fs/nfs/callback_proc.c

   1 /*
   2  * linux/fs/nfs/callback_proc.c
   3  *
   4  * Copyright (C) 2004 Trond Myklebust
   5  *
   6  * NFSv4 callback procedures
   7  */
   8 #include <linux/nfs4.h>
   9 #include <linux/nfs_fs.h>
  10 #include <linux/slab.h>
  11 #include <linux/rcupdate.h>
  12 #include "nfs4_fs.h"
  13 #include "callback.h"
  14 #include "delegation.h"
  15 #include "internal.h"
  16 #include "pnfs.h"
  17 #include "nfs4session.h"
  18 #include "nfs4trace.h"
  19
  20 #ifdef NFS_DEBUG
  21 #define NFSDBG_FACILITY NFSDBG_CALLBACK
  22 #endif
  23
  24 __be32 nfs4_callback_getattr(struct cb_getattrargs *args,
  25                              struct cb_getattrres *res,
  26                              struct cb_process_state *cps)
  27 {
  28         struct nfs_delegation *delegation;
  29         struct nfs_inode *nfsi;
  30         struct inode *inode;
  31
  32         res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
  33         if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
  34                 goto out;
  35
  36         res->bitmap[0] = res->bitmap[1] = 0;
  37         res->status = htonl(NFS4ERR_BADHANDLE);
  38
  39         dprintk_rcu("NFS: GETATTR callback request from %s\n",
  40                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
  41
  42         inode = nfs_delegation_find_inode(cps->clp, &args->fh);
  43         if (inode == NULL) {
  44                 trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
  45                                 -ntohl(res->status));
  46                 goto out;
  47         }
  48         nfsi = NFS_I(inode);
  49         rcu_read_lock();
  50         delegation = rcu_dereference(nfsi->delegation);
  51         if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
  52                 goto out_iput;
  53         res->size = i_size_read(inode);
  54         res->change_attr = delegation->change_attr;
  55         if (nfsi->nrequests != 0)
  56                 res->change_attr++;
  57         res->ctime = inode->i_ctime;
  58         res->mtime = inode->i_mtime;
  59         res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
  60                 args->bitmap[0];
  61         res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
  62                 args->bitmap[1];
  63         res->status = 0;
  64 out_iput:
  65         rcu_read_unlock();
  66         trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
  67         iput(inode);
  68 out:
  69         dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
  70         return res->status;
  71 }
  72
  73 __be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
  74                             struct cb_process_state *cps)
  75 {
  76         struct inode *inode;
  77         __be32 res;
  78
  79         res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
  80         if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
  81                 goto out;
  82
  83         dprintk_rcu("NFS: RECALL callback request from %s\n",
  84                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
  85
  86         res = htonl(NFS4ERR_BADHANDLE);
  87         inode = nfs_delegation_find_inode(cps->clp, &args->fh);
  88         if (inode == NULL)
  89                 goto out;
  90         /* Set up a helper thread to actually return the delegation */
  91         switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
  92         case 0:
  93                 res = 0;
  94                 break;
  95         case -ENOENT:
  96                 res = htonl(NFS4ERR_BAD_STATEID);
  97                 break;
  98         default:
  99                 res = htonl(NFS4ERR_RESOURCE);
 100         }
 101         trace_nfs4_recall_delegation(inode, -ntohl(res));
 102         iput(inode);
 103 out:
 104         dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
 105         return res;
 106 }
 107
 108 #if defined(CONFIG_NFS_V4_1)
 109
 110 /*
 111  * Lookup a layout by filehandle.
 112  *
 113  * Note: gets a refcount on the layout hdr and on its respective inode.
 114  * Caller must put the layout hdr and the inode.
 115  *
 116  * TODO: keep track of all layouts (and delegations) in a hash table
 117  * hashed by filehandle.
 118  */
 119 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
 120                 struct nfs_fh *fh, nfs4_stateid *stateid)
 121 {
 122         struct nfs_server *server;
 123         struct inode *ino;
 124         struct pnfs_layout_hdr *lo;
 125
 126         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 127                 list_for_each_entry(lo, &server->layouts, plh_layouts) {
 128                         if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
 129                                 continue;
 130                         if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
 131                                 continue;
 132                         ino = igrab(lo->plh_inode);
 133                         if (!ino)
 134                                 break;
 135                         spin_lock(&ino->i_lock);
 136                         /* Is this layout in the process of being freed? */
 137                         if (NFS_I(ino)->layout != lo) {
 138                                 spin_unlock(&ino->i_lock);
 139                                 iput(ino);
 140                                 break;
 141                         }
 142                         pnfs_get_layout_hdr(lo);
 143                         spin_unlock(&ino->i_lock);
 144                         return lo;
 145                 }
 146         }
 147
 148         return NULL;
 149 }
 150
 151 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
 152                 struct nfs_fh *fh, nfs4_stateid *stateid)
 153 {
 154         struct pnfs_layout_hdr *lo;
 155
 156         spin_lock(&clp->cl_lock);
 157         rcu_read_lock();
 158         lo = get_layout_by_fh_locked(clp, fh, stateid);
 159         rcu_read_unlock();
 160         spin_unlock(&clp->cl_lock);
 161
 162         return lo;
 163 }
 164
 165 static u32 initiate_file_draining(struct nfs_client *clp,
 166                                   struct cb_layoutrecallargs *args)
 167 {
 168         struct inode *ino;
 169         struct pnfs_layout_hdr *lo;
 170         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 171         LIST_HEAD(free_me_list);
 172
 173         lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
 174         if (!lo)
 175                 goto out;
 176
 177         ino = lo->plh_inode;
 178
 179         spin_lock(&ino->i_lock);
 180         pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
 181         spin_unlock(&ino->i_lock);
 182
 183         pnfs_layoutcommit_inode(ino, false);
 184
 185         spin_lock(&ino->i_lock);
 186         if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
 187             pnfs_mark_matching_lsegs_invalid(lo, &free_me_list,
 188                                         &args->cbl_range)) {
 189                 rv = NFS4ERR_DELAY;
 190                 goto unlock;
 191         }
 192
 193         if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
 194                 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
 195                         &args->cbl_range);
 196         }
 197 unlock:
 198         spin_unlock(&ino->i_lock);
 199         pnfs_free_lseg_list(&free_me_list);
 200         pnfs_put_layout_hdr(lo);
 201         trace_nfs4_cb_layoutrecall_inode(clp, &args->cbl_fh, ino, -rv);
 202         iput(ino);
 203 out:
 204         return rv;
 205 }
 206
 207 static u32 initiate_bulk_draining(struct nfs_client *clp,
 208                                   struct cb_layoutrecallargs *args)
 209 {
 210         int stat;
 211
 212         if (args->cbl_recall_type == RETURN_FSID)
 213                 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
 214         else
 215                 stat = pnfs_destroy_layouts_byclid(clp, true);
 216         if (stat != 0)
 217                 return NFS4ERR_DELAY;
 218         return NFS4ERR_NOMATCHING_LAYOUT;
 219 }
 220
 221 static u32 do_callback_layoutrecall(struct nfs_client *clp,
 222                                     struct cb_layoutrecallargs *args)
 223 {
 224         u32 res;
 225
 226         dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
 227         if (args->cbl_recall_type == RETURN_FILE)
 228                 res = initiate_file_draining(clp, args);
 229         else
 230                 res = initiate_bulk_draining(clp, args);
 231         dprintk("%s returning %i\n", __func__, res);
 232         return res;
 233
 234 }
 235
 236 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
 237                                   void *dummy, struct cb_process_state *cps)
 238 {
 239         u32 res;
 240
 241         dprintk("%s: -->\n", __func__);
 242
 243         if (cps->clp)
 244                 res = do_callback_layoutrecall(cps->clp, args);
 245         else
 246                 res = NFS4ERR_OP_NOT_IN_SESSION;
 247
 248         dprintk("%s: exit with status = %d\n", __func__, res);
 249         return cpu_to_be32(res);
 250 }
 251
 252 static void pnfs_recall_all_layouts(struct nfs_client *clp)
 253 {
 254         struct cb_layoutrecallargs args;
 255
 256         /* Pretend we got a CB_LAYOUTRECALL(ALL) */
 257         memset(&args, 0, sizeof(args));
 258         args.cbl_recall_type = RETURN_ALL;
 259         /* FIXME we ignore errors, what should we do? */
 260         do_callback_layoutrecall(clp, &args);
 261 }
 262
 263 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
 264                                   void *dummy, struct cb_process_state *cps)
 265 {
 266         int i;
 267         __be32 res = 0;
 268         struct nfs_client *clp = cps->clp;
 269         struct nfs_server *server = NULL;
 270
 271         dprintk("%s: -->\n", __func__);
 272
 273         if (!clp) {
 274                 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 275                 goto out;
 276         }
 277
 278         for (i = 0; i < args->ndevs; i++) {
 279                 struct cb_devicenotifyitem *dev = &args->devs[i];
 280
 281                 if (!server ||
 282                     server->pnfs_curr_ld->id != dev->cbd_layout_type) {
 283                         rcu_read_lock();
 284                         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 285                                 if (server->pnfs_curr_ld &&
 286                                     server->pnfs_curr_ld->id == dev->cbd_layout_type) {
 287                                         rcu_read_unlock();
 288                                         goto found;
 289                                 }
 290                         rcu_read_unlock();
 291                         dprintk("%s: layout type %u not found\n",
 292                                 __func__, dev->cbd_layout_type);
 293                         continue;
 294                 }
 295
 296         found:
 297                 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
 298         }
 299
 300 out:
 301         kfree(args->devs);
 302         dprintk("%s: exit with status = %u\n",
 303                 __func__, be32_to_cpu(res));
 304         return res;
 305 }
 306
 307 /*
 308  * Validate the sequenceID sent by the server.
 309  * Return success if the sequenceID is one more than what we last saw on
 310  * this slot, accounting for wraparound.  Increments the slot's sequence.
 311  *
 312  * We don't yet implement a duplicate request cache, instead we set the
 313  * back channel ca_maxresponsesize_cached to zero. This is OK for now
 314  * since we only currently implement idempotent callbacks anyway.
 315  *
 316  * We have a single slot backchannel at this time, so we don't bother
 317  * checking the used_slots bit array on the table.  The lower layer guarantees
 318  * a single outstanding callback request at a time.
 319  */
 320 static __be32
 321 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
 322 {
 323         struct nfs4_slot *slot;
 324
 325         dprintk("%s enter. slotid %u seqid %u\n",
 326                 __func__, args->csa_slotid, args->csa_sequenceid);
 327
 328         if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
 329                 return htonl(NFS4ERR_BADSLOT);
 330
 331         slot = tbl->slots + args->csa_slotid;
 332         dprintk("%s slot table seqid: %u\n", __func__, slot->seq_nr);
 333
 334         /* Normal */
 335         if (likely(args->csa_sequenceid == slot->seq_nr + 1))
 336                 goto out_ok;
 337
 338         /* Replay */
 339         if (args->csa_sequenceid == slot->seq_nr) {
 340                 dprintk("%s seqid %u is a replay\n",
 341                         __func__, args->csa_sequenceid);
 342                 /* Signal process_op to set this error on next op */
 343                 if (args->csa_cachethis == 0)
 344                         return htonl(NFS4ERR_RETRY_UNCACHED_REP);
 345
 346                 /* The ca_maxresponsesize_cached is 0 with no DRC */
 347                 else if (args->csa_cachethis == 1)
 348                         return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
 349         }
 350
 351         /* Wraparound */
 352         if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
 353                 slot->seq_nr = 1;
 354                 goto out_ok;
 355         }
 356
 357         /* Misordered request */
 358         return htonl(NFS4ERR_SEQ_MISORDERED);
 359 out_ok:
 360         tbl->highest_used_slotid = args->csa_slotid;
 361         return htonl(NFS4_OK);
 362 }
 363
 364 /*
 365  * For each referring call triple, check the session's slot table for
 366  * a match.  If the slot is in use and the sequence numbers match, the
 367  * client is still waiting for a response to the original request.
 368  */
 369 static bool referring_call_exists(struct nfs_client *clp,
 370                                   uint32_t nrclists,
 371                                   struct referring_call_list *rclists)
 372 {
 373         bool status = 0;
 374         int i, j;
 375         struct nfs4_session *session;
 376         struct nfs4_slot_table *tbl;
 377         struct referring_call_list *rclist;
 378         struct referring_call *ref;
 379
 380         /*
 381          * XXX When client trunking is implemented, this becomes
 382          * a session lookup from within the loop
 383          */
 384         session = clp->cl_session;
 385         tbl = &session->fc_slot_table;
 386
 387         for (i = 0; i < nrclists; i++) {
 388                 rclist = &rclists[i];
 389                 if (memcmp(session->sess_id.data,
 390                            rclist->rcl_sessionid.data,
 391                            NFS4_MAX_SESSIONID_LEN) != 0)
 392                         continue;
 393
 394                 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
 395                         ref = &rclist->rcl_refcalls[j];
 396
 397                         dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
 398                                 "slotid %u\n", __func__,
 399                                 ((u32 *)&rclist->rcl_sessionid.data)[0],
 400                                 ((u32 *)&rclist->rcl_sessionid.data)[1],
 401                                 ((u32 *)&rclist->rcl_sessionid.data)[2],
 402                                 ((u32 *)&rclist->rcl_sessionid.data)[3],
 403                                 ref->rc_sequenceid, ref->rc_slotid);
 404
 405                         spin_lock(&tbl->slot_tbl_lock);
 406                         status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
 407                                   tbl->slots[ref->rc_slotid].seq_nr ==
 408                                         ref->rc_sequenceid);
 409                         spin_unlock(&tbl->slot_tbl_lock);
 410                         if (status)
 411                                 goto out;
 412                 }
 413         }
 414
 415 out:
 416         return status;
 417 }
 418
 419 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
 420                               struct cb_sequenceres *res,
 421                               struct cb_process_state *cps)
 422 {
 423         struct nfs4_slot_table *tbl;
 424         struct nfs4_slot *slot;
 425         struct nfs_client *clp;
 426         int i;
 427         __be32 status = htonl(NFS4ERR_BADSESSION);
 428
 429         clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
 430                                          &args->csa_sessionid, cps->minorversion);
 431         if (clp == NULL)
 432                 goto out;
 433
 434         if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
 435                 goto out;
 436
 437         tbl = &clp->cl_session->bc_slot_table;
 438         slot = tbl->slots + args->csa_slotid;
 439
 440         spin_lock(&tbl->slot_tbl_lock);
 441         /* state manager is resetting the session */
 442         if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 443                 status = htonl(NFS4ERR_DELAY);
 444                 /* Return NFS4ERR_BADSESSION if we're draining the session
 445                  * in order to reset it.
 446                  */
 447                 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
 448                         status = htonl(NFS4ERR_BADSESSION);
 449                 goto out_unlock;
 450         }
 451
 452         memcpy(&res->csr_sessionid, &args->csa_sessionid,
 453                sizeof(res->csr_sessionid));
 454         res->csr_sequenceid = args->csa_sequenceid;
 455         res->csr_slotid = args->csa_slotid;
 456         res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 457         res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 458
 459         status = validate_seqid(tbl, args);
 460         if (status)
 461                 goto out_unlock;
 462
 463         cps->slotid = args->csa_slotid;
 464
 465         /*
 466          * Check for pending referring calls.  If a match is found, a
 467          * related callback was received before the response to the original
 468          * call.
 469          */
 470         if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
 471                 status = htonl(NFS4ERR_DELAY);
 472                 goto out_unlock;
 473         }
 474
 475         /*
 476          * RFC5661 20.9.3
 477          * If CB_SEQUENCE returns an error, then the state of the slot
 478          * (sequence ID, cached reply) MUST NOT change.
 479          */
 480         slot->seq_nr++;
 481 out_unlock:
 482         spin_unlock(&tbl->slot_tbl_lock);
 483
 484 out:
 485         cps->clp = clp; /* put in nfs4_callback_compound */
 486         for (i = 0; i < args->csa_nrclists; i++)
 487                 kfree(args->csa_rclists[i].rcl_refcalls);
 488         kfree(args->csa_rclists);
 489
 490         if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
 491                 cps->drc_status = status;
 492                 status = 0;
 493         } else
 494                 res->csr_status = status;
 495
 496         trace_nfs4_cb_sequence(args, res, status);
 497         dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
 498                 ntohl(status), ntohl(res->csr_status));
 499         return status;
 500 }
 501
 502 static bool
 503 validate_bitmap_values(unsigned long mask)
 504 {
 505         return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
 506 }
 507
 508 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
 509                                struct cb_process_state *cps)
 510 {
 511         __be32 status;
 512         fmode_t flags = 0;
 513
 514         status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 515         if (!cps->clp) /* set in cb_sequence */
 516                 goto out;
 517
 518         dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
 519                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
 520
 521         status = cpu_to_be32(NFS4ERR_INVAL);
 522         if (!validate_bitmap_values(args->craa_type_mask))
 523                 goto out;
 524
 525         status = cpu_to_be32(NFS4_OK);
 526         if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
 527                      &args->craa_type_mask))
 528                 flags = FMODE_READ;
 529         if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
 530                      &args->craa_type_mask))
 531                 flags |= FMODE_WRITE;
 532         if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
 533                      &args->craa_type_mask))
 534                 pnfs_recall_all_layouts(cps->clp);
 535         if (flags)
 536                 nfs_expire_unused_delegation_types(cps->clp, flags);
 537 out:
 538         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 539         return status;
 540 }
 541
 542 /* Reduce the fore channel's max_slots to the target value */
 543 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 544                                 struct cb_process_state *cps)
 545 {
 546         struct nfs4_slot_table *fc_tbl;
 547         __be32 status;
 548
 549         status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
 550         if (!cps->clp) /* set in cb_sequence */
 551                 goto out;
 552
 553         dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
 554                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
 555                 args->crsa_target_highest_slotid);
 556
 557         fc_tbl = &cps->clp->cl_session->fc_slot_table;
 558
 559         status = htonl(NFS4_OK);
 560
 561         nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
 562         nfs41_notify_server(cps->clp);
 563 out:
 564         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 565         return status;
 566 }
 567 #endif /* CONFIG_NFS_V4_1 */