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