fs/nfs/flexfilelayout/flexfilelayoutdev.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Device operations for the pnfs nfs4 file layout driver.
   4  *
   5  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   6  *
   7  * Tao Peng <bergwolf@primarydata.com>
   8  */
   9
  10 #include <linux/nfs_fs.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/module.h>
  13 #include <linux/sunrpc/addr.h>
  14
  15 #include "../internal.h"
  16 #include "../nfs4session.h"
  17 #include "flexfilelayout.h"
  18
  19 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  20
  21 static unsigned int dataserver_timeo = NFS_DEF_TCP_TIMEO;
  22 static unsigned int dataserver_retrans;
  23
  24 static bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg);
  25
  26 void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
  27 {
  28         if (!IS_ERR_OR_NULL(mirror_ds))
  29                 nfs4_put_deviceid_node(&mirror_ds->id_node);
  30 }
  31
  32 void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
  33 {
  34         nfs4_print_deviceid(&mirror_ds->id_node.deviceid);
  35         nfs4_pnfs_ds_put(mirror_ds->ds);
  36         kfree(mirror_ds->ds_versions);
  37         kfree_rcu(mirror_ds, id_node.rcu);
  38 }
  39
  40 /* Decode opaque device data and construct new_ds using it */
  41 struct nfs4_ff_layout_ds *
  42 nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
  43                             gfp_t gfp_flags)
  44 {
  45         struct xdr_stream stream;
  46         struct xdr_buf buf;
  47         struct page *scratch;
  48         struct list_head dsaddrs;
  49         struct nfs4_pnfs_ds_addr *da;
  50         struct nfs4_ff_layout_ds *new_ds = NULL;
  51         struct nfs4_ff_ds_version *ds_versions = NULL;
  52         u32 mp_count;
  53         u32 version_count;
  54         __be32 *p;
  55         int i, ret = -ENOMEM;
  56
  57         /* set up xdr stream */
  58         scratch = alloc_page(gfp_flags);
  59         if (!scratch)
  60                 goto out_err;
  61
  62         new_ds = kzalloc(sizeof(struct nfs4_ff_layout_ds), gfp_flags);
  63         if (!new_ds)
  64                 goto out_scratch;
  65
  66         nfs4_init_deviceid_node(&new_ds->id_node,
  67                                 server,
  68                                 &pdev->dev_id);
  69         INIT_LIST_HEAD(&dsaddrs);
  70
  71         xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
  72         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
  73
  74         /* multipath count */
  75         p = xdr_inline_decode(&stream, 4);
  76         if (unlikely(!p))
  77                 goto out_err_drain_dsaddrs;
  78         mp_count = be32_to_cpup(p);
  79         dprintk("%s: multipath ds count %d\n", __func__, mp_count);
  80
  81         for (i = 0; i < mp_count; i++) {
  82                 /* multipath ds */
  83                 da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
  84                                             &stream, gfp_flags);
  85                 if (da)
  86                         list_add_tail(&da->da_node, &dsaddrs);
  87         }
  88         if (list_empty(&dsaddrs)) {
  89                 dprintk("%s: no suitable DS addresses found\n",
  90                         __func__);
  91                 ret = -ENOMEDIUM;
  92                 goto out_err_drain_dsaddrs;
  93         }
  94
  95         /* version count */
  96         p = xdr_inline_decode(&stream, 4);
  97         if (unlikely(!p))
  98                 goto out_err_drain_dsaddrs;
  99         version_count = be32_to_cpup(p);
 100         dprintk("%s: version count %d\n", __func__, version_count);
 101
 102         ds_versions = kzalloc(version_count * sizeof(struct nfs4_ff_ds_version),
 103                               gfp_flags);
 104         if (!ds_versions)
 105                 goto out_scratch;
 106
 107         for (i = 0; i < version_count; i++) {
 108                 /* 20 = version(4) + minor_version(4) + rsize(4) + wsize(4) +
 109                  * tightly_coupled(4) */
 110                 p = xdr_inline_decode(&stream, 20);
 111                 if (unlikely(!p))
 112                         goto out_err_drain_dsaddrs;
 113                 ds_versions[i].version = be32_to_cpup(p++);
 114                 ds_versions[i].minor_version = be32_to_cpup(p++);
 115                 ds_versions[i].rsize = nfs_block_size(be32_to_cpup(p++), NULL);
 116                 ds_versions[i].wsize = nfs_block_size(be32_to_cpup(p++), NULL);
 117                 ds_versions[i].tightly_coupled = be32_to_cpup(p);
 118
 119                 if (ds_versions[i].rsize > NFS_MAX_FILE_IO_SIZE)
 120                         ds_versions[i].rsize = NFS_MAX_FILE_IO_SIZE;
 121                 if (ds_versions[i].wsize > NFS_MAX_FILE_IO_SIZE)
 122                         ds_versions[i].wsize = NFS_MAX_FILE_IO_SIZE;
 123
 124                 /*
 125                  * check for valid major/minor combination.
 126                  * currently we support dataserver which talk:
 127                  *   v3, v4.0, v4.1, v4.2
 128                  */
 129                 if (!((ds_versions[i].version == 3 && ds_versions[i].minor_version == 0) ||
 130                         (ds_versions[i].version == 4 && ds_versions[i].minor_version < 3))) {
 131                         dprintk("%s: [%d] unsupported ds version %d-%d\n", __func__,
 132                                 i, ds_versions[i].version,
 133                                 ds_versions[i].minor_version);
 134                         ret = -EPROTONOSUPPORT;
 135                         goto out_err_drain_dsaddrs;
 136                 }
 137
 138                 dprintk("%s: [%d] vers %u minor_ver %u rsize %u wsize %u coupled %d\n",
 139                         __func__, i, ds_versions[i].version,
 140                         ds_versions[i].minor_version,
 141                         ds_versions[i].rsize,
 142                         ds_versions[i].wsize,
 143                         ds_versions[i].tightly_coupled);
 144         }
 145
 146         new_ds->ds_versions = ds_versions;
 147         new_ds->ds_versions_cnt = version_count;
 148
 149         new_ds->ds = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
 150         if (!new_ds->ds)
 151                 goto out_err_drain_dsaddrs;
 152
 153         /* If DS was already in cache, free ds addrs */
 154         while (!list_empty(&dsaddrs)) {
 155                 da = list_first_entry(&dsaddrs,
 156                                       struct nfs4_pnfs_ds_addr,
 157                                       da_node);
 158                 list_del_init(&da->da_node);
 159                 kfree(da->da_remotestr);
 160                 kfree(da);
 161         }
 162
 163         __free_page(scratch);
 164         return new_ds;
 165
 166 out_err_drain_dsaddrs:
 167         while (!list_empty(&dsaddrs)) {
 168                 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
 169                                       da_node);
 170                 list_del_init(&da->da_node);
 171                 kfree(da->da_remotestr);
 172                 kfree(da);
 173         }
 174
 175         kfree(ds_versions);
 176 out_scratch:
 177         __free_page(scratch);
 178 out_err:
 179         kfree(new_ds);
 180
 181         dprintk("%s ERROR: returning %d\n", __func__, ret);
 182         return NULL;
 183 }
 184
 185 static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg,
 186                 struct nfs4_deviceid_node *devid)
 187 {
 188         nfs4_delete_deviceid(devid->ld, devid->nfs_client, &devid->deviceid);
 189         if (!ff_layout_has_available_ds(lseg))
 190                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
 191                                 lseg);
 192 }
 193
 194 static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
 195                                    struct nfs4_ff_layout_mirror *mirror,
 196                                    bool create)
 197 {
 198         if (mirror == NULL || IS_ERR(mirror->mirror_ds))
 199                 goto outerr;
 200         if (mirror->mirror_ds == NULL) {
 201                 if (create) {
 202                         struct nfs4_deviceid_node *node;
 203                         struct pnfs_layout_hdr *lh = lseg->pls_layout;
 204                         struct nfs4_ff_layout_ds *mirror_ds = ERR_PTR(-ENODEV);
 205
 206                         node = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 207                                         &mirror->devid, lh->plh_lc_cred,
 208                                         GFP_KERNEL);
 209                         if (node)
 210                                 mirror_ds = FF_LAYOUT_MIRROR_DS(node);
 211
 212                         /* check for race with another call to this function */
 213                         if (cmpxchg(&mirror->mirror_ds, NULL, mirror_ds) &&
 214                             mirror_ds != ERR_PTR(-ENODEV))
 215                                 nfs4_put_deviceid_node(node);
 216                 } else
 217                         goto outerr;
 218         }
 219
 220         if (IS_ERR(mirror->mirror_ds))
 221                 goto outerr;
 222
 223         if (mirror->mirror_ds->ds == NULL) {
 224                 struct nfs4_deviceid_node *devid;
 225                 devid = &mirror->mirror_ds->id_node;
 226                 ff_layout_mark_devid_invalid(lseg, devid);
 227                 return false;
 228         }
 229         return true;
 230 outerr:
 231         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
 232         return false;
 233 }
 234
 235 static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
 236                             u64 offset, u64 length)
 237 {
 238         u64 end;
 239
 240         end = max_t(u64, pnfs_end_offset(err->offset, err->length),
 241                     pnfs_end_offset(offset, length));
 242         err->offset = min_t(u64, err->offset, offset);
 243         err->length = end - err->offset;
 244 }
 245
 246 static int
 247 ff_ds_error_match(const struct nfs4_ff_layout_ds_err *e1,
 248                 const struct nfs4_ff_layout_ds_err *e2)
 249 {
 250         int ret;
 251
 252         if (e1->opnum != e2->opnum)
 253                 return e1->opnum < e2->opnum ? -1 : 1;
 254         if (e1->status != e2->status)
 255                 return e1->status < e2->status ? -1 : 1;
 256         ret = memcmp(e1->stateid.data, e2->stateid.data,
 257                         sizeof(e1->stateid.data));
 258         if (ret != 0)
 259                 return ret;
 260         ret = memcmp(&e1->deviceid, &e2->deviceid, sizeof(e1->deviceid));
 261         if (ret != 0)
 262                 return ret;
 263         if (pnfs_end_offset(e1->offset, e1->length) < e2->offset)
 264                 return -1;
 265         if (e1->offset > pnfs_end_offset(e2->offset, e2->length))
 266                 return 1;
 267         /* If ranges overlap or are contiguous, they are the same */
 268         return 0;
 269 }
 270
 271 static void
 272 ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo,
 273                               struct nfs4_ff_layout_ds_err *dserr)
 274 {
 275         struct nfs4_ff_layout_ds_err *err, *tmp;
 276         struct list_head *head = &flo->error_list;
 277         int match;
 278
 279         /* Do insertion sort w/ merges */
 280         list_for_each_entry_safe(err, tmp, &flo->error_list, list) {
 281                 match = ff_ds_error_match(err, dserr);
 282                 if (match < 0)
 283                         continue;
 284                 if (match > 0) {
 285                         /* Add entry "dserr" _before_ entry "err" */
 286                         head = &err->list;
 287                         break;
 288                 }
 289                 /* Entries match, so merge "err" into "dserr" */
 290                 extend_ds_error(dserr, err->offset, err->length);
 291                 list_replace(&err->list, &dserr->list);
 292                 kfree(err);
 293                 return;
 294         }
 295
 296         list_add_tail(&dserr->list, head);
 297 }
 298
 299 int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
 300                              struct nfs4_ff_layout_mirror *mirror, u64 offset,
 301                              u64 length, int status, enum nfs_opnum4 opnum,
 302                              gfp_t gfp_flags)
 303 {
 304         struct nfs4_ff_layout_ds_err *dserr;
 305
 306         if (status == 0)
 307                 return 0;
 308
 309         if (IS_ERR_OR_NULL(mirror->mirror_ds))
 310                 return -EINVAL;
 311
 312         dserr = kmalloc(sizeof(*dserr), gfp_flags);
 313         if (!dserr)
 314                 return -ENOMEM;
 315
 316         INIT_LIST_HEAD(&dserr->list);
 317         dserr->offset = offset;
 318         dserr->length = length;
 319         dserr->status = status;
 320         dserr->opnum = opnum;
 321         nfs4_stateid_copy(&dserr->stateid, &mirror->stateid);
 322         memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid,
 323                NFS4_DEVICEID4_SIZE);
 324
 325         spin_lock(&flo->generic_hdr.plh_inode->i_lock);
 326         ff_layout_add_ds_error_locked(flo, dserr);
 327         spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
 328
 329         return 0;
 330 }
 331
 332 static struct rpc_cred *
 333 ff_layout_get_mirror_cred(struct nfs4_ff_layout_mirror *mirror, u32 iomode)
 334 {
 335         struct rpc_cred *cred, __rcu **pcred;
 336
 337         if (iomode == IOMODE_READ)
 338                 pcred = &mirror->ro_cred;
 339         else
 340                 pcred = &mirror->rw_cred;
 341
 342         rcu_read_lock();
 343         do {
 344                 cred = rcu_dereference(*pcred);
 345                 if (!cred)
 346                         break;
 347
 348                 cred = get_rpccred_rcu(cred);
 349         } while(!cred);
 350         rcu_read_unlock();
 351         return cred;
 352 }
 353
 354 struct nfs_fh *
 355 nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
 356 {
 357         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
 358         struct nfs_fh *fh = NULL;
 359
 360         if (!ff_layout_mirror_valid(lseg, mirror, false)) {
 361                 pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
 362                         __func__, mirror_idx);
 363                 goto out;
 364         }
 365
 366         /* FIXME: For now assume there is only 1 version available for the DS */
 367         fh = &mirror->fh_versions[0];
 368 out:
 369         return fh;
 370 }
 371
 372 int
 373 nfs4_ff_layout_select_ds_stateid(struct pnfs_layout_segment *lseg,
 374                                 u32 mirror_idx,
 375                                 nfs4_stateid *stateid)
 376 {
 377         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
 378
 379         if (!ff_layout_mirror_valid(lseg, mirror, false)) {
 380                 pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
 381                         __func__, mirror_idx);
 382                 goto out;
 383         }
 384
 385         nfs4_stateid_copy(stateid, &mirror->stateid);
 386         return 1;
 387 out:
 388         return 0;
 389 }
 390
 391 /**
 392  * nfs4_ff_layout_prepare_ds - prepare a DS connection for an RPC call
 393  * @lseg: the layout segment we're operating on
 394  * @ds_idx: index of the DS to use
 395  * @fail_return: return layout on connect failure?
 396  *
 397  * Try to prepare a DS connection to accept an RPC call. This involves
 398  * selecting a mirror to use and connecting the client to it if it's not
 399  * already connected.
 400  *
 401  * Since we only need a single functioning mirror to satisfy a read, we don't
 402  * want to return the layout if there is one. For writes though, any down
 403  * mirror should result in a LAYOUTRETURN. @fail_return is how we distinguish
 404  * between the two cases.
 405  *
 406  * Returns a pointer to a connected DS object on success or NULL on failure.
 407  */
 408 struct nfs4_pnfs_ds *
 409 nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
 410                           bool fail_return)
 411 {
 412         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 413         struct nfs4_pnfs_ds *ds = NULL;
 414         struct nfs4_deviceid_node *devid;
 415         struct inode *ino = lseg->pls_layout->plh_inode;
 416         struct nfs_server *s = NFS_SERVER(ino);
 417         unsigned int max_payload;
 418         int status;
 419
 420         if (!ff_layout_mirror_valid(lseg, mirror, true)) {
 421                 pr_err_ratelimited("NFS: %s: No data server for offset index %d\n",
 422                         __func__, ds_idx);
 423                 goto out;
 424         }
 425
 426         devid = &mirror->mirror_ds->id_node;
 427         if (ff_layout_test_devid_unavailable(devid))
 428                 goto out_fail;
 429
 430         ds = mirror->mirror_ds->ds;
 431         if (READ_ONCE(ds->ds_clp))
 432                 goto out;
 433         /* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
 434         smp_rmb();
 435
 436         /* FIXME: For now we assume the server sent only one version of NFS
 437          * to use for the DS.
 438          */
 439         status = nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
 440                              dataserver_retrans,
 441                              mirror->mirror_ds->ds_versions[0].version,
 442                              mirror->mirror_ds->ds_versions[0].minor_version);
 443
 444         /* connect success, check rsize/wsize limit */
 445         if (!status) {
 446                 max_payload =
 447                         nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
 448                                        NULL);
 449                 if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
 450                         mirror->mirror_ds->ds_versions[0].rsize = max_payload;
 451                 if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
 452                         mirror->mirror_ds->ds_versions[0].wsize = max_payload;
 453                 goto out;
 454         }
 455 out_fail:
 456         ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
 457                                  mirror, lseg->pls_range.offset,
 458                                  lseg->pls_range.length, NFS4ERR_NXIO,
 459                                  OP_ILLEGAL, GFP_NOIO);
 460         if (fail_return || !ff_layout_has_available_ds(lseg))
 461                 pnfs_error_mark_layout_for_return(ino, lseg);
 462         ds = NULL;
 463 out:
 464         return ds;
 465 }
 466
 467 struct rpc_cred *
 468 ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
 469                       struct rpc_cred *mdscred)
 470 {
 471         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 472         struct rpc_cred *cred;
 473
 474         if (mirror) {
 475                 cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
 476                 if (!cred)
 477                         cred = get_rpccred(mdscred);
 478         } else {
 479                 cred = get_rpccred(mdscred);
 480         }
 481         return cred;
 482 }
 483
 484 /**
 485 * Find or create a DS rpc client with th MDS server rpc client auth flavor
 486 * in the nfs_client cl_ds_clients list.
 487 */
 488 struct rpc_clnt *
 489 nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
 490                                  struct nfs_client *ds_clp, struct inode *inode)
 491 {
 492         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 493
 494         switch (mirror->mirror_ds->ds_versions[0].version) {
 495         case 3:
 496                 /* For NFSv3 DS, flavor is set when creating DS connections */
 497                 return ds_clp->cl_rpcclient;
 498         case 4:
 499                 return nfs4_find_or_create_ds_client(ds_clp, inode);
 500         default:
 501                 BUG();
 502         }
 503 }
 504
 505 void ff_layout_free_ds_ioerr(struct list_head *head)
 506 {
 507         struct nfs4_ff_layout_ds_err *err;
 508
 509         while (!list_empty(head)) {
 510                 err = list_first_entry(head,
 511                                 struct nfs4_ff_layout_ds_err,
 512                                 list);
 513                 list_del(&err->list);
 514                 kfree(err);
 515         }
 516 }
 517
 518 /* called with inode i_lock held */
 519 int ff_layout_encode_ds_ioerr(struct xdr_stream *xdr, const struct list_head *head)
 520 {
 521         struct nfs4_ff_layout_ds_err *err;
 522         __be32 *p;
 523
 524         list_for_each_entry(err, head, list) {
 525                 /* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
 526                  * + array length + deviceid(NFS4_DEVICEID4_SIZE)
 527                  * + status(4) + opnum(4)
 528                  */
 529                 p = xdr_reserve_space(xdr,
 530                                 28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
 531                 if (unlikely(!p))
 532                         return -ENOBUFS;
 533                 p = xdr_encode_hyper(p, err->offset);
 534                 p = xdr_encode_hyper(p, err->length);
 535                 p = xdr_encode_opaque_fixed(p, &err->stateid,
 536                                             NFS4_STATEID_SIZE);
 537                 /* Encode 1 error */
 538                 *p++ = cpu_to_be32(1);
 539                 p = xdr_encode_opaque_fixed(p, &err->deviceid,
 540                                             NFS4_DEVICEID4_SIZE);
 541                 *p++ = cpu_to_be32(err->status);
 542                 *p++ = cpu_to_be32(err->opnum);
 543                 dprintk("%s: offset %llu length %llu status %d op %d\n",
 544                         __func__, err->offset, err->length, err->status,
 545                         err->opnum);
 546         }
 547
 548         return 0;
 549 }
 550
 551 static
 552 unsigned int do_layout_fetch_ds_ioerr(struct pnfs_layout_hdr *lo,
 553                                       const struct pnfs_layout_range *range,
 554                                       struct list_head *head,
 555                                       unsigned int maxnum)
 556 {
 557         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
 558         struct inode *inode = lo->plh_inode;
 559         struct nfs4_ff_layout_ds_err *err, *n;
 560         unsigned int ret = 0;
 561
 562         spin_lock(&inode->i_lock);
 563         list_for_each_entry_safe(err, n, &flo->error_list, list) {
 564                 if (!pnfs_is_range_intersecting(err->offset,
 565                                 pnfs_end_offset(err->offset, err->length),
 566                                 range->offset,
 567                                 pnfs_end_offset(range->offset, range->length)))
 568                         continue;
 569                 if (!maxnum)
 570                         break;
 571                 list_move(&err->list, head);
 572                 maxnum--;
 573                 ret++;
 574         }
 575         spin_unlock(&inode->i_lock);
 576         return ret;
 577 }
 578
 579 unsigned int ff_layout_fetch_ds_ioerr(struct pnfs_layout_hdr *lo,
 580                                       const struct pnfs_layout_range *range,
 581                                       struct list_head *head,
 582                                       unsigned int maxnum)
 583 {
 584         unsigned int ret;
 585
 586         ret = do_layout_fetch_ds_ioerr(lo, range, head, maxnum);
 587         /* If we're over the max, discard all remaining entries */
 588         if (ret == maxnum) {
 589                 LIST_HEAD(discard);
 590                 do_layout_fetch_ds_ioerr(lo, range, &discard, -1);
 591                 ff_layout_free_ds_ioerr(&discard);
 592         }
 593         return ret;
 594 }
 595
 596 static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 597 {
 598         struct nfs4_ff_layout_mirror *mirror;
 599         struct nfs4_deviceid_node *devid;
 600         u32 idx;
 601
 602         for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
 603                 mirror = FF_LAYOUT_COMP(lseg, idx);
 604                 if (mirror) {
 605                         if (!mirror->mirror_ds)
 606                                 return true;
 607                         if (IS_ERR(mirror->mirror_ds))
 608                                 continue;
 609                         devid = &mirror->mirror_ds->id_node;
 610                         if (!ff_layout_test_devid_unavailable(devid))
 611                                 return true;
 612                 }
 613         }
 614
 615         return false;
 616 }
 617
 618 static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 619 {
 620         struct nfs4_ff_layout_mirror *mirror;
 621         struct nfs4_deviceid_node *devid;
 622         u32 idx;
 623
 624         for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
 625                 mirror = FF_LAYOUT_COMP(lseg, idx);
 626                 if (!mirror || IS_ERR(mirror->mirror_ds))
 627                         return false;
 628                 if (!mirror->mirror_ds)
 629                         continue;
 630                 devid = &mirror->mirror_ds->id_node;
 631                 if (ff_layout_test_devid_unavailable(devid))
 632                         return false;
 633         }
 634
 635         return FF_LAYOUT_MIRROR_COUNT(lseg) != 0;
 636 }
 637
 638 static bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 639 {
 640         if (lseg->pls_range.iomode == IOMODE_READ)
 641                 return  ff_read_layout_has_available_ds(lseg);
 642         /* Note: RW layout needs all mirrors available */
 643         return ff_rw_layout_has_available_ds(lseg);
 644 }
 645
 646 bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg)
 647 {
 648         return ff_layout_no_fallback_to_mds(lseg) ||
 649                ff_layout_has_available_ds(lseg);
 650 }
 651
 652 bool ff_layout_avoid_read_on_rw(struct pnfs_layout_segment *lseg)
 653 {
 654         return lseg->pls_range.iomode == IOMODE_RW &&
 655                ff_layout_no_read_on_rw(lseg);
 656 }
 657
 658 module_param(dataserver_retrans, uint, 0644);
 659 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
 660                         "retries a request before it attempts further "
 661                         " recovery  action.");
 662 module_param(dataserver_timeo, uint, 0644);
 663 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
 664                         "NFSv4.1  client  waits for a response from a "
 665                         " data server before it retries an NFS request.");