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
  18 #ifdef NFS_DEBUG
  19 #define NFSDBG_FACILITY NFSDBG_CALLBACK
  20 #endif
  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                 goto out;
  43         nfsi = NFS_I(inode);
  44         rcu_read_lock();
  45         delegation = rcu_dereference(nfsi->delegation);
  46         if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
  47                 goto out_iput;
  48         res->size = i_size_read(inode);
  49         res->change_attr = delegation->change_attr;
  50         if (nfsi->npages != 0)
  51                 res->change_attr++;
  52         res->ctime = inode->i_ctime;
  53         res->mtime = inode->i_mtime;
  54         res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
  55                 args->bitmap[0];
  56         res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
  57                 args->bitmap[1];
  58         res->status = 0;
  59 out_iput:
  60         rcu_read_unlock();
  61         iput(inode);
  62 out:
  63         dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
  64         return res->status;
  65 }
  66
  67 __be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
  68                             struct cb_process_state *cps)
  69 {
  70         struct inode *inode;
  71         __be32 res;
  72
  73         res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
  74         if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
  75                 goto out;
  76
  77         dprintk_rcu("NFS: RECALL callback request from %s\n",
  78                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
  79
  80         res = htonl(NFS4ERR_BADHANDLE);
  81         inode = nfs_delegation_find_inode(cps->clp, &args->fh);
  82         if (inode == NULL)
  83                 goto out;
  84         /* Set up a helper thread to actually return the delegation */
  85         switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
  86         case 0:
  87                 res = 0;
  88                 break;
  89         case -ENOENT:
  90                 res = htonl(NFS4ERR_BAD_STATEID);
  91                 break;
  92         default:
  93                 res = htonl(NFS4ERR_RESOURCE);
  94         }
  95         iput(inode);
  96 out:
  97         dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
  98         return res;
  99 }
 100
 101 #if defined(CONFIG_NFS_V4_1)
 102
 103 /*
 104  * Lookup a layout by filehandle.
 105  *
 106  * Note: gets a refcount on the layout hdr and on its respective inode.
 107  * Caller must put the layout hdr and the inode.
 108  *
 109  * TODO: keep track of all layouts (and delegations) in a hash table
 110  * hashed by filehandle.
 111  */
 112 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp, struct nfs_fh *fh)
 113 {
 114         struct nfs_server *server;
 115         struct inode *ino;
 116         struct pnfs_layout_hdr *lo;
 117
 118         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 119                 list_for_each_entry(lo, &server->layouts, plh_layouts) {
 120                         if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
 121                                 continue;
 122                         ino = igrab(lo->plh_inode);
 123                         if (!ino)
 124                                 continue;
 125                         spin_lock(&ino->i_lock);
 126                         /* Is this layout in the process of being freed? */
 127                         if (NFS_I(ino)->layout != lo) {
 128                                 spin_unlock(&ino->i_lock);
 129                                 iput(ino);
 130                                 continue;
 131                         }
 132                         pnfs_get_layout_hdr(lo);
 133                         spin_unlock(&ino->i_lock);
 134                         return lo;
 135                 }
 136         }
 137
 138         return NULL;
 139 }
 140
 141 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp, struct nfs_fh *fh)
 142 {
 143         struct pnfs_layout_hdr *lo;
 144
 145         spin_lock(&clp->cl_lock);
 146         rcu_read_lock();
 147         lo = get_layout_by_fh_locked(clp, fh);
 148         rcu_read_unlock();
 149         spin_unlock(&clp->cl_lock);
 150
 151         return lo;
 152 }
 153
 154 static u32 initiate_file_draining(struct nfs_client *clp,
 155                                   struct cb_layoutrecallargs *args)
 156 {
 157         struct inode *ino;
 158         struct pnfs_layout_hdr *lo;
 159         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 160         LIST_HEAD(free_me_list);
 161
 162         lo = get_layout_by_fh(clp, &args->cbl_fh);
 163         if (!lo)
 164                 return NFS4ERR_NOMATCHING_LAYOUT;
 165
 166         ino = lo->plh_inode;
 167         spin_lock(&ino->i_lock);
 168         if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
 169             pnfs_mark_matching_lsegs_invalid(lo, &free_me_list,
 170                                         &args->cbl_range))
 171                 rv = NFS4ERR_DELAY;
 172         else
 173                 rv = NFS4ERR_NOMATCHING_LAYOUT;
 174         pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
 175         spin_unlock(&ino->i_lock);
 176         pnfs_free_lseg_list(&free_me_list);
 177         pnfs_put_layout_hdr(lo);
 178         iput(ino);
 179         return rv;
 180 }
 181
 182 static u32 initiate_bulk_draining(struct nfs_client *clp,
 183                                   struct cb_layoutrecallargs *args)
 184 {
 185         struct nfs_server *server;
 186         struct pnfs_layout_hdr *lo;
 187         struct inode *ino;
 188         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 189         struct pnfs_layout_hdr *tmp;
 190         LIST_HEAD(recall_list);
 191         LIST_HEAD(free_me_list);
 192         struct pnfs_layout_range range = {
 193                 .iomode = IOMODE_ANY,
 194                 .offset = 0,
 195                 .length = NFS4_MAX_UINT64,
 196         };
 197
 198         spin_lock(&clp->cl_lock);
 199         rcu_read_lock();
 200         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 201                 if ((args->cbl_recall_type == RETURN_FSID) &&
 202                     memcmp(&server->fsid, &args->cbl_fsid,
 203                            sizeof(struct nfs_fsid)))
 204                         continue;
 205
 206                 list_for_each_entry(lo, &server->layouts, plh_layouts) {
 207                         ino = igrab(lo->plh_inode);
 208                         if (ino)
 209                                 continue;
 210                         spin_lock(&ino->i_lock);
 211                         /* Is this layout in the process of being freed? */
 212                         if (NFS_I(ino)->layout != lo) {
 213                                 spin_unlock(&ino->i_lock);
 214                                 iput(ino);
 215                                 continue;
 216                         }
 217                         pnfs_get_layout_hdr(lo);
 218                         spin_unlock(&ino->i_lock);
 219                         BUG_ON(!list_empty(&lo->plh_bulk_recall));
 220                         list_add(&lo->plh_bulk_recall, &recall_list);
 221                 }
 222         }
 223         rcu_read_unlock();
 224         spin_unlock(&clp->cl_lock);
 225
 226         list_for_each_entry_safe(lo, tmp,
 227                                  &recall_list, plh_bulk_recall) {
 228                 ino = lo->plh_inode;
 229                 spin_lock(&ino->i_lock);
 230                 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
 231                 if (pnfs_mark_matching_lsegs_invalid(lo, &free_me_list, &range))
 232                         rv = NFS4ERR_DELAY;
 233                 list_del_init(&lo->plh_bulk_recall);
 234                 spin_unlock(&ino->i_lock);
 235                 pnfs_free_lseg_list(&free_me_list);
 236                 pnfs_put_layout_hdr(lo);
 237                 iput(ino);
 238         }
 239         return rv;
 240 }
 241
 242 static u32 do_callback_layoutrecall(struct nfs_client *clp,
 243                                     struct cb_layoutrecallargs *args)
 244 {
 245         u32 res;
 246
 247         dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
 248         if (args->cbl_recall_type == RETURN_FILE)
 249                 res = initiate_file_draining(clp, args);
 250         else
 251                 res = initiate_bulk_draining(clp, args);
 252         dprintk("%s returning %i\n", __func__, res);
 253         return res;
 254
 255 }
 256
 257 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
 258                                   void *dummy, struct cb_process_state *cps)
 259 {
 260         u32 res;
 261
 262         dprintk("%s: -->\n", __func__);
 263
 264         if (cps->clp)
 265                 res = do_callback_layoutrecall(cps->clp, args);
 266         else
 267                 res = NFS4ERR_OP_NOT_IN_SESSION;
 268
 269         dprintk("%s: exit with status = %d\n", __func__, res);
 270         return cpu_to_be32(res);
 271 }
 272
 273 static void pnfs_recall_all_layouts(struct nfs_client *clp)
 274 {
 275         struct cb_layoutrecallargs args;
 276
 277         /* Pretend we got a CB_LAYOUTRECALL(ALL) */
 278         memset(&args, 0, sizeof(args));
 279         args.cbl_recall_type = RETURN_ALL;
 280         /* FIXME we ignore errors, what should we do? */
 281         do_callback_layoutrecall(clp, &args);
 282 }
 283
 284 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
 285                                   void *dummy, struct cb_process_state *cps)
 286 {
 287         int i;
 288         __be32 res = 0;
 289         struct nfs_client *clp = cps->clp;
 290         struct nfs_server *server = NULL;
 291
 292         dprintk("%s: -->\n", __func__);
 293
 294         if (!clp) {
 295                 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 296                 goto out;
 297         }
 298
 299         for (i = 0; i < args->ndevs; i++) {
 300                 struct cb_devicenotifyitem *dev = &args->devs[i];
 301
 302                 if (!server ||
 303                     server->pnfs_curr_ld->id != dev->cbd_layout_type) {
 304                         rcu_read_lock();
 305                         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 306                                 if (server->pnfs_curr_ld &&
 307                                     server->pnfs_curr_ld->id == dev->cbd_layout_type) {
 308                                         rcu_read_unlock();
 309                                         goto found;
 310                                 }
 311                         rcu_read_unlock();
 312                         dprintk("%s: layout type %u not found\n",
 313                                 __func__, dev->cbd_layout_type);
 314                         continue;
 315                 }
 316
 317         found:
 318                 if (dev->cbd_notify_type == NOTIFY_DEVICEID4_CHANGE)
 319                         dprintk("%s: NOTIFY_DEVICEID4_CHANGE not supported, "
 320                                 "deleting instead\n", __func__);
 321                 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
 322         }
 323
 324 out:
 325         kfree(args->devs);
 326         dprintk("%s: exit with status = %u\n",
 327                 __func__, be32_to_cpu(res));
 328         return res;
 329 }
 330
 331 /*
 332  * Validate the sequenceID sent by the server.
 333  * Return success if the sequenceID is one more than what we last saw on
 334  * this slot, accounting for wraparound.  Increments the slot's sequence.
 335  *
 336  * We don't yet implement a duplicate request cache, instead we set the
 337  * back channel ca_maxresponsesize_cached to zero. This is OK for now
 338  * since we only currently implement idempotent callbacks anyway.
 339  *
 340  * We have a single slot backchannel at this time, so we don't bother
 341  * checking the used_slots bit array on the table.  The lower layer guarantees
 342  * a single outstanding callback request at a time.
 343  */
 344 static __be32
 345 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
 346 {
 347         struct nfs4_slot *slot;
 348
 349         dprintk("%s enter. slotid %d seqid %d\n",
 350                 __func__, args->csa_slotid, args->csa_sequenceid);
 351
 352         if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
 353                 return htonl(NFS4ERR_BADSLOT);
 354
 355         slot = tbl->slots + args->csa_slotid;
 356         dprintk("%s slot table seqid: %d\n", __func__, slot->seq_nr);
 357
 358         /* Normal */
 359         if (likely(args->csa_sequenceid == slot->seq_nr + 1)) {
 360                 slot->seq_nr++;
 361                 goto out_ok;
 362         }
 363
 364         /* Replay */
 365         if (args->csa_sequenceid == slot->seq_nr) {
 366                 dprintk("%s seqid %d is a replay\n",
 367                         __func__, args->csa_sequenceid);
 368                 /* Signal process_op to set this error on next op */
 369                 if (args->csa_cachethis == 0)
 370                         return htonl(NFS4ERR_RETRY_UNCACHED_REP);
 371
 372                 /* The ca_maxresponsesize_cached is 0 with no DRC */
 373                 else if (args->csa_cachethis == 1)
 374                         return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
 375         }
 376
 377         /* Wraparound */
 378         if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
 379                 slot->seq_nr = 1;
 380                 goto out_ok;
 381         }
 382
 383         /* Misordered request */
 384         return htonl(NFS4ERR_SEQ_MISORDERED);
 385 out_ok:
 386         tbl->highest_used_slotid = args->csa_slotid;
 387         return htonl(NFS4_OK);
 388 }
 389
 390 /*
 391  * For each referring call triple, check the session's slot table for
 392  * a match.  If the slot is in use and the sequence numbers match, the
 393  * client is still waiting for a response to the original request.
 394  */
 395 static bool referring_call_exists(struct nfs_client *clp,
 396                                   uint32_t nrclists,
 397                                   struct referring_call_list *rclists)
 398 {
 399         bool status = 0;
 400         int i, j;
 401         struct nfs4_session *session;
 402         struct nfs4_slot_table *tbl;
 403         struct referring_call_list *rclist;
 404         struct referring_call *ref;
 405
 406         /*
 407          * XXX When client trunking is implemented, this becomes
 408          * a session lookup from within the loop
 409          */
 410         session = clp->cl_session;
 411         tbl = &session->fc_slot_table;
 412
 413         for (i = 0; i < nrclists; i++) {
 414                 rclist = &rclists[i];
 415                 if (memcmp(session->sess_id.data,
 416                            rclist->rcl_sessionid.data,
 417                            NFS4_MAX_SESSIONID_LEN) != 0)
 418                         continue;
 419
 420                 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
 421                         ref = &rclist->rcl_refcalls[j];
 422
 423                         dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
 424                                 "slotid %u\n", __func__,
 425                                 ((u32 *)&rclist->rcl_sessionid.data)[0],
 426                                 ((u32 *)&rclist->rcl_sessionid.data)[1],
 427                                 ((u32 *)&rclist->rcl_sessionid.data)[2],
 428                                 ((u32 *)&rclist->rcl_sessionid.data)[3],
 429                                 ref->rc_sequenceid, ref->rc_slotid);
 430
 431                         spin_lock(&tbl->slot_tbl_lock);
 432                         status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
 433                                   tbl->slots[ref->rc_slotid].seq_nr ==
 434                                         ref->rc_sequenceid);
 435                         spin_unlock(&tbl->slot_tbl_lock);
 436                         if (status)
 437                                 goto out;
 438                 }
 439         }
 440
 441 out:
 442         return status;
 443 }
 444
 445 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
 446                               struct cb_sequenceres *res,
 447                               struct cb_process_state *cps)
 448 {
 449         struct nfs4_slot_table *tbl;
 450         struct nfs_client *clp;
 451         int i;
 452         __be32 status = htonl(NFS4ERR_BADSESSION);
 453
 454         clp = nfs4_find_client_sessionid(cps->net, args->csa_addr, &args->csa_sessionid);
 455         if (clp == NULL)
 456                 goto out;
 457
 458         tbl = &clp->cl_session->bc_slot_table;
 459
 460         spin_lock(&tbl->slot_tbl_lock);
 461         /* state manager is resetting the session */
 462         if (test_bit(NFS4_SESSION_DRAINING, &clp->cl_session->session_state)) {
 463                 spin_unlock(&tbl->slot_tbl_lock);
 464                 status = htonl(NFS4ERR_DELAY);
 465                 /* Return NFS4ERR_BADSESSION if we're draining the session
 466                  * in order to reset it.
 467                  */
 468                 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
 469                         status = htonl(NFS4ERR_BADSESSION);
 470                 goto out;
 471         }
 472
 473         status = validate_seqid(&clp->cl_session->bc_slot_table, args);
 474         spin_unlock(&tbl->slot_tbl_lock);
 475         if (status)
 476                 goto out;
 477
 478         cps->slotid = args->csa_slotid;
 479
 480         /*
 481          * Check for pending referring calls.  If a match is found, a
 482          * related callback was received before the response to the original
 483          * call.
 484          */
 485         if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
 486                 status = htonl(NFS4ERR_DELAY);
 487                 goto out;
 488         }
 489
 490         memcpy(&res->csr_sessionid, &args->csa_sessionid,
 491                sizeof(res->csr_sessionid));
 492         res->csr_sequenceid = args->csa_sequenceid;
 493         res->csr_slotid = args->csa_slotid;
 494         res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 495         res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 496
 497 out:
 498         cps->clp = clp; /* put in nfs4_callback_compound */
 499         for (i = 0; i < args->csa_nrclists; i++)
 500                 kfree(args->csa_rclists[i].rcl_refcalls);
 501         kfree(args->csa_rclists);
 502
 503         if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
 504                 cps->drc_status = status;
 505                 status = 0;
 506         } else
 507                 res->csr_status = status;
 508
 509         dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
 510                 ntohl(status), ntohl(res->csr_status));
 511         return status;
 512 }
 513
 514 static bool
 515 validate_bitmap_values(unsigned long mask)
 516 {
 517         return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
 518 }
 519
 520 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
 521                                struct cb_process_state *cps)
 522 {
 523         __be32 status;
 524         fmode_t flags = 0;
 525
 526         status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 527         if (!cps->clp) /* set in cb_sequence */
 528                 goto out;
 529
 530         dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
 531                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
 532
 533         status = cpu_to_be32(NFS4ERR_INVAL);
 534         if (!validate_bitmap_values(args->craa_type_mask))
 535                 goto out;
 536
 537         status = cpu_to_be32(NFS4_OK);
 538         if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
 539                      &args->craa_type_mask))
 540                 flags = FMODE_READ;
 541         if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
 542                      &args->craa_type_mask))
 543                 flags |= FMODE_WRITE;
 544         if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
 545                      &args->craa_type_mask))
 546                 pnfs_recall_all_layouts(cps->clp);
 547         if (flags)
 548                 nfs_expire_all_delegation_types(cps->clp, flags);
 549 out:
 550         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 551         return status;
 552 }
 553
 554 /* Reduce the fore channel's max_slots to the target value */
 555 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 556                                 struct cb_process_state *cps)
 557 {
 558         struct nfs4_slot_table *fc_tbl;
 559         __be32 status;
 560
 561         status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
 562         if (!cps->clp) /* set in cb_sequence */
 563                 goto out;
 564
 565         dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target max slots %d\n",
 566                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
 567                 args->crsa_target_max_slots);
 568
 569         fc_tbl = &cps->clp->cl_session->fc_slot_table;
 570
 571         status = htonl(NFS4ERR_BAD_HIGH_SLOT);
 572         if (args->crsa_target_max_slots > fc_tbl->max_slots ||
 573             args->crsa_target_max_slots < 1)
 574                 goto out;
 575
 576         status = htonl(NFS4_OK);
 577         if (args->crsa_target_max_slots == fc_tbl->max_slots)
 578                 goto out;
 579
 580         fc_tbl->target_max_slots = args->crsa_target_max_slots;
 581         nfs41_handle_recall_slot(cps->clp);
 582 out:
 583         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 584         return status;
 585 }
 586 #endif /* CONFIG_NFS_V4_1 */