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