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                 goto out;
  45         nfsi = NFS_I(inode);
  46         rcu_read_lock();
  47         delegation = rcu_dereference(nfsi->delegation);
  48         if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
  49                 goto out_iput;
  50         res->size = i_size_read(inode);
  51         res->change_attr = delegation->change_attr;
  52         if (nfsi->nrequests != 0)
  53                 res->change_attr++;
  54         res->ctime = inode->i_ctime;
  55         res->mtime = inode->i_mtime;
  56         res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
  57                 args->bitmap[0];
  58         res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
  59                 args->bitmap[1];
  60         res->status = 0;
  61 out_iput:
  62         rcu_read_unlock();
  63         iput(inode);
  64 out:
  65         dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
  66         return res->status;
  67 }
  68
  69 __be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
  70                             struct cb_process_state *cps)
  71 {
  72         struct inode *inode;
  73         __be32 res;
  74
  75         res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
  76         if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
  77                 goto out;
  78
  79         dprintk_rcu("NFS: RECALL callback request from %s\n",
  80                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
  81
  82         res = htonl(NFS4ERR_BADHANDLE);
  83         inode = nfs_delegation_find_inode(cps->clp, &args->fh);
  84         if (inode == NULL)
  85                 goto out;
  86         /* Set up a helper thread to actually return the delegation */
  87         switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
  88         case 0:
  89                 res = 0;
  90                 break;
  91         case -ENOENT:
  92                 res = htonl(NFS4ERR_BAD_STATEID);
  93                 break;
  94         default:
  95                 res = htonl(NFS4ERR_RESOURCE);
  96         }
  97         trace_nfs4_recall_delegation(inode, -ntohl(res));
  98         iput(inode);
  99 out:
 100         dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
 101         return res;
 102 }
 103
 104 #if defined(CONFIG_NFS_V4_1)
 105
 106 /*
 107  * Lookup a layout by filehandle.
 108  *
 109  * Note: gets a refcount on the layout hdr and on its respective inode.
 110  * Caller must put the layout hdr and the inode.
 111  *
 112  * TODO: keep track of all layouts (and delegations) in a hash table
 113  * hashed by filehandle.
 114  */
 115 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
 116                 struct nfs_fh *fh, nfs4_stateid *stateid)
 117 {
 118         struct nfs_server *server;
 119         struct inode *ino;
 120         struct pnfs_layout_hdr *lo;
 121
 122         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 123                 list_for_each_entry(lo, &server->layouts, plh_layouts) {
 124                         if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
 125                                 continue;
 126                         if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
 127                                 continue;
 128                         ino = igrab(lo->plh_inode);
 129                         if (!ino)
 130                                 break;
 131                         spin_lock(&ino->i_lock);
 132                         /* Is this layout in the process of being freed? */
 133                         if (NFS_I(ino)->layout != lo) {
 134                                 spin_unlock(&ino->i_lock);
 135                                 iput(ino);
 136                                 break;
 137                         }
 138                         pnfs_get_layout_hdr(lo);
 139                         spin_unlock(&ino->i_lock);
 140                         return lo;
 141                 }
 142         }
 143
 144         return NULL;
 145 }
 146
 147 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
 148                 struct nfs_fh *fh, nfs4_stateid *stateid)
 149 {
 150         struct pnfs_layout_hdr *lo;
 151
 152         spin_lock(&clp->cl_lock);
 153         rcu_read_lock();
 154         lo = get_layout_by_fh_locked(clp, fh, stateid);
 155         rcu_read_unlock();
 156         spin_unlock(&clp->cl_lock);
 157
 158         return lo;
 159 }
 160
 161 static u32 initiate_file_draining(struct nfs_client *clp,
 162                                   struct cb_layoutrecallargs *args)
 163 {
 164         struct inode *ino;
 165         struct pnfs_layout_hdr *lo;
 166         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 167         LIST_HEAD(free_me_list);
 168
 169         lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
 170         if (!lo)
 171                 goto out;
 172
 173         ino = lo->plh_inode;
 174
 175         spin_lock(&ino->i_lock);
 176         pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
 177         spin_unlock(&ino->i_lock);
 178
 179         pnfs_layoutcommit_inode(ino, false);
 180
 181         spin_lock(&ino->i_lock);
 182         if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
 183             pnfs_mark_matching_lsegs_invalid(lo, &free_me_list,
 184                                         &args->cbl_range)) {
 185                 rv = NFS4ERR_DELAY;
 186                 goto unlock;
 187         }
 188
 189         if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
 190                 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
 191                         &args->cbl_range);
 192         }
 193 unlock:
 194         spin_unlock(&ino->i_lock);
 195         pnfs_free_lseg_list(&free_me_list);
 196         pnfs_put_layout_hdr(lo);
 197         iput(ino);
 198 out:
 199         return rv;
 200 }
 201
 202 static u32 initiate_bulk_draining(struct nfs_client *clp,
 203                                   struct cb_layoutrecallargs *args)
 204 {
 205         int stat;
 206
 207         if (args->cbl_recall_type == RETURN_FSID)
 208                 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
 209         else
 210                 stat = pnfs_destroy_layouts_byclid(clp, true);
 211         if (stat != 0)
 212                 return NFS4ERR_DELAY;
 213         return NFS4ERR_NOMATCHING_LAYOUT;
 214 }
 215
 216 static u32 do_callback_layoutrecall(struct nfs_client *clp,
 217                                     struct cb_layoutrecallargs *args)
 218 {
 219         u32 res;
 220
 221         dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
 222         if (args->cbl_recall_type == RETURN_FILE)
 223                 res = initiate_file_draining(clp, args);
 224         else
 225                 res = initiate_bulk_draining(clp, args);
 226         dprintk("%s returning %i\n", __func__, res);
 227         return res;
 228
 229 }
 230
 231 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
 232                                   void *dummy, struct cb_process_state *cps)
 233 {
 234         u32 res;
 235
 236         dprintk("%s: -->\n", __func__);
 237
 238         if (cps->clp)
 239                 res = do_callback_layoutrecall(cps->clp, args);
 240         else
 241                 res = NFS4ERR_OP_NOT_IN_SESSION;
 242
 243         dprintk("%s: exit with status = %d\n", __func__, res);
 244         return cpu_to_be32(res);
 245 }
 246
 247 static void pnfs_recall_all_layouts(struct nfs_client *clp)
 248 {
 249         struct cb_layoutrecallargs args;
 250
 251         /* Pretend we got a CB_LAYOUTRECALL(ALL) */
 252         memset(&args, 0, sizeof(args));
 253         args.cbl_recall_type = RETURN_ALL;
 254         /* FIXME we ignore errors, what should we do? */
 255         do_callback_layoutrecall(clp, &args);
 256 }
 257
 258 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
 259                                   void *dummy, struct cb_process_state *cps)
 260 {
 261         int i;
 262         __be32 res = 0;
 263         struct nfs_client *clp = cps->clp;
 264         struct nfs_server *server = NULL;
 265
 266         dprintk("%s: -->\n", __func__);
 267
 268         if (!clp) {
 269                 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 270                 goto out;
 271         }
 272
 273         for (i = 0; i < args->ndevs; i++) {
 274                 struct cb_devicenotifyitem *dev = &args->devs[i];
 275
 276                 if (!server ||
 277                     server->pnfs_curr_ld->id != dev->cbd_layout_type) {
 278                         rcu_read_lock();
 279                         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 280                                 if (server->pnfs_curr_ld &&
 281                                     server->pnfs_curr_ld->id == dev->cbd_layout_type) {
 282                                         rcu_read_unlock();
 283                                         goto found;
 284                                 }
 285                         rcu_read_unlock();
 286                         dprintk("%s: layout type %u not found\n",
 287                                 __func__, dev->cbd_layout_type);
 288                         continue;
 289                 }
 290
 291         found:
 292                 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
 293         }
 294
 295 out:
 296         kfree(args->devs);
 297         dprintk("%s: exit with status = %u\n",
 298                 __func__, be32_to_cpu(res));
 299         return res;
 300 }
 301
 302 /*
 303  * Validate the sequenceID sent by the server.
 304  * Return success if the sequenceID is one more than what we last saw on
 305  * this slot, accounting for wraparound.  Increments the slot's sequence.
 306  *
 307  * We don't yet implement a duplicate request cache, instead we set the
 308  * back channel ca_maxresponsesize_cached to zero. This is OK for now
 309  * since we only currently implement idempotent callbacks anyway.
 310  *
 311  * We have a single slot backchannel at this time, so we don't bother
 312  * checking the used_slots bit array on the table.  The lower layer guarantees
 313  * a single outstanding callback request at a time.
 314  */
 315 static __be32
 316 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
 317 {
 318         struct nfs4_slot *slot;
 319
 320         dprintk("%s enter. slotid %u seqid %u\n",
 321                 __func__, args->csa_slotid, args->csa_sequenceid);
 322
 323         if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
 324                 return htonl(NFS4ERR_BADSLOT);
 325
 326         slot = tbl->slots + args->csa_slotid;
 327         dprintk("%s slot table seqid: %u\n", __func__, slot->seq_nr);
 328
 329         /* Normal */
 330         if (likely(args->csa_sequenceid == slot->seq_nr + 1)) {
 331                 slot->seq_nr++;
 332                 goto out_ok;
 333         }
 334
 335         /* Replay */
 336         if (args->csa_sequenceid == slot->seq_nr) {
 337                 dprintk("%s seqid %u is a replay\n",
 338                         __func__, args->csa_sequenceid);
 339                 /* Signal process_op to set this error on next op */
 340                 if (args->csa_cachethis == 0)
 341                         return htonl(NFS4ERR_RETRY_UNCACHED_REP);
 342
 343                 /* The ca_maxresponsesize_cached is 0 with no DRC */
 344                 else if (args->csa_cachethis == 1)
 345                         return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
 346         }
 347
 348         /* Wraparound */
 349         if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
 350                 slot->seq_nr = 1;
 351                 goto out_ok;
 352         }
 353
 354         /* Misordered request */
 355         return htonl(NFS4ERR_SEQ_MISORDERED);
 356 out_ok:
 357         tbl->highest_used_slotid = args->csa_slotid;
 358         return htonl(NFS4_OK);
 359 }
 360
 361 /*
 362  * For each referring call triple, check the session's slot table for
 363  * a match.  If the slot is in use and the sequence numbers match, the
 364  * client is still waiting for a response to the original request.
 365  */
 366 static bool referring_call_exists(struct nfs_client *clp,
 367                                   uint32_t nrclists,
 368                                   struct referring_call_list *rclists)
 369 {
 370         bool status = 0;
 371         int i, j;
 372         struct nfs4_session *session;
 373         struct nfs4_slot_table *tbl;
 374         struct referring_call_list *rclist;
 375         struct referring_call *ref;
 376
 377         /*
 378          * XXX When client trunking is implemented, this becomes
 379          * a session lookup from within the loop
 380          */
 381         session = clp->cl_session;
 382         tbl = &session->fc_slot_table;
 383
 384         for (i = 0; i < nrclists; i++) {
 385                 rclist = &rclists[i];
 386                 if (memcmp(session->sess_id.data,
 387                            rclist->rcl_sessionid.data,
 388                            NFS4_MAX_SESSIONID_LEN) != 0)
 389                         continue;
 390
 391                 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
 392                         ref = &rclist->rcl_refcalls[j];
 393
 394                         dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
 395                                 "slotid %u\n", __func__,
 396                                 ((u32 *)&rclist->rcl_sessionid.data)[0],
 397                                 ((u32 *)&rclist->rcl_sessionid.data)[1],
 398                                 ((u32 *)&rclist->rcl_sessionid.data)[2],
 399                                 ((u32 *)&rclist->rcl_sessionid.data)[3],
 400                                 ref->rc_sequenceid, ref->rc_slotid);
 401
 402                         spin_lock(&tbl->slot_tbl_lock);
 403                         status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
 404                                   tbl->slots[ref->rc_slotid].seq_nr ==
 405                                         ref->rc_sequenceid);
 406                         spin_unlock(&tbl->slot_tbl_lock);
 407                         if (status)
 408                                 goto out;
 409                 }
 410         }
 411
 412 out:
 413         return status;
 414 }
 415
 416 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
 417                               struct cb_sequenceres *res,
 418                               struct cb_process_state *cps)
 419 {
 420         struct nfs4_slot_table *tbl;
 421         struct nfs_client *clp;
 422         int i;
 423         __be32 status = htonl(NFS4ERR_BADSESSION);
 424
 425         clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
 426                                          &args->csa_sessionid, cps->minorversion);
 427         if (clp == NULL)
 428                 goto out;
 429
 430         tbl = &clp->cl_session->bc_slot_table;
 431
 432         spin_lock(&tbl->slot_tbl_lock);
 433         /* state manager is resetting the session */
 434         if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 435                 spin_unlock(&tbl->slot_tbl_lock);
 436                 status = htonl(NFS4ERR_DELAY);
 437                 /* Return NFS4ERR_BADSESSION if we're draining the session
 438                  * in order to reset it.
 439                  */
 440                 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
 441                         status = htonl(NFS4ERR_BADSESSION);
 442                 goto out;
 443         }
 444
 445         status = validate_seqid(&clp->cl_session->bc_slot_table, args);
 446         spin_unlock(&tbl->slot_tbl_lock);
 447         if (status)
 448                 goto out;
 449
 450         cps->slotid = args->csa_slotid;
 451
 452         /*
 453          * Check for pending referring calls.  If a match is found, a
 454          * related callback was received before the response to the original
 455          * call.
 456          */
 457         if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
 458                 status = htonl(NFS4ERR_DELAY);
 459                 goto out;
 460         }
 461
 462         memcpy(&res->csr_sessionid, &args->csa_sessionid,
 463                sizeof(res->csr_sessionid));
 464         res->csr_sequenceid = args->csa_sequenceid;
 465         res->csr_slotid = args->csa_slotid;
 466         res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 467         res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 468
 469 out:
 470         cps->clp = clp; /* put in nfs4_callback_compound */
 471         for (i = 0; i < args->csa_nrclists; i++)
 472                 kfree(args->csa_rclists[i].rcl_refcalls);
 473         kfree(args->csa_rclists);
 474
 475         if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
 476                 cps->drc_status = status;
 477                 status = 0;
 478         } else
 479                 res->csr_status = status;
 480
 481         trace_nfs4_cb_sequence(args, res, status);
 482         dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
 483                 ntohl(status), ntohl(res->csr_status));
 484         return status;
 485 }
 486
 487 static bool
 488 validate_bitmap_values(unsigned long mask)
 489 {
 490         return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
 491 }
 492
 493 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
 494                                struct cb_process_state *cps)
 495 {
 496         __be32 status;
 497         fmode_t flags = 0;
 498
 499         status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 500         if (!cps->clp) /* set in cb_sequence */
 501                 goto out;
 502
 503         dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
 504                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
 505
 506         status = cpu_to_be32(NFS4ERR_INVAL);
 507         if (!validate_bitmap_values(args->craa_type_mask))
 508                 goto out;
 509
 510         status = cpu_to_be32(NFS4_OK);
 511         if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
 512                      &args->craa_type_mask))
 513                 flags = FMODE_READ;
 514         if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
 515                      &args->craa_type_mask))
 516                 flags |= FMODE_WRITE;
 517         if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
 518                      &args->craa_type_mask))
 519                 pnfs_recall_all_layouts(cps->clp);
 520         if (flags)
 521                 nfs_expire_unused_delegation_types(cps->clp, flags);
 522 out:
 523         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 524         return status;
 525 }
 526
 527 /* Reduce the fore channel's max_slots to the target value */
 528 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 529                                 struct cb_process_state *cps)
 530 {
 531         struct nfs4_slot_table *fc_tbl;
 532         __be32 status;
 533
 534         status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
 535         if (!cps->clp) /* set in cb_sequence */
 536                 goto out;
 537
 538         dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
 539                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
 540                 args->crsa_target_highest_slotid);
 541
 542         fc_tbl = &cps->clp->cl_session->fc_slot_table;
 543
 544         status = htonl(NFS4_OK);
 545
 546         nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
 547         nfs41_server_notify_target_slotid_update(cps->clp);
 548 out:
 549         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 550         return status;
 551 }
 552 #endif /* CONFIG_NFS_V4_1 */