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