fs/nfs/flexfilelayout/flexfilelayout.c

   1 /*
   2  * Module for pnfs flexfile layout driver.
   3  *
   4  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   5  *
   6  * Tao Peng <bergwolf@primarydata.com>
   7  */
   8
   9 #include <linux/nfs_fs.h>
  10 #include <linux/nfs_page.h>
  11 #include <linux/module.h>
  12
  13 #include <linux/sunrpc/metrics.h>
  14
  15 #include "flexfilelayout.h"
  16 #include "../nfs4session.h"
  17 #include "../nfs4idmap.h"
  18 #include "../internal.h"
  19 #include "../delegation.h"
  20 #include "../nfs4trace.h"
  21 #include "../iostat.h"
  22 #include "../nfs.h"
  23 #include "../nfs42.h"
  24
  25 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  26
  27 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  28
  29 static struct pnfs_layout_hdr *
  30 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  31 {
  32         struct nfs4_flexfile_layout *ffl;
  33
  34         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  35         if (ffl) {
  36                 INIT_LIST_HEAD(&ffl->error_list);
  37                 INIT_LIST_HEAD(&ffl->mirrors);
  38                 return &ffl->generic_hdr;
  39         } else
  40                 return NULL;
  41 }
  42
  43 static void
  44 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  45 {
  46         struct nfs4_ff_layout_ds_err *err, *n;
  47
  48         list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
  49                                  list) {
  50                 list_del(&err->list);
  51                 kfree(err);
  52         }
  53         kfree(FF_LAYOUT_FROM_HDR(lo));
  54 }
  55
  56 static int decode_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  57 {
  58         __be32 *p;
  59
  60         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  61         if (unlikely(p == NULL))
  62                 return -ENOBUFS;
  63         memcpy(stateid, p, NFS4_STATEID_SIZE);
  64         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  65                 p[0], p[1], p[2], p[3]);
  66         return 0;
  67 }
  68
  69 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  70 {
  71         __be32 *p;
  72
  73         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  74         if (unlikely(!p))
  75                 return -ENOBUFS;
  76         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  77         nfs4_print_deviceid(devid);
  78         return 0;
  79 }
  80
  81 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
  82 {
  83         __be32 *p;
  84
  85         p = xdr_inline_decode(xdr, 4);
  86         if (unlikely(!p))
  87                 return -ENOBUFS;
  88         fh->size = be32_to_cpup(p++);
  89         if (fh->size > sizeof(struct nfs_fh)) {
  90                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
  91                        fh->size);
  92                 return -EOVERFLOW;
  93         }
  94         /* fh.data */
  95         p = xdr_inline_decode(xdr, fh->size);
  96         if (unlikely(!p))
  97                 return -ENOBUFS;
  98         memcpy(&fh->data, p, fh->size);
  99         dprintk("%s: fh len %d\n", __func__, fh->size);
 100
 101         return 0;
 102 }
 103
 104 /*
 105  * Currently only stringified uids and gids are accepted.
 106  * I.e., kerberos is not supported to the DSes, so no pricipals.
 107  *
 108  * That means that one common function will suffice, but when
 109  * principals are added, this should be split to accomodate
 110  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 111  */
 112 static int
 113 decode_name(struct xdr_stream *xdr, u32 *id)
 114 {
 115         __be32 *p;
 116         int len;
 117
 118         /* opaque_length(4)*/
 119         p = xdr_inline_decode(xdr, 4);
 120         if (unlikely(!p))
 121                 return -ENOBUFS;
 122         len = be32_to_cpup(p++);
 123         if (len < 0)
 124                 return -EINVAL;
 125
 126         dprintk("%s: len %u\n", __func__, len);
 127
 128         /* opaque body */
 129         p = xdr_inline_decode(xdr, len);
 130         if (unlikely(!p))
 131                 return -ENOBUFS;
 132
 133         if (!nfs_map_string_to_numeric((char *)p, len, id))
 134                 return -EINVAL;
 135
 136         return 0;
 137 }
 138
 139 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 140                 const struct nfs4_ff_layout_mirror *m2)
 141 {
 142         int i, j;
 143
 144         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 145                 return false;
 146         for (i = 0; i < m1->fh_versions_cnt; i++) {
 147                 bool found_fh = false;
 148                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 149                         if (nfs_compare_fh(&m1->fh_versions[i],
 150                                         &m2->fh_versions[j]) == 0) {
 151                                 found_fh = true;
 152                                 break;
 153                         }
 154                 }
 155                 if (!found_fh)
 156                         return false;
 157         }
 158         return true;
 159 }
 160
 161 static struct nfs4_ff_layout_mirror *
 162 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 163                 struct nfs4_ff_layout_mirror *mirror)
 164 {
 165         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 166         struct nfs4_ff_layout_mirror *pos;
 167         struct inode *inode = lo->plh_inode;
 168
 169         spin_lock(&inode->i_lock);
 170         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 171                 if (mirror->mirror_ds != pos->mirror_ds)
 172                         continue;
 173                 if (!ff_mirror_match_fh(mirror, pos))
 174                         continue;
 175                 if (atomic_inc_not_zero(&pos->ref)) {
 176                         spin_unlock(&inode->i_lock);
 177                         return pos;
 178                 }
 179         }
 180         list_add(&mirror->mirrors, &ff_layout->mirrors);
 181         mirror->layout = lo;
 182         spin_unlock(&inode->i_lock);
 183         return mirror;
 184 }
 185
 186 static void
 187 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 188 {
 189         struct inode *inode;
 190         if (mirror->layout == NULL)
 191                 return;
 192         inode = mirror->layout->plh_inode;
 193         spin_lock(&inode->i_lock);
 194         list_del(&mirror->mirrors);
 195         spin_unlock(&inode->i_lock);
 196         mirror->layout = NULL;
 197 }
 198
 199 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 200 {
 201         struct nfs4_ff_layout_mirror *mirror;
 202
 203         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 204         if (mirror != NULL) {
 205                 spin_lock_init(&mirror->lock);
 206                 atomic_set(&mirror->ref, 1);
 207                 INIT_LIST_HEAD(&mirror->mirrors);
 208         }
 209         return mirror;
 210 }
 211
 212 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 213 {
 214         ff_layout_remove_mirror(mirror);
 215         kfree(mirror->fh_versions);
 216         if (mirror->cred)
 217                 put_rpccred(mirror->cred);
 218         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 219         kfree(mirror);
 220 }
 221
 222 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 223 {
 224         if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
 225                 ff_layout_free_mirror(mirror);
 226 }
 227
 228 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 229 {
 230         int i;
 231
 232         if (fls->mirror_array) {
 233                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 234                         /* normally mirror_ds is freed in
 235                          * .free_deviceid_node but we still do it here
 236                          * for .alloc_lseg error path */
 237                         ff_layout_put_mirror(fls->mirror_array[i]);
 238                 }
 239                 kfree(fls->mirror_array);
 240                 fls->mirror_array = NULL;
 241         }
 242 }
 243
 244 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 245 {
 246         int ret = 0;
 247
 248         dprintk("--> %s\n", __func__);
 249
 250         /* FIXME: remove this check when layout segment support is added */
 251         if (lgr->range.offset != 0 ||
 252             lgr->range.length != NFS4_MAX_UINT64) {
 253                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 254                         __func__);
 255                 ret = -EINVAL;
 256         }
 257
 258         dprintk("--> %s returns %d\n", __func__, ret);
 259         return ret;
 260 }
 261
 262 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 263 {
 264         if (fls) {
 265                 ff_layout_free_mirror_array(fls);
 266                 kfree(fls);
 267         }
 268 }
 269
 270 static bool
 271 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 272                 const struct pnfs_layout_range *l2)
 273 {
 274         u64 end1, end2;
 275
 276         if (l1->iomode != l2->iomode)
 277                 return l1->iomode != IOMODE_READ;
 278         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 279         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 280         if (end1 < l2->offset)
 281                 return false;
 282         if (end2 < l1->offset)
 283                 return true;
 284         return l2->offset <= l1->offset;
 285 }
 286
 287 static bool
 288 ff_lseg_merge(struct pnfs_layout_segment *new,
 289                 struct pnfs_layout_segment *old)
 290 {
 291         u64 new_end, old_end;
 292
 293         if (new->pls_range.iomode != old->pls_range.iomode)
 294                 return false;
 295         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 296                         old->pls_range.length);
 297         if (old_end < new->pls_range.offset)
 298                 return false;
 299         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 300                         new->pls_range.length);
 301         if (new_end < old->pls_range.offset)
 302                 return false;
 303
 304         /* Mergeable: copy info from 'old' to 'new' */
 305         if (new_end < old_end)
 306                 new_end = old_end;
 307         if (new->pls_range.offset < old->pls_range.offset)
 308                 new->pls_range.offset = old->pls_range.offset;
 309         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 310                         new_end);
 311         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 312                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 313         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 314                 set_bit(NFS_LSEG_LAYOUTRETURN, &new->pls_flags);
 315         return true;
 316 }
 317
 318 static void
 319 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 320                 struct pnfs_layout_segment *lseg,
 321                 struct list_head *free_me)
 322 {
 323         pnfs_generic_layout_insert_lseg(lo, lseg,
 324                         ff_lseg_range_is_after,
 325                         ff_lseg_merge,
 326                         free_me);
 327 }
 328
 329 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 330 {
 331         int i, j;
 332
 333         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 334                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 335                         if (fls->mirror_array[i]->efficiency <
 336                             fls->mirror_array[j]->efficiency)
 337                                 swap(fls->mirror_array[i],
 338                                      fls->mirror_array[j]);
 339         }
 340 }
 341
 342 static void ff_layout_mark_devices_valid(struct nfs4_ff_layout_segment *fls)
 343 {
 344         struct nfs4_deviceid_node *node;
 345         int i;
 346
 347         if (!(fls->flags & FF_FLAGS_NO_IO_THRU_MDS))
 348                 return;
 349         for (i = 0; i < fls->mirror_array_cnt; i++) {
 350                 node = &fls->mirror_array[i]->mirror_ds->id_node;
 351                 clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
 352         }
 353 }
 354
 355 static struct pnfs_layout_segment *
 356 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 357                      struct nfs4_layoutget_res *lgr,
 358                      gfp_t gfp_flags)
 359 {
 360         struct pnfs_layout_segment *ret;
 361         struct nfs4_ff_layout_segment *fls = NULL;
 362         struct xdr_stream stream;
 363         struct xdr_buf buf;
 364         struct page *scratch;
 365         u64 stripe_unit;
 366         u32 mirror_array_cnt;
 367         __be32 *p;
 368         int i, rc;
 369
 370         dprintk("--> %s\n", __func__);
 371         scratch = alloc_page(gfp_flags);
 372         if (!scratch)
 373                 return ERR_PTR(-ENOMEM);
 374
 375         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 376                               lgr->layoutp->len);
 377         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 378
 379         /* stripe unit and mirror_array_cnt */
 380         rc = -EIO;
 381         p = xdr_inline_decode(&stream, 8 + 4);
 382         if (!p)
 383                 goto out_err_free;
 384
 385         p = xdr_decode_hyper(p, &stripe_unit);
 386         mirror_array_cnt = be32_to_cpup(p++);
 387         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 388                 stripe_unit, mirror_array_cnt);
 389
 390         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 391             mirror_array_cnt == 0)
 392                 goto out_err_free;
 393
 394         rc = -ENOMEM;
 395         fls = kzalloc(sizeof(*fls), gfp_flags);
 396         if (!fls)
 397                 goto out_err_free;
 398
 399         fls->mirror_array_cnt = mirror_array_cnt;
 400         fls->stripe_unit = stripe_unit;
 401         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 402                                     sizeof(fls->mirror_array[0]), gfp_flags);
 403         if (fls->mirror_array == NULL)
 404                 goto out_err_free;
 405
 406         for (i = 0; i < fls->mirror_array_cnt; i++) {
 407                 struct nfs4_ff_layout_mirror *mirror;
 408                 struct nfs4_deviceid devid;
 409                 struct nfs4_deviceid_node *idnode;
 410                 u32 ds_count;
 411                 u32 fh_count;
 412                 int j;
 413
 414                 rc = -EIO;
 415                 p = xdr_inline_decode(&stream, 4);
 416                 if (!p)
 417                         goto out_err_free;
 418                 ds_count = be32_to_cpup(p);
 419
 420                 /* FIXME: allow for striping? */
 421                 if (ds_count != 1)
 422                         goto out_err_free;
 423
 424                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 425                 if (fls->mirror_array[i] == NULL) {
 426                         rc = -ENOMEM;
 427                         goto out_err_free;
 428                 }
 429
 430                 fls->mirror_array[i]->ds_count = ds_count;
 431
 432                 /* deviceid */
 433                 rc = decode_deviceid(&stream, &devid);
 434                 if (rc)
 435                         goto out_err_free;
 436
 437                 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 438                                                 &devid, lh->plh_lc_cred,
 439                                                 gfp_flags);
 440                 /*
 441                  * upon success, mirror_ds is allocated by previous
 442                  * getdeviceinfo, or newly by .alloc_deviceid_node
 443                  * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
 444                  */
 445                 if (idnode)
 446                         fls->mirror_array[i]->mirror_ds =
 447                                 FF_LAYOUT_MIRROR_DS(idnode);
 448                 else
 449                         goto out_err_free;
 450
 451                 /* efficiency */
 452                 rc = -EIO;
 453                 p = xdr_inline_decode(&stream, 4);
 454                 if (!p)
 455                         goto out_err_free;
 456                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 457
 458                 /* stateid */
 459                 rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
 460                 if (rc)
 461                         goto out_err_free;
 462
 463                 /* fh */
 464                 p = xdr_inline_decode(&stream, 4);
 465                 if (!p)
 466                         goto out_err_free;
 467                 fh_count = be32_to_cpup(p);
 468
 469                 fls->mirror_array[i]->fh_versions =
 470                         kzalloc(fh_count * sizeof(struct nfs_fh),
 471                                 gfp_flags);
 472                 if (fls->mirror_array[i]->fh_versions == NULL) {
 473                         rc = -ENOMEM;
 474                         goto out_err_free;
 475                 }
 476
 477                 for (j = 0; j < fh_count; j++) {
 478                         rc = decode_nfs_fh(&stream,
 479                                            &fls->mirror_array[i]->fh_versions[j]);
 480                         if (rc)
 481                                 goto out_err_free;
 482                 }
 483
 484                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 485
 486                 /* user */
 487                 rc = decode_name(&stream, &fls->mirror_array[i]->uid);
 488                 if (rc)
 489                         goto out_err_free;
 490
 491                 /* group */
 492                 rc = decode_name(&stream, &fls->mirror_array[i]->gid);
 493                 if (rc)
 494                         goto out_err_free;
 495
 496                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 497                 if (mirror != fls->mirror_array[i]) {
 498                         ff_layout_free_mirror(fls->mirror_array[i]);
 499                         fls->mirror_array[i] = mirror;
 500                 }
 501
 502                 dprintk("%s: uid %d gid %d\n", __func__,
 503                         fls->mirror_array[i]->uid,
 504                         fls->mirror_array[i]->gid);
 505         }
 506
 507         p = xdr_inline_decode(&stream, 4);
 508         if (!p)
 509                 goto out_sort_mirrors;
 510         fls->flags = be32_to_cpup(p);
 511
 512         p = xdr_inline_decode(&stream, 4);
 513         if (!p)
 514                 goto out_sort_mirrors;
 515         for (i=0; i < fls->mirror_array_cnt; i++)
 516                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 517
 518 out_sort_mirrors:
 519         ff_layout_sort_mirrors(fls);
 520         rc = ff_layout_check_layout(lgr);
 521         if (rc)
 522                 goto out_err_free;
 523         ff_layout_mark_devices_valid(fls);
 524
 525         ret = &fls->generic_hdr;
 526         dprintk("<-- %s (success)\n", __func__);
 527 out_free_page:
 528         __free_page(scratch);
 529         return ret;
 530 out_err_free:
 531         _ff_layout_free_lseg(fls);
 532         ret = ERR_PTR(rc);
 533         dprintk("<-- %s (%d)\n", __func__, rc);
 534         goto out_free_page;
 535 }
 536
 537 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 538 {
 539         struct pnfs_layout_segment *lseg;
 540
 541         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 542                 if (lseg->pls_range.iomode == IOMODE_RW)
 543                         return true;
 544
 545         return false;
 546 }
 547
 548 static void
 549 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 550 {
 551         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 552
 553         dprintk("--> %s\n", __func__);
 554
 555         if (lseg->pls_range.iomode == IOMODE_RW) {
 556                 struct nfs4_flexfile_layout *ffl;
 557                 struct inode *inode;
 558
 559                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 560                 inode = ffl->generic_hdr.plh_inode;
 561                 spin_lock(&inode->i_lock);
 562                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 563                         ffl->commit_info.nbuckets = 0;
 564                         kfree(ffl->commit_info.buckets);
 565                         ffl->commit_info.buckets = NULL;
 566                 }
 567                 spin_unlock(&inode->i_lock);
 568         }
 569         _ff_layout_free_lseg(fls);
 570 }
 571
 572 /* Return 1 until we have multiple lsegs support */
 573 static int
 574 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 575 {
 576         return 1;
 577 }
 578
 579 static void
 580 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 581 {
 582         /* first IO request? */
 583         if (atomic_inc_return(&timer->n_ops) == 1) {
 584                 timer->start_time = now;
 585         }
 586 }
 587
 588 static ktime_t
 589 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 590 {
 591         ktime_t start;
 592
 593         if (atomic_dec_return(&timer->n_ops) < 0)
 594                 WARN_ON_ONCE(1);
 595
 596         start = timer->start_time;
 597         timer->start_time = now;
 598         return ktime_sub(now, start);
 599 }
 600
 601 static bool
 602 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 603                             struct nfs4_ff_layoutstat *layoutstat,
 604                             ktime_t now)
 605 {
 606         static const ktime_t notime = {0};
 607         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 608
 609         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 610         if (ktime_equal(mirror->start_time, notime))
 611                 mirror->start_time = now;
 612         if (ktime_equal(mirror->last_report_time, notime))
 613                 mirror->last_report_time = now;
 614         if (mirror->report_interval != 0)
 615                 report_interval = (s64)mirror->report_interval * 1000LL;
 616         else if (layoutstats_timer != 0)
 617                 report_interval = (s64)layoutstats_timer * 1000LL;
 618         if (ktime_to_ms(ktime_sub(now, mirror->last_report_time)) >=
 619                         report_interval) {
 620                 mirror->last_report_time = now;
 621                 return true;
 622         }
 623
 624         return false;
 625 }
 626
 627 static void
 628 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 629                 __u64 requested)
 630 {
 631         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 632
 633         iostat->ops_requested++;
 634         iostat->bytes_requested += requested;
 635 }
 636
 637 static void
 638 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 639                 __u64 requested,
 640                 __u64 completed,
 641                 ktime_t time_completed,
 642                 ktime_t time_started)
 643 {
 644         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 645         ktime_t completion_time = ktime_sub(time_completed, time_started);
 646         ktime_t timer;
 647
 648         iostat->ops_completed++;
 649         iostat->bytes_completed += completed;
 650         iostat->bytes_not_delivered += requested - completed;
 651
 652         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 653         iostat->total_busy_time =
 654                         ktime_add(iostat->total_busy_time, timer);
 655         iostat->aggregate_completion_time =
 656                         ktime_add(iostat->aggregate_completion_time,
 657                                         completion_time);
 658 }
 659
 660 static void
 661 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 662                 struct nfs4_ff_layout_mirror *mirror,
 663                 __u64 requested, ktime_t now)
 664 {
 665         bool report;
 666
 667         spin_lock(&mirror->lock);
 668         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 669         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 670         spin_unlock(&mirror->lock);
 671
 672         if (report)
 673                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 674 }
 675
 676 static void
 677 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 678                 struct nfs4_ff_layout_mirror *mirror,
 679                 __u64 requested,
 680                 __u64 completed)
 681 {
 682         spin_lock(&mirror->lock);
 683         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 684                         requested, completed,
 685                         ktime_get(), task->tk_start);
 686         spin_unlock(&mirror->lock);
 687 }
 688
 689 static void
 690 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 691                 struct nfs4_ff_layout_mirror *mirror,
 692                 __u64 requested, ktime_t now)
 693 {
 694         bool report;
 695
 696         spin_lock(&mirror->lock);
 697         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 698         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 699         spin_unlock(&mirror->lock);
 700
 701         if (report)
 702                 pnfs_report_layoutstat(inode, GFP_NOIO);
 703 }
 704
 705 static void
 706 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 707                 struct nfs4_ff_layout_mirror *mirror,
 708                 __u64 requested,
 709                 __u64 completed,
 710                 enum nfs3_stable_how committed)
 711 {
 712         if (committed == NFS_UNSTABLE)
 713                 requested = completed = 0;
 714
 715         spin_lock(&mirror->lock);
 716         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 717                         requested, completed, ktime_get(), task->tk_start);
 718         spin_unlock(&mirror->lock);
 719 }
 720
 721 static int
 722 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 723                             struct nfs_commit_info *cinfo,
 724                             gfp_t gfp_flags)
 725 {
 726         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 727         struct pnfs_commit_bucket *buckets;
 728         int size;
 729
 730         if (cinfo->ds->nbuckets != 0) {
 731                 /* This assumes there is only one RW lseg per file.
 732                  * To support multiple lseg per file, we need to
 733                  * change struct pnfs_commit_bucket to allow dynamic
 734                  * increasing nbuckets.
 735                  */
 736                 return 0;
 737         }
 738
 739         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 740
 741         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 742                           gfp_flags);
 743         if (!buckets)
 744                 return -ENOMEM;
 745         else {
 746                 int i;
 747
 748                 spin_lock(cinfo->lock);
 749                 if (cinfo->ds->nbuckets != 0)
 750                         kfree(buckets);
 751                 else {
 752                         cinfo->ds->buckets = buckets;
 753                         cinfo->ds->nbuckets = size;
 754                         for (i = 0; i < size; i++) {
 755                                 INIT_LIST_HEAD(&buckets[i].written);
 756                                 INIT_LIST_HEAD(&buckets[i].committing);
 757                                 /* mark direct verifier as unset */
 758                                 buckets[i].direct_verf.committed =
 759                                         NFS_INVALID_STABLE_HOW;
 760                         }
 761                 }
 762                 spin_unlock(cinfo->lock);
 763                 return 0;
 764         }
 765 }
 766
 767 static struct nfs4_pnfs_ds *
 768 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 769                                   int start_idx,
 770                                   int *best_idx)
 771 {
 772         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 773         struct nfs4_pnfs_ds *ds;
 774         int idx;
 775
 776         /* mirrors are sorted by efficiency */
 777         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 778                 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
 779                 if (ds) {
 780                         *best_idx = idx;
 781                         return ds;
 782                 }
 783         }
 784
 785         return NULL;
 786 }
 787
 788 static void
 789 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 790                         struct nfs_page *req)
 791 {
 792         struct nfs_pgio_mirror *pgm;
 793         struct nfs4_ff_layout_mirror *mirror;
 794         struct nfs4_pnfs_ds *ds;
 795         int ds_idx;
 796
 797         /* Use full layout for now */
 798         if (!pgio->pg_lseg) {
 799                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 800                                                    req->wb_context,
 801                                                    0,
 802                                                    NFS4_MAX_UINT64,
 803                                                    IOMODE_READ,
 804                                                    GFP_KERNEL);
 805                 if (IS_ERR(pgio->pg_lseg)) {
 806                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 807                         pgio->pg_lseg = NULL;
 808                         return;
 809                 }
 810         }
 811         /* If no lseg, fall back to read through mds */
 812         if (pgio->pg_lseg == NULL)
 813                 goto out_mds;
 814
 815         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 816         if (!ds)
 817                 goto out_mds;
 818         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 819
 820         pgio->pg_mirror_idx = ds_idx;
 821
 822         /* read always uses only one mirror - idx 0 for pgio layer */
 823         pgm = &pgio->pg_mirrors[0];
 824         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 825
 826         return;
 827 out_mds:
 828         pnfs_put_lseg(pgio->pg_lseg);
 829         pgio->pg_lseg = NULL;
 830         nfs_pageio_reset_read_mds(pgio);
 831 }
 832
 833 static void
 834 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 835                         struct nfs_page *req)
 836 {
 837         struct nfs4_ff_layout_mirror *mirror;
 838         struct nfs_pgio_mirror *pgm;
 839         struct nfs_commit_info cinfo;
 840         struct nfs4_pnfs_ds *ds;
 841         int i;
 842         int status;
 843
 844         if (!pgio->pg_lseg) {
 845                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 846                                                    req->wb_context,
 847                                                    0,
 848                                                    NFS4_MAX_UINT64,
 849                                                    IOMODE_RW,
 850                                                    GFP_NOFS);
 851                 if (IS_ERR(pgio->pg_lseg)) {
 852                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 853                         pgio->pg_lseg = NULL;
 854                         return;
 855                 }
 856         }
 857         /* If no lseg, fall back to write through mds */
 858         if (pgio->pg_lseg == NULL)
 859                 goto out_mds;
 860
 861         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 862         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 863         if (status < 0)
 864                 goto out_mds;
 865
 866         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 867         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 868             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 869                 goto out_mds;
 870
 871         for (i = 0; i < pgio->pg_mirror_count; i++) {
 872                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 873                 if (!ds)
 874                         goto out_mds;
 875                 pgm = &pgio->pg_mirrors[i];
 876                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 877                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 878         }
 879
 880         return;
 881
 882 out_mds:
 883         pnfs_put_lseg(pgio->pg_lseg);
 884         pgio->pg_lseg = NULL;
 885         nfs_pageio_reset_write_mds(pgio);
 886 }
 887
 888 static unsigned int
 889 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 890                                     struct nfs_page *req)
 891 {
 892         if (!pgio->pg_lseg) {
 893                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 894                                                    req->wb_context,
 895                                                    0,
 896                                                    NFS4_MAX_UINT64,
 897                                                    IOMODE_RW,
 898                                                    GFP_NOFS);
 899                 if (IS_ERR(pgio->pg_lseg)) {
 900                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 901                         pgio->pg_lseg = NULL;
 902                         goto out;
 903                 }
 904         }
 905         if (pgio->pg_lseg)
 906                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 907
 908         /* no lseg means that pnfs is not in use, so no mirroring here */
 909         nfs_pageio_reset_write_mds(pgio);
 910 out:
 911         return 1;
 912 }
 913
 914 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 915         .pg_init = ff_layout_pg_init_read,
 916         .pg_test = pnfs_generic_pg_test,
 917         .pg_doio = pnfs_generic_pg_readpages,
 918         .pg_cleanup = pnfs_generic_pg_cleanup,
 919 };
 920
 921 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 922         .pg_init = ff_layout_pg_init_write,
 923         .pg_test = pnfs_generic_pg_test,
 924         .pg_doio = pnfs_generic_pg_writepages,
 925         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 926         .pg_cleanup = pnfs_generic_pg_cleanup,
 927 };
 928
 929 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 930 {
 931         struct rpc_task *task = &hdr->task;
 932
 933         pnfs_layoutcommit_inode(hdr->inode, false);
 934
 935         if (retry_pnfs) {
 936                 dprintk("%s Reset task %5u for i/o through pNFS "
 937                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 938                         hdr->task.tk_pid,
 939                         hdr->inode->i_sb->s_id,
 940                         (unsigned long long)NFS_FILEID(hdr->inode),
 941                         hdr->args.count,
 942                         (unsigned long long)hdr->args.offset);
 943
 944                 hdr->completion_ops->reschedule_io(hdr);
 945                 return;
 946         }
 947
 948         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 949                 dprintk("%s Reset task %5u for i/o through MDS "
 950                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 951                         hdr->task.tk_pid,
 952                         hdr->inode->i_sb->s_id,
 953                         (unsigned long long)NFS_FILEID(hdr->inode),
 954                         hdr->args.count,
 955                         (unsigned long long)hdr->args.offset);
 956
 957                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
 958         }
 959 }
 960
 961 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
 962 {
 963         struct rpc_task *task = &hdr->task;
 964
 965         pnfs_layoutcommit_inode(hdr->inode, false);
 966
 967         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 968                 dprintk("%s Reset task %5u for i/o through MDS "
 969                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 970                         hdr->task.tk_pid,
 971                         hdr->inode->i_sb->s_id,
 972                         (unsigned long long)NFS_FILEID(hdr->inode),
 973                         hdr->args.count,
 974                         (unsigned long long)hdr->args.offset);
 975
 976                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
 977         }
 978 }
 979
 980 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
 981                                            struct nfs4_state *state,
 982                                            struct nfs_client *clp,
 983                                            struct pnfs_layout_segment *lseg,
 984                                            int idx)
 985 {
 986         struct pnfs_layout_hdr *lo = lseg->pls_layout;
 987         struct inode *inode = lo->plh_inode;
 988         struct nfs_server *mds_server = NFS_SERVER(inode);
 989
 990         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 991         struct nfs_client *mds_client = mds_server->nfs_client;
 992         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 993
 994         if (task->tk_status >= 0)
 995                 return 0;
 996
 997         switch (task->tk_status) {
 998         /* MDS state errors */
 999         case -NFS4ERR_DELEG_REVOKED:
1000         case -NFS4ERR_ADMIN_REVOKED:
1001         case -NFS4ERR_BAD_STATEID:
1002                 if (state == NULL)
1003                         break;
1004                 nfs_remove_bad_delegation(state->inode);
1005         case -NFS4ERR_OPENMODE:
1006                 if (state == NULL)
1007                         break;
1008                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1009                         goto out_bad_stateid;
1010                 goto wait_on_recovery;
1011         case -NFS4ERR_EXPIRED:
1012                 if (state != NULL) {
1013                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1014                                 goto out_bad_stateid;
1015                 }
1016                 nfs4_schedule_lease_recovery(mds_client);
1017                 goto wait_on_recovery;
1018         /* DS session errors */
1019         case -NFS4ERR_BADSESSION:
1020         case -NFS4ERR_BADSLOT:
1021         case -NFS4ERR_BAD_HIGH_SLOT:
1022         case -NFS4ERR_DEADSESSION:
1023         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1024         case -NFS4ERR_SEQ_FALSE_RETRY:
1025         case -NFS4ERR_SEQ_MISORDERED:
1026                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1027                         "flags 0x%x\n", __func__, task->tk_status,
1028                         clp->cl_exchange_flags);
1029                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1030                 break;
1031         case -NFS4ERR_DELAY:
1032         case -NFS4ERR_GRACE:
1033                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1034                 break;
1035         case -NFS4ERR_RETRY_UNCACHED_REP:
1036                 break;
1037         /* Invalidate Layout errors */
1038         case -NFS4ERR_PNFS_NO_LAYOUT:
1039         case -ESTALE:           /* mapped NFS4ERR_STALE */
1040         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1041         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1042         case -NFS4ERR_FHEXPIRED:
1043         case -NFS4ERR_WRONG_TYPE:
1044                 dprintk("%s Invalid layout error %d\n", __func__,
1045                         task->tk_status);
1046                 /*
1047                  * Destroy layout so new i/o will get a new layout.
1048                  * Layout will not be destroyed until all current lseg
1049                  * references are put. Mark layout as invalid to resend failed
1050                  * i/o and all i/o waiting on the slot table to the MDS until
1051                  * layout is destroyed and a new valid layout is obtained.
1052                  */
1053                 pnfs_destroy_layout(NFS_I(inode));
1054                 rpc_wake_up(&tbl->slot_tbl_waitq);
1055                 goto reset;
1056         /* RPC connection errors */
1057         case -ECONNREFUSED:
1058         case -EHOSTDOWN:
1059         case -EHOSTUNREACH:
1060         case -ENETUNREACH:
1061         case -EIO:
1062         case -ETIMEDOUT:
1063         case -EPIPE:
1064                 dprintk("%s DS connection error %d\n", __func__,
1065                         task->tk_status);
1066                 nfs4_mark_deviceid_unavailable(devid);
1067                 rpc_wake_up(&tbl->slot_tbl_waitq);
1068                 /* fall through */
1069         default:
1070                 if (ff_layout_no_fallback_to_mds(lseg) ||
1071                     ff_layout_has_available_ds(lseg))
1072                         return -NFS4ERR_RESET_TO_PNFS;
1073 reset:
1074                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1075                         task->tk_status);
1076                 return -NFS4ERR_RESET_TO_MDS;
1077         }
1078 out:
1079         task->tk_status = 0;
1080         return -EAGAIN;
1081 out_bad_stateid:
1082         task->tk_status = -EIO;
1083         return 0;
1084 wait_on_recovery:
1085         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
1086         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
1087                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
1088         goto out;
1089 }
1090
1091 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1092 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1093                                            struct pnfs_layout_segment *lseg,
1094                                            int idx)
1095 {
1096         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1097
1098         if (task->tk_status >= 0)
1099                 return 0;
1100
1101         switch (task->tk_status) {
1102         /* File access problems. Don't mark the device as unavailable */
1103         case -EACCES:
1104         case -ESTALE:
1105         case -EISDIR:
1106         case -EBADHANDLE:
1107         case -ELOOP:
1108         case -ENOSPC:
1109                 break;
1110         case -EJUKEBOX:
1111                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1112                 goto out_retry;
1113         default:
1114                 dprintk("%s DS connection error %d\n", __func__,
1115                         task->tk_status);
1116                 nfs4_mark_deviceid_unavailable(devid);
1117         }
1118         /* FIXME: Need to prevent infinite looping here. */
1119         return -NFS4ERR_RESET_TO_PNFS;
1120 out_retry:
1121         task->tk_status = 0;
1122         rpc_restart_call_prepare(task);
1123         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1124         return -EAGAIN;
1125 }
1126
1127 static int ff_layout_async_handle_error(struct rpc_task *task,
1128                                         struct nfs4_state *state,
1129                                         struct nfs_client *clp,
1130                                         struct pnfs_layout_segment *lseg,
1131                                         int idx)
1132 {
1133         int vers = clp->cl_nfs_mod->rpc_vers->number;
1134
1135         switch (vers) {
1136         case 3:
1137                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1138         case 4:
1139                 return ff_layout_async_handle_error_v4(task, state, clp,
1140                                                        lseg, idx);
1141         default:
1142                 /* should never happen */
1143                 WARN_ON_ONCE(1);
1144                 return 0;
1145         }
1146 }
1147
1148 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1149                                         int idx, u64 offset, u64 length,
1150                                         u32 status, int opnum, int error)
1151 {
1152         struct nfs4_ff_layout_mirror *mirror;
1153         int err;
1154
1155         if (status == 0) {
1156                 switch (error) {
1157                 case -ETIMEDOUT:
1158                 case -EPFNOSUPPORT:
1159                 case -EPROTONOSUPPORT:
1160                 case -EOPNOTSUPP:
1161                 case -ECONNREFUSED:
1162                 case -ECONNRESET:
1163                 case -EHOSTDOWN:
1164                 case -EHOSTUNREACH:
1165                 case -ENETUNREACH:
1166                 case -EADDRINUSE:
1167                 case -ENOBUFS:
1168                 case -EPIPE:
1169                 case -EPERM:
1170                         status = NFS4ERR_NXIO;
1171                         break;
1172                 case -EACCES:
1173                         status = NFS4ERR_ACCESS;
1174                         break;
1175                 default:
1176                         return;
1177                 }
1178         }
1179
1180         switch (status) {
1181         case NFS4ERR_DELAY:
1182         case NFS4ERR_GRACE:
1183                 return;
1184         default:
1185                 break;
1186         }
1187
1188         mirror = FF_LAYOUT_COMP(lseg, idx);
1189         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1190                                        mirror, offset, length, status, opnum,
1191                                        GFP_NOIO);
1192         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1193         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1194 }
1195
1196 /* NFS_PROTO call done callback routines */
1197 static int ff_layout_read_done_cb(struct rpc_task *task,
1198                                 struct nfs_pgio_header *hdr)
1199 {
1200         int err;
1201
1202         trace_nfs4_pnfs_read(hdr, task->tk_status);
1203         if (task->tk_status < 0)
1204                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1205                                             hdr->args.offset, hdr->args.count,
1206                                             hdr->res.op_status, OP_READ,
1207                                             task->tk_status);
1208         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1209                                            hdr->ds_clp, hdr->lseg,
1210                                            hdr->pgio_mirror_idx);
1211
1212         switch (err) {
1213         case -NFS4ERR_RESET_TO_PNFS:
1214                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1215                                         hdr->pgio_mirror_idx + 1,
1216                                         &hdr->pgio_mirror_idx))
1217                         goto out_eagain;
1218                 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1219                         &hdr->lseg->pls_layout->plh_flags);
1220                 pnfs_read_resend_pnfs(hdr);
1221                 return task->tk_status;
1222         case -NFS4ERR_RESET_TO_MDS:
1223                 ff_layout_reset_read(hdr);
1224                 return task->tk_status;
1225         case -EAGAIN:
1226                 goto out_eagain;
1227         }
1228
1229         return 0;
1230 out_eagain:
1231         rpc_restart_call_prepare(task);
1232         return -EAGAIN;
1233 }
1234
1235 static bool
1236 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1237 {
1238         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1239 }
1240
1241 /*
1242  * We reference the rpc_cred of the first WRITE that triggers the need for
1243  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1244  * rfc5661 is not clear about which credential should be used.
1245  *
1246  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1247  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1248  * we always send layoutcommit after DS writes.
1249  */
1250 static void
1251 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
1252 {
1253         if (!ff_layout_need_layoutcommit(hdr->lseg))
1254                 return;
1255
1256         pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1257                         hdr->mds_offset + hdr->res.count);
1258         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
1259                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
1260 }
1261
1262 static bool
1263 ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
1264 {
1265         /* No mirroring for now */
1266         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1267
1268         return ff_layout_test_devid_unavailable(node);
1269 }
1270
1271 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1272                 struct nfs_pgio_header *hdr)
1273 {
1274         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1275                 return;
1276         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1277                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1278                         hdr->args.count,
1279                         task->tk_start);
1280 }
1281
1282 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1283                 struct nfs_pgio_header *hdr)
1284 {
1285         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1286                 return;
1287         nfs4_ff_layout_stat_io_end_read(task,
1288                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1289                         hdr->args.count,
1290                         hdr->res.count);
1291 }
1292
1293 static int ff_layout_read_prepare_common(struct rpc_task *task,
1294                                          struct nfs_pgio_header *hdr)
1295 {
1296         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1297                 rpc_exit(task, -EIO);
1298                 return -EIO;
1299         }
1300         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1301                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
1302                 if (ff_layout_has_available_ds(hdr->lseg))
1303                         pnfs_read_resend_pnfs(hdr);
1304                 else
1305                         ff_layout_reset_read(hdr);
1306                 rpc_exit(task, 0);
1307                 return -EAGAIN;
1308         }
1309         hdr->pgio_done_cb = ff_layout_read_done_cb;
1310
1311         ff_layout_read_record_layoutstats_start(task, hdr);
1312         return 0;
1313 }
1314
1315 /*
1316  * Call ops for the async read/write cases
1317  * In the case of dense layouts, the offset needs to be reset to its
1318  * original value.
1319  */
1320 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1321 {
1322         struct nfs_pgio_header *hdr = data;
1323
1324         if (ff_layout_read_prepare_common(task, hdr))
1325                 return;
1326
1327         rpc_call_start(task);
1328 }
1329
1330 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1331                                     struct nfs4_sequence_args *args,
1332                                     struct nfs4_sequence_res *res,
1333                                     struct rpc_task *task)
1334 {
1335         if (ds_clp->cl_session)
1336                 return nfs41_setup_sequence(ds_clp->cl_session,
1337                                            args,
1338                                            res,
1339                                            task);
1340         return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1341                                    args,
1342                                    res,
1343                                    task);
1344 }
1345
1346 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1347 {
1348         struct nfs_pgio_header *hdr = data;
1349
1350         if (ff_layout_setup_sequence(hdr->ds_clp,
1351                                      &hdr->args.seq_args,
1352                                      &hdr->res.seq_res,
1353                                      task))
1354                 return;
1355
1356         if (ff_layout_read_prepare_common(task, hdr))
1357                 return;
1358
1359         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1360                         hdr->args.lock_context, FMODE_READ) == -EIO)
1361                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1362 }
1363
1364 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1365 {
1366         struct nfs_pgio_header *hdr = data;
1367
1368         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1369
1370         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1371             task->tk_status == 0) {
1372                 nfs4_sequence_done(task, &hdr->res.seq_res);
1373                 return;
1374         }
1375
1376         /* Note this may cause RPC to be resent */
1377         hdr->mds_ops->rpc_call_done(task, hdr);
1378 }
1379
1380 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1381 {
1382         struct nfs_pgio_header *hdr = data;
1383
1384         ff_layout_read_record_layoutstats_done(task, hdr);
1385         rpc_count_iostats_metrics(task,
1386             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1387 }
1388
1389 static void ff_layout_read_release(void *data)
1390 {
1391         struct nfs_pgio_header *hdr = data;
1392
1393         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1394         pnfs_generic_rw_release(data);
1395 }
1396
1397
1398 static int ff_layout_write_done_cb(struct rpc_task *task,
1399                                 struct nfs_pgio_header *hdr)
1400 {
1401         int err;
1402
1403         trace_nfs4_pnfs_write(hdr, task->tk_status);
1404         if (task->tk_status < 0)
1405                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1406                                             hdr->args.offset, hdr->args.count,
1407                                             hdr->res.op_status, OP_WRITE,
1408                                             task->tk_status);
1409         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1410                                            hdr->ds_clp, hdr->lseg,
1411                                            hdr->pgio_mirror_idx);
1412
1413         switch (err) {
1414         case -NFS4ERR_RESET_TO_PNFS:
1415                 ff_layout_reset_write(hdr, true);
1416                 return task->tk_status;
1417         case -NFS4ERR_RESET_TO_MDS:
1418                 ff_layout_reset_write(hdr, false);
1419                 return task->tk_status;
1420         case -EAGAIN:
1421                 return -EAGAIN;
1422         }
1423
1424         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1425             hdr->res.verf->committed == NFS_DATA_SYNC)
1426                 ff_layout_set_layoutcommit(hdr);
1427
1428         /* zero out fattr since we don't care DS attr at all */
1429         hdr->fattr.valid = 0;
1430         if (task->tk_status >= 0)
1431                 nfs_writeback_update_inode(hdr);
1432
1433         return 0;
1434 }
1435
1436 static int ff_layout_commit_done_cb(struct rpc_task *task,
1437                                      struct nfs_commit_data *data)
1438 {
1439         int err;
1440
1441         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1442         if (task->tk_status < 0)
1443                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1444                                             data->args.offset, data->args.count,
1445                                             data->res.op_status, OP_COMMIT,
1446                                             task->tk_status);
1447         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1448                                            data->lseg, data->ds_commit_index);
1449
1450         switch (err) {
1451         case -NFS4ERR_RESET_TO_PNFS:
1452                 pnfs_generic_prepare_to_resend_writes(data);
1453                 return -EAGAIN;
1454         case -NFS4ERR_RESET_TO_MDS:
1455                 pnfs_generic_prepare_to_resend_writes(data);
1456                 return -EAGAIN;
1457         case -EAGAIN:
1458                 rpc_restart_call_prepare(task);
1459                 return -EAGAIN;
1460         }
1461
1462         if (data->verf.committed == NFS_UNSTABLE
1463             && ff_layout_need_layoutcommit(data->lseg))
1464                 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
1465
1466         return 0;
1467 }
1468
1469 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1470                 struct nfs_pgio_header *hdr)
1471 {
1472         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1473                 return;
1474         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1475                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1476                         hdr->args.count,
1477                         task->tk_start);
1478 }
1479
1480 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1481                 struct nfs_pgio_header *hdr)
1482 {
1483         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1484                 return;
1485         nfs4_ff_layout_stat_io_end_write(task,
1486                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1487                         hdr->args.count, hdr->res.count,
1488                         hdr->res.verf->committed);
1489 }
1490
1491 static int ff_layout_write_prepare_common(struct rpc_task *task,
1492                                           struct nfs_pgio_header *hdr)
1493 {
1494         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1495                 rpc_exit(task, -EIO);
1496                 return -EIO;
1497         }
1498
1499         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1500                 bool retry_pnfs;
1501
1502                 retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1503                 dprintk("%s task %u reset io to %s\n", __func__,
1504                         task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1505                 ff_layout_reset_write(hdr, retry_pnfs);
1506                 rpc_exit(task, 0);
1507                 return -EAGAIN;
1508         }
1509
1510         ff_layout_write_record_layoutstats_start(task, hdr);
1511         return 0;
1512 }
1513
1514 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1515 {
1516         struct nfs_pgio_header *hdr = data;
1517
1518         if (ff_layout_write_prepare_common(task, hdr))
1519                 return;
1520
1521         rpc_call_start(task);
1522 }
1523
1524 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1525 {
1526         struct nfs_pgio_header *hdr = data;
1527
1528         if (ff_layout_setup_sequence(hdr->ds_clp,
1529                                      &hdr->args.seq_args,
1530                                      &hdr->res.seq_res,
1531                                      task))
1532                 return;
1533
1534         if (ff_layout_write_prepare_common(task, hdr))
1535                 return;
1536
1537         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1538                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1539                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1540 }
1541
1542 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1543 {
1544         struct nfs_pgio_header *hdr = data;
1545
1546         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1547             task->tk_status == 0) {
1548                 nfs4_sequence_done(task, &hdr->res.seq_res);
1549                 return;
1550         }
1551
1552         /* Note this may cause RPC to be resent */
1553         hdr->mds_ops->rpc_call_done(task, hdr);
1554 }
1555
1556 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1557 {
1558         struct nfs_pgio_header *hdr = data;
1559
1560         ff_layout_write_record_layoutstats_done(task, hdr);
1561         rpc_count_iostats_metrics(task,
1562             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1563 }
1564
1565 static void ff_layout_write_release(void *data)
1566 {
1567         struct nfs_pgio_header *hdr = data;
1568
1569         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1570         pnfs_generic_rw_release(data);
1571 }
1572
1573 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1574                 struct nfs_commit_data *cdata)
1575 {
1576         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1577                 return;
1578         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1579                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1580                         0, task->tk_start);
1581 }
1582
1583 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1584                 struct nfs_commit_data *cdata)
1585 {
1586         struct nfs_page *req;
1587         __u64 count = 0;
1588
1589         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1590                 return;
1591
1592         if (task->tk_status == 0) {
1593                 list_for_each_entry(req, &cdata->pages, wb_list)
1594                         count += req->wb_bytes;
1595         }
1596         nfs4_ff_layout_stat_io_end_write(task,
1597                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1598                         count, count, NFS_FILE_SYNC);
1599 }
1600
1601 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1602                 struct nfs_commit_data *cdata)
1603 {
1604         ff_layout_commit_record_layoutstats_start(task, cdata);
1605 }
1606
1607 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1608 {
1609         ff_layout_commit_prepare_common(task, data);
1610         rpc_call_start(task);
1611 }
1612
1613 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1614 {
1615         struct nfs_commit_data *wdata = data;
1616
1617         if (ff_layout_setup_sequence(wdata->ds_clp,
1618                                  &wdata->args.seq_args,
1619                                  &wdata->res.seq_res,
1620                                  task))
1621                 return;
1622         ff_layout_commit_prepare_common(task, data);
1623 }
1624
1625 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1626 {
1627         pnfs_generic_write_commit_done(task, data);
1628 }
1629
1630 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1631 {
1632         struct nfs_commit_data *cdata = data;
1633
1634         ff_layout_commit_record_layoutstats_done(task, cdata);
1635         rpc_count_iostats_metrics(task,
1636             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1637 }
1638
1639 static void ff_layout_commit_release(void *data)
1640 {
1641         struct nfs_commit_data *cdata = data;
1642
1643         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1644         pnfs_generic_commit_release(data);
1645 }
1646
1647 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1648         .rpc_call_prepare = ff_layout_read_prepare_v3,
1649         .rpc_call_done = ff_layout_read_call_done,
1650         .rpc_count_stats = ff_layout_read_count_stats,
1651         .rpc_release = ff_layout_read_release,
1652 };
1653
1654 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1655         .rpc_call_prepare = ff_layout_read_prepare_v4,
1656         .rpc_call_done = ff_layout_read_call_done,
1657         .rpc_count_stats = ff_layout_read_count_stats,
1658         .rpc_release = ff_layout_read_release,
1659 };
1660
1661 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1662         .rpc_call_prepare = ff_layout_write_prepare_v3,
1663         .rpc_call_done = ff_layout_write_call_done,
1664         .rpc_count_stats = ff_layout_write_count_stats,
1665         .rpc_release = ff_layout_write_release,
1666 };
1667
1668 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1669         .rpc_call_prepare = ff_layout_write_prepare_v4,
1670         .rpc_call_done = ff_layout_write_call_done,
1671         .rpc_count_stats = ff_layout_write_count_stats,
1672         .rpc_release = ff_layout_write_release,
1673 };
1674
1675 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1676         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1677         .rpc_call_done = ff_layout_commit_done,
1678         .rpc_count_stats = ff_layout_commit_count_stats,
1679         .rpc_release = ff_layout_commit_release,
1680 };
1681
1682 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1683         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1684         .rpc_call_done = ff_layout_commit_done,
1685         .rpc_count_stats = ff_layout_commit_count_stats,
1686         .rpc_release = ff_layout_commit_release,
1687 };
1688
1689 static enum pnfs_try_status
1690 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1691 {
1692         struct pnfs_layout_segment *lseg = hdr->lseg;
1693         struct nfs4_pnfs_ds *ds;
1694         struct rpc_clnt *ds_clnt;
1695         struct rpc_cred *ds_cred;
1696         loff_t offset = hdr->args.offset;
1697         u32 idx = hdr->pgio_mirror_idx;
1698         int vers;
1699         struct nfs_fh *fh;
1700
1701         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1702                 __func__, hdr->inode->i_ino,
1703                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1704
1705         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1706         if (!ds)
1707                 goto out_failed;
1708
1709         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1710                                                    hdr->inode);
1711         if (IS_ERR(ds_clnt))
1712                 goto out_failed;
1713
1714         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1715         if (IS_ERR(ds_cred))
1716                 goto out_failed;
1717
1718         vers = nfs4_ff_layout_ds_version(lseg, idx);
1719
1720         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1721                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1722
1723         atomic_inc(&ds->ds_clp->cl_count);
1724         hdr->ds_clp = ds->ds_clp;
1725         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1726         if (fh)
1727                 hdr->args.fh = fh;
1728         /*
1729          * Note that if we ever decide to split across DSes,
1730          * then we may need to handle dense-like offsets.
1731          */
1732         hdr->args.offset = offset;
1733         hdr->mds_offset = offset;
1734
1735         /* Perform an asynchronous read to ds */
1736         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1737                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1738                                       &ff_layout_read_call_ops_v4,
1739                           0, RPC_TASK_SOFTCONN);
1740
1741         return PNFS_ATTEMPTED;
1742
1743 out_failed:
1744         if (ff_layout_has_available_ds(lseg))
1745                 return PNFS_TRY_AGAIN;
1746         return PNFS_NOT_ATTEMPTED;
1747 }
1748
1749 /* Perform async writes. */
1750 static enum pnfs_try_status
1751 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1752 {
1753         struct pnfs_layout_segment *lseg = hdr->lseg;
1754         struct nfs4_pnfs_ds *ds;
1755         struct rpc_clnt *ds_clnt;
1756         struct rpc_cred *ds_cred;
1757         loff_t offset = hdr->args.offset;
1758         int vers;
1759         struct nfs_fh *fh;
1760         int idx = hdr->pgio_mirror_idx;
1761
1762         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1763         if (!ds)
1764                 return PNFS_NOT_ATTEMPTED;
1765
1766         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1767                                                    hdr->inode);
1768         if (IS_ERR(ds_clnt))
1769                 return PNFS_NOT_ATTEMPTED;
1770
1771         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1772         if (IS_ERR(ds_cred))
1773                 return PNFS_NOT_ATTEMPTED;
1774
1775         vers = nfs4_ff_layout_ds_version(lseg, idx);
1776
1777         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1778                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1779                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1780                 vers);
1781
1782         hdr->pgio_done_cb = ff_layout_write_done_cb;
1783         atomic_inc(&ds->ds_clp->cl_count);
1784         hdr->ds_clp = ds->ds_clp;
1785         hdr->ds_commit_idx = idx;
1786         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1787         if (fh)
1788                 hdr->args.fh = fh;
1789
1790         /*
1791          * Note that if we ever decide to split across DSes,
1792          * then we may need to handle dense-like offsets.
1793          */
1794         hdr->args.offset = offset;
1795
1796         /* Perform an asynchronous write */
1797         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1798                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1799                                       &ff_layout_write_call_ops_v4,
1800                           sync, RPC_TASK_SOFTCONN);
1801         return PNFS_ATTEMPTED;
1802 }
1803
1804 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1805 {
1806         return i;
1807 }
1808
1809 static struct nfs_fh *
1810 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1811 {
1812         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1813
1814         /* FIXME: Assume that there is only one NFS version available
1815          * for the DS.
1816          */
1817         return &flseg->mirror_array[i]->fh_versions[0];
1818 }
1819
1820 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1821 {
1822         struct pnfs_layout_segment *lseg = data->lseg;
1823         struct nfs4_pnfs_ds *ds;
1824         struct rpc_clnt *ds_clnt;
1825         struct rpc_cred *ds_cred;
1826         u32 idx;
1827         int vers;
1828         struct nfs_fh *fh;
1829
1830         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1831         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1832         if (!ds)
1833                 goto out_err;
1834
1835         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1836                                                    data->inode);
1837         if (IS_ERR(ds_clnt))
1838                 goto out_err;
1839
1840         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1841         if (IS_ERR(ds_cred))
1842                 goto out_err;
1843
1844         vers = nfs4_ff_layout_ds_version(lseg, idx);
1845
1846         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1847                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1848                 vers);
1849         data->commit_done_cb = ff_layout_commit_done_cb;
1850         data->cred = ds_cred;
1851         atomic_inc(&ds->ds_clp->cl_count);
1852         data->ds_clp = ds->ds_clp;
1853         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1854         if (fh)
1855                 data->args.fh = fh;
1856
1857         return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1858                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1859                                                &ff_layout_commit_call_ops_v4,
1860                                    how, RPC_TASK_SOFTCONN);
1861 out_err:
1862         pnfs_generic_prepare_to_resend_writes(data);
1863         pnfs_generic_commit_release(data);
1864         return -EAGAIN;
1865 }
1866
1867 static int
1868 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1869                            int how, struct nfs_commit_info *cinfo)
1870 {
1871         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1872                                             ff_layout_initiate_commit);
1873 }
1874
1875 static struct pnfs_ds_commit_info *
1876 ff_layout_get_ds_info(struct inode *inode)
1877 {
1878         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1879
1880         if (layout == NULL)
1881                 return NULL;
1882
1883         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1884 }
1885
1886 static void
1887 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1888 {
1889         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1890                                                   id_node));
1891 }
1892
1893 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1894                                   struct xdr_stream *xdr,
1895                                   const struct nfs4_layoutreturn_args *args)
1896 {
1897         struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1898         __be32 *start;
1899         int count = 0, ret = 0;
1900
1901         start = xdr_reserve_space(xdr, 4);
1902         if (unlikely(!start))
1903                 return -E2BIG;
1904
1905         /* This assume we always return _ALL_ layouts */
1906         spin_lock(&hdr->plh_inode->i_lock);
1907         ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1908         spin_unlock(&hdr->plh_inode->i_lock);
1909
1910         *start = cpu_to_be32(count);
1911
1912         return ret;
1913 }
1914
1915 /* report nothing for now */
1916 static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1917                                      struct xdr_stream *xdr,
1918                                      const struct nfs4_layoutreturn_args *args)
1919 {
1920         __be32 *p;
1921
1922         p = xdr_reserve_space(xdr, 4);
1923         if (likely(p))
1924                 *p = cpu_to_be32(0);
1925 }
1926
1927 static struct nfs4_deviceid_node *
1928 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1929                               struct pnfs_device *pdev, gfp_t gfp_flags)
1930 {
1931         struct nfs4_ff_layout_ds *dsaddr;
1932
1933         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1934         if (!dsaddr)
1935                 return NULL;
1936         return &dsaddr->id_node;
1937 }
1938
1939 static void
1940 ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1941                               struct xdr_stream *xdr,
1942                               const struct nfs4_layoutreturn_args *args)
1943 {
1944         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1945         __be32 *start;
1946
1947         dprintk("%s: Begin\n", __func__);
1948         start = xdr_reserve_space(xdr, 4);
1949         BUG_ON(!start);
1950
1951         ff_layout_encode_ioerr(flo, xdr, args);
1952         ff_layout_encode_iostats(flo, xdr, args);
1953
1954         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1955         dprintk("%s: Return\n", __func__);
1956 }
1957
1958 static int
1959 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
1960 {
1961         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
1962
1963         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
1964 }
1965
1966 static size_t
1967 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
1968                           const int buflen)
1969 {
1970         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
1971         const struct in6_addr *addr = &sin6->sin6_addr;
1972
1973         /*
1974          * RFC 4291, Section 2.2.2
1975          *
1976          * Shorthanded ANY address
1977          */
1978         if (ipv6_addr_any(addr))
1979                 return snprintf(buf, buflen, "::");
1980
1981         /*
1982          * RFC 4291, Section 2.2.2
1983          *
1984          * Shorthanded loopback address
1985          */
1986         if (ipv6_addr_loopback(addr))
1987                 return snprintf(buf, buflen, "::1");
1988
1989         /*
1990          * RFC 4291, Section 2.2.3
1991          *
1992          * Special presentation address format for mapped v4
1993          * addresses.
1994          */
1995         if (ipv6_addr_v4mapped(addr))
1996                 return snprintf(buf, buflen, "::ffff:%pI4",
1997                                         &addr->s6_addr32[3]);
1998
1999         /*
2000          * RFC 4291, Section 2.2.1
2001          */
2002         return snprintf(buf, buflen, "%pI6c", addr);
2003 }
2004
2005 /* Derived from rpc_sockaddr2uaddr */
2006 static void
2007 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2008 {
2009         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2010         char portbuf[RPCBIND_MAXUADDRPLEN];
2011         char addrbuf[RPCBIND_MAXUADDRLEN];
2012         char *netid;
2013         unsigned short port;
2014         int len, netid_len;
2015         __be32 *p;
2016
2017         switch (sap->sa_family) {
2018         case AF_INET:
2019                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2020                         return;
2021                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2022                 netid = "tcp";
2023                 netid_len = 3;
2024                 break;
2025         case AF_INET6:
2026                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2027                         return;
2028                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2029                 netid = "tcp6";
2030                 netid_len = 4;
2031                 break;
2032         default:
2033                 /* we only support tcp and tcp6 */
2034                 WARN_ON_ONCE(1);
2035                 return;
2036         }
2037
2038         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2039         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2040
2041         p = xdr_reserve_space(xdr, 4 + netid_len);
2042         xdr_encode_opaque(p, netid, netid_len);
2043
2044         p = xdr_reserve_space(xdr, 4 + len);
2045         xdr_encode_opaque(p, addrbuf, len);
2046 }
2047
2048 static void
2049 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2050                          ktime_t t)
2051 {
2052         struct timespec64 ts;
2053         __be32 *p;
2054
2055         p = xdr_reserve_space(xdr, 12);
2056         ts = ktime_to_timespec64(t);
2057         p = xdr_encode_hyper(p, ts.tv_sec);
2058         *p++ = cpu_to_be32(ts.tv_nsec);
2059 }
2060
2061 static void
2062 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2063                             struct nfs4_ff_io_stat *stat)
2064 {
2065         __be32 *p;
2066
2067         p = xdr_reserve_space(xdr, 5 * 8);
2068         p = xdr_encode_hyper(p, stat->ops_requested);
2069         p = xdr_encode_hyper(p, stat->bytes_requested);
2070         p = xdr_encode_hyper(p, stat->ops_completed);
2071         p = xdr_encode_hyper(p, stat->bytes_completed);
2072         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2073         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2074         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2075 }
2076
2077 static void
2078 ff_layout_encode_layoutstats(struct xdr_stream *xdr,
2079                              struct nfs42_layoutstat_args *args,
2080                              struct nfs42_layoutstat_devinfo *devinfo)
2081 {
2082         struct nfs4_ff_layout_mirror *mirror = devinfo->layout_private;
2083         struct nfs4_pnfs_ds_addr *da;
2084         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2085         struct nfs_fh *fh = &mirror->fh_versions[0];
2086         __be32 *p, *start;
2087
2088         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2089         dprintk("%s: DS %s: encoding address %s\n",
2090                 __func__, ds->ds_remotestr, da->da_remotestr);
2091         /* layoutupdate length */
2092         start = xdr_reserve_space(xdr, 4);
2093         /* netaddr4 */
2094         ff_layout_encode_netaddr(xdr, da);
2095         /* nfs_fh4 */
2096         p = xdr_reserve_space(xdr, 4 + fh->size);
2097         xdr_encode_opaque(p, fh->data, fh->size);
2098         /* ff_io_latency4 read */
2099         spin_lock(&mirror->lock);
2100         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2101         /* ff_io_latency4 write */
2102         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2103         spin_unlock(&mirror->lock);
2104         /* nfstime4 */
2105         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2106         /* bool */
2107         p = xdr_reserve_space(xdr, 4);
2108         *p = cpu_to_be32(false);
2109
2110         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2111 }
2112
2113 static int
2114 ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
2115                                struct pnfs_layout_hdr *lo,
2116                                int dev_limit)
2117 {
2118         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2119         struct nfs4_ff_layout_mirror *mirror;
2120         struct nfs4_deviceid_node *dev;
2121         struct nfs42_layoutstat_devinfo *devinfo;
2122         int i = 0;
2123
2124         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2125                 if (i >= dev_limit)
2126                         break;
2127                 if (!mirror->mirror_ds)
2128                         continue;
2129                 /* mirror refcount put in cleanup_layoutstats */
2130                 if (!atomic_inc_not_zero(&mirror->ref))
2131                         continue;
2132                 dev = &mirror->mirror_ds->id_node;
2133                 devinfo = &args->devinfo[i];
2134                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2135                 devinfo->offset = 0;
2136                 devinfo->length = NFS4_MAX_UINT64;
2137                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2138                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2139                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2140                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2141                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2142                 devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
2143                 devinfo->layout_private = mirror;
2144
2145                 i++;
2146         }
2147         return i;
2148 }
2149
2150 static int
2151 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2152 {
2153         struct nfs4_flexfile_layout *ff_layout;
2154         struct nfs4_ff_layout_mirror *mirror;
2155         int dev_count = 0;
2156
2157         spin_lock(&args->inode->i_lock);
2158         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2159         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2160                 if (atomic_read(&mirror->ref) != 0)
2161                         dev_count ++;
2162         }
2163         spin_unlock(&args->inode->i_lock);
2164         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2165         if (dev_count > PNFS_LAYOUTSTATS_MAXDEV) {
2166                 dprintk("%s: truncating devinfo to limit (%d:%d)\n",
2167                         __func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
2168                 dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2169         }
2170         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2171         if (!args->devinfo)
2172                 return -ENOMEM;
2173
2174         spin_lock(&args->inode->i_lock);
2175         args->num_dev = ff_layout_mirror_prepare_stats(args,
2176                         &ff_layout->generic_hdr, dev_count);
2177         spin_unlock(&args->inode->i_lock);
2178
2179         return 0;
2180 }
2181
2182 static void
2183 ff_layout_cleanup_layoutstats(struct nfs42_layoutstat_data *data)
2184 {
2185         struct nfs4_ff_layout_mirror *mirror;
2186         int i;
2187
2188         for (i = 0; i < data->args.num_dev; i++) {
2189                 mirror = data->args.devinfo[i].layout_private;
2190                 data->args.devinfo[i].layout_private = NULL;
2191                 ff_layout_put_mirror(mirror);
2192         }
2193 }
2194
2195 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2196         .id                     = LAYOUT_FLEX_FILES,
2197         .name                   = "LAYOUT_FLEX_FILES",
2198         .owner                  = THIS_MODULE,
2199         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2200         .free_layout_hdr        = ff_layout_free_layout_hdr,
2201         .alloc_lseg             = ff_layout_alloc_lseg,
2202         .free_lseg              = ff_layout_free_lseg,
2203         .add_lseg               = ff_layout_add_lseg,
2204         .pg_read_ops            = &ff_layout_pg_read_ops,
2205         .pg_write_ops           = &ff_layout_pg_write_ops,
2206         .get_ds_info            = ff_layout_get_ds_info,
2207         .free_deviceid_node     = ff_layout_free_deviceid_node,
2208         .mark_request_commit    = pnfs_layout_mark_request_commit,
2209         .clear_request_commit   = pnfs_generic_clear_request_commit,
2210         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2211         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2212         .commit_pagelist        = ff_layout_commit_pagelist,
2213         .read_pagelist          = ff_layout_read_pagelist,
2214         .write_pagelist         = ff_layout_write_pagelist,
2215         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2216         .encode_layoutreturn    = ff_layout_encode_layoutreturn,
2217         .sync                   = pnfs_nfs_generic_sync,
2218         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2219         .cleanup_layoutstats    = ff_layout_cleanup_layoutstats,
2220 };
2221
2222 static int __init nfs4flexfilelayout_init(void)
2223 {
2224         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2225                __func__);
2226         return pnfs_register_layoutdriver(&flexfilelayout_type);
2227 }
2228
2229 static void __exit nfs4flexfilelayout_exit(void)
2230 {
2231         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2232                __func__);
2233         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2234 }
2235
2236 MODULE_ALIAS("nfs-layouttype4-4");
2237
2238 MODULE_LICENSE("GPL");
2239 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2240
2241 module_init(nfs4flexfilelayout_init);
2242 module_exit(nfs4flexfilelayout_exit);