475                         goto out_err_free;
 476
 477                 uid = make_kuid(&init_user_ns, id);
 478
 479                 /* group */
 480                 rc = decode_name(&stream, &id);
 481                 if (rc)
 482                         goto out_err_free;
 483
 484                 gid = make_kgid(&init_user_ns, id);
 485
 486                 if (gfp_flags & __GFP_FS)
 487                         kcred = prepare_kernel_cred(NULL);
 488                 else {
 489                         unsigned int nofs_flags = memalloc_nofs_save();
 490                         kcred = prepare_kernel_cred(NULL);
 491                         memalloc_nofs_restore(nofs_flags);
 492                 }
 493                 rc = -ENOMEM;
 494                 if (!kcred)
 495                         goto out_err_free;
 496                 kcred->fsuid = uid;
 497                 kcred->fsgid = gid;
 498                 cred = RCU_INITIALIZER(kcred);
 499
 500                 if (lgr->range.iomode == IOMODE_READ)
 501                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 502                 else
 503                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 504
 505                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 506                 if (mirror != fls->mirror_array[i]) {
 507                         /* swap cred ptrs so free_mirror will clean up old */
 508                         if (lgr->range.iomode == IOMODE_READ) {
 509                                 cred = xchg(&mirror->ro_cred, cred);
 510                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 511                         } else {
 512                                 cred = xchg(&mirror->rw_cred, cred);
 513                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 514                         }
 515                         ff_layout_free_mirror(fls->mirror_array[i]);
 516                         fls->mirror_array[i] = mirror;
 517                 }
 518
 519                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 520                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 521                         from_kuid(&init_user_ns, uid),
 522                         from_kgid(&init_user_ns, gid));
 523         }
 524
 525         p = xdr_inline_decode(&stream, 4);
 526         if (!p)
 527                 goto out_sort_mirrors;
 528         fls->flags = be32_to_cpup(p);
 529
 530         p = xdr_inline_decode(&stream, 4);
 531         if (!p)
 532                 goto out_sort_mirrors;
 533         for (i=0; i < fls->mirror_array_cnt; i++)
 534                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 535
 536 out_sort_mirrors:
 537         ff_layout_sort_mirrors(fls);
 538         rc = ff_layout_check_layout(lgr);
 539         if (rc)
 540                 goto out_err_free;
 541         ret = &fls->generic_hdr;
 542         dprintk("<-- %s (success)\n", __func__);
 543 out_free_page:
 544         __free_page(scratch);
 545         return ret;
 546 out_err_free:
 547         _ff_layout_free_lseg(fls);
 548         ret = ERR_PTR(rc);
 549         dprintk("<-- %s (%d)\n", __func__, rc);
 550         goto out_free_page;
 551 }
 552
 553 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 554 {
 555         struct pnfs_layout_segment *lseg;
 556
 557         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 558                 if (lseg->pls_range.iomode == IOMODE_RW)
 559                         return true;
 560
 561         return false;
 562 }
 563
 564 static void
 565 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 566 {
 567         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 568
 569         dprintk("--> %s\n", __func__);
 570
 571         if (lseg->pls_range.iomode == IOMODE_RW) {
 572                 struct nfs4_flexfile_layout *ffl;
 573                 struct inode *inode;
 574
 575                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 576                 inode = ffl->generic_hdr.plh_inode;
 577                 spin_lock(&inode->i_lock);
 578                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 579                         ffl->commit_info.nbuckets = 0;
 580                         kfree(ffl->commit_info.buckets);
 581                         ffl->commit_info.buckets = NULL;
 582                 }
 583                 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
 584                 spin_unlock(&inode->i_lock);
 585         }
 586         _ff_layout_free_lseg(fls);
 587 }
 588
 589 /* Return 1 until we have multiple lsegs support */
 590 static int
 591 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 592 {
 593         return 1;
 594 }
 595
 596 static void
 597 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 598 {
 599         /* first IO request? */
 600         if (atomic_inc_return(&timer->n_ops) == 1) {
 601                 timer->start_time = now;
 602         }
 603 }
 604
 605 static ktime_t
 606 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 607 {
 608         ktime_t start;
 609
 610         if (atomic_dec_return(&timer->n_ops) < 0)
 611                 WARN_ON_ONCE(1);
 612
 613         start = timer->start_time;
 614         timer->start_time = now;
 615         return ktime_sub(now, start);
 616 }
 617
 618 static bool
 619 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 620                             struct nfs4_ff_layoutstat *layoutstat,
 621                             ktime_t now)
 622 {
 623         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 624         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
 625
 626         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 627         if (!mirror->start_time)
 628                 mirror->start_time = now;
 629         if (mirror->report_interval != 0)
 630                 report_interval = (s64)mirror->report_interval * 1000LL;
 631         else if (layoutstats_timer != 0)
 632                 report_interval = (s64)layoutstats_timer * 1000LL;
 633         if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
 634                         report_interval) {
 635                 ffl->last_report_time = now;
 636                 return true;
 637         }
 638
 639         return false;
 640 }
 641
 642 static void
 643 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 644                 __u64 requested)
 645 {
 646         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 647
 648         iostat->ops_requested++;
 649         iostat->bytes_requested += requested;
 650 }
 651
 652 static void
 653 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 654                 __u64 requested,
 655                 __u64 completed,
 656                 ktime_t time_completed,
 657                 ktime_t time_started)
 658 {
 659         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 660         ktime_t completion_time = ktime_sub(time_completed, time_started);
 661         ktime_t timer;
 662
 663         iostat->ops_completed++;
 664         iostat->bytes_completed += completed;
 665         iostat->bytes_not_delivered += requested - completed;
 666
 667         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 668         iostat->total_busy_time =
 669                         ktime_add(iostat->total_busy_time, timer);
 670         iostat->aggregate_completion_time =
 671                         ktime_add(iostat->aggregate_completion_time,
 672                                         completion_time);
 673 }
 674
 675 static void
 676 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 677                 struct nfs4_ff_layout_mirror *mirror,
 678                 __u64 requested, ktime_t now)
 679 {
 680         bool report;
 681
 682         spin_lock(&mirror->lock);
 683         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 684         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 685         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 686         spin_unlock(&mirror->lock);
 687
 688         if (report)
 689                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 690 }
 691
 692 static void
 693 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 694                 struct nfs4_ff_layout_mirror *mirror,
 695                 __u64 requested,
 696                 __u64 completed)
 697 {
 698         spin_lock(&mirror->lock);
 699         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 700                         requested, completed,
 701                         ktime_get(), task->tk_start);
 702         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 703         spin_unlock(&mirror->lock);
 704 }
 705
 706 static void
 707 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 708                 struct nfs4_ff_layout_mirror *mirror,
 709                 __u64 requested, ktime_t now)
 710 {
 711         bool report;
 712
 713         spin_lock(&mirror->lock);
 714         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 715         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 716         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 717         spin_unlock(&mirror->lock);
 718
 719         if (report)
 720                 pnfs_report_layoutstat(inode, GFP_NOIO);
 721 }
 722
 723 static void
 724 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 725                 struct nfs4_ff_layout_mirror *mirror,
 726                 __u64 requested,
 727                 __u64 completed,
 728                 enum nfs3_stable_how committed)
 729 {
 730         if (committed == NFS_UNSTABLE)
 731                 requested = completed = 0;
 732
 733         spin_lock(&mirror->lock);
 734         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 735                         requested, completed, ktime_get(), task->tk_start);
 736         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 737         spin_unlock(&mirror->lock);
 738 }
 739
 740 static int
 741 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 742                             struct nfs_commit_info *cinfo,
 743                             gfp_t gfp_flags)
 744 {
 745         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 746         struct pnfs_commit_bucket *buckets;
 747         int size;
 748
 749         if (cinfo->ds->nbuckets != 0) {
 750                 /* This assumes there is only one RW lseg per file.
 751                  * To support multiple lseg per file, we need to
 752                  * change struct pnfs_commit_bucket to allow dynamic
 753                  * increasing nbuckets.
 754                  */
 755                 return 0;
 756         }
 757
 758         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 759
 760         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 761                           gfp_flags);
 762         if (!buckets)
 763                 return -ENOMEM;
 764         else {
 765                 int i;
 766
 767                 spin_lock(&cinfo->inode->i_lock);
 768                 if (cinfo->ds->nbuckets != 0)
 769                         kfree(buckets);
 770                 else {
 771                         cinfo->ds->buckets = buckets;
 772                         cinfo->ds->nbuckets = size;
 773                         for (i = 0; i < size; i++) {
 774                                 INIT_LIST_HEAD(&buckets[i].written);
 775                                 INIT_LIST_HEAD(&buckets[i].committing);
 776                                 /* mark direct verifier as unset */
 777                                 buckets[i].direct_verf.committed =
 778                                         NFS_INVALID_STABLE_HOW;
 779                         }
 780                 }
 781                 spin_unlock(&cinfo->inode->i_lock);
 782                 return 0;
 783         }
 784 }
 785
 786 static void
 787 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
 788 {
 789         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 790
 791         if (devid)
 792                 nfs4_mark_deviceid_unavailable(devid);
 793 }
 794
 795 static void
 796 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
 797 {
 798         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 799
 800         if (devid)
 801                 nfs4_mark_deviceid_available(devid);
 802 }
 803
 804 static struct nfs4_pnfs_ds *
 805 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
 806                              int start_idx, int *best_idx,
 807                              bool check_device)
 808 {
 809         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 810         struct nfs4_ff_layout_mirror *mirror;
 811         struct nfs4_pnfs_ds *ds;
 812         bool fail_return = false;
 813         int idx;
 814
 815         /* mirrors are initially sorted by efficiency */
 816         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 817                 if (idx+1 == fls->mirror_array_cnt)
 818                         fail_return = !check_device;
 819
 820                 mirror = FF_LAYOUT_COMP(lseg, idx);
 821                 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
 822                 if (!ds)
 823                         continue;
 824
 825                 if (check_device &&
 826                     nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
 827                         continue;
 828
 829                 *best_idx = idx;
 830                 return ds;
 831         }
 832
 833         return NULL;
 834 }
 835
 836 static struct nfs4_pnfs_ds *
 837 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
 838                                  int start_idx, int *best_idx)
 839 {
 840         return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
 841 }
 842
 843 static struct nfs4_pnfs_ds *
 844 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
 845                                    int start_idx, int *best_idx)
 846 {
 847         return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
 848 }
 849
 850 static struct nfs4_pnfs_ds *
 851 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 852                                   int start_idx, int *best_idx)
 853 {
 854         struct nfs4_pnfs_ds *ds;
 855
 856         ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
 857         if (ds)
 858                 return ds;
 859         return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
 860 }
 861
 862 static void
 863 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 864                       struct nfs_page *req,
 865                       bool strict_iomode)
 866 {
 867         pnfs_put_lseg(pgio->pg_lseg);
 868         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 869                                            nfs_req_openctx(req),
 870                                            0,
 871                                            NFS4_MAX_UINT64,
 872                                            IOMODE_READ,
 873                                            strict_iomode,
 874                                            GFP_KERNEL);
 875         if (IS_ERR(pgio->pg_lseg)) {
 876                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 877                 pgio->pg_lseg = NULL;
 878         }
 879 }
 880
 881 static void
 882 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 883                         struct nfs_page *req)
 884 {
 885         struct nfs_pgio_mirror *pgm;
 886         struct nfs4_ff_layout_mirror *mirror;
 887         struct nfs4_pnfs_ds *ds;
 888         int ds_idx;
 889
 890 retry:
 891         pnfs_generic_pg_check_layout(pgio);
 892         /* Use full layout for now */
 893         if (!pgio->pg_lseg) {
 894                 ff_layout_pg_get_read(pgio, req, false);
 895                 if (!pgio->pg_lseg)
 896                         goto out_nolseg;
 897         }
 898         if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 899                 ff_layout_pg_get_read(pgio, req, true);
 900                 if (!pgio->pg_lseg)
 901                         goto out_nolseg;
 902         }
 903
 904         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 905         if (!ds) {
 906                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 907                         goto out_mds;
 908                 pnfs_put_lseg(pgio->pg_lseg);
 909                 pgio->pg_lseg = NULL;
 910                 /* Sleep for 1 second before retrying */
 911                 ssleep(1);
 912                 goto retry;
 913         }
 914
 915         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 916
 917         pgio->pg_mirror_idx = ds_idx;
 918
 919         /* read always uses only one mirror - idx 0 for pgio layer */
 920         pgm = &pgio->pg_mirrors[0];
 921         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 922
 923         if (NFS_SERVER(pgio->pg_inode)->flags &
 924                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 925                 pgio->pg_maxretrans = io_maxretrans;
 926         return;
 927 out_nolseg:
 928         if (pgio->pg_error < 0)
 929                 return;
 930 out_mds:
 931         trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
 932                         0, NFS4_MAX_UINT64, IOMODE_READ,
 933                         NFS_I(pgio->pg_inode)->layout,
 934                         pgio->pg_lseg);
 935         pnfs_put_lseg(pgio->pg_lseg);
 936         pgio->pg_lseg = NULL;
 937         pgio->pg_maxretrans = 0;
 938         nfs_pageio_reset_read_mds(pgio);
 939 }
 940
 941 static void
 942 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 943                         struct nfs_page *req)
 944 {
 945         struct nfs4_ff_layout_mirror *mirror;
 946         struct nfs_pgio_mirror *pgm;
 947         struct nfs_commit_info cinfo;
 948         struct nfs4_pnfs_ds *ds;
 949         int i;
 950         int status;
 951
 952 retry:
 953         pnfs_generic_pg_check_layout(pgio);
 954         if (!pgio->pg_lseg) {
 955                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 956                                                    nfs_req_openctx(req),
 957                                                    0,
 958                                                    NFS4_MAX_UINT64,
 959                                                    IOMODE_RW,
 960                                                    false,
 961                                                    GFP_NOFS);
 962                 if (IS_ERR(pgio->pg_lseg)) {
 963                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 964                         pgio->pg_lseg = NULL;
 965                         return;
 966                 }
 967         }
 968         /* If no lseg, fall back to write through mds */
 969         if (pgio->pg_lseg == NULL)
 970                 goto out_mds;
 971
 972         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 973         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 974         if (status < 0)
 975                 goto out_mds;
 976
 977         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 978         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 979             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 980                 goto out_mds;
 981
 982         for (i = 0; i < pgio->pg_mirror_count; i++) {
 983                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 984                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
 985                 if (!ds) {
 986                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 987                                 goto out_mds;
 988                         pnfs_put_lseg(pgio->pg_lseg);
 989                         pgio->pg_lseg = NULL;
 990                         /* Sleep for 1 second before retrying */
 991                         ssleep(1);
 992                         goto retry;
 993                 }
 994                 pgm = &pgio->pg_mirrors[i];
 995                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 996         }
 997
 998         if (NFS_SERVER(pgio->pg_inode)->flags &
 999                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
1000                 pgio->pg_maxretrans = io_maxretrans;
1001         return;
1002
1003 out_mds:
1004         trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
1005                         0, NFS4_MAX_UINT64, IOMODE_RW,
1006                         NFS_I(pgio->pg_inode)->layout,
1007                         pgio->pg_lseg);
1008         pnfs_put_lseg(pgio->pg_lseg);
1009         pgio->pg_lseg = NULL;
1010         pgio->pg_maxretrans = 0;
1011         nfs_pageio_reset_write_mds(pgio);
1012 }
1013
1014 static unsigned int
1015 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
1016                                     struct nfs_page *req)
1017 {
1018         if (!pgio->pg_lseg) {
1019                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1020                                                    nfs_req_openctx(req),
1021                                                    0,
1022                                                    NFS4_MAX_UINT64,
1023                                                    IOMODE_RW,
1024                                                    false,
1025                                                    GFP_NOFS);
1026                 if (IS_ERR(pgio->pg_lseg)) {
1027                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1028                         pgio->pg_lseg = NULL;
1029                         goto out;
1030                 }
1031         }
1032         if (pgio->pg_lseg)
1033                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1034
1035         trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
1036                         0, NFS4_MAX_UINT64, IOMODE_RW,
1037                         NFS_I(pgio->pg_inode)->layout,
1038                         pgio->pg_lseg);
1039         /* no lseg means that pnfs is not in use, so no mirroring here */
1040         nfs_pageio_reset_write_mds(pgio);
1041 out:
1042         return 1;
1043 }
1044
1045 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1046         .pg_init = ff_layout_pg_init_read,
1047         .pg_test = pnfs_generic_pg_test,
1048         .pg_doio = pnfs_generic_pg_readpages,
1049         .pg_cleanup = pnfs_generic_pg_cleanup,
1050 };
1051
1052 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1053         .pg_init = ff_layout_pg_init_write,
1054         .pg_test = pnfs_generic_pg_test,
1055         .pg_doio = pnfs_generic_pg_writepages,
1056         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1057         .pg_cleanup = pnfs_generic_pg_cleanup,
1058 };
1059
1060 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1061 {
1062         struct rpc_task *task = &hdr->task;
1063
1064         pnfs_layoutcommit_inode(hdr->inode, false);
1065
1066         if (retry_pnfs) {
1067                 dprintk("%s Reset task %5u for i/o through pNFS "
1068                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1069                         hdr->task.tk_pid,
1070                         hdr->inode->i_sb->s_id,
1071                         (unsigned long long)NFS_FILEID(hdr->inode),
1072                         hdr->args.count,
1073                         (unsigned long long)hdr->args.offset);
1074
1075                 hdr->completion_ops->reschedule_io(hdr);
1076                 return;
1077         }
1078
1079         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1080                 dprintk("%s Reset task %5u for i/o through MDS "
1081                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1082                         hdr->task.tk_pid,
1083                         hdr->inode->i_sb->s_id,
1084                         (unsigned long long)NFS_FILEID(hdr->inode),
1085                         hdr->args.count,
1086                         (unsigned long long)hdr->args.offset);
1087
1088                 trace_pnfs_mds_fallback_write_done(hdr->inode,
1089                                 hdr->args.offset, hdr->args.count,
1090                                 IOMODE_RW, NFS_I(hdr->inode)->layout,
1091                                 hdr->lseg);
1092                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1093         }
1094 }
1095
1096 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1097 {
1098         struct rpc_task *task = &hdr->task;
1099
1100         pnfs_layoutcommit_inode(hdr->inode, false);
1101
1102         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1103                 dprintk("%s Reset task %5u for i/o through MDS "
1104                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1105                         hdr->task.tk_pid,
1106                         hdr->inode->i_sb->s_id,
1107                         (unsigned long long)NFS_FILEID(hdr->inode),
1108                         hdr->args.count,
1109                         (unsigned long long)hdr->args.offset);
1110
1111                 trace_pnfs_mds_fallback_read_done(hdr->inode,
1112                                 hdr->args.offset, hdr->args.count,
1113                                 IOMODE_READ, NFS_I(hdr->inode)->layout,
1114                                 hdr->lseg);
1115                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1116         }
1117 }
1118
1119 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1120                                            struct nfs4_state *state,
1121                                            struct nfs_client *clp,
1122                                            struct pnfs_layout_segment *lseg,
1123                                            int idx)
1124 {
1125         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1126         struct inode *inode = lo->plh_inode;
1127         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1128         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1129
1130         switch (task->tk_status) {
1131         case -NFS4ERR_BADSESSION:
1132         case -NFS4ERR_BADSLOT:
1133         case -NFS4ERR_BAD_HIGH_SLOT:
1134         case -NFS4ERR_DEADSESSION:
1135         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1136         case -NFS4ERR_SEQ_FALSE_RETRY:
1137         case -NFS4ERR_SEQ_MISORDERED:
1138                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1139                         "flags 0x%x\n", __func__, task->tk_status,
1140                         clp->cl_exchange_flags);
1141                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1142                 break;
1143         case -NFS4ERR_DELAY:
1144         case -NFS4ERR_GRACE:
1145                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1146                 break;
1147         case -NFS4ERR_RETRY_UNCACHED_REP:
1148                 break;
1149         /* Invalidate Layout errors */
1150         case -NFS4ERR_PNFS_NO_LAYOUT:
1151         case -ESTALE:           /* mapped NFS4ERR_STALE */
1152         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1153         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1154         case -NFS4ERR_FHEXPIRED:
1155         case -NFS4ERR_WRONG_TYPE:
1156                 dprintk("%s Invalid layout error %d\n", __func__,
1157                         task->tk_status);
1158                 /*
1159                  * Destroy layout so new i/o will get a new layout.
1160                  * Layout will not be destroyed until all current lseg
1161                  * references are put. Mark layout as invalid to resend failed
1162                  * i/o and all i/o waiting on the slot table to the MDS until
1163                  * layout is destroyed and a new valid layout is obtained.
1164                  */
1165                 pnfs_destroy_layout(NFS_I(inode));
1166                 rpc_wake_up(&tbl->slot_tbl_waitq);
1167                 goto reset;
1168         /* RPC connection errors */
1169         case -ECONNREFUSED:
1170         case -EHOSTDOWN:
1171         case -EHOSTUNREACH:
1172         case -ENETUNREACH:
1173         case -EIO:
1174         case -ETIMEDOUT:
1175         case -EPIPE:
1176                 dprintk("%s DS connection error %d\n", __func__,
1177                         task->tk_status);
1178                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1179                                 &devid->deviceid);
1180                 rpc_wake_up(&tbl->slot_tbl_waitq);
1181                 /* fall through */
1182         default:
1183                 if (ff_layout_avoid_mds_available_ds(lseg))
1184                         return -NFS4ERR_RESET_TO_PNFS;
1185 reset:
1186                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1187                         task->tk_status);
1188                 return -NFS4ERR_RESET_TO_MDS;
1189         }
1190         task->tk_status = 0;
1191         return -EAGAIN;
1192 }
1193
1194 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1195 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1196                                            struct pnfs_layout_segment *lseg,
1197                                            int idx)
1198 {
1199         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1200
1201         switch (task->tk_status) {
1202         /* File access problems. Don't mark the device as unavailable */
1203         case -EACCES:
1204         case -ESTALE:
1205         case -EISDIR:
1206         case -EBADHANDLE:
1207         case -ELOOP:
1208         case -ENOSPC:
1209                 break;
1210         case -EJUKEBOX:
1211                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1212                 goto out_retry;
1213         default:
1214                 dprintk("%s DS connection error %d\n", __func__,
1215                         task->tk_status);
1216                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1217                                 &devid->deviceid);
1218         }
1219         /* FIXME: Need to prevent infinite looping here. */
1220         return -NFS4ERR_RESET_TO_PNFS;
1221 out_retry:
1222         task->tk_status = 0;
1223         rpc_restart_call_prepare(task);
1224         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1225         return -EAGAIN;
1226 }
1227
1228 static int ff_layout_async_handle_error(struct rpc_task *task,
1229                                         struct nfs4_state *state,
1230                                         struct nfs_client *clp,
1231                                         struct pnfs_layout_segment *lseg,
1232                                         int idx)
1233 {
1234         int vers = clp->cl_nfs_mod->rpc_vers->number;
1235
1236         if (task->tk_status >= 0) {
1237                 ff_layout_mark_ds_reachable(lseg, idx);
1238                 return 0;
1239         }
1240
1241         /* Handle the case of an invalid layout segment */
1242         if (!pnfs_is_valid_lseg(lseg))
1243                 return -NFS4ERR_RESET_TO_PNFS;
1244
1245         switch (vers) {
1246         case 3:
1247                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1248         case 4:
1249                 return ff_layout_async_handle_error_v4(task, state, clp,
1250                                                        lseg, idx);
1251         default:
1252                 /* should never happen */
1253                 WARN_ON_ONCE(1);
1254                 return 0;
1255         }
1256 }
1257
1258 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1259                                         int idx, u64 offset, u64 length,
1260                                         u32 *op_status, int opnum, int error)
1261 {
1262         struct nfs4_ff_layout_mirror *mirror;
1263         u32 status = *op_status;
1264         int err;
1265
1266         if (status == 0) {
1267                 switch (error) {
1268                 case -ETIMEDOUT:
1269                 case -EPFNOSUPPORT:
1270                 case -EPROTONOSUPPORT:
1271                 case -EOPNOTSUPP:
1272                 case -ECONNREFUSED:
1273                 case -ECONNRESET:
1274                 case -EHOSTDOWN:
1275                 case -EHOSTUNREACH:
1276                 case -ENETUNREACH:
1277                 case -EADDRINUSE:
1278                 case -ENOBUFS:
1279                 case -EPIPE:
1280                 case -EPERM:
1281                         *op_status = status = NFS4ERR_NXIO;
1282                         break;
1283                 case -EACCES:
1284                         *op_status = status = NFS4ERR_ACCESS;
1285                         break;
1286                 default:
1287                         return;
1288                 }
1289         }
1290
1291         mirror = FF_LAYOUT_COMP(lseg, idx);
1292         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1293                                        mirror, offset, length, status, opnum,
1294                                        GFP_NOIO);
1295
1296         switch (status) {
1297         case NFS4ERR_DELAY:
1298         case NFS4ERR_GRACE:
1299                 break;
1300         case NFS4ERR_NXIO:
1301                 ff_layout_mark_ds_unreachable(lseg, idx);
1302                 /* Fallthrough */
1303         default:
1304                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1305                                                   lseg);
1306         }
1307
1308         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1309 }
1310
1311 /* NFS_PROTO call done callback routines */
1312 static int ff_layout_read_done_cb(struct rpc_task *task,
1313                                 struct nfs_pgio_header *hdr)
1314 {
1315         int new_idx = hdr->pgio_mirror_idx;
1316         int err;
1317
1318         if (task->tk_status < 0) {
1319                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1320                                             hdr->args.offset, hdr->args.count,
1321                                             &hdr->res.op_status, OP_READ,
1322                                             task->tk_status);
1323                 trace_ff_layout_read_error(hdr);
1324         }
1325
1326         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1327                                            hdr->ds_clp, hdr->lseg,
1328                                            hdr->pgio_mirror_idx);
1329
1330         trace_nfs4_pnfs_read(hdr, err);
1331         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1332         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1333         switch (err) {
1334         case -NFS4ERR_RESET_TO_PNFS:
1335                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1336                                         hdr->pgio_mirror_idx + 1,
1337                                         &new_idx))
1338                         goto out_layouterror;
1339                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1340                 return task->tk_status;
1341         case -NFS4ERR_RESET_TO_MDS:
1342                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1343                 return task->tk_status;
1344         case -EAGAIN:
1345                 goto out_eagain;
1346         }
1347
1348         return 0;
1349 out_layouterror:
1350         ff_layout_read_record_layoutstats_done(task, hdr);
1351         ff_layout_send_layouterror(hdr->lseg);
1352         hdr->pgio_mirror_idx = new_idx;
1353 out_eagain:
1354         rpc_restart_call_prepare(task);
1355         return -EAGAIN;
1356 }
1357
1358 static bool
1359 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1360 {
1361         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1362 }
1363
1364 /*
1365  * We reference the rpc_cred of the first WRITE that triggers the need for
1366  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1367  * rfc5661 is not clear about which credential should be used.
1368  *
1369  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1370  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1371  * we always send layoutcommit after DS writes.
1372  */
1373 static void
1374 ff_layout_set_layoutcommit(struct inode *inode,
1375                 struct pnfs_layout_segment *lseg,
1376                 loff_t end_offset)
1377 {
1378         if (!ff_layout_need_layoutcommit(lseg))
1379                 return;
1380
1381         pnfs_set_layoutcommit(inode, lseg, end_offset);
1382         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1383                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1384 }
1385
1386 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1387                 struct nfs_pgio_header *hdr)
1388 {
1389         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1390                 return;
1391         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1392                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1393                         hdr->args.count,
1394                         task->tk_start);
1395 }
1396
1397 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1398                 struct nfs_pgio_header *hdr)
1399 {
1400         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1401                 return;
1402         nfs4_ff_layout_stat_io_end_read(task,
1403                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1404                         hdr->args.count,
1405                         hdr->res.count);
1406         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1407 }
1408
1409 static int ff_layout_read_prepare_common(struct rpc_task *task,
1410                                          struct nfs_pgio_header *hdr)
1411 {
1412         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1413                 rpc_exit(task, -EIO);
1414                 return -EIO;
1415         }
1416
1417         ff_layout_read_record_layoutstats_start(task, hdr);
1418         return 0;
1419 }
1420
1421 /*
1422  * Call ops for the async read/write cases
1423  * In the case of dense layouts, the offset needs to be reset to its
1424  * original value.
1425  */
1426 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1427 {
1428         struct nfs_pgio_header *hdr = data;
1429
1430         if (ff_layout_read_prepare_common(task, hdr))
1431                 return;
1432
1433         rpc_call_start(task);
1434 }
1435
1436 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1437 {
1438         struct nfs_pgio_header *hdr = data;
1439
1440         if (nfs4_setup_sequence(hdr->ds_clp,
1441                                 &hdr->args.seq_args,
1442                                 &hdr->res.seq_res,
1443                                 task))
1444                 return;
1445
1446         ff_layout_read_prepare_common(task, hdr);
1447 }
1448
1449 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1450 {
1451         struct nfs_pgio_header *hdr = data;
1452
1453         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1454
1455         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1456             task->tk_status == 0) {
1457                 nfs4_sequence_done(task, &hdr->res.seq_res);
1458                 return;
1459         }
1460
1461         /* Note this may cause RPC to be resent */
1462         hdr->mds_ops->rpc_call_done(task, hdr);
1463 }
1464
1465 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1466 {
1467         struct nfs_pgio_header *hdr = data;
1468
1469         ff_layout_read_record_layoutstats_done(task, hdr);
1470         rpc_count_iostats_metrics(task,
1471             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1472 }
1473
1474 static void ff_layout_read_release(void *data)
1475 {
1476         struct nfs_pgio_header *hdr = data;
1477
1478         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1479         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1480                 ff_layout_send_layouterror(hdr->lseg);
1481                 pnfs_read_resend_pnfs(hdr);
1482         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1483                 ff_layout_reset_read(hdr);
1484         pnfs_generic_rw_release(data);
1485 }
1486
1487
1488 static int ff_layout_write_done_cb(struct rpc_task *task,
1489                                 struct nfs_pgio_header *hdr)
1490 {
1491         loff_t end_offs = 0;
1492         int err;
1493
1494         if (task->tk_status < 0) {
1495                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1496                                             hdr->args.offset, hdr->args.count,
1497                                             &hdr->res.op_status, OP_WRITE,
1498                                             task->tk_status);
1499                 trace_ff_layout_write_error(hdr);
1500         }
1501
1502         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1503                                            hdr->ds_clp, hdr->lseg,
1504                                            hdr->pgio_mirror_idx);
1505
1506         trace_nfs4_pnfs_write(hdr, err);
1507         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1508         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1509         switch (err) {
1510         case -NFS4ERR_RESET_TO_PNFS:
1511                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1512                 return task->tk_status;
1513         case -NFS4ERR_RESET_TO_MDS:
1514                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1515                 return task->tk_status;
1516         case -EAGAIN:
1517                 return -EAGAIN;
1518         }
1519
1520         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1521             hdr->res.verf->committed == NFS_DATA_SYNC)
1522                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1523
1524         /* Note: if the write is unstable, don't set end_offs until commit */
1525         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1526
1527         /* zero out fattr since we don't care DS attr at all */
1528         hdr->fattr.valid = 0;
1529         if (task->tk_status >= 0)
1530                 nfs_writeback_update_inode(hdr);
1531
1532         return 0;
1533 }
1534
1535 static int ff_layout_commit_done_cb(struct rpc_task *task,
1536                                      struct nfs_commit_data *data)
1537 {
1538         int err;
1539
1540         if (task->tk_status < 0) {
1541                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1542                                             data->args.offset, data->args.count,
1543                                             &data->res.op_status, OP_COMMIT,
1544                                             task->tk_status);
1545                 trace_ff_layout_commit_error(data);
1546         }
1547
1548         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1549                                            data->lseg, data->ds_commit_index);
1550
1551         trace_nfs4_pnfs_commit_ds(data, err);
1552         switch (err) {
1553         case -NFS4ERR_RESET_TO_PNFS:
1554                 pnfs_generic_prepare_to_resend_writes(data);
1555                 return -EAGAIN;
1556         case -NFS4ERR_RESET_TO_MDS:
1557                 pnfs_generic_prepare_to_resend_writes(data);
1558                 return -EAGAIN;
1559         case -EAGAIN:
1560                 rpc_restart_call_prepare(task);
1561                 return -EAGAIN;
1562         }
1563
1564         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1565
1566         return 0;
1567 }
1568
1569 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1570                 struct nfs_pgio_header *hdr)
1571 {
1572         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1573                 return;
1574         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1575                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1576                         hdr->args.count,
1577                         task->tk_start);
1578 }
1579
1580 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1581                 struct nfs_pgio_header *hdr)
1582 {
1583         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1584                 return;
1585         nfs4_ff_layout_stat_io_end_write(task,
1586                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1587                         hdr->args.count, hdr->res.count,
1588                         hdr->res.verf->committed);
1589         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1590 }
1591
1592 static int ff_layout_write_prepare_common(struct rpc_task *task,
1593                                           struct nfs_pgio_header *hdr)
1594 {
1595         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1596                 rpc_exit(task, -EIO);
1597                 return -EIO;
1598         }
1599
1600         ff_layout_write_record_layoutstats_start(task, hdr);
1601         return 0;
1602 }
1603
1604 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1605 {
1606         struct nfs_pgio_header *hdr = data;
1607
1608         if (ff_layout_write_prepare_common(task, hdr))
1609                 return;
1610
1611         rpc_call_start(task);
1612 }
1613
1614 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1615 {
1616         struct nfs_pgio_header *hdr = data;
1617
1618         if (nfs4_setup_sequence(hdr->ds_clp,
1619                                 &hdr->args.seq_args,
1620                                 &hdr->res.seq_res,
1621                                 task))
1622                 return;
1623
1624         ff_layout_write_prepare_common(task, hdr);
1625 }
1626
1627 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1628 {
1629         struct nfs_pgio_header *hdr = data;
1630
1631         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1632             task->tk_status == 0) {
1633                 nfs4_sequence_done(task, &hdr->res.seq_res);
1634                 return;
1635         }
1636
1637         /* Note this may cause RPC to be resent */
1638         hdr->mds_ops->rpc_call_done(task, hdr);
1639 }
1640
1641 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1642 {
1643         struct nfs_pgio_header *hdr = data;
1644
1645         ff_layout_write_record_layoutstats_done(task, hdr);
1646         rpc_count_iostats_metrics(task,
1647             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1648 }
1649
1650 static void ff_layout_write_release(void *data)
1651 {
1652         struct nfs_pgio_header *hdr = data;
1653
1654         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1655         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1656                 ff_layout_send_layouterror(hdr->lseg);
1657                 ff_layout_reset_write(hdr, true);
1658         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1659                 ff_layout_reset_write(hdr, false);
1660         pnfs_generic_rw_release(data);
1661 }
1662
1663 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1664                 struct nfs_commit_data *cdata)
1665 {
1666         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1667                 return;
1668         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1669                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1670                         0, task->tk_start);
1671 }
1672
1673 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1674                 struct nfs_commit_data *cdata)
1675 {
1676         struct nfs_page *req;
1677         __u64 count = 0;
1678
1679         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1680                 return;
1681
1682         if (task->tk_status == 0) {
1683                 list_for_each_entry(req, &cdata->pages, wb_list)
1684                         count += req->wb_bytes;
1685         }
1686         nfs4_ff_layout_stat_io_end_write(task,
1687                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1688                         count, count, NFS_FILE_SYNC);
1689         set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1690 }
1691
1692 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1693                 struct nfs_commit_data *cdata)
1694 {
1695         ff_layout_commit_record_layoutstats_start(task, cdata);
1696 }
1697
1698 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1699 {
1700         ff_layout_commit_prepare_common(task, data);
1701         rpc_call_start(task);
1702 }
1703
1704 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1705 {
1706         struct nfs_commit_data *wdata = data;
1707
1708         if (nfs4_setup_sequence(wdata->ds_clp,
1709                                 &wdata->args.seq_args,
1710                                 &wdata->res.seq_res,
1711                                 task))
1712                 return;
1713         ff_layout_commit_prepare_common(task, data);
1714 }
1715
1716 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1717 {
1718         pnfs_generic_write_commit_done(task, data);
1719 }
1720
1721 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1722 {
1723         struct nfs_commit_data *cdata = data;
1724
1725         ff_layout_commit_record_layoutstats_done(task, cdata);
1726         rpc_count_iostats_metrics(task,
1727             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1728 }
1729
1730 static void ff_layout_commit_release(void *data)
1731 {
1732         struct nfs_commit_data *cdata = data;
1733
1734         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1735         pnfs_generic_commit_release(data);
1736 }
1737
1738 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1739         .rpc_call_prepare = ff_layout_read_prepare_v3,
1740         .rpc_call_done = ff_layout_read_call_done,
1741         .rpc_count_stats = ff_layout_read_count_stats,
1742         .rpc_release = ff_layout_read_release,
1743 };
1744
1745 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1746         .rpc_call_prepare = ff_layout_read_prepare_v4,
1747         .rpc_call_done = ff_layout_read_call_done,
1748         .rpc_count_stats = ff_layout_read_count_stats,
1749         .rpc_release = ff_layout_read_release,
1750 };
1751
1752 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1753         .rpc_call_prepare = ff_layout_write_prepare_v3,
1754         .rpc_call_done = ff_layout_write_call_done,
1755         .rpc_count_stats = ff_layout_write_count_stats,
1756         .rpc_release = ff_layout_write_release,
1757 };
1758
1759 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1760         .rpc_call_prepare = ff_layout_write_prepare_v4,
1761         .rpc_call_done = ff_layout_write_call_done,
1762         .rpc_count_stats = ff_layout_write_count_stats,
1763         .rpc_release = ff_layout_write_release,
1764 };
1765
1766 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1767         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1768         .rpc_call_done = ff_layout_commit_done,
1769         .rpc_count_stats = ff_layout_commit_count_stats,
1770         .rpc_release = ff_layout_commit_release,
1771 };
1772
1773 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1774         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1775         .rpc_call_done = ff_layout_commit_done,
1776         .rpc_count_stats = ff_layout_commit_count_stats,
1777         .rpc_release = ff_layout_commit_release,
1778 };
1779
1780 static enum pnfs_try_status
1781 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1782 {
1783         struct pnfs_layout_segment *lseg = hdr->lseg;
1784         struct nfs4_pnfs_ds *ds;
1785         struct rpc_clnt *ds_clnt;
1786         struct nfs4_ff_layout_mirror *mirror;
1787         const struct cred *ds_cred;
1788         loff_t offset = hdr->args.offset;
1789         u32 idx = hdr->pgio_mirror_idx;
1790         int vers;
1791         struct nfs_fh *fh;
1792
1793         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1794                 __func__, hdr->inode->i_ino,
1795                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1796
1797         mirror = FF_LAYOUT_COMP(lseg, idx);
1798         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1799         if (!ds)
1800                 goto out_failed;
1801
1802         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1803                                                    hdr->inode);
1804         if (IS_ERR(ds_clnt))
1805                 goto out_failed;
1806
1807         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1808         if (!ds_cred)
1809                 goto out_failed;
1810
1811         vers = nfs4_ff_layout_ds_version(mirror);
1812
1813         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1814                 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1815
1816         hdr->pgio_done_cb = ff_layout_read_done_cb;
1817         refcount_inc(&ds->ds_clp->cl_count);
1818         hdr->ds_clp = ds->ds_clp;
1819         fh = nfs4_ff_layout_select_ds_fh(mirror);
1820         if (fh)
1821                 hdr->args.fh = fh;
1822
1823         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1824
1825         /*
1826          * Note that if we ever decide to split across DSes,
1827          * then we may need to handle dense-like offsets.
1828          */
1829         hdr->args.offset = offset;
1830         hdr->mds_offset = offset;
1831
1832         /* Perform an asynchronous read to ds */
1833         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1834                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1835                                       &ff_layout_read_call_ops_v4,
1836                           0, RPC_TASK_SOFTCONN);
1837         put_cred(ds_cred);
1838         return PNFS_ATTEMPTED;
1839
1840 out_failed:
1841         if (ff_layout_avoid_mds_available_ds(lseg))
1842                 return PNFS_TRY_AGAIN;
1843         trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1844                         hdr->args.offset, hdr->args.count,
1845                         IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1846         return PNFS_NOT_ATTEMPTED;
1847 }
1848
1849 /* Perform async writes. */
1850 static enum pnfs_try_status
1851 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1852 {
1853         struct pnfs_layout_segment *lseg = hdr->lseg;
1854         struct nfs4_pnfs_ds *ds;
1855         struct rpc_clnt *ds_clnt;
1856         struct nfs4_ff_layout_mirror *mirror;
1857         const struct cred *ds_cred;
1858         loff_t offset = hdr->args.offset;
1859         int vers;
1860         struct nfs_fh *fh;
1861         int idx = hdr->pgio_mirror_idx;
1862
1863         mirror = FF_LAYOUT_COMP(lseg, idx);
1864         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1865         if (!ds)
1866                 goto out_failed;
1867
1868         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1869                                                    hdr->inode);
1870         if (IS_ERR(ds_clnt))
1871                 goto out_failed;
1872
1873         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1874         if (!ds_cred)
1875                 goto out_failed;
1876
1877         vers = nfs4_ff_layout_ds_version(mirror);
1878
1879         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1880                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1881                 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1882                 vers);
1883
1884         hdr->pgio_done_cb = ff_layout_write_done_cb;
1885         refcount_inc(&ds->ds_clp->cl_count);
1886         hdr->ds_clp = ds->ds_clp;
1887         hdr->ds_commit_idx = idx;
1888         fh = nfs4_ff_layout_select_ds_fh(mirror);
1889         if (fh)
1890                 hdr->args.fh = fh;
1891
1892         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1893
1894         /*
1895          * Note that if we ever decide to split across DSes,
1896          * then we may need to handle dense-like offsets.
1897          */
1898         hdr->args.offset = offset;
1899
1900         /* Perform an asynchronous write */
1901         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1902                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1903                                       &ff_layout_write_call_ops_v4,
1904                           sync, RPC_TASK_SOFTCONN);
1905         put_cred(ds_cred);
1906         return PNFS_ATTEMPTED;
1907
1908 out_failed:
1909         if (ff_layout_avoid_mds_available_ds(lseg))
1910                 return PNFS_TRY_AGAIN;
1911         trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1912                         hdr->args.offset, hdr->args.count,
1913                         IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1914         return PNFS_NOT_ATTEMPTED;
1915 }
1916
1917 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1918 {
1919         return i;
1920 }
1921
1922 static struct nfs_fh *
1923 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1924 {
1925         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1926
1927         /* FIXME: Assume that there is only one NFS version available
1928          * for the DS.
1929          */
1930         return &flseg->mirror_array[i]->fh_versions[0];
1931 }
1932
1933 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1934 {
1935         struct pnfs_layout_segment *lseg = data->lseg;
1936         struct nfs4_pnfs_ds *ds;
1937         struct rpc_clnt *ds_clnt;
1938         struct nfs4_ff_layout_mirror *mirror;
1939         const struct cred *ds_cred;
1940         u32 idx;
1941         int vers, ret;
1942         struct nfs_fh *fh;
1943
1944         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1945             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1946                 goto out_err;
1947
1948         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1949         mirror = FF_LAYOUT_COMP(lseg, idx);
1950         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1951         if (!ds)
1952                 goto out_err;
1953
1954         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1955                                                    data->inode);
1956         if (IS_ERR(ds_clnt))
1957                 goto out_err;
1958
1959         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1960         if (!ds_cred)
1961                 goto out_err;
1962
1963         vers = nfs4_ff_layout_ds_version(mirror);
1964
1965         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1966                 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1967                 vers);
1968         data->commit_done_cb = ff_layout_commit_done_cb;
1969         data->cred = ds_cred;
1970         refcount_inc(&ds->ds_clp->cl_count);
1971         data->ds_clp = ds->ds_clp;
1972         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1973         if (fh)
1974                 data->args.fh = fh;
1975
1976         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1977                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1978                                                &ff_layout_commit_call_ops_v4,
1979                                    how, RPC_TASK_SOFTCONN);
1980         put_cred(ds_cred);
1981         return ret;
1982 out_err:
1983         pnfs_generic_prepare_to_resend_writes(data);
1984         pnfs_generic_commit_release(data);
1985         return -EAGAIN;
1986 }
1987
1988 static int
1989 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1990                            int how, struct nfs_commit_info *cinfo)
1991 {
1992         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1993                                             ff_layout_initiate_commit);
1994 }
1995
1996 static struct pnfs_ds_commit_info *
1997 ff_layout_get_ds_info(struct inode *inode)
1998 {
1999         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2000
2001         if (layout == NULL)
2002                 return NULL;
2003
2004         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2005 }
2006
2007 static void
2008 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2009                 struct inode *inode)
2010 {
2011         spin_lock(&inode->i_lock);
2012         pnfs_generic_ds_cinfo_destroy(fl_cinfo);
2013         spin_unlock(&inode->i_lock);
2014 }
2015
2016 static void
2017 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2018 {
2019         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2020                                                   id_node));
2021 }
2022
2023 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2024                                   const struct nfs4_layoutreturn_args *args,
2025                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
2026 {
2027         __be32 *start;
2028
2029         start = xdr_reserve_space(xdr, 4);
2030         if (unlikely(!start))
2031                 return -E2BIG;
2032
2033         *start = cpu_to_be32(ff_args->num_errors);
2034         /* This assume we always return _ALL_ layouts */
2035         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2036 }
2037
2038 static void
2039 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2040 {
2041         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2042 }
2043
2044 static void
2045 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2046                             const nfs4_stateid *stateid,
2047                             const struct nfs42_layoutstat_devinfo *devinfo)
2048 {
2049         __be32 *p;
2050
2051         p = xdr_reserve_space(xdr, 8 + 8);
2052         p = xdr_encode_hyper(p, devinfo->offset);
2053         p = xdr_encode_hyper(p, devinfo->length);
2054         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2055         p = xdr_reserve_space(xdr, 4*8);
2056         p = xdr_encode_hyper(p, devinfo->read_count);
2057         p = xdr_encode_hyper(p, devinfo->read_bytes);
2058         p = xdr_encode_hyper(p, devinfo->write_count);
2059         p = xdr_encode_hyper(p, devinfo->write_bytes);
2060         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2061 }
2062
2063 static void
2064 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2065                             const nfs4_stateid *stateid,
2066                             const struct nfs42_layoutstat_devinfo *devinfo)
2067 {
2068         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2069         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2070                         devinfo->ld_private.data);
2071 }
2072
2073 /* report nothing for now */
2074 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2075                 const struct nfs4_layoutreturn_args *args,
2076                 struct nfs4_flexfile_layoutreturn_args *ff_args)
2077 {
2078         __be32 *p;
2079         int i;
2080
2081         p = xdr_reserve_space(xdr, 4);
2082         *p = cpu_to_be32(ff_args->num_dev);
2083         for (i = 0; i < ff_args->num_dev; i++)
2084                 ff_layout_encode_ff_iostat(xdr,
2085                                 &args->layout->plh_stateid,
2086                                 &ff_args->devinfo[i]);
2087 }
2088
2089 static void
2090 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2091                 unsigned int num_entries)
2092 {
2093         unsigned int i;
2094
2095         for (i = 0; i < num_entries; i++) {
2096                 if (!devinfo[i].ld_private.ops)
2097                         continue;
2098                 if (!devinfo[i].ld_private.ops->free)
2099                         continue;
2100                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2101         }
2102 }
2103
2104 static struct nfs4_deviceid_node *
2105 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2106                               struct pnfs_device *pdev, gfp_t gfp_flags)
2107 {
2108         struct nfs4_ff_layout_ds *dsaddr;
2109
2110         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2111         if (!dsaddr)
2112                 return NULL;
2113         return &dsaddr->id_node;
2114 }
2115
2116 static void
2117 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2118                 const void *voidargs,
2119                 const struct nfs4_xdr_opaque_data *ff_opaque)
2120 {
2121         const struct nfs4_layoutreturn_args *args = voidargs;
2122         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2123         struct xdr_buf tmp_buf = {
2124                 .head = {
2125                         [0] = {
2126                                 .iov_base = page_address(ff_args->pages[0]),
2127                         },
2128                 },
2129                 .buflen = PAGE_SIZE,
2130         };
2131         struct xdr_stream tmp_xdr;
2132         __be32 *start;
2133
2134         dprintk("%s: Begin\n", __func__);
2135
2136         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2137
2138         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2139         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2140
2141         start = xdr_reserve_space(xdr, 4);
2142         *start = cpu_to_be32(tmp_buf.len);
2143         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2144
2145         dprintk("%s: Return\n", __func__);
2146 }
2147
2148 static void
2149 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2150 {
2151         struct nfs4_flexfile_layoutreturn_args *ff_args;
2152
2153         if (!args->data)
2154                 return;
2155         ff_args = args->data;
2156         args->data = NULL;
2157
2158         ff_layout_free_ds_ioerr(&ff_args->errors);
2159         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2160
2161         put_page(ff_args->pages[0]);
2162         kfree(ff_args);
2163 }
2164
2165 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2166         .encode = ff_layout_encode_layoutreturn,
2167         .free = ff_layout_free_layoutreturn,
2168 };
2169
2170 static int
2171 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2172 {
2173         struct nfs4_flexfile_layoutreturn_args *ff_args;
2174         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2175
2176         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2177         if (!ff_args)
2178                 goto out_nomem;
2179         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2180         if (!ff_args->pages[0])
2181                 goto out_nomem_free;
2182
2183         INIT_LIST_HEAD(&ff_args->errors);
2184         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2185                         &args->range, &ff_args->errors,
2186                         FF_LAYOUTRETURN_MAXERR);
2187
2188         spin_lock(&args->inode->i_lock);
2189         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2190                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2191         spin_unlock(&args->inode->i_lock);
2192
2193         args->ld_private->ops = &layoutreturn_ops;
2194         args->ld_private->data = ff_args;
2195         return 0;
2196 out_nomem_free:
2197         kfree(ff_args);
2198 out_nomem:
2199         return -ENOMEM;
2200 }
2201
2202 #ifdef CONFIG_NFS_V4_2
2203 void
2204 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2205 {
2206         struct pnfs_layout_hdr *lo = lseg->pls_layout;
2207         struct nfs42_layout_error *errors;
2208         LIST_HEAD(head);
2209
2210         if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2211                 return;
2212         ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2213         if (list_empty(&head))
2214                 return;
2215
2216         errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2217                         sizeof(*errors), GFP_NOFS);
2218         if (errors != NULL) {
2219                 const struct nfs4_ff_layout_ds_err *pos;
2220                 size_t n = 0;
2221
2222                 list_for_each_entry(pos, &head, list) {
2223                         errors[n].offset = pos->offset;
2224                         errors[n].length = pos->length;
2225                         nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2226                         errors[n].errors[0].dev_id = pos->deviceid;
2227                         errors[n].errors[0].status = pos->status;
2228                         errors[n].errors[0].opnum = pos->opnum;
2229                         n++;
2230                         if (!list_is_last(&pos->list, &head) &&
2231                             n < NFS42_LAYOUTERROR_MAX)
2232                                 continue;
2233                         if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2234                                 break;
2235                         n = 0;
2236                 }
2237                 kfree(errors);
2238         }
2239         ff_layout_free_ds_ioerr(&head);
2240 }
2241 #else
2242 void
2243 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2244 {
2245 }
2246 #endif
2247
2248 static int
2249 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2250 {
2251         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2252
2253         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2254 }
2255
2256 static size_t
2257 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2258                           const int buflen)
2259 {
2260         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2261         const struct in6_addr *addr = &sin6->sin6_addr;
2262
2263         /*
2264          * RFC 4291, Section 2.2.2
2265          *
2266          * Shorthanded ANY address
2267          */
2268         if (ipv6_addr_any(addr))
2269                 return snprintf(buf, buflen, "::");
2270
2271         /*
2272          * RFC 4291, Section 2.2.2
2273          *
2274          * Shorthanded loopback address
2275          */
2276         if (ipv6_addr_loopback(addr))
2277                 return snprintf(buf, buflen, "::1");
2278
2279         /*
2280          * RFC 4291, Section 2.2.3
2281          *
2282          * Special presentation address format for mapped v4
2283          * addresses.
2284          */
2285         if (ipv6_addr_v4mapped(addr))
2286                 return snprintf(buf, buflen, "::ffff:%pI4",
2287                                         &addr->s6_addr32[3]);
2288
2289         /*
2290          * RFC 4291, Section 2.2.1
2291          */
2292         return snprintf(buf, buflen, "%pI6c", addr);
2293 }
2294
2295 /* Derived from rpc_sockaddr2uaddr */
2296 static void
2297 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2298 {
2299         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2300         char portbuf[RPCBIND_MAXUADDRPLEN];
2301         char addrbuf[RPCBIND_MAXUADDRLEN];
2302         char *netid;
2303         unsigned short port;
2304         int len, netid_len;
2305         __be32 *p;
2306
2307         switch (sap->sa_family) {
2308         case AF_INET:
2309                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2310                         return;
2311                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2312                 netid = "tcp";
2313                 netid_len = 3;
2314                 break;
2315         case AF_INET6:
2316                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2317                         return;
2318                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2319                 netid = "tcp6";
2320                 netid_len = 4;
2321                 break;
2322         default:
2323                 /* we only support tcp and tcp6 */
2324                 WARN_ON_ONCE(1);
2325                 return;
2326         }
2327
2328         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2329         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2330
2331         p = xdr_reserve_space(xdr, 4 + netid_len);
2332         xdr_encode_opaque(p, netid, netid_len);
2333
2334         p = xdr_reserve_space(xdr, 4 + len);
2335         xdr_encode_opaque(p, addrbuf, len);
2336 }
2337
2338 static void
2339 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2340                          ktime_t t)
2341 {
2342         struct timespec64 ts;
2343         __be32 *p;
2344
2345         p = xdr_reserve_space(xdr, 12);
2346         ts = ktime_to_timespec64(t);
2347         p = xdr_encode_hyper(p, ts.tv_sec);
2348         *p++ = cpu_to_be32(ts.tv_nsec);
2349 }
2350
2351 static void
2352 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2353                             struct nfs4_ff_io_stat *stat)
2354 {
2355         __be32 *p;
2356
2357         p = xdr_reserve_space(xdr, 5 * 8);
2358         p = xdr_encode_hyper(p, stat->ops_requested);
2359         p = xdr_encode_hyper(p, stat->bytes_requested);
2360         p = xdr_encode_hyper(p, stat->ops_completed);
2361         p = xdr_encode_hyper(p, stat->bytes_completed);
2362         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2363         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2364         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2365 }
2366
2367 static void
2368 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2369                               const struct nfs42_layoutstat_devinfo *devinfo,
2370                               struct nfs4_ff_layout_mirror *mirror)
2371 {
2372         struct nfs4_pnfs_ds_addr *da;
2373         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2374         struct nfs_fh *fh = &mirror->fh_versions[0];
2375         __be32 *p;
2376
2377         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2378         dprintk("%s: DS %s: encoding address %s\n",
2379                 __func__, ds->ds_remotestr, da->da_remotestr);
2380         /* netaddr4 */
2381         ff_layout_encode_netaddr(xdr, da);
2382         /* nfs_fh4 */
2383         p = xdr_reserve_space(xdr, 4 + fh->size);
2384         xdr_encode_opaque(p, fh->data, fh->size);
2385         /* ff_io_latency4 read */
2386         spin_lock(&mirror->lock);
2387         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2388         /* ff_io_latency4 write */
2389         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2390         spin_unlock(&mirror->lock);
2391         /* nfstime4 */
2392         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2393         /* bool */
2394         p = xdr_reserve_space(xdr, 4);
2395         *p = cpu_to_be32(false);
2396 }
2397
2398 static void
2399 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2400                              const struct nfs4_xdr_opaque_data *opaque)
2401 {
2402         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2403                         struct nfs42_layoutstat_devinfo, ld_private);
2404         __be32 *start;
2405
2406         /* layoutupdate length */
2407         start = xdr_reserve_space(xdr, 4);
2408         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2409
2410         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2411 }
2412
2413 static void
2414 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2415 {
2416         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2417
2418         ff_layout_put_mirror(mirror);
2419 }
2420
2421 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2422         .encode = ff_layout_encode_layoutstats,
2423         .free   = ff_layout_free_layoutstats,
2424 };
2425
2426 static int
2427 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2428                                struct nfs42_layoutstat_devinfo *devinfo,
2429                                int dev_limit)
2430 {
2431         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2432         struct nfs4_ff_layout_mirror *mirror;
2433         struct nfs4_deviceid_node *dev;
2434         int i = 0;
2435
2436         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2437                 if (i >= dev_limit)
2438                         break;
2439                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2440                         continue;
2441                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2442                         continue;
2443                 /* mirror refcount put in cleanup_layoutstats */
2444                 if (!refcount_inc_not_zero(&mirror->ref))
2445                         continue;
2446                 dev = &mirror->mirror_ds->id_node;
2447                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2448                 devinfo->offset = 0;
2449                 devinfo->length = NFS4_MAX_UINT64;
2450                 spin_lock(&mirror->lock);
2451                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2452                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2453                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2454                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2455                 spin_unlock(&mirror->lock);
2456                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2457                 devinfo->ld_private.ops = &layoutstat_ops;
2458                 devinfo->ld_private.data = mirror;
2459
2460                 devinfo++;
2461                 i++;
2462         }
2463         return i;
2464 }
2465
2466 static int
2467 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2468 {
2469         struct nfs4_flexfile_layout *ff_layout;
2470         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2471
2472         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2473         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2474         if (!args->devinfo)
2475                 return -ENOMEM;
2476
2477         spin_lock(&args->inode->i_lock);
2478         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2479         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2480                         &args->devinfo[0], dev_count);
2481         spin_unlock(&args->inode->i_lock);
2482         if (!args->num_dev) {
2483                 kfree(args->devinfo);
2484                 args->devinfo = NULL;
2485                 return -ENOENT;
2486         }
2487
2488         return 0;
2489 }
2490
2491 static int
2492 ff_layout_set_layoutdriver(struct nfs_server *server,
2493                 const struct nfs_fh *dummy)
2494 {
2495 #if IS_ENABLED(CONFIG_NFS_V4_2)
2496         server->caps |= NFS_CAP_LAYOUTSTATS;
2497 #endif
2498         return 0;
2499 }
2500
2501 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2502         .id                     = LAYOUT_FLEX_FILES,
2503         .name                   = "LAYOUT_FLEX_FILES",
2504         .owner                  = THIS_MODULE,
2505         .flags                  = PNFS_LAYOUTGET_ON_OPEN,
2506         .max_layoutget_response = 4096, /* 1 page or so... */
2507         .set_layoutdriver       = ff_layout_set_layoutdriver,
2508         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2509         .free_layout_hdr        = ff_layout_free_layout_hdr,
2510         .alloc_lseg             = ff_layout_alloc_lseg,
2511         .free_lseg              = ff_layout_free_lseg,
2512         .add_lseg               = ff_layout_add_lseg,
2513         .pg_read_ops            = &ff_layout_pg_read_ops,
2514         .pg_write_ops           = &ff_layout_pg_write_ops,
2515         .get_ds_info            = ff_layout_get_ds_info,
2516         .release_ds_info        = ff_layout_release_ds_info,
2517         .free_deviceid_node     = ff_layout_free_deviceid_node,
2518         .mark_request_commit    = pnfs_layout_mark_request_commit,
2519         .clear_request_commit   = pnfs_generic_clear_request_commit,
2520         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2521         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2522         .commit_pagelist        = ff_layout_commit_pagelist,
2523         .read_pagelist          = ff_layout_read_pagelist,
2524         .write_pagelist         = ff_layout_write_pagelist,
2525         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2526         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2527         .sync                   = pnfs_nfs_generic_sync,
2528         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2529 };
2530
2531 static int __init nfs4flexfilelayout_init(void)
2532 {
2533         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2534                __func__);
2535         return pnfs_register_layoutdriver(&flexfilelayout_type);
2536 }
2537
2538 static void __exit nfs4flexfilelayout_exit(void)
2539 {
2540         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2541                __func__);
2542         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2543 }
2544
2545 MODULE_ALIAS("nfs-layouttype4-4");
2546
2547 MODULE_LICENSE("GPL");
2548 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2549
2550 module_init(nfs4flexfilelayout_init);
2551 module_exit(nfs4flexfilelayout_exit);
2552
2553 module_param(io_maxretrans, ushort, 0644);
2554 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2555                         "retries an I/O request before returning an error. ");