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                 trace_nfs4_cb_recall(cps->clp, &args->fh, NULL,
  88                                 &args->stateid, -ntohl(res));
  89                 goto out;
  90         }
  91         /* Set up a helper thread to actually return the delegation */
  92         switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
  93         case 0:
  94                 res = 0;
  95                 break;
  96         case -ENOENT:
  97                 res = htonl(NFS4ERR_BAD_STATEID);
  98                 break;
  99         default:
 100                 res = htonl(NFS4ERR_RESOURCE);
 101         }
 102         trace_nfs4_cb_recall(cps->clp, &args->fh, inode,
 103                         &args->stateid, -ntohl(res));
 104         iput(inode);
 105 out:
 106         dprintk("%s: exit with status = %d\n", __func__, ntohl(res));
 107         return res;
 108 }
 109
 110 #if defined(CONFIG_NFS_V4_1)
 111
 112 /*
 113  * Lookup a layout by filehandle.
 114  *
 115  * Note: gets a refcount on the layout hdr and on its respective inode.
 116  * Caller must put the layout hdr and the inode.
 117  *
 118  * TODO: keep track of all layouts (and delegations) in a hash table
 119  * hashed by filehandle.
 120  */
 121 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
 122                 struct nfs_fh *fh, nfs4_stateid *stateid)
 123 {
 124         struct nfs_server *server;
 125         struct inode *ino;
 126         struct pnfs_layout_hdr *lo;
 127
 128         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 129                 list_for_each_entry(lo, &server->layouts, plh_layouts) {
 130                         if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
 131                                 continue;
 132                         if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
 133                                 continue;
 134                         ino = igrab(lo->plh_inode);
 135                         if (!ino)
 136                                 break;
 137                         spin_lock(&ino->i_lock);
 138                         /* Is this layout in the process of being freed? */
 139                         if (NFS_I(ino)->layout != lo) {
 140                                 spin_unlock(&ino->i_lock);
 141                                 iput(ino);
 142                                 break;
 143                         }
 144                         pnfs_get_layout_hdr(lo);
 145                         spin_unlock(&ino->i_lock);
 146                         return lo;
 147                 }
 148         }
 149
 150         return NULL;
 151 }
 152
 153 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
 154                 struct nfs_fh *fh, nfs4_stateid *stateid)
 155 {
 156         struct pnfs_layout_hdr *lo;
 157
 158         spin_lock(&clp->cl_lock);
 159         rcu_read_lock();
 160         lo = get_layout_by_fh_locked(clp, fh, stateid);
 161         rcu_read_unlock();
 162         spin_unlock(&clp->cl_lock);
 163
 164         return lo;
 165 }
 166
 167 /*
 168  * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
 169  */
 170 static bool pnfs_check_stateid_sequence(struct pnfs_layout_hdr *lo,
 171                                         const nfs4_stateid *new)
 172 {
 173         u32 oldseq, newseq;
 174
 175         oldseq = be32_to_cpu(lo->plh_stateid.seqid);
 176         newseq = be32_to_cpu(new->seqid);
 177
 178         if (newseq > oldseq + 1)
 179                 return false;
 180         return true;
 181 }
 182
 183 static u32 initiate_file_draining(struct nfs_client *clp,
 184                                   struct cb_layoutrecallargs *args)
 185 {
 186         struct inode *ino;
 187         struct pnfs_layout_hdr *lo;
 188         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 189         LIST_HEAD(free_me_list);
 190
 191         lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
 192         if (!lo) {
 193                 trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, NULL,
 194                                 &args->cbl_stateid, -rv);
 195                 goto out;
 196         }
 197
 198         ino = lo->plh_inode;
 199
 200         spin_lock(&ino->i_lock);
 201         if (!pnfs_check_stateid_sequence(lo, &args->cbl_stateid)) {
 202                 rv = NFS4ERR_DELAY;
 203                 goto unlock;
 204         }
 205         pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
 206         spin_unlock(&ino->i_lock);
 207
 208         pnfs_layoutcommit_inode(ino, false);
 209
 210         spin_lock(&ino->i_lock);
 211         /*
 212          * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
 213          */
 214         if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
 215                 rv = NFS4ERR_DELAY;
 216                 goto unlock;
 217         }
 218
 219         if (pnfs_mark_matching_lsegs_return(lo, &free_me_list,
 220                                         &args->cbl_range)) {
 221                 rv = NFS4_OK;
 222                 goto unlock;
 223         }
 224
 225         if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
 226                 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
 227                         &args->cbl_range);
 228         }
 229         pnfs_mark_layout_returned_if_empty(lo);
 230 unlock:
 231         spin_unlock(&ino->i_lock);
 232         pnfs_free_lseg_list(&free_me_list);
 233         /* Free all lsegs that are attached to commit buckets */
 234         nfs_commit_inode(ino, 0);
 235         pnfs_put_layout_hdr(lo);
 236         trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, ino,
 237                         &args->cbl_stateid, -rv);
 238         iput(ino);
 239 out:
 240         return rv;
 241 }
 242
 243 static u32 initiate_bulk_draining(struct nfs_client *clp,
 244                                   struct cb_layoutrecallargs *args)
 245 {
 246         int stat;
 247
 248         if (args->cbl_recall_type == RETURN_FSID)
 249                 stat = pnfs_destroy_layouts_byfsid(clp, &args->cbl_fsid, true);
 250         else
 251                 stat = pnfs_destroy_layouts_byclid(clp, true);
 252         if (stat != 0)
 253                 return NFS4ERR_DELAY;
 254         return NFS4ERR_NOMATCHING_LAYOUT;
 255 }
 256
 257 static u32 do_callback_layoutrecall(struct nfs_client *clp,
 258                                     struct cb_layoutrecallargs *args)
 259 {
 260         u32 res;
 261
 262         dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
 263         if (args->cbl_recall_type == RETURN_FILE)
 264                 res = initiate_file_draining(clp, args);
 265         else
 266                 res = initiate_bulk_draining(clp, args);
 267         dprintk("%s returning %i\n", __func__, res);
 268         return res;
 269
 270 }
 271
 272 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
 273                                   void *dummy, struct cb_process_state *cps)
 274 {
 275         u32 res;
 276
 277         dprintk("%s: -->\n", __func__);
 278
 279         if (cps->clp)
 280                 res = do_callback_layoutrecall(cps->clp, args);
 281         else
 282                 res = NFS4ERR_OP_NOT_IN_SESSION;
 283
 284         dprintk("%s: exit with status = %d\n", __func__, res);
 285         return cpu_to_be32(res);
 286 }
 287
 288 static void pnfs_recall_all_layouts(struct nfs_client *clp)
 289 {
 290         struct cb_layoutrecallargs args;
 291
 292         /* Pretend we got a CB_LAYOUTRECALL(ALL) */
 293         memset(&args, 0, sizeof(args));
 294         args.cbl_recall_type = RETURN_ALL;
 295         /* FIXME we ignore errors, what should we do? */
 296         do_callback_layoutrecall(clp, &args);
 297 }
 298
 299 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
 300                                   void *dummy, struct cb_process_state *cps)
 301 {
 302         int i;
 303         __be32 res = 0;
 304         struct nfs_client *clp = cps->clp;
 305         struct nfs_server *server = NULL;
 306
 307         dprintk("%s: -->\n", __func__);
 308
 309         if (!clp) {
 310                 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 311                 goto out;
 312         }
 313
 314         for (i = 0; i < args->ndevs; i++) {
 315                 struct cb_devicenotifyitem *dev = &args->devs[i];
 316
 317                 if (!server ||
 318                     server->pnfs_curr_ld->id != dev->cbd_layout_type) {
 319                         rcu_read_lock();
 320                         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 321                                 if (server->pnfs_curr_ld &&
 322                                     server->pnfs_curr_ld->id == dev->cbd_layout_type) {
 323                                         rcu_read_unlock();
 324                                         goto found;
 325                                 }
 326                         rcu_read_unlock();
 327                         dprintk("%s: layout type %u not found\n",
 328                                 __func__, dev->cbd_layout_type);
 329                         continue;
 330                 }
 331
 332         found:
 333                 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
 334         }
 335
 336 out:
 337         kfree(args->devs);
 338         dprintk("%s: exit with status = %u\n",
 339                 __func__, be32_to_cpu(res));
 340         return res;
 341 }
 342
 343 /*
 344  * Validate the sequenceID sent by the server.
 345  * Return success if the sequenceID is one more than what we last saw on
 346  * this slot, accounting for wraparound.  Increments the slot's sequence.
 347  *
 348  * We don't yet implement a duplicate request cache, instead we set the
 349  * back channel ca_maxresponsesize_cached to zero. This is OK for now
 350  * since we only currently implement idempotent callbacks anyway.
 351  *
 352  * We have a single slot backchannel at this time, so we don't bother
 353  * checking the used_slots bit array on the table.  The lower layer guarantees
 354  * a single outstanding callback request at a time.
 355  */
 356 static __be32
 357 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
 358 {
 359         struct nfs4_slot *slot;
 360
 361         dprintk("%s enter. slotid %u seqid %u\n",
 362                 __func__, args->csa_slotid, args->csa_sequenceid);
 363
 364         if (args->csa_slotid >= NFS41_BC_MAX_CALLBACKS)
 365                 return htonl(NFS4ERR_BADSLOT);
 366
 367         slot = tbl->slots + args->csa_slotid;
 368         dprintk("%s slot table seqid: %u\n", __func__, slot->seq_nr);
 369
 370         /* Normal */
 371         if (likely(args->csa_sequenceid == slot->seq_nr + 1))
 372                 goto out_ok;
 373
 374         /* Replay */
 375         if (args->csa_sequenceid == slot->seq_nr) {
 376                 dprintk("%s seqid %u is a replay\n",
 377                         __func__, args->csa_sequenceid);
 378                 /* Signal process_op to set this error on next op */
 379                 if (args->csa_cachethis == 0)
 380                         return htonl(NFS4ERR_RETRY_UNCACHED_REP);
 381
 382                 /* The ca_maxresponsesize_cached is 0 with no DRC */
 383                 else if (args->csa_cachethis == 1)
 384                         return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
 385         }
 386
 387         /* Wraparound */
 388         if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
 389                 slot->seq_nr = 1;
 390                 goto out_ok;
 391         }
 392
 393         /* Misordered request */
 394         return htonl(NFS4ERR_SEQ_MISORDERED);
 395 out_ok:
 396         tbl->highest_used_slotid = args->csa_slotid;
 397         return htonl(NFS4_OK);
 398 }
 399
 400 /*
 401  * For each referring call triple, check the session's slot table for
 402  * a match.  If the slot is in use and the sequence numbers match, the
 403  * client is still waiting for a response to the original request.
 404  */
 405 static bool referring_call_exists(struct nfs_client *clp,
 406                                   uint32_t nrclists,
 407                                   struct referring_call_list *rclists)
 408 {
 409         bool status = 0;
 410         int i, j;
 411         struct nfs4_session *session;
 412         struct nfs4_slot_table *tbl;
 413         struct referring_call_list *rclist;
 414         struct referring_call *ref;
 415
 416         /*
 417          * XXX When client trunking is implemented, this becomes
 418          * a session lookup from within the loop
 419          */
 420         session = clp->cl_session;
 421         tbl = &session->fc_slot_table;
 422
 423         for (i = 0; i < nrclists; i++) {
 424                 rclist = &rclists[i];
 425                 if (memcmp(session->sess_id.data,
 426                            rclist->rcl_sessionid.data,
 427                            NFS4_MAX_SESSIONID_LEN) != 0)
 428                         continue;
 429
 430                 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
 431                         ref = &rclist->rcl_refcalls[j];
 432
 433                         dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
 434                                 "slotid %u\n", __func__,
 435                                 ((u32 *)&rclist->rcl_sessionid.data)[0],
 436                                 ((u32 *)&rclist->rcl_sessionid.data)[1],
 437                                 ((u32 *)&rclist->rcl_sessionid.data)[2],
 438                                 ((u32 *)&rclist->rcl_sessionid.data)[3],
 439                                 ref->rc_sequenceid, ref->rc_slotid);
 440
 441                         spin_lock(&tbl->slot_tbl_lock);
 442                         status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
 443                                   tbl->slots[ref->rc_slotid].seq_nr ==
 444                                         ref->rc_sequenceid);
 445                         spin_unlock(&tbl->slot_tbl_lock);
 446                         if (status)
 447                                 goto out;
 448                 }
 449         }
 450
 451 out:
 452         return status;
 453 }
 454
 455 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
 456                               struct cb_sequenceres *res,
 457                               struct cb_process_state *cps)
 458 {
 459         struct nfs4_slot_table *tbl;
 460         struct nfs4_slot *slot;
 461         struct nfs_client *clp;
 462         int i;
 463         __be32 status = htonl(NFS4ERR_BADSESSION);
 464
 465         clp = nfs4_find_client_sessionid(cps->net, args->csa_addr,
 466                                          &args->csa_sessionid, cps->minorversion);
 467         if (clp == NULL)
 468                 goto out;
 469
 470         if (!(clp->cl_session->flags & SESSION4_BACK_CHAN))
 471                 goto out;
 472
 473         tbl = &clp->cl_session->bc_slot_table;
 474         slot = tbl->slots + args->csa_slotid;
 475
 476         spin_lock(&tbl->slot_tbl_lock);
 477         /* state manager is resetting the session */
 478         if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 479                 status = htonl(NFS4ERR_DELAY);
 480                 /* Return NFS4ERR_BADSESSION if we're draining the session
 481                  * in order to reset it.
 482                  */
 483                 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
 484                         status = htonl(NFS4ERR_BADSESSION);
 485                 goto out_unlock;
 486         }
 487
 488         memcpy(&res->csr_sessionid, &args->csa_sessionid,
 489                sizeof(res->csr_sessionid));
 490         res->csr_sequenceid = args->csa_sequenceid;
 491         res->csr_slotid = args->csa_slotid;
 492         res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 493         res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 494
 495         status = validate_seqid(tbl, args);
 496         if (status)
 497                 goto out_unlock;
 498
 499         cps->slotid = args->csa_slotid;
 500
 501         /*
 502          * Check for pending referring calls.  If a match is found, a
 503          * related callback was received before the response to the original
 504          * call.
 505          */
 506         if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
 507                 status = htonl(NFS4ERR_DELAY);
 508                 goto out_unlock;
 509         }
 510
 511         /*
 512          * RFC5661 20.9.3
 513          * If CB_SEQUENCE returns an error, then the state of the slot
 514          * (sequence ID, cached reply) MUST NOT change.
 515          */
 516         slot->seq_nr++;
 517 out_unlock:
 518         spin_unlock(&tbl->slot_tbl_lock);
 519
 520 out:
 521         cps->clp = clp; /* put in nfs4_callback_compound */
 522         for (i = 0; i < args->csa_nrclists; i++)
 523                 kfree(args->csa_rclists[i].rcl_refcalls);
 524         kfree(args->csa_rclists);
 525
 526         if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
 527                 cps->drc_status = status;
 528                 status = 0;
 529         } else
 530                 res->csr_status = status;
 531
 532         trace_nfs4_cb_sequence(args, res, status);
 533         dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
 534                 ntohl(status), ntohl(res->csr_status));
 535         return status;
 536 }
 537
 538 static bool
 539 validate_bitmap_values(unsigned long mask)
 540 {
 541         return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
 542 }
 543
 544 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
 545                                struct cb_process_state *cps)
 546 {
 547         __be32 status;
 548         fmode_t flags = 0;
 549
 550         status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 551         if (!cps->clp) /* set in cb_sequence */
 552                 goto out;
 553
 554         dprintk_rcu("NFS: RECALL_ANY callback request from %s\n",
 555                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
 556
 557         status = cpu_to_be32(NFS4ERR_INVAL);
 558         if (!validate_bitmap_values(args->craa_type_mask))
 559                 goto out;
 560
 561         status = cpu_to_be32(NFS4_OK);
 562         if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
 563                      &args->craa_type_mask))
 564                 flags = FMODE_READ;
 565         if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
 566                      &args->craa_type_mask))
 567                 flags |= FMODE_WRITE;
 568         if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
 569                      &args->craa_type_mask))
 570                 pnfs_recall_all_layouts(cps->clp);
 571         if (flags)
 572                 nfs_expire_unused_delegation_types(cps->clp, flags);
 573 out:
 574         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 575         return status;
 576 }
 577
 578 /* Reduce the fore channel's max_slots to the target value */
 579 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 580                                 struct cb_process_state *cps)
 581 {
 582         struct nfs4_slot_table *fc_tbl;
 583         __be32 status;
 584
 585         status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
 586         if (!cps->clp) /* set in cb_sequence */
 587                 goto out;
 588
 589         dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %u\n",
 590                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
 591                 args->crsa_target_highest_slotid);
 592
 593         fc_tbl = &cps->clp->cl_session->fc_slot_table;
 594
 595         status = htonl(NFS4_OK);
 596
 597         nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
 598         nfs41_notify_server(cps->clp);
 599 out:
 600         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 601         return status;
 602 }
 603 #endif /* CONFIG_NFS_V4_1 */