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 "nfs4_fs.h"
  12 #include "callback.h"
  13 #include "delegation.h"
  14 #include "internal.h"
  15 #include "pnfs.h"
  16
  17 #ifdef NFS_DEBUG
  18 #define NFSDBG_FACILITY NFSDBG_CALLBACK
  19 #endif
  20
  21 __be32 nfs4_callback_getattr(struct cb_getattrargs *args,
  22                              struct cb_getattrres *res,
  23                              struct cb_process_state *cps)
  24 {
  25         struct nfs_delegation *delegation;
  26         struct nfs_inode *nfsi;
  27         struct inode *inode;
  28
  29         res->status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
  30         if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
  31                 goto out;
  32
  33         res->bitmap[0] = res->bitmap[1] = 0;
  34         res->status = htonl(NFS4ERR_BADHANDLE);
  35
  36         dprintk("NFS: GETATTR callback request from %s\n",
  37                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
  38
  39         inode = nfs_delegation_find_inode(cps->clp, &args->fh);
  40         if (inode == NULL)
  41                 goto out;
  42         nfsi = NFS_I(inode);
  43         rcu_read_lock();
  44         delegation = rcu_dereference(nfsi->delegation);
  45         if (delegation == NULL || (delegation->type & FMODE_WRITE) == 0)
  46                 goto out_iput;
  47         res->size = i_size_read(inode);
  48         res->change_attr = delegation->change_attr;
  49         if (nfsi->npages != 0)
  50                 res->change_attr++;
  51         res->ctime = inode->i_ctime;
  52         res->mtime = inode->i_mtime;
  53         res->bitmap[0] = (FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE) &
  54                 args->bitmap[0];
  55         res->bitmap[1] = (FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY) &
  56                 args->bitmap[1];
  57         res->status = 0;
  58 out_iput:
  59         rcu_read_unlock();
  60         iput(inode);
  61 out:
  62         dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
  63         return res->status;
  64 }
  65
  66 __be32 nfs4_callback_recall(struct cb_recallargs *args, void *dummy,
  67                             struct cb_process_state *cps)
  68 {
  69         struct inode *inode;
  70         __be32 res;
  71
  72         res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
  73         if (!cps->clp) /* Always set for v4.0. Set in cb_sequence for v4.1 */
  74                 goto out;
  75
  76         dprintk("NFS: RECALL callback request from %s\n",
  77                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
  78
  79         res = htonl(NFS4ERR_BADHANDLE);
  80         inode = nfs_delegation_find_inode(cps->clp, &args->fh);
  81         if (inode == NULL)
  82                 goto out;
  83         /* Set up a helper thread to actually return the delegation */
  84         switch (nfs_async_inode_return_delegation(inode, &args->stateid)) {
  85         case 0:
  86                 res = 0;
  87                 break;
  88         case -ENOENT:
  89                 if (res != 0)
  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 int nfs4_validate_delegation_stateid(struct nfs_delegation *delegation, const nfs4_stateid *stateid)
 102 {
 103         if (delegation == NULL || memcmp(delegation->stateid.data, stateid->data,
 104                                          sizeof(delegation->stateid.data)) != 0)
 105                 return 0;
 106         return 1;
 107 }
 108
 109 #if defined(CONFIG_NFS_V4_1)
 110
 111 static u32 initiate_file_draining(struct nfs_client *clp,
 112                                   struct cb_layoutrecallargs *args)
 113 {
 114         struct pnfs_layout_hdr *lo;
 115         struct inode *ino;
 116         bool found = false;
 117         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 118         LIST_HEAD(free_me_list);
 119
 120         spin_lock(&clp->cl_lock);
 121         list_for_each_entry(lo, &clp->cl_layouts, plh_layouts) {
 122                 if (nfs_compare_fh(&args->cbl_fh,
 123                                    &NFS_I(lo->plh_inode)->fh))
 124                         continue;
 125                 ino = igrab(lo->plh_inode);
 126                 if (!ino)
 127                         continue;
 128                 found = true;
 129                 /* Without this, layout can be freed as soon
 130                  * as we release cl_lock.
 131                  */
 132                 get_layout_hdr(lo);
 133                 break;
 134         }
 135         spin_unlock(&clp->cl_lock);
 136         if (!found)
 137                 return NFS4ERR_NOMATCHING_LAYOUT;
 138
 139         spin_lock(&ino->i_lock);
 140         if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
 141             mark_matching_lsegs_invalid(lo, &free_me_list,
 142                                         &args->cbl_range))
 143                 rv = NFS4ERR_DELAY;
 144         else
 145                 rv = NFS4ERR_NOMATCHING_LAYOUT;
 146         pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
 147         spin_unlock(&ino->i_lock);
 148         pnfs_free_lseg_list(&free_me_list);
 149         put_layout_hdr(lo);
 150         iput(ino);
 151         return rv;
 152 }
 153
 154 static u32 initiate_bulk_draining(struct nfs_client *clp,
 155                                   struct cb_layoutrecallargs *args)
 156 {
 157         struct pnfs_layout_hdr *lo;
 158         struct inode *ino;
 159         u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 160         struct pnfs_layout_hdr *tmp;
 161         LIST_HEAD(recall_list);
 162         LIST_HEAD(free_me_list);
 163         struct pnfs_layout_range range = {
 164                 .iomode = IOMODE_ANY,
 165                 .offset = 0,
 166                 .length = NFS4_MAX_UINT64,
 167         };
 168
 169         spin_lock(&clp->cl_lock);
 170         list_for_each_entry(lo, &clp->cl_layouts, plh_layouts) {
 171                 if ((args->cbl_recall_type == RETURN_FSID) &&
 172                     memcmp(&NFS_SERVER(lo->plh_inode)->fsid,
 173                            &args->cbl_fsid, sizeof(struct nfs_fsid)))
 174                         continue;
 175                 if (!igrab(lo->plh_inode))
 176                         continue;
 177                 get_layout_hdr(lo);
 178                 BUG_ON(!list_empty(&lo->plh_bulk_recall));
 179                 list_add(&lo->plh_bulk_recall, &recall_list);
 180         }
 181         spin_unlock(&clp->cl_lock);
 182         list_for_each_entry_safe(lo, tmp,
 183                                  &recall_list, plh_bulk_recall) {
 184                 ino = lo->plh_inode;
 185                 spin_lock(&ino->i_lock);
 186                 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
 187                 if (mark_matching_lsegs_invalid(lo, &free_me_list, &range))
 188                         rv = NFS4ERR_DELAY;
 189                 list_del_init(&lo->plh_bulk_recall);
 190                 spin_unlock(&ino->i_lock);
 191                 pnfs_free_lseg_list(&free_me_list);
 192                 put_layout_hdr(lo);
 193                 iput(ino);
 194         }
 195         return rv;
 196 }
 197
 198 static u32 do_callback_layoutrecall(struct nfs_client *clp,
 199                                     struct cb_layoutrecallargs *args)
 200 {
 201         u32 res = NFS4ERR_DELAY;
 202
 203         dprintk("%s enter, type=%i\n", __func__, args->cbl_recall_type);
 204         if (test_and_set_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state))
 205                 goto out;
 206         if (args->cbl_recall_type == RETURN_FILE)
 207                 res = initiate_file_draining(clp, args);
 208         else
 209                 res = initiate_bulk_draining(clp, args);
 210         clear_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state);
 211 out:
 212         dprintk("%s returning %i\n", __func__, res);
 213         return res;
 214
 215 }
 216
 217 __be32 nfs4_callback_layoutrecall(struct cb_layoutrecallargs *args,
 218                                   void *dummy, struct cb_process_state *cps)
 219 {
 220         u32 res;
 221
 222         dprintk("%s: -->\n", __func__);
 223
 224         if (cps->clp)
 225                 res = do_callback_layoutrecall(cps->clp, args);
 226         else
 227                 res = NFS4ERR_OP_NOT_IN_SESSION;
 228
 229         dprintk("%s: exit with status = %d\n", __func__, res);
 230         return cpu_to_be32(res);
 231 }
 232
 233 static void pnfs_recall_all_layouts(struct nfs_client *clp)
 234 {
 235         struct cb_layoutrecallargs args;
 236
 237         /* Pretend we got a CB_LAYOUTRECALL(ALL) */
 238         memset(&args, 0, sizeof(args));
 239         args.cbl_recall_type = RETURN_ALL;
 240         /* FIXME we ignore errors, what should we do? */
 241         do_callback_layoutrecall(clp, &args);
 242 }
 243
 244 __be32 nfs4_callback_devicenotify(struct cb_devicenotifyargs *args,
 245                                   void *dummy, struct cb_process_state *cps)
 246 {
 247         int i;
 248         __be32 res = 0;
 249         struct nfs_client *clp = cps->clp;
 250         struct nfs_server *server = NULL;
 251
 252         dprintk("%s: -->\n", __func__);
 253
 254         if (!clp) {
 255                 res = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 256                 goto out;
 257         }
 258
 259         for (i = 0; i < args->ndevs; i++) {
 260                 struct cb_devicenotifyitem *dev = &args->devs[i];
 261
 262                 if (!server ||
 263                     server->pnfs_curr_ld->id != dev->cbd_layout_type) {
 264                         rcu_read_lock();
 265                         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 266                                 if (server->pnfs_curr_ld &&
 267                                     server->pnfs_curr_ld->id == dev->cbd_layout_type) {
 268                                         rcu_read_unlock();
 269                                         goto found;
 270                                 }
 271                         rcu_read_unlock();
 272                         dprintk("%s: layout type %u not found\n",
 273                                 __func__, dev->cbd_layout_type);
 274                         continue;
 275                 }
 276
 277         found:
 278                 if (dev->cbd_notify_type == NOTIFY_DEVICEID4_CHANGE)
 279                         dprintk("%s: NOTIFY_DEVICEID4_CHANGE not supported, "
 280                                 "deleting instead\n", __func__);
 281                 nfs4_delete_deviceid(server->pnfs_curr_ld, clp, &dev->cbd_dev_id);
 282         }
 283
 284 out:
 285         kfree(args->devs);
 286         dprintk("%s: exit with status = %u\n",
 287                 __func__, be32_to_cpu(res));
 288         return res;
 289 }
 290
 291 int nfs41_validate_delegation_stateid(struct nfs_delegation *delegation, const nfs4_stateid *stateid)
 292 {
 293         if (delegation == NULL)
 294                 return 0;
 295
 296         if (stateid->stateid.seqid != 0)
 297                 return 0;
 298         if (memcmp(&delegation->stateid.stateid.other,
 299                    &stateid->stateid.other,
 300                    NFS4_STATEID_OTHER_SIZE))
 301                 return 0;
 302
 303         return 1;
 304 }
 305
 306 /*
 307  * Validate the sequenceID sent by the server.
 308  * Return success if the sequenceID is one more than what we last saw on
 309  * this slot, accounting for wraparound.  Increments the slot's sequence.
 310  *
 311  * We don't yet implement a duplicate request cache, instead we set the
 312  * back channel ca_maxresponsesize_cached to zero. This is OK for now
 313  * since we only currently implement idempotent callbacks anyway.
 314  *
 315  * We have a single slot backchannel at this time, so we don't bother
 316  * checking the used_slots bit array on the table.  The lower layer guarantees
 317  * a single outstanding callback request at a time.
 318  */
 319 static __be32
 320 validate_seqid(struct nfs4_slot_table *tbl, struct cb_sequenceargs * args)
 321 {
 322         struct nfs4_slot *slot;
 323
 324         dprintk("%s enter. slotid %d seqid %d\n",
 325                 __func__, args->csa_slotid, args->csa_sequenceid);
 326
 327         if (args->csa_slotid > NFS41_BC_MAX_CALLBACKS)
 328                 return htonl(NFS4ERR_BADSLOT);
 329
 330         slot = tbl->slots + args->csa_slotid;
 331         dprintk("%s slot table seqid: %d\n", __func__, slot->seq_nr);
 332
 333         /* Normal */
 334         if (likely(args->csa_sequenceid == slot->seq_nr + 1)) {
 335                 slot->seq_nr++;
 336                 goto out_ok;
 337         }
 338
 339         /* Replay */
 340         if (args->csa_sequenceid == slot->seq_nr) {
 341                 dprintk("%s seqid %d is a replay\n",
 342                         __func__, args->csa_sequenceid);
 343                 /* Signal process_op to set this error on next op */
 344                 if (args->csa_cachethis == 0)
 345                         return htonl(NFS4ERR_RETRY_UNCACHED_REP);
 346
 347                 /* The ca_maxresponsesize_cached is 0 with no DRC */
 348                 else if (args->csa_cachethis == 1)
 349                         return htonl(NFS4ERR_REP_TOO_BIG_TO_CACHE);
 350         }
 351
 352         /* Wraparound */
 353         if (args->csa_sequenceid == 1 && (slot->seq_nr + 1) == 0) {
 354                 slot->seq_nr = 1;
 355                 goto out_ok;
 356         }
 357
 358         /* Misordered request */
 359         return htonl(NFS4ERR_SEQ_MISORDERED);
 360 out_ok:
 361         tbl->highest_used_slotid = args->csa_slotid;
 362         return htonl(NFS4_OK);
 363 }
 364
 365 /*
 366  * For each referring call triple, check the session's slot table for
 367  * a match.  If the slot is in use and the sequence numbers match, the
 368  * client is still waiting for a response to the original request.
 369  */
 370 static bool referring_call_exists(struct nfs_client *clp,
 371                                   uint32_t nrclists,
 372                                   struct referring_call_list *rclists)
 373 {
 374         bool status = 0;
 375         int i, j;
 376         struct nfs4_session *session;
 377         struct nfs4_slot_table *tbl;
 378         struct referring_call_list *rclist;
 379         struct referring_call *ref;
 380
 381         /*
 382          * XXX When client trunking is implemented, this becomes
 383          * a session lookup from within the loop
 384          */
 385         session = clp->cl_session;
 386         tbl = &session->fc_slot_table;
 387
 388         for (i = 0; i < nrclists; i++) {
 389                 rclist = &rclists[i];
 390                 if (memcmp(session->sess_id.data,
 391                            rclist->rcl_sessionid.data,
 392                            NFS4_MAX_SESSIONID_LEN) != 0)
 393                         continue;
 394
 395                 for (j = 0; j < rclist->rcl_nrefcalls; j++) {
 396                         ref = &rclist->rcl_refcalls[j];
 397
 398                         dprintk("%s: sessionid %x:%x:%x:%x sequenceid %u "
 399                                 "slotid %u\n", __func__,
 400                                 ((u32 *)&rclist->rcl_sessionid.data)[0],
 401                                 ((u32 *)&rclist->rcl_sessionid.data)[1],
 402                                 ((u32 *)&rclist->rcl_sessionid.data)[2],
 403                                 ((u32 *)&rclist->rcl_sessionid.data)[3],
 404                                 ref->rc_sequenceid, ref->rc_slotid);
 405
 406                         spin_lock(&tbl->slot_tbl_lock);
 407                         status = (test_bit(ref->rc_slotid, tbl->used_slots) &&
 408                                   tbl->slots[ref->rc_slotid].seq_nr ==
 409                                         ref->rc_sequenceid);
 410                         spin_unlock(&tbl->slot_tbl_lock);
 411                         if (status)
 412                                 goto out;
 413                 }
 414         }
 415
 416 out:
 417         return status;
 418 }
 419
 420 __be32 nfs4_callback_sequence(struct cb_sequenceargs *args,
 421                               struct cb_sequenceres *res,
 422                               struct cb_process_state *cps)
 423 {
 424         struct nfs4_slot_table *tbl;
 425         struct nfs_client *clp;
 426         int i;
 427         __be32 status = htonl(NFS4ERR_BADSESSION);
 428
 429         clp = nfs4_find_client_sessionid(args->csa_addr, &args->csa_sessionid);
 430         if (clp == NULL)
 431                 goto out;
 432
 433         tbl = &clp->cl_session->bc_slot_table;
 434
 435         spin_lock(&tbl->slot_tbl_lock);
 436         /* state manager is resetting the session */
 437         if (test_bit(NFS4_SESSION_DRAINING, &clp->cl_session->session_state)) {
 438                 spin_unlock(&tbl->slot_tbl_lock);
 439                 status = htonl(NFS4ERR_DELAY);
 440                 /* Return NFS4ERR_BADSESSION if we're draining the session
 441                  * in order to reset it.
 442                  */
 443                 if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
 444                         status = htonl(NFS4ERR_BADSESSION);
 445                 goto out;
 446         }
 447
 448         status = validate_seqid(&clp->cl_session->bc_slot_table, args);
 449         spin_unlock(&tbl->slot_tbl_lock);
 450         if (status)
 451                 goto out;
 452
 453         cps->slotid = args->csa_slotid;
 454
 455         /*
 456          * Check for pending referring calls.  If a match is found, a
 457          * related callback was received before the response to the original
 458          * call.
 459          */
 460         if (referring_call_exists(clp, args->csa_nrclists, args->csa_rclists)) {
 461                 status = htonl(NFS4ERR_DELAY);
 462                 goto out;
 463         }
 464
 465         memcpy(&res->csr_sessionid, &args->csa_sessionid,
 466                sizeof(res->csr_sessionid));
 467         res->csr_sequenceid = args->csa_sequenceid;
 468         res->csr_slotid = args->csa_slotid;
 469         res->csr_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 470         res->csr_target_highestslotid = NFS41_BC_MAX_CALLBACKS - 1;
 471
 472 out:
 473         cps->clp = clp; /* put in nfs4_callback_compound */
 474         for (i = 0; i < args->csa_nrclists; i++)
 475                 kfree(args->csa_rclists[i].rcl_refcalls);
 476         kfree(args->csa_rclists);
 477
 478         if (status == htonl(NFS4ERR_RETRY_UNCACHED_REP)) {
 479                 cps->drc_status = status;
 480                 status = 0;
 481         } else
 482                 res->csr_status = status;
 483
 484         dprintk("%s: exit with status = %d res->csr_status %d\n", __func__,
 485                 ntohl(status), ntohl(res->csr_status));
 486         return status;
 487 }
 488
 489 static bool
 490 validate_bitmap_values(unsigned long mask)
 491 {
 492         return (mask & ~RCA4_TYPE_MASK_ALL) == 0;
 493 }
 494
 495 __be32 nfs4_callback_recallany(struct cb_recallanyargs *args, void *dummy,
 496                                struct cb_process_state *cps)
 497 {
 498         __be32 status;
 499         fmode_t flags = 0;
 500
 501         status = cpu_to_be32(NFS4ERR_OP_NOT_IN_SESSION);
 502         if (!cps->clp) /* set in cb_sequence */
 503                 goto out;
 504
 505         dprintk("NFS: RECALL_ANY callback request from %s\n",
 506                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
 507
 508         status = cpu_to_be32(NFS4ERR_INVAL);
 509         if (!validate_bitmap_values(args->craa_type_mask))
 510                 goto out;
 511
 512         status = cpu_to_be32(NFS4_OK);
 513         if (test_bit(RCA4_TYPE_MASK_RDATA_DLG, (const unsigned long *)
 514                      &args->craa_type_mask))
 515                 flags = FMODE_READ;
 516         if (test_bit(RCA4_TYPE_MASK_WDATA_DLG, (const unsigned long *)
 517                      &args->craa_type_mask))
 518                 flags |= FMODE_WRITE;
 519         if (test_bit(RCA4_TYPE_MASK_FILE_LAYOUT, (const unsigned long *)
 520                      &args->craa_type_mask))
 521                 pnfs_recall_all_layouts(cps->clp);
 522         if (flags)
 523                 nfs_expire_all_delegation_types(cps->clp, flags);
 524 out:
 525         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 526         return status;
 527 }
 528
 529 /* Reduce the fore channel's max_slots to the target value */
 530 __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 531                                 struct cb_process_state *cps)
 532 {
 533         struct nfs4_slot_table *fc_tbl;
 534         __be32 status;
 535
 536         status = htonl(NFS4ERR_OP_NOT_IN_SESSION);
 537         if (!cps->clp) /* set in cb_sequence */
 538                 goto out;
 539
 540         dprintk("NFS: CB_RECALL_SLOT request from %s target max slots %d\n",
 541                 rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
 542                 args->crsa_target_max_slots);
 543
 544         fc_tbl = &cps->clp->cl_session->fc_slot_table;
 545
 546         status = htonl(NFS4ERR_BAD_HIGH_SLOT);
 547         if (args->crsa_target_max_slots > fc_tbl->max_slots ||
 548             args->crsa_target_max_slots < 1)
 549                 goto out;
 550
 551         status = htonl(NFS4_OK);
 552         if (args->crsa_target_max_slots == fc_tbl->max_slots)
 553                 goto out;
 554
 555         fc_tbl->target_max_slots = args->crsa_target_max_slots;
 556         nfs41_handle_recall_slot(cps->clp);
 557 out:
 558         dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 559         return status;
 560 }
 561 #endif /* CONFIG_NFS_V4_1 */