fs/nfs/flexfilelayout/flexfilelayoutdev.c

   1 /*
   2  * Device operations for the pnfs nfs4 file 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/vmalloc.h>
  11 #include <linux/module.h>
  12 #include <linux/sunrpc/addr.h>
  13
  14 #include "../internal.h"
  15 #include "../nfs4session.h"
  16 #include "flexfilelayout.h"
  17
  18 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  19
  20 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
  21 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
  22
  23 void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
  24 {
  25         if (mirror_ds)
  26                 nfs4_put_deviceid_node(&mirror_ds->id_node);
  27 }
  28
  29 void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
  30 {
  31         nfs4_print_deviceid(&mirror_ds->id_node.deviceid);
  32         nfs4_pnfs_ds_put(mirror_ds->ds);
  33         kfree_rcu(mirror_ds, id_node.rcu);
  34 }
  35
  36 /* Decode opaque device data and construct new_ds using it */
  37 struct nfs4_ff_layout_ds *
  38 nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
  39                             gfp_t gfp_flags)
  40 {
  41         struct xdr_stream stream;
  42         struct xdr_buf buf;
  43         struct page *scratch;
  44         struct list_head dsaddrs;
  45         struct nfs4_pnfs_ds_addr *da;
  46         struct nfs4_ff_layout_ds *new_ds = NULL;
  47         struct nfs4_ff_ds_version *ds_versions = NULL;
  48         u32 mp_count;
  49         u32 version_count;
  50         __be32 *p;
  51         int i, ret = -ENOMEM;
  52
  53         /* set up xdr stream */
  54         scratch = alloc_page(gfp_flags);
  55         if (!scratch)
  56                 goto out_err;
  57
  58         new_ds = kzalloc(sizeof(struct nfs4_ff_layout_ds), gfp_flags);
  59         if (!new_ds)
  60                 goto out_scratch;
  61
  62         nfs4_init_deviceid_node(&new_ds->id_node,
  63                                 server,
  64                                 &pdev->dev_id);
  65         INIT_LIST_HEAD(&dsaddrs);
  66
  67         xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
  68         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
  69
  70         /* multipath count */
  71         p = xdr_inline_decode(&stream, 4);
  72         if (unlikely(!p))
  73                 goto out_err_drain_dsaddrs;
  74         mp_count = be32_to_cpup(p);
  75         dprintk("%s: multipath ds count %d\n", __func__, mp_count);
  76
  77         for (i = 0; i < mp_count; i++) {
  78                 /* multipath ds */
  79                 da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
  80                                             &stream, gfp_flags);
  81                 if (da)
  82                         list_add_tail(&da->da_node, &dsaddrs);
  83         }
  84         if (list_empty(&dsaddrs)) {
  85                 dprintk("%s: no suitable DS addresses found\n",
  86                         __func__);
  87                 ret = -ENOMEDIUM;
  88                 goto out_err_drain_dsaddrs;
  89         }
  90
  91         /* version count */
  92         p = xdr_inline_decode(&stream, 4);
  93         if (unlikely(!p))
  94                 goto out_err_drain_dsaddrs;
  95         version_count = be32_to_cpup(p);
  96         dprintk("%s: version count %d\n", __func__, version_count);
  97
  98         ds_versions = kzalloc(version_count * sizeof(struct nfs4_ff_ds_version),
  99                               gfp_flags);
 100         if (!ds_versions)
 101                 goto out_scratch;
 102
 103         for (i = 0; i < version_count; i++) {
 104                 /* 20 = version(4) + minor_version(4) + rsize(4) + wsize(4) +
 105                  * tightly_coupled(4) */
 106                 p = xdr_inline_decode(&stream, 20);
 107                 if (unlikely(!p))
 108                         goto out_err_drain_dsaddrs;
 109                 ds_versions[i].version = be32_to_cpup(p++);
 110                 ds_versions[i].minor_version = be32_to_cpup(p++);
 111                 ds_versions[i].rsize = nfs_block_size(be32_to_cpup(p++), NULL);
 112                 ds_versions[i].wsize = nfs_block_size(be32_to_cpup(p++), NULL);
 113                 ds_versions[i].tightly_coupled = be32_to_cpup(p);
 114
 115                 if (ds_versions[i].rsize > NFS_MAX_FILE_IO_SIZE)
 116                         ds_versions[i].rsize = NFS_MAX_FILE_IO_SIZE;
 117                 if (ds_versions[i].wsize > NFS_MAX_FILE_IO_SIZE)
 118                         ds_versions[i].wsize = NFS_MAX_FILE_IO_SIZE;
 119
 120                 if (ds_versions[i].version != 3 || ds_versions[i].minor_version != 0) {
 121                         dprintk("%s: [%d] unsupported ds version %d-%d\n", __func__,
 122                                 i, ds_versions[i].version,
 123                                 ds_versions[i].minor_version);
 124                         ret = -EPROTONOSUPPORT;
 125                         goto out_err_drain_dsaddrs;
 126                 }
 127
 128                 dprintk("%s: [%d] vers %u minor_ver %u rsize %u wsize %u coupled %d\n",
 129                         __func__, i, ds_versions[i].version,
 130                         ds_versions[i].minor_version,
 131                         ds_versions[i].rsize,
 132                         ds_versions[i].wsize,
 133                         ds_versions[i].tightly_coupled);
 134         }
 135
 136         new_ds->ds_versions = ds_versions;
 137         new_ds->ds_versions_cnt = version_count;
 138
 139         new_ds->ds = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
 140         if (!new_ds->ds)
 141                 goto out_err_drain_dsaddrs;
 142
 143         /* If DS was already in cache, free ds addrs */
 144         while (!list_empty(&dsaddrs)) {
 145                 da = list_first_entry(&dsaddrs,
 146                                       struct nfs4_pnfs_ds_addr,
 147                                       da_node);
 148                 list_del_init(&da->da_node);
 149                 kfree(da->da_remotestr);
 150                 kfree(da);
 151         }
 152
 153         __free_page(scratch);
 154         return new_ds;
 155
 156 out_err_drain_dsaddrs:
 157         while (!list_empty(&dsaddrs)) {
 158                 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
 159                                       da_node);
 160                 list_del_init(&da->da_node);
 161                 kfree(da->da_remotestr);
 162                 kfree(da);
 163         }
 164
 165         kfree(ds_versions);
 166 out_scratch:
 167         __free_page(scratch);
 168 out_err:
 169         kfree(new_ds);
 170
 171         dprintk("%s ERROR: returning %d\n", __func__, ret);
 172         return NULL;
 173 }
 174
 175 static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg,
 176                 struct nfs4_deviceid_node *devid)
 177 {
 178         nfs4_mark_deviceid_unavailable(devid);
 179         if (!ff_layout_has_available_ds(lseg))
 180                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
 181                                 lseg);
 182 }
 183
 184 static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
 185                 struct nfs4_ff_layout_mirror *mirror)
 186 {
 187         if (mirror == NULL || mirror->mirror_ds == NULL) {
 188                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
 189                                         lseg);
 190                 return false;
 191         }
 192         if (mirror->mirror_ds->ds == NULL) {
 193                 struct nfs4_deviceid_node *devid;
 194                 devid = &mirror->mirror_ds->id_node;
 195                 ff_layout_mark_devid_invalid(lseg, devid);
 196                 return false;
 197         }
 198         return true;
 199 }
 200
 201 static u64
 202 end_offset(u64 start, u64 len)
 203 {
 204         u64 end;
 205
 206         end = start + len;
 207         return end >= start ? end : NFS4_MAX_UINT64;
 208 }
 209
 210 static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
 211                             u64 offset, u64 length)
 212 {
 213         u64 end;
 214
 215         end = max_t(u64, end_offset(err->offset, err->length),
 216                     end_offset(offset, length));
 217         err->offset = min_t(u64, err->offset, offset);
 218         err->length = end - err->offset;
 219 }
 220
 221 static int
 222 ff_ds_error_match(const struct nfs4_ff_layout_ds_err *e1,
 223                 const struct nfs4_ff_layout_ds_err *e2)
 224 {
 225         int ret;
 226
 227         if (e1->opnum != e2->opnum)
 228                 return e1->opnum < e2->opnum ? -1 : 1;
 229         if (e1->status != e2->status)
 230                 return e1->status < e2->status ? -1 : 1;
 231         ret = memcmp(e1->stateid.data, e2->stateid.data,
 232                         sizeof(e1->stateid.data));
 233         if (ret != 0)
 234                 return ret;
 235         ret = memcmp(&e1->deviceid, &e2->deviceid, sizeof(e1->deviceid));
 236         if (ret != 0)
 237                 return ret;
 238         if (end_offset(e1->offset, e1->length) < e2->offset)
 239                 return -1;
 240         if (e1->offset > end_offset(e2->offset, e2->length))
 241                 return 1;
 242         /* If ranges overlap or are contiguous, they are the same */
 243         return 0;
 244 }
 245
 246 static void
 247 ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo,
 248                               struct nfs4_ff_layout_ds_err *dserr)
 249 {
 250         struct nfs4_ff_layout_ds_err *err, *tmp;
 251         struct list_head *head = &flo->error_list;
 252         int match;
 253
 254         /* Do insertion sort w/ merges */
 255         list_for_each_entry_safe(err, tmp, &flo->error_list, list) {
 256                 match = ff_ds_error_match(err, dserr);
 257                 if (match < 0)
 258                         continue;
 259                 if (match > 0) {
 260                         /* Add entry "dserr" _before_ entry "err" */
 261                         head = &err->list;
 262                         break;
 263                 }
 264                 /* Entries match, so merge "err" into "dserr" */
 265                 extend_ds_error(dserr, err->offset, err->length);
 266                 list_del(&err->list);
 267                 kfree(err);
 268         }
 269
 270         list_add_tail(&dserr->list, head);
 271 }
 272
 273 int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
 274                              struct nfs4_ff_layout_mirror *mirror, u64 offset,
 275                              u64 length, int status, enum nfs_opnum4 opnum,
 276                              gfp_t gfp_flags)
 277 {
 278         struct nfs4_ff_layout_ds_err *dserr;
 279
 280         if (status == 0)
 281                 return 0;
 282
 283         if (mirror->mirror_ds == NULL)
 284                 return -EINVAL;
 285
 286         dserr = kmalloc(sizeof(*dserr), gfp_flags);
 287         if (!dserr)
 288                 return -ENOMEM;
 289
 290         INIT_LIST_HEAD(&dserr->list);
 291         dserr->offset = offset;
 292         dserr->length = length;
 293         dserr->status = status;
 294         dserr->opnum = opnum;
 295         nfs4_stateid_copy(&dserr->stateid, &mirror->stateid);
 296         memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid,
 297                NFS4_DEVICEID4_SIZE);
 298
 299         spin_lock(&flo->generic_hdr.plh_inode->i_lock);
 300         ff_layout_add_ds_error_locked(flo, dserr);
 301         spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
 302
 303         return 0;
 304 }
 305
 306 static struct rpc_cred *
 307 ff_layout_get_mirror_cred(struct nfs4_ff_layout_mirror *mirror, u32 iomode)
 308 {
 309         struct rpc_cred *cred, __rcu **pcred;
 310
 311         if (iomode == IOMODE_READ)
 312                 pcred = &mirror->ro_cred;
 313         else
 314                 pcred = &mirror->rw_cred;
 315
 316         rcu_read_lock();
 317         do {
 318                 cred = rcu_dereference(*pcred);
 319                 if (!cred)
 320                         break;
 321
 322                 cred = get_rpccred_rcu(cred);
 323         } while(!cred);
 324         rcu_read_unlock();
 325         return cred;
 326 }
 327
 328 struct nfs_fh *
 329 nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
 330 {
 331         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
 332         struct nfs_fh *fh = NULL;
 333
 334         if (!ff_layout_mirror_valid(lseg, mirror)) {
 335                 pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
 336                         __func__, mirror_idx);
 337                 goto out;
 338         }
 339
 340         /* FIXME: For now assume there is only 1 version available for the DS */
 341         fh = &mirror->fh_versions[0];
 342 out:
 343         return fh;
 344 }
 345
 346 /* Upon return, either ds is connected, or ds is NULL */
 347 struct nfs4_pnfs_ds *
 348 nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
 349                           bool fail_return)
 350 {
 351         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 352         struct nfs4_pnfs_ds *ds = NULL;
 353         struct nfs4_deviceid_node *devid;
 354         struct inode *ino = lseg->pls_layout->plh_inode;
 355         struct nfs_server *s = NFS_SERVER(ino);
 356         unsigned int max_payload;
 357
 358         if (!ff_layout_mirror_valid(lseg, mirror)) {
 359                 pr_err_ratelimited("NFS: %s: No data server for offset index %d\n",
 360                         __func__, ds_idx);
 361                 goto out;
 362         }
 363
 364         devid = &mirror->mirror_ds->id_node;
 365         if (ff_layout_test_devid_unavailable(devid))
 366                 goto out;
 367
 368         ds = mirror->mirror_ds->ds;
 369         /* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
 370         smp_rmb();
 371         if (ds->ds_clp)
 372                 goto out;
 373
 374         /* FIXME: For now we assume the server sent only one version of NFS
 375          * to use for the DS.
 376          */
 377         nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
 378                              dataserver_retrans,
 379                              mirror->mirror_ds->ds_versions[0].version,
 380                              mirror->mirror_ds->ds_versions[0].minor_version,
 381                              RPC_AUTH_UNIX);
 382
 383         /* connect success, check rsize/wsize limit */
 384         if (ds->ds_clp) {
 385                 max_payload =
 386                         nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
 387                                        NULL);
 388                 if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
 389                         mirror->mirror_ds->ds_versions[0].rsize = max_payload;
 390                 if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
 391                         mirror->mirror_ds->ds_versions[0].wsize = max_payload;
 392         } else {
 393                 ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
 394                                          mirror, lseg->pls_range.offset,
 395                                          lseg->pls_range.length, NFS4ERR_NXIO,
 396                                          OP_ILLEGAL, GFP_NOIO);
 397                 if (!fail_return) {
 398                         if (ff_layout_has_available_ds(lseg))
 399                                 set_bit(NFS_LAYOUT_RETURN_REQUESTED,
 400                                         &lseg->pls_layout->plh_flags);
 401                         else
 402                                 pnfs_error_mark_layout_for_return(ino, lseg);
 403                 } else
 404                         pnfs_error_mark_layout_for_return(ino, lseg);
 405                 ds = NULL;
 406         }
 407 out:
 408         return ds;
 409 }
 410
 411 struct rpc_cred *
 412 ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
 413                       struct rpc_cred *mdscred)
 414 {
 415         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 416         struct rpc_cred *cred;
 417
 418         if (mirror) {
 419                 cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
 420                 if (!cred)
 421                         cred = get_rpccred(mdscred);
 422         } else {
 423                 cred = get_rpccred(mdscred);
 424         }
 425         return cred;
 426 }
 427
 428 /**
 429 * Find or create a DS rpc client with th MDS server rpc client auth flavor
 430 * in the nfs_client cl_ds_clients list.
 431 */
 432 struct rpc_clnt *
 433 nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
 434                                  struct nfs_client *ds_clp, struct inode *inode)
 435 {
 436         struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 437
 438         switch (mirror->mirror_ds->ds_versions[0].version) {
 439         case 3:
 440                 /* For NFSv3 DS, flavor is set when creating DS connections */
 441                 return ds_clp->cl_rpcclient;
 442         case 4:
 443                 return nfs4_find_or_create_ds_client(ds_clp, inode);
 444         default:
 445                 BUG();
 446         }
 447 }
 448
 449 static bool is_range_intersecting(u64 offset1, u64 length1,
 450                                   u64 offset2, u64 length2)
 451 {
 452         u64 end1 = end_offset(offset1, length1);
 453         u64 end2 = end_offset(offset2, length2);
 454
 455         return (end1 == NFS4_MAX_UINT64 || end1 > offset2) &&
 456                (end2 == NFS4_MAX_UINT64 || end2 > offset1);
 457 }
 458
 459 /* called with inode i_lock held */
 460 int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
 461                               struct xdr_stream *xdr, int *count,
 462                               const struct pnfs_layout_range *range)
 463 {
 464         struct nfs4_ff_layout_ds_err *err, *n;
 465         __be32 *p;
 466
 467         list_for_each_entry_safe(err, n, &flo->error_list, list) {
 468                 if (!is_range_intersecting(err->offset, err->length,
 469                                            range->offset, range->length))
 470                         continue;
 471                 /* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
 472                  * + array length + deviceid(NFS4_DEVICEID4_SIZE)
 473                  * + status(4) + opnum(4)
 474                  */
 475                 p = xdr_reserve_space(xdr,
 476                                 28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
 477                 if (unlikely(!p))
 478                         return -ENOBUFS;
 479                 p = xdr_encode_hyper(p, err->offset);
 480                 p = xdr_encode_hyper(p, err->length);
 481                 p = xdr_encode_opaque_fixed(p, &err->stateid,
 482                                             NFS4_STATEID_SIZE);
 483                 /* Encode 1 error */
 484                 *p++ = cpu_to_be32(1);
 485                 p = xdr_encode_opaque_fixed(p, &err->deviceid,
 486                                             NFS4_DEVICEID4_SIZE);
 487                 *p++ = cpu_to_be32(err->status);
 488                 *p++ = cpu_to_be32(err->opnum);
 489                 *count += 1;
 490                 list_del(&err->list);
 491                 dprintk("%s: offset %llu length %llu status %d op %d count %d\n",
 492                         __func__, err->offset, err->length, err->status,
 493                         err->opnum, *count);
 494                 kfree(err);
 495         }
 496
 497         return 0;
 498 }
 499
 500 static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 501 {
 502         struct nfs4_ff_layout_mirror *mirror;
 503         struct nfs4_deviceid_node *devid;
 504         u32 idx;
 505
 506         for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
 507                 mirror = FF_LAYOUT_COMP(lseg, idx);
 508                 if (mirror && mirror->mirror_ds) {
 509                         devid = &mirror->mirror_ds->id_node;
 510                         if (!ff_layout_test_devid_unavailable(devid))
 511                                 return true;
 512                 }
 513         }
 514
 515         return false;
 516 }
 517
 518 static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 519 {
 520         struct nfs4_ff_layout_mirror *mirror;
 521         struct nfs4_deviceid_node *devid;
 522         u32 idx;
 523
 524         for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
 525                 mirror = FF_LAYOUT_COMP(lseg, idx);
 526                 if (!mirror || !mirror->mirror_ds)
 527                         return false;
 528                 devid = &mirror->mirror_ds->id_node;
 529                 if (ff_layout_test_devid_unavailable(devid))
 530                         return false;
 531         }
 532
 533         return FF_LAYOUT_MIRROR_COUNT(lseg) != 0;
 534 }
 535
 536 bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 537 {
 538         if (lseg->pls_range.iomode == IOMODE_READ)
 539                 return  ff_read_layout_has_available_ds(lseg);
 540         /* Note: RW layout needs all mirrors available */
 541         return ff_rw_layout_has_available_ds(lseg);
 542 }
 543
 544 bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg)
 545 {
 546         return ff_layout_no_fallback_to_mds(lseg) ||
 547                ff_layout_has_available_ds(lseg);
 548 }
 549
 550 module_param(dataserver_retrans, uint, 0644);
 551 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
 552                         "retries a request before it attempts further "
 553                         " recovery  action.");
 554 module_param(dataserver_timeo, uint, 0644);
 555 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
 556                         "NFSv4.1  client  waits for a response from a "
 557                         " data server before it retries an NFS request.");