fs/nfs/delegation.c

   1 /*
   2  * linux/fs/nfs/delegation.c
   3  *
   4  * Copyright (C) 2004 Trond Myklebust
   5  *
   6  * NFS file delegation management
   7  *
   8  */
   9 #include <linux/completion.h>
  10 #include <linux/kthread.h>
  11 #include <linux/module.h>
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/smp_lock.h>
  15 #include <linux/spinlock.h>
  16
  17 #include <linux/nfs4.h>
  18 #include <linux/nfs_fs.h>
  19 #include <linux/nfs_xdr.h>
  20
  21 #include "nfs4_fs.h"
  22 #include "delegation.h"
  23 #include "internal.h"
  24
  25 static void nfs_do_free_delegation(struct nfs_delegation *delegation)
  26 {
  27         kfree(delegation);
  28 }
  29
  30 static void nfs_free_delegation_callback(struct rcu_head *head)
  31 {
  32         struct nfs_delegation *delegation = container_of(head, struct nfs_delegation, rcu);
  33
  34         nfs_do_free_delegation(delegation);
  35 }
  36
  37 static void nfs_free_delegation(struct nfs_delegation *delegation)
  38 {
  39         struct rpc_cred *cred;
  40
  41         cred = rcu_dereference(delegation->cred);
  42         rcu_assign_pointer(delegation->cred, NULL);
  43         call_rcu(&delegation->rcu, nfs_free_delegation_callback);
  44         if (cred)
  45                 put_rpccred(cred);
  46 }
  47
  48 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
  49 {
  50         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
  51 }
  52
  53 int nfs_have_delegation(struct inode *inode, fmode_t flags)
  54 {
  55         struct nfs_delegation *delegation;
  56         int ret = 0;
  57
  58         flags &= FMODE_READ|FMODE_WRITE;
  59         rcu_read_lock();
  60         delegation = rcu_dereference(NFS_I(inode)->delegation);
  61         if (delegation != NULL && (delegation->type & flags) == flags) {
  62                 nfs_mark_delegation_referenced(delegation);
  63                 ret = 1;
  64         }
  65         rcu_read_unlock();
  66         return ret;
  67 }
  68
  69 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
  70 {
  71         struct inode *inode = state->inode;
  72         struct file_lock *fl;
  73         int status = 0;
  74
  75         if (inode->i_flock == NULL)
  76                 goto out;
  77
  78         /* Protect inode->i_flock using the BKL */
  79         lock_kernel();
  80         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
  81                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
  82                         continue;
  83                 if (nfs_file_open_context(fl->fl_file) != ctx)
  84                         continue;
  85                 unlock_kernel();
  86                 status = nfs4_lock_delegation_recall(state, fl);
  87                 if (status < 0)
  88                         goto out;
  89                 lock_kernel();
  90         }
  91         unlock_kernel();
  92 out:
  93         return status;
  94 }
  95
  96 static int nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
  97 {
  98         struct nfs_inode *nfsi = NFS_I(inode);
  99         struct nfs_open_context *ctx;
 100         struct nfs4_state *state;
 101         int err;
 102
 103 again:
 104         spin_lock(&inode->i_lock);
 105         list_for_each_entry(ctx, &nfsi->open_files, list) {
 106                 state = ctx->state;
 107                 if (state == NULL)
 108                         continue;
 109                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 110                         continue;
 111                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
 112                         continue;
 113                 get_nfs_open_context(ctx);
 114                 spin_unlock(&inode->i_lock);
 115                 err = nfs4_open_delegation_recall(ctx, state, stateid);
 116                 if (err >= 0)
 117                         err = nfs_delegation_claim_locks(ctx, state);
 118                 put_nfs_open_context(ctx);
 119                 if (err != 0)
 120                         return err;
 121                 goto again;
 122         }
 123         spin_unlock(&inode->i_lock);
 124         return 0;
 125 }
 126
 127 /*
 128  * Set up a delegation on an inode
 129  */
 130 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
 131 {
 132         struct nfs_delegation *delegation;
 133         struct rpc_cred *oldcred = NULL;
 134
 135         rcu_read_lock();
 136         delegation = rcu_dereference(NFS_I(inode)->delegation);
 137         if (delegation != NULL) {
 138                 spin_lock(&delegation->lock);
 139                 if (delegation->inode != NULL) {
 140                         memcpy(delegation->stateid.data, res->delegation.data,
 141                                sizeof(delegation->stateid.data));
 142                         delegation->type = res->delegation_type;
 143                         delegation->maxsize = res->maxsize;
 144                         oldcred = delegation->cred;
 145                         delegation->cred = get_rpccred(cred);
 146                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
 147                                   &delegation->flags);
 148                         NFS_I(inode)->delegation_state = delegation->type;
 149                         spin_unlock(&delegation->lock);
 150                         put_rpccred(oldcred);
 151                         rcu_read_unlock();
 152                 } else {
 153                         /* We appear to have raced with a delegation return. */
 154                         spin_unlock(&delegation->lock);
 155                         rcu_read_unlock();
 156                         nfs_inode_set_delegation(inode, cred, res);
 157                 }
 158         } else {
 159                 rcu_read_unlock();
 160         }
 161 }
 162
 163 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 164 {
 165         int res = 0;
 166
 167         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
 168         nfs_free_delegation(delegation);
 169         return res;
 170 }
 171
 172 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
 173 {
 174         struct inode *inode = NULL;
 175
 176         spin_lock(&delegation->lock);
 177         if (delegation->inode != NULL)
 178                 inode = igrab(delegation->inode);
 179         spin_unlock(&delegation->lock);
 180         return inode;
 181 }
 182
 183 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
 184 {
 185         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
 186
 187         if (delegation == NULL)
 188                 goto nomatch;
 189         spin_lock(&delegation->lock);
 190         if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
 191                                 sizeof(delegation->stateid.data)) != 0)
 192                 goto nomatch_unlock;
 193         list_del_rcu(&delegation->super_list);
 194         delegation->inode = NULL;
 195         nfsi->delegation_state = 0;
 196         rcu_assign_pointer(nfsi->delegation, NULL);
 197         spin_unlock(&delegation->lock);
 198         return delegation;
 199 nomatch_unlock:
 200         spin_unlock(&delegation->lock);
 201 nomatch:
 202         return NULL;
 203 }
 204
 205 /*
 206  * Set up a delegation on an inode
 207  */
 208 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
 209 {
 210         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 211         struct nfs_inode *nfsi = NFS_I(inode);
 212         struct nfs_delegation *delegation;
 213         struct nfs_delegation *freeme = NULL;
 214         int status = 0;
 215
 216         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
 217         if (delegation == NULL)
 218                 return -ENOMEM;
 219         memcpy(delegation->stateid.data, res->delegation.data,
 220                         sizeof(delegation->stateid.data));
 221         delegation->type = res->delegation_type;
 222         delegation->maxsize = res->maxsize;
 223         delegation->change_attr = nfsi->change_attr;
 224         delegation->cred = get_rpccred(cred);
 225         delegation->inode = inode;
 226         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
 227         spin_lock_init(&delegation->lock);
 228
 229         spin_lock(&clp->cl_lock);
 230         if (rcu_dereference(nfsi->delegation) != NULL) {
 231                 if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
 232                                         sizeof(delegation->stateid)) == 0 &&
 233                                 delegation->type == nfsi->delegation->type) {
 234                         goto out;
 235                 }
 236                 /*
 237                  * Deal with broken servers that hand out two
 238                  * delegations for the same file.
 239                  */
 240                 dfprintk(FILE, "%s: server %s handed out "
 241                                 "a duplicate delegation!\n",
 242                                 __func__, clp->cl_hostname);
 243                 if (delegation->type <= nfsi->delegation->type) {
 244                         freeme = delegation;
 245                         delegation = NULL;
 246                         goto out;
 247                 }
 248                 freeme = nfs_detach_delegation_locked(nfsi, NULL);
 249         }
 250         list_add_rcu(&delegation->super_list, &clp->cl_delegations);
 251         nfsi->delegation_state = delegation->type;
 252         rcu_assign_pointer(nfsi->delegation, delegation);
 253         delegation = NULL;
 254
 255         /* Ensure we revalidate the attributes and page cache! */
 256         spin_lock(&inode->i_lock);
 257         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
 258         spin_unlock(&inode->i_lock);
 259
 260 out:
 261         spin_unlock(&clp->cl_lock);
 262         if (delegation != NULL)
 263                 nfs_free_delegation(delegation);
 264         if (freeme != NULL)
 265                 nfs_do_return_delegation(inode, freeme, 0);
 266         return status;
 267 }
 268
 269 /* Sync all data to disk upon delegation return */
 270 static void nfs_msync_inode(struct inode *inode)
 271 {
 272         filemap_fdatawrite(inode->i_mapping);
 273         nfs_wb_all(inode);
 274         filemap_fdatawait(inode->i_mapping);
 275 }
 276
 277 /*
 278  * Basic procedure for returning a delegation to the server
 279  */
 280 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 281 {
 282         struct nfs_inode *nfsi = NFS_I(inode);
 283         int err;
 284
 285         /*
 286          * Guard against new delegated open/lock/unlock calls and against
 287          * state recovery
 288          */
 289         down_write(&nfsi->rwsem);
 290         err = nfs_delegation_claim_opens(inode, &delegation->stateid);
 291         up_write(&nfsi->rwsem);
 292         if (err)
 293                 goto out;
 294
 295         err = nfs_do_return_delegation(inode, delegation, issync);
 296 out:
 297         return err;
 298 }
 299
 300 /*
 301  * Return all delegations that have been marked for return
 302  */
 303 int nfs_client_return_marked_delegations(struct nfs_client *clp)
 304 {
 305         struct nfs_delegation *delegation;
 306         struct inode *inode;
 307         int err = 0;
 308
 309 restart:
 310         rcu_read_lock();
 311         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
 312                 if (!test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
 313                         continue;
 314                 inode = nfs_delegation_grab_inode(delegation);
 315                 if (inode == NULL)
 316                         continue;
 317                 spin_lock(&clp->cl_lock);
 318                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
 319                 spin_unlock(&clp->cl_lock);
 320                 rcu_read_unlock();
 321                 if (delegation != NULL) {
 322                         filemap_flush(inode->i_mapping);
 323                         err = __nfs_inode_return_delegation(inode, delegation, 0);
 324                 }
 325                 iput(inode);
 326                 if (!err)
 327                         goto restart;
 328                 set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 329                 return err;
 330         }
 331         rcu_read_unlock();
 332         return 0;
 333 }
 334
 335 /*
 336  * This function returns the delegation without reclaiming opens
 337  * or protecting against delegation reclaims.
 338  * It is therefore really only safe to be called from
 339  * nfs4_clear_inode()
 340  */
 341 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
 342 {
 343         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 344         struct nfs_inode *nfsi = NFS_I(inode);
 345         struct nfs_delegation *delegation;
 346
 347         if (rcu_dereference(nfsi->delegation) != NULL) {
 348                 spin_lock(&clp->cl_lock);
 349                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
 350                 spin_unlock(&clp->cl_lock);
 351                 if (delegation != NULL)
 352                         nfs_do_return_delegation(inode, delegation, 0);
 353         }
 354 }
 355
 356 int nfs_inode_return_delegation(struct inode *inode)
 357 {
 358         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 359         struct nfs_inode *nfsi = NFS_I(inode);
 360         struct nfs_delegation *delegation;
 361         int err = 0;
 362
 363         if (rcu_dereference(nfsi->delegation) != NULL) {
 364                 spin_lock(&clp->cl_lock);
 365                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
 366                 spin_unlock(&clp->cl_lock);
 367                 if (delegation != NULL) {
 368                         nfs_msync_inode(inode);
 369                         err = __nfs_inode_return_delegation(inode, delegation, 1);
 370                 }
 371         }
 372         return err;
 373 }
 374
 375 static void nfs_mark_return_delegation(struct nfs_client *clp, struct nfs_delegation *delegation)
 376 {
 377         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 378         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 379 }
 380
 381 /*
 382  * Return all delegations associated to a super block
 383  */
 384 void nfs_super_return_all_delegations(struct super_block *sb)
 385 {
 386         struct nfs_client *clp = NFS_SB(sb)->nfs_client;
 387         struct nfs_delegation *delegation;
 388
 389         if (clp == NULL)
 390                 return;
 391         rcu_read_lock();
 392         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
 393                 spin_lock(&delegation->lock);
 394                 if (delegation->inode != NULL && delegation->inode->i_sb == sb)
 395                         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 396                 spin_unlock(&delegation->lock);
 397         }
 398         rcu_read_unlock();
 399         if (nfs_client_return_marked_delegations(clp) != 0)
 400                 nfs4_schedule_state_manager(clp);
 401 }
 402
 403 static
 404 void nfs_client_mark_return_all_delegation_types(struct nfs_client *clp, fmode_t flags)
 405 {
 406         struct nfs_delegation *delegation;
 407
 408         rcu_read_lock();
 409         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
 410                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
 411                         continue;
 412                 if (delegation->type & flags)
 413                         nfs_mark_return_delegation(clp, delegation);
 414         }
 415         rcu_read_unlock();
 416 }
 417
 418 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
 419 {
 420         nfs_client_mark_return_all_delegation_types(clp, FMODE_READ|FMODE_WRITE);
 421 }
 422
 423 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
 424 {
 425         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
 426                 nfs4_schedule_state_manager(clp);
 427 }
 428
 429 void nfs_expire_all_delegation_types(struct nfs_client *clp, fmode_t flags)
 430 {
 431         nfs_client_mark_return_all_delegation_types(clp, flags);
 432         nfs_delegation_run_state_manager(clp);
 433 }
 434
 435 void nfs_expire_all_delegations(struct nfs_client *clp)
 436 {
 437         nfs_expire_all_delegation_types(clp, FMODE_READ|FMODE_WRITE);
 438 }
 439
 440 /*
 441  * Return all delegations following an NFS4ERR_CB_PATH_DOWN error.
 442  */
 443 void nfs_handle_cb_pathdown(struct nfs_client *clp)
 444 {
 445         if (clp == NULL)
 446                 return;
 447         nfs_client_mark_return_all_delegations(clp);
 448 }
 449
 450 static void nfs_client_mark_return_unreferenced_delegations(struct nfs_client *clp)
 451 {
 452         struct nfs_delegation *delegation;
 453
 454         rcu_read_lock();
 455         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
 456                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
 457                         continue;
 458                 nfs_mark_return_delegation(clp, delegation);
 459         }
 460         rcu_read_unlock();
 461 }
 462
 463 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
 464 {
 465         nfs_client_mark_return_unreferenced_delegations(clp);
 466         nfs_delegation_run_state_manager(clp);
 467 }
 468
 469 /*
 470  * Asynchronous delegation recall!
 471  */
 472 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid,
 473                                       int (*validate_stateid)(struct nfs_delegation *delegation,
 474                                                               const nfs4_stateid *stateid))
 475 {
 476         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 477         struct nfs_delegation *delegation;
 478
 479         rcu_read_lock();
 480         delegation = rcu_dereference(NFS_I(inode)->delegation);
 481
 482         if (!validate_stateid(delegation, stateid)) {
 483                 rcu_read_unlock();
 484                 return -ENOENT;
 485         }
 486
 487         nfs_mark_return_delegation(clp, delegation);
 488         rcu_read_unlock();
 489         nfs_delegation_run_state_manager(clp);
 490         return 0;
 491 }
 492
 493 /*
 494  * Retrieve the inode associated with a delegation
 495  */
 496 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle)
 497 {
 498         struct nfs_delegation *delegation;
 499         struct inode *res = NULL;
 500         rcu_read_lock();
 501         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
 502                 spin_lock(&delegation->lock);
 503                 if (delegation->inode != NULL &&
 504                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
 505                         res = igrab(delegation->inode);
 506                 }
 507                 spin_unlock(&delegation->lock);
 508                 if (res != NULL)
 509                         break;
 510         }
 511         rcu_read_unlock();
 512         return res;
 513 }
 514
 515 /*
 516  * Mark all delegations as needing to be reclaimed
 517  */
 518 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
 519 {
 520         struct nfs_delegation *delegation;
 521         rcu_read_lock();
 522         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list)
 523                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 524         rcu_read_unlock();
 525 }
 526
 527 /*
 528  * Reap all unclaimed delegations after reboot recovery is done
 529  */
 530 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
 531 {
 532         struct nfs_delegation *delegation;
 533         struct inode *inode;
 534 restart:
 535         rcu_read_lock();
 536         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
 537                 if (test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) == 0)
 538                         continue;
 539                 inode = nfs_delegation_grab_inode(delegation);
 540                 if (inode == NULL)
 541                         continue;
 542                 spin_lock(&clp->cl_lock);
 543                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
 544                 spin_unlock(&clp->cl_lock);
 545                 rcu_read_unlock();
 546                 if (delegation != NULL)
 547                         nfs_free_delegation(delegation);
 548                 iput(inode);
 549                 goto restart;
 550         }
 551         rcu_read_unlock();
 552 }
 553
 554 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
 555 {
 556         struct nfs_inode *nfsi = NFS_I(inode);
 557         struct nfs_delegation *delegation;
 558         int ret = 0;
 559
 560         rcu_read_lock();
 561         delegation = rcu_dereference(nfsi->delegation);
 562         if (delegation != NULL) {
 563                 memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
 564                 ret = 1;
 565         }
 566         rcu_read_unlock();
 567         return ret;
 568 }