fs/nfs/nfs4proc.c

   1 /*
   2  *  fs/nfs/nfs4proc.c
   3  *
   4  *  Client-side procedure declarations for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *  Andy Adamson   <andros@umich.edu>
  11  *
  12  *  Redistribution and use in source and binary forms, with or without
  13  *  modification, are permitted provided that the following conditions
  14  *  are met:
  15  *
  16  *  1. Redistributions of source code must retain the above copyright
  17  *     notice, this list of conditions and the following disclaimer.
  18  *  2. Redistributions in binary form must reproduce the above copyright
  19  *     notice, this list of conditions and the following disclaimer in the
  20  *     documentation and/or other materials provided with the distribution.
  21  *  3. Neither the name of the University nor the names of its
  22  *     contributors may be used to endorse or promote products derived
  23  *     from this software without specific prior written permission.
  24  *
  25  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  26  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  27  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  28  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  30  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  31  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  32  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  33  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  34  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  35  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  36  */
  37
  38 #include <linux/mm.h>
  39 #include <linux/delay.h>
  40 #include <linux/errno.h>
  41 #include <linux/string.h>
  42 #include <linux/ratelimit.h>
  43 #include <linux/printk.h>
  44 #include <linux/slab.h>
  45 #include <linux/sunrpc/clnt.h>
  46 #include <linux/sunrpc/gss_api.h>
  47 #include <linux/nfs.h>
  48 #include <linux/nfs4.h>
  49 #include <linux/nfs_fs.h>
  50 #include <linux/nfs_page.h>
  51 #include <linux/nfs_mount.h>
  52 #include <linux/namei.h>
  53 #include <linux/mount.h>
  54 #include <linux/module.h>
  55 #include <linux/sunrpc/bc_xprt.h>
  56 #include <linux/xattr.h>
  57 #include <linux/utsname.h>
  58
  59 #include "nfs4_fs.h"
  60 #include "delegation.h"
  61 #include "internal.h"
  62 #include "iostat.h"
  63 #include "callback.h"
  64 #include "pnfs.h"
  65
  66 #define NFSDBG_FACILITY         NFSDBG_PROC
  67
  68 #define NFS4_POLL_RETRY_MIN     (HZ/10)
  69 #define NFS4_POLL_RETRY_MAX     (15*HZ)
  70
  71 #define NFS4_MAX_LOOP_ON_RECOVER (10)
  72
  73 struct nfs4_opendata;
  74 static int _nfs4_proc_open(struct nfs4_opendata *data);
  75 static int _nfs4_recover_proc_open(struct nfs4_opendata *data);
  76 static int nfs4_do_fsinfo(struct nfs_server *, struct nfs_fh *, struct nfs_fsinfo *);
  77 static int nfs4_async_handle_error(struct rpc_task *, const struct nfs_server *, struct nfs4_state *);
  78 static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fattr *fattr);
  79 static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
  80                             struct nfs_fattr *fattr, struct iattr *sattr,
  81                             struct nfs4_state *state);
  82 #ifdef CONFIG_NFS_V4_1
  83 static int nfs41_test_stateid(struct nfs_server *, struct nfs4_state *);
  84 static int nfs41_free_stateid(struct nfs_server *, struct nfs4_state *);
  85 #endif
  86 /* Prevent leaks of NFSv4 errors into userland */
  87 static int nfs4_map_errors(int err)
  88 {
  89         if (err >= -1000)
  90                 return err;
  91         switch (err) {
  92         case -NFS4ERR_RESOURCE:
  93                 return -EREMOTEIO;
  94         case -NFS4ERR_WRONGSEC:
  95                 return -EPERM;
  96         case -NFS4ERR_BADOWNER:
  97         case -NFS4ERR_BADNAME:
  98                 return -EINVAL;
  99         case -NFS4ERR_SHARE_DENIED:
 100                 return -EACCES;
 101         default:
 102                 dprintk("%s could not handle NFSv4 error %d\n",
 103                                 __func__, -err);
 104                 break;
 105         }
 106         return -EIO;
 107 }
 108
 109 /*
 110  * This is our standard bitmap for GETATTR requests.
 111  */
 112 const u32 nfs4_fattr_bitmap[2] = {
 113         FATTR4_WORD0_TYPE
 114         | FATTR4_WORD0_CHANGE
 115         | FATTR4_WORD0_SIZE
 116         | FATTR4_WORD0_FSID
 117         | FATTR4_WORD0_FILEID,
 118         FATTR4_WORD1_MODE
 119         | FATTR4_WORD1_NUMLINKS
 120         | FATTR4_WORD1_OWNER
 121         | FATTR4_WORD1_OWNER_GROUP
 122         | FATTR4_WORD1_RAWDEV
 123         | FATTR4_WORD1_SPACE_USED
 124         | FATTR4_WORD1_TIME_ACCESS
 125         | FATTR4_WORD1_TIME_METADATA
 126         | FATTR4_WORD1_TIME_MODIFY
 127 };
 128
 129 const u32 nfs4_statfs_bitmap[2] = {
 130         FATTR4_WORD0_FILES_AVAIL
 131         | FATTR4_WORD0_FILES_FREE
 132         | FATTR4_WORD0_FILES_TOTAL,
 133         FATTR4_WORD1_SPACE_AVAIL
 134         | FATTR4_WORD1_SPACE_FREE
 135         | FATTR4_WORD1_SPACE_TOTAL
 136 };
 137
 138 const u32 nfs4_pathconf_bitmap[2] = {
 139         FATTR4_WORD0_MAXLINK
 140         | FATTR4_WORD0_MAXNAME,
 141         0
 142 };
 143
 144 const u32 nfs4_fsinfo_bitmap[3] = { FATTR4_WORD0_MAXFILESIZE
 145                         | FATTR4_WORD0_MAXREAD
 146                         | FATTR4_WORD0_MAXWRITE
 147                         | FATTR4_WORD0_LEASE_TIME,
 148                         FATTR4_WORD1_TIME_DELTA
 149                         | FATTR4_WORD1_FS_LAYOUT_TYPES,
 150                         FATTR4_WORD2_LAYOUT_BLKSIZE
 151 };
 152
 153 const u32 nfs4_fs_locations_bitmap[2] = {
 154         FATTR4_WORD0_TYPE
 155         | FATTR4_WORD0_CHANGE
 156         | FATTR4_WORD0_SIZE
 157         | FATTR4_WORD0_FSID
 158         | FATTR4_WORD0_FILEID
 159         | FATTR4_WORD0_FS_LOCATIONS,
 160         FATTR4_WORD1_MODE
 161         | FATTR4_WORD1_NUMLINKS
 162         | FATTR4_WORD1_OWNER
 163         | FATTR4_WORD1_OWNER_GROUP
 164         | FATTR4_WORD1_RAWDEV
 165         | FATTR4_WORD1_SPACE_USED
 166         | FATTR4_WORD1_TIME_ACCESS
 167         | FATTR4_WORD1_TIME_METADATA
 168         | FATTR4_WORD1_TIME_MODIFY
 169         | FATTR4_WORD1_MOUNTED_ON_FILEID
 170 };
 171
 172 static void nfs4_setup_readdir(u64 cookie, __be32 *verifier, struct dentry *dentry,
 173                 struct nfs4_readdir_arg *readdir)
 174 {
 175         __be32 *start, *p;
 176
 177         BUG_ON(readdir->count < 80);
 178         if (cookie > 2) {
 179                 readdir->cookie = cookie;
 180                 memcpy(&readdir->verifier, verifier, sizeof(readdir->verifier));
 181                 return;
 182         }
 183
 184         readdir->cookie = 0;
 185         memset(&readdir->verifier, 0, sizeof(readdir->verifier));
 186         if (cookie == 2)
 187                 return;
 188
 189         /*
 190          * NFSv4 servers do not return entries for '.' and '..'
 191          * Therefore, we fake these entries here.  We let '.'
 192          * have cookie 0 and '..' have cookie 1.  Note that
 193          * when talking to the server, we always send cookie 0
 194          * instead of 1 or 2.
 195          */
 196         start = p = kmap_atomic(*readdir->pages, KM_USER0);
 197
 198         if (cookie == 0) {
 199                 *p++ = xdr_one;                                  /* next */
 200                 *p++ = xdr_zero;                   /* cookie, first word */
 201                 *p++ = xdr_one;                   /* cookie, second word */
 202                 *p++ = xdr_one;                             /* entry len */
 203                 memcpy(p, ".\0\0\0", 4);                        /* entry */
 204                 p++;
 205                 *p++ = xdr_one;                         /* bitmap length */
 206                 *p++ = htonl(FATTR4_WORD0_FILEID);             /* bitmap */
 207                 *p++ = htonl(8);              /* attribute buffer length */
 208                 p = xdr_encode_hyper(p, NFS_FILEID(dentry->d_inode));
 209         }
 210
 211         *p++ = xdr_one;                                  /* next */
 212         *p++ = xdr_zero;                   /* cookie, first word */
 213         *p++ = xdr_two;                   /* cookie, second word */
 214         *p++ = xdr_two;                             /* entry len */
 215         memcpy(p, "..\0\0", 4);                         /* entry */
 216         p++;
 217         *p++ = xdr_one;                         /* bitmap length */
 218         *p++ = htonl(FATTR4_WORD0_FILEID);             /* bitmap */
 219         *p++ = htonl(8);              /* attribute buffer length */
 220         p = xdr_encode_hyper(p, NFS_FILEID(dentry->d_parent->d_inode));
 221
 222         readdir->pgbase = (char *)p - (char *)start;
 223         readdir->count -= readdir->pgbase;
 224         kunmap_atomic(start, KM_USER0);
 225 }
 226
 227 static int nfs4_wait_clnt_recover(struct nfs_client *clp)
 228 {
 229         int res;
 230
 231         might_sleep();
 232
 233         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
 234                         nfs_wait_bit_killable, TASK_KILLABLE);
 235         return res;
 236 }
 237
 238 static int nfs4_delay(struct rpc_clnt *clnt, long *timeout)
 239 {
 240         int res = 0;
 241
 242         might_sleep();
 243
 244         if (*timeout <= 0)
 245                 *timeout = NFS4_POLL_RETRY_MIN;
 246         if (*timeout > NFS4_POLL_RETRY_MAX)
 247                 *timeout = NFS4_POLL_RETRY_MAX;
 248         schedule_timeout_killable(*timeout);
 249         if (fatal_signal_pending(current))
 250                 res = -ERESTARTSYS;
 251         *timeout <<= 1;
 252         return res;
 253 }
 254
 255 /* This is the error handling routine for processes that are allowed
 256  * to sleep.
 257  */
 258 static int nfs4_handle_exception(struct nfs_server *server, int errorcode, struct nfs4_exception *exception)
 259 {
 260         struct nfs_client *clp = server->nfs_client;
 261         struct nfs4_state *state = exception->state;
 262         struct inode *inode = exception->inode;
 263         int ret = errorcode;
 264
 265         exception->retry = 0;
 266         switch(errorcode) {
 267                 case 0:
 268                         return 0;
 269                 case -NFS4ERR_OPENMODE:
 270                         if (nfs_have_delegation(inode, FMODE_READ)) {
 271                                 nfs_inode_return_delegation(inode);
 272                                 exception->retry = 1;
 273                                 return 0;
 274                         }
 275                         if (state == NULL)
 276                                 break;
 277                         nfs4_schedule_stateid_recovery(server, state);
 278                         goto wait_on_recovery;
 279                 case -NFS4ERR_DELEG_REVOKED:
 280                 case -NFS4ERR_ADMIN_REVOKED:
 281                 case -NFS4ERR_BAD_STATEID:
 282                         if (state != NULL)
 283                                 nfs_remove_bad_delegation(state->inode);
 284                         if (state == NULL)
 285                                 break;
 286                         nfs4_schedule_stateid_recovery(server, state);
 287                         goto wait_on_recovery;
 288                 case -NFS4ERR_EXPIRED:
 289                         if (state != NULL)
 290                                 nfs4_schedule_stateid_recovery(server, state);
 291                 case -NFS4ERR_STALE_STATEID:
 292                 case -NFS4ERR_STALE_CLIENTID:
 293                         nfs4_schedule_lease_recovery(clp);
 294                         goto wait_on_recovery;
 295 #if defined(CONFIG_NFS_V4_1)
 296                 case -NFS4ERR_BADSESSION:
 297                 case -NFS4ERR_BADSLOT:
 298                 case -NFS4ERR_BAD_HIGH_SLOT:
 299                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 300                 case -NFS4ERR_DEADSESSION:
 301                 case -NFS4ERR_SEQ_FALSE_RETRY:
 302                 case -NFS4ERR_SEQ_MISORDERED:
 303                         dprintk("%s ERROR: %d Reset session\n", __func__,
 304                                 errorcode);
 305                         nfs4_schedule_session_recovery(clp->cl_session);
 306                         goto wait_on_recovery;
 307 #endif /* defined(CONFIG_NFS_V4_1) */
 308                 case -NFS4ERR_FILE_OPEN:
 309                         if (exception->timeout > HZ) {
 310                                 /* We have retried a decent amount, time to
 311                                  * fail
 312                                  */
 313                                 ret = -EBUSY;
 314                                 break;
 315                         }
 316                 case -NFS4ERR_GRACE:
 317                 case -NFS4ERR_DELAY:
 318                 case -EKEYEXPIRED:
 319                         ret = nfs4_delay(server->client, &exception->timeout);
 320                         if (ret != 0)
 321                                 break;
 322                 case -NFS4ERR_RETRY_UNCACHED_REP:
 323                 case -NFS4ERR_OLD_STATEID:
 324                         exception->retry = 1;
 325                         break;
 326                 case -NFS4ERR_BADOWNER:
 327                         /* The following works around a Linux server bug! */
 328                 case -NFS4ERR_BADNAME:
 329                         if (server->caps & NFS_CAP_UIDGID_NOMAP) {
 330                                 server->caps &= ~NFS_CAP_UIDGID_NOMAP;
 331                                 exception->retry = 1;
 332                                 printk(KERN_WARNING "NFS: v4 server %s "
 333                                                 "does not accept raw "
 334                                                 "uid/gids. "
 335                                                 "Reenabling the idmapper.\n",
 336                                                 server->nfs_client->cl_hostname);
 337                         }
 338         }
 339         /* We failed to handle the error */
 340         return nfs4_map_errors(ret);
 341 wait_on_recovery:
 342         ret = nfs4_wait_clnt_recover(clp);
 343         if (ret == 0)
 344                 exception->retry = 1;
 345         return ret;
 346 }
 347
 348
 349 static void do_renew_lease(struct nfs_client *clp, unsigned long timestamp)
 350 {
 351         spin_lock(&clp->cl_lock);
 352         if (time_before(clp->cl_last_renewal,timestamp))
 353                 clp->cl_last_renewal = timestamp;
 354         spin_unlock(&clp->cl_lock);
 355 }
 356
 357 static void renew_lease(const struct nfs_server *server, unsigned long timestamp)
 358 {
 359         do_renew_lease(server->nfs_client, timestamp);
 360 }
 361
 362 #if defined(CONFIG_NFS_V4_1)
 363
 364 /*
 365  * nfs4_free_slot - free a slot and efficiently update slot table.
 366  *
 367  * freeing a slot is trivially done by clearing its respective bit
 368  * in the bitmap.
 369  * If the freed slotid equals highest_used_slotid we want to update it
 370  * so that the server would be able to size down the slot table if needed,
 371  * otherwise we know that the highest_used_slotid is still in use.
 372  * When updating highest_used_slotid there may be "holes" in the bitmap
 373  * so we need to scan down from highest_used_slotid to 0 looking for the now
 374  * highest slotid in use.
 375  * If none found, highest_used_slotid is set to -1.
 376  *
 377  * Must be called while holding tbl->slot_tbl_lock
 378  */
 379 static void
 380 nfs4_free_slot(struct nfs4_slot_table *tbl, struct nfs4_slot *free_slot)
 381 {
 382         int free_slotid = free_slot - tbl->slots;
 383         int slotid = free_slotid;
 384
 385         BUG_ON(slotid < 0 || slotid >= NFS4_MAX_SLOT_TABLE);
 386         /* clear used bit in bitmap */
 387         __clear_bit(slotid, tbl->used_slots);
 388
 389         /* update highest_used_slotid when it is freed */
 390         if (slotid == tbl->highest_used_slotid) {
 391                 slotid = find_last_bit(tbl->used_slots, tbl->max_slots);
 392                 if (slotid < tbl->max_slots)
 393                         tbl->highest_used_slotid = slotid;
 394                 else
 395                         tbl->highest_used_slotid = -1;
 396         }
 397         dprintk("%s: free_slotid %u highest_used_slotid %d\n", __func__,
 398                 free_slotid, tbl->highest_used_slotid);
 399 }
 400
 401 /*
 402  * Signal state manager thread if session fore channel is drained
 403  */
 404 static void nfs4_check_drain_fc_complete(struct nfs4_session *ses)
 405 {
 406         struct rpc_task *task;
 407
 408         if (!test_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 409                 task = rpc_wake_up_next(&ses->fc_slot_table.slot_tbl_waitq);
 410                 if (task)
 411                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 412                 return;
 413         }
 414
 415         if (ses->fc_slot_table.highest_used_slotid != -1)
 416                 return;
 417
 418         dprintk("%s COMPLETE: Session Fore Channel Drained\n", __func__);
 419         complete(&ses->fc_slot_table.complete);
 420 }
 421
 422 /*
 423  * Signal state manager thread if session back channel is drained
 424  */
 425 void nfs4_check_drain_bc_complete(struct nfs4_session *ses)
 426 {
 427         if (!test_bit(NFS4_SESSION_DRAINING, &ses->session_state) ||
 428             ses->bc_slot_table.highest_used_slotid != -1)
 429                 return;
 430         dprintk("%s COMPLETE: Session Back Channel Drained\n", __func__);
 431         complete(&ses->bc_slot_table.complete);
 432 }
 433
 434 static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 435 {
 436         struct nfs4_slot_table *tbl;
 437
 438         tbl = &res->sr_session->fc_slot_table;
 439         if (!res->sr_slot) {
 440                 /* just wake up the next guy waiting since
 441                  * we may have not consumed a slot after all */
 442                 dprintk("%s: No slot\n", __func__);
 443                 return;
 444         }
 445
 446         spin_lock(&tbl->slot_tbl_lock);
 447         nfs4_free_slot(tbl, res->sr_slot);
 448         nfs4_check_drain_fc_complete(res->sr_session);
 449         spin_unlock(&tbl->slot_tbl_lock);
 450         res->sr_slot = NULL;
 451 }
 452
 453 static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
 454 {
 455         unsigned long timestamp;
 456         struct nfs_client *clp;
 457
 458         /*
 459          * sr_status remains 1 if an RPC level error occurred. The server
 460          * may or may not have processed the sequence operation..
 461          * Proceed as if the server received and processed the sequence
 462          * operation.
 463          */
 464         if (res->sr_status == 1)
 465                 res->sr_status = NFS_OK;
 466
 467         /* don't increment the sequence number if the task wasn't sent */
 468         if (!RPC_WAS_SENT(task))
 469                 goto out;
 470
 471         /* Check the SEQUENCE operation status */
 472         switch (res->sr_status) {
 473         case 0:
 474                 /* Update the slot's sequence and clientid lease timer */
 475                 ++res->sr_slot->seq_nr;
 476                 timestamp = res->sr_renewal_time;
 477                 clp = res->sr_session->clp;
 478                 do_renew_lease(clp, timestamp);
 479                 /* Check sequence flags */
 480                 if (res->sr_status_flags != 0)
 481                         nfs4_schedule_lease_recovery(clp);
 482                 break;
 483         case -NFS4ERR_DELAY:
 484                 /* The server detected a resend of the RPC call and
 485                  * returned NFS4ERR_DELAY as per Section 2.10.6.2
 486                  * of RFC5661.
 487                  */
 488                 dprintk("%s: slot=%td seq=%d: Operation in progress\n",
 489                         __func__,
 490                         res->sr_slot - res->sr_session->fc_slot_table.slots,
 491                         res->sr_slot->seq_nr);
 492                 goto out_retry;
 493         default:
 494                 /* Just update the slot sequence no. */
 495                 ++res->sr_slot->seq_nr;
 496         }
 497 out:
 498         /* The session may be reset by one of the error handlers. */
 499         dprintk("%s: Error %d free the slot \n", __func__, res->sr_status);
 500         nfs41_sequence_free_slot(res);
 501         return 1;
 502 out_retry:
 503         if (!rpc_restart_call(task))
 504                 goto out;
 505         rpc_delay(task, NFS4_POLL_RETRY_MAX);
 506         return 0;
 507 }
 508
 509 static int nfs4_sequence_done(struct rpc_task *task,
 510                                struct nfs4_sequence_res *res)
 511 {
 512         if (res->sr_session == NULL)
 513                 return 1;
 514         return nfs41_sequence_done(task, res);
 515 }
 516
 517 /*
 518  * nfs4_find_slot - efficiently look for a free slot
 519  *
 520  * nfs4_find_slot looks for an unset bit in the used_slots bitmap.
 521  * If found, we mark the slot as used, update the highest_used_slotid,
 522  * and respectively set up the sequence operation args.
 523  * The slot number is returned if found, or NFS4_MAX_SLOT_TABLE otherwise.
 524  *
 525  * Note: must be called with under the slot_tbl_lock.
 526  */
 527 static u8
 528 nfs4_find_slot(struct nfs4_slot_table *tbl)
 529 {
 530         int slotid;
 531         u8 ret_id = NFS4_MAX_SLOT_TABLE;
 532         BUILD_BUG_ON((u8)NFS4_MAX_SLOT_TABLE != (int)NFS4_MAX_SLOT_TABLE);
 533
 534         dprintk("--> %s used_slots=%04lx highest_used=%d max_slots=%d\n",
 535                 __func__, tbl->used_slots[0], tbl->highest_used_slotid,
 536                 tbl->max_slots);
 537         slotid = find_first_zero_bit(tbl->used_slots, tbl->max_slots);
 538         if (slotid >= tbl->max_slots)
 539                 goto out;
 540         __set_bit(slotid, tbl->used_slots);
 541         if (slotid > tbl->highest_used_slotid)
 542                 tbl->highest_used_slotid = slotid;
 543         ret_id = slotid;
 544 out:
 545         dprintk("<-- %s used_slots=%04lx highest_used=%d slotid=%d \n",
 546                 __func__, tbl->used_slots[0], tbl->highest_used_slotid, ret_id);
 547         return ret_id;
 548 }
 549
 550 int nfs41_setup_sequence(struct nfs4_session *session,
 551                                 struct nfs4_sequence_args *args,
 552                                 struct nfs4_sequence_res *res,
 553                                 int cache_reply,
 554                                 struct rpc_task *task)
 555 {
 556         struct nfs4_slot *slot;
 557         struct nfs4_slot_table *tbl;
 558         u8 slotid;
 559
 560         dprintk("--> %s\n", __func__);
 561         /* slot already allocated? */
 562         if (res->sr_slot != NULL)
 563                 return 0;
 564
 565         tbl = &session->fc_slot_table;
 566
 567         spin_lock(&tbl->slot_tbl_lock);
 568         if (test_bit(NFS4_SESSION_DRAINING, &session->session_state) &&
 569             !rpc_task_has_priority(task, RPC_PRIORITY_PRIVILEGED)) {
 570                 /*
 571                  * The state manager will wait until the slot table is empty.
 572                  * Schedule the reset thread
 573                  */
 574                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
 575                 spin_unlock(&tbl->slot_tbl_lock);
 576                 dprintk("%s Schedule Session Reset\n", __func__);
 577                 return -EAGAIN;
 578         }
 579
 580         if (!rpc_queue_empty(&tbl->slot_tbl_waitq) &&
 581             !rpc_task_has_priority(task, RPC_PRIORITY_PRIVILEGED)) {
 582                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
 583                 spin_unlock(&tbl->slot_tbl_lock);
 584                 dprintk("%s enforce FIFO order\n", __func__);
 585                 return -EAGAIN;
 586         }
 587
 588         slotid = nfs4_find_slot(tbl);
 589         if (slotid == NFS4_MAX_SLOT_TABLE) {
 590                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
 591                 spin_unlock(&tbl->slot_tbl_lock);
 592                 dprintk("<-- %s: no free slots\n", __func__);
 593                 return -EAGAIN;
 594         }
 595         spin_unlock(&tbl->slot_tbl_lock);
 596
 597         rpc_task_set_priority(task, RPC_PRIORITY_NORMAL);
 598         slot = tbl->slots + slotid;
 599         args->sa_session = session;
 600         args->sa_slotid = slotid;
 601         args->sa_cache_this = cache_reply;
 602
 603         dprintk("<-- %s slotid=%d seqid=%d\n", __func__, slotid, slot->seq_nr);
 604
 605         res->sr_session = session;
 606         res->sr_slot = slot;
 607         res->sr_renewal_time = jiffies;
 608         res->sr_status_flags = 0;
 609         /*
 610          * sr_status is only set in decode_sequence, and so will remain
 611          * set to 1 if an rpc level failure occurs.
 612          */
 613         res->sr_status = 1;
 614         return 0;
 615 }
 616 EXPORT_SYMBOL_GPL(nfs41_setup_sequence);
 617
 618 int nfs4_setup_sequence(const struct nfs_server *server,
 619                         struct nfs4_sequence_args *args,
 620                         struct nfs4_sequence_res *res,
 621                         int cache_reply,
 622                         struct rpc_task *task)
 623 {
 624         struct nfs4_session *session = nfs4_get_session(server);
 625         int ret = 0;
 626
 627         if (session == NULL) {
 628                 args->sa_session = NULL;
 629                 res->sr_session = NULL;
 630                 goto out;
 631         }
 632
 633         dprintk("--> %s clp %p session %p sr_slot %td\n",
 634                 __func__, session->clp, session, res->sr_slot ?
 635                         res->sr_slot - session->fc_slot_table.slots : -1);
 636
 637         ret = nfs41_setup_sequence(session, args, res, cache_reply,
 638                                    task);
 639 out:
 640         dprintk("<-- %s status=%d\n", __func__, ret);
 641         return ret;
 642 }
 643
 644 struct nfs41_call_sync_data {
 645         const struct nfs_server *seq_server;
 646         struct nfs4_sequence_args *seq_args;
 647         struct nfs4_sequence_res *seq_res;
 648         int cache_reply;
 649 };
 650
 651 static void nfs41_call_sync_prepare(struct rpc_task *task, void *calldata)
 652 {
 653         struct nfs41_call_sync_data *data = calldata;
 654
 655         dprintk("--> %s data->seq_server %p\n", __func__, data->seq_server);
 656
 657         if (nfs4_setup_sequence(data->seq_server, data->seq_args,
 658                                 data->seq_res, data->cache_reply, task))
 659                 return;
 660         rpc_call_start(task);
 661 }
 662
 663 static void nfs41_call_priv_sync_prepare(struct rpc_task *task, void *calldata)
 664 {
 665         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 666         nfs41_call_sync_prepare(task, calldata);
 667 }
 668
 669 static void nfs41_call_sync_done(struct rpc_task *task, void *calldata)
 670 {
 671         struct nfs41_call_sync_data *data = calldata;
 672
 673         nfs41_sequence_done(task, data->seq_res);
 674 }
 675
 676 struct rpc_call_ops nfs41_call_sync_ops = {
 677         .rpc_call_prepare = nfs41_call_sync_prepare,
 678         .rpc_call_done = nfs41_call_sync_done,
 679 };
 680
 681 struct rpc_call_ops nfs41_call_priv_sync_ops = {
 682         .rpc_call_prepare = nfs41_call_priv_sync_prepare,
 683         .rpc_call_done = nfs41_call_sync_done,
 684 };
 685
 686 static int nfs4_call_sync_sequence(struct rpc_clnt *clnt,
 687                                    struct nfs_server *server,
 688                                    struct rpc_message *msg,
 689                                    struct nfs4_sequence_args *args,
 690                                    struct nfs4_sequence_res *res,
 691                                    int cache_reply,
 692                                    int privileged)
 693 {
 694         int ret;
 695         struct rpc_task *task;
 696         struct nfs41_call_sync_data data = {
 697                 .seq_server = server,
 698                 .seq_args = args,
 699                 .seq_res = res,
 700                 .cache_reply = cache_reply,
 701         };
 702         struct rpc_task_setup task_setup = {
 703                 .rpc_client = clnt,
 704                 .rpc_message = msg,
 705                 .callback_ops = &nfs41_call_sync_ops,
 706                 .callback_data = &data
 707         };
 708
 709         res->sr_slot = NULL;
 710         if (privileged)
 711                 task_setup.callback_ops = &nfs41_call_priv_sync_ops;
 712         task = rpc_run_task(&task_setup);
 713         if (IS_ERR(task))
 714                 ret = PTR_ERR(task);
 715         else {
 716                 ret = task->tk_status;
 717                 rpc_put_task(task);
 718         }
 719         return ret;
 720 }
 721
 722 int _nfs4_call_sync_session(struct rpc_clnt *clnt,
 723                             struct nfs_server *server,
 724                             struct rpc_message *msg,
 725                             struct nfs4_sequence_args *args,
 726                             struct nfs4_sequence_res *res,
 727                             int cache_reply)
 728 {
 729         return nfs4_call_sync_sequence(clnt, server, msg, args, res, cache_reply, 0);
 730 }
 731
 732 #else
 733 static int nfs4_sequence_done(struct rpc_task *task,
 734                                struct nfs4_sequence_res *res)
 735 {
 736         return 1;
 737 }
 738 #endif /* CONFIG_NFS_V4_1 */
 739
 740 int _nfs4_call_sync(struct rpc_clnt *clnt,
 741                     struct nfs_server *server,
 742                     struct rpc_message *msg,
 743                     struct nfs4_sequence_args *args,
 744                     struct nfs4_sequence_res *res,
 745                     int cache_reply)
 746 {
 747         args->sa_session = res->sr_session = NULL;
 748         return rpc_call_sync(clnt, msg, 0);
 749 }
 750
 751 static inline
 752 int nfs4_call_sync(struct rpc_clnt *clnt,
 753                    struct nfs_server *server,
 754                    struct rpc_message *msg,
 755                    struct nfs4_sequence_args *args,
 756                    struct nfs4_sequence_res *res,
 757                    int cache_reply)
 758 {
 759         return server->nfs_client->cl_mvops->call_sync(clnt, server, msg,
 760                                                 args, res, cache_reply);
 761 }
 762
 763 static void update_changeattr(struct inode *dir, struct nfs4_change_info *cinfo)
 764 {
 765         struct nfs_inode *nfsi = NFS_I(dir);
 766
 767         spin_lock(&dir->i_lock);
 768         nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA;
 769         if (!cinfo->atomic || cinfo->before != dir->i_version)
 770                 nfs_force_lookup_revalidate(dir);
 771         dir->i_version = cinfo->after;
 772         spin_unlock(&dir->i_lock);
 773 }
 774
 775 struct nfs4_opendata {
 776         struct kref kref;
 777         struct nfs_openargs o_arg;
 778         struct nfs_openres o_res;
 779         struct nfs_open_confirmargs c_arg;
 780         struct nfs_open_confirmres c_res;
 781         struct nfs_fattr f_attr;
 782         struct nfs_fattr dir_attr;
 783         struct dentry *dir;
 784         struct dentry *dentry;
 785         struct nfs4_state_owner *owner;
 786         struct nfs4_state *state;
 787         struct iattr attrs;
 788         unsigned long timestamp;
 789         unsigned int rpc_done : 1;
 790         int rpc_status;
 791         int cancelled;
 792 };
 793
 794
 795 static void nfs4_init_opendata_res(struct nfs4_opendata *p)
 796 {
 797         p->o_res.f_attr = &p->f_attr;
 798         p->o_res.dir_attr = &p->dir_attr;
 799         p->o_res.seqid = p->o_arg.seqid;
 800         p->c_res.seqid = p->c_arg.seqid;
 801         p->o_res.server = p->o_arg.server;
 802         nfs_fattr_init(&p->f_attr);
 803         nfs_fattr_init(&p->dir_attr);
 804 }
 805
 806 static struct nfs4_opendata *nfs4_opendata_alloc(struct dentry *dentry,
 807                 struct nfs4_state_owner *sp, fmode_t fmode, int flags,
 808                 const struct iattr *attrs,
 809                 gfp_t gfp_mask)
 810 {
 811         struct dentry *parent = dget_parent(dentry);
 812         struct inode *dir = parent->d_inode;
 813         struct nfs_server *server = NFS_SERVER(dir);
 814         struct nfs4_opendata *p;
 815
 816         p = kzalloc(sizeof(*p), gfp_mask);
 817         if (p == NULL)
 818                 goto err;
 819         p->o_arg.seqid = nfs_alloc_seqid(&sp->so_seqid, gfp_mask);
 820         if (p->o_arg.seqid == NULL)
 821                 goto err_free;
 822         nfs_sb_active(dentry->d_sb);
 823         p->dentry = dget(dentry);
 824         p->dir = parent;
 825         p->owner = sp;
 826         atomic_inc(&sp->so_count);
 827         p->o_arg.fh = NFS_FH(dir);
 828         p->o_arg.open_flags = flags;
 829         p->o_arg.fmode = fmode & (FMODE_READ|FMODE_WRITE);
 830         p->o_arg.clientid = server->nfs_client->cl_clientid;
 831         p->o_arg.id = sp->so_owner_id.id;
 832         p->o_arg.name = &dentry->d_name;
 833         p->o_arg.server = server;
 834         p->o_arg.bitmask = server->attr_bitmask;
 835         p->o_arg.claim = NFS4_OPEN_CLAIM_NULL;
 836         if (flags & O_CREAT) {
 837                 u32 *s;
 838
 839                 p->o_arg.u.attrs = &p->attrs;
 840                 memcpy(&p->attrs, attrs, sizeof(p->attrs));
 841                 s = (u32 *) p->o_arg.u.verifier.data;
 842                 s[0] = jiffies;
 843                 s[1] = current->pid;
 844         }
 845         p->c_arg.fh = &p->o_res.fh;
 846         p->c_arg.stateid = &p->o_res.stateid;
 847         p->c_arg.seqid = p->o_arg.seqid;
 848         nfs4_init_opendata_res(p);
 849         kref_init(&p->kref);
 850         return p;
 851 err_free:
 852         kfree(p);
 853 err:
 854         dput(parent);
 855         return NULL;
 856 }
 857
 858 static void nfs4_opendata_free(struct kref *kref)
 859 {
 860         struct nfs4_opendata *p = container_of(kref,
 861                         struct nfs4_opendata, kref);
 862         struct super_block *sb = p->dentry->d_sb;
 863
 864         nfs_free_seqid(p->o_arg.seqid);
 865         if (p->state != NULL)
 866                 nfs4_put_open_state(p->state);
 867         nfs4_put_state_owner(p->owner);
 868         dput(p->dir);
 869         dput(p->dentry);
 870         nfs_sb_deactive(sb);
 871         kfree(p);
 872 }
 873
 874 static void nfs4_opendata_put(struct nfs4_opendata *p)
 875 {
 876         if (p != NULL)
 877                 kref_put(&p->kref, nfs4_opendata_free);
 878 }
 879
 880 static int nfs4_wait_for_completion_rpc_task(struct rpc_task *task)
 881 {
 882         int ret;
 883
 884         ret = rpc_wait_for_completion_task(task);
 885         return ret;
 886 }
 887
 888 static int can_open_cached(struct nfs4_state *state, fmode_t mode, int open_mode)
 889 {
 890         int ret = 0;
 891
 892         if (open_mode & O_EXCL)
 893                 goto out;
 894         switch (mode & (FMODE_READ|FMODE_WRITE)) {
 895                 case FMODE_READ:
 896                         ret |= test_bit(NFS_O_RDONLY_STATE, &state->flags) != 0
 897                                 && state->n_rdonly != 0;
 898                         break;
 899                 case FMODE_WRITE:
 900                         ret |= test_bit(NFS_O_WRONLY_STATE, &state->flags) != 0
 901                                 && state->n_wronly != 0;
 902                         break;
 903                 case FMODE_READ|FMODE_WRITE:
 904                         ret |= test_bit(NFS_O_RDWR_STATE, &state->flags) != 0
 905                                 && state->n_rdwr != 0;
 906         }
 907 out:
 908         return ret;
 909 }
 910
 911 static int can_open_delegated(struct nfs_delegation *delegation, fmode_t fmode)
 912 {
 913         if (delegation == NULL)
 914                 return 0;
 915         if ((delegation->type & fmode) != fmode)
 916                 return 0;
 917         if (test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags))
 918                 return 0;
 919         nfs_mark_delegation_referenced(delegation);
 920         return 1;
 921 }
 922
 923 static void update_open_stateflags(struct nfs4_state *state, fmode_t fmode)
 924 {
 925         switch (fmode) {
 926                 case FMODE_WRITE:
 927                         state->n_wronly++;
 928                         break;
 929                 case FMODE_READ:
 930                         state->n_rdonly++;
 931                         break;
 932                 case FMODE_READ|FMODE_WRITE:
 933                         state->n_rdwr++;
 934         }
 935         nfs4_state_set_mode_locked(state, state->state | fmode);
 936 }
 937
 938 static void nfs_set_open_stateid_locked(struct nfs4_state *state, nfs4_stateid *stateid, fmode_t fmode)
 939 {
 940         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
 941                 memcpy(state->stateid.data, stateid->data, sizeof(state->stateid.data));
 942         memcpy(state->open_stateid.data, stateid->data, sizeof(state->open_stateid.data));
 943         switch (fmode) {
 944                 case FMODE_READ:
 945                         set_bit(NFS_O_RDONLY_STATE, &state->flags);
 946                         break;
 947                 case FMODE_WRITE:
 948                         set_bit(NFS_O_WRONLY_STATE, &state->flags);
 949                         break;
 950                 case FMODE_READ|FMODE_WRITE:
 951                         set_bit(NFS_O_RDWR_STATE, &state->flags);
 952         }
 953 }
 954
 955 static void nfs_set_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, fmode_t fmode)
 956 {
 957         write_seqlock(&state->seqlock);
 958         nfs_set_open_stateid_locked(state, stateid, fmode);
 959         write_sequnlock(&state->seqlock);
 960 }
 961
 962 static void __update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, const nfs4_stateid *deleg_stateid, fmode_t fmode)
 963 {
 964         /*
 965          * Protect the call to nfs4_state_set_mode_locked and
 966          * serialise the stateid update
 967          */
 968         write_seqlock(&state->seqlock);
 969         if (deleg_stateid != NULL) {
 970                 memcpy(state->stateid.data, deleg_stateid->data, sizeof(state->stateid.data));
 971                 set_bit(NFS_DELEGATED_STATE, &state->flags);
 972         }
 973         if (open_stateid != NULL)
 974                 nfs_set_open_stateid_locked(state, open_stateid, fmode);
 975         write_sequnlock(&state->seqlock);
 976         spin_lock(&state->owner->so_lock);
 977         update_open_stateflags(state, fmode);
 978         spin_unlock(&state->owner->so_lock);
 979 }
 980
 981 static int update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, nfs4_stateid *delegation, fmode_t fmode)
 982 {
 983         struct nfs_inode *nfsi = NFS_I(state->inode);
 984         struct nfs_delegation *deleg_cur;
 985         int ret = 0;
 986
 987         fmode &= (FMODE_READ|FMODE_WRITE);
 988
 989         rcu_read_lock();
 990         deleg_cur = rcu_dereference(nfsi->delegation);
 991         if (deleg_cur == NULL)
 992                 goto no_delegation;
 993
 994         spin_lock(&deleg_cur->lock);
 995         if (nfsi->delegation != deleg_cur ||
 996             (deleg_cur->type & fmode) != fmode)
 997                 goto no_delegation_unlock;
 998
 999         if (delegation == NULL)
1000                 delegation = &deleg_cur->stateid;
1001         else if (memcmp(deleg_cur->stateid.data, delegation->data, NFS4_STATEID_SIZE) != 0)
1002                 goto no_delegation_unlock;
1003
1004         nfs_mark_delegation_referenced(deleg_cur);
1005         __update_open_stateid(state, open_stateid, &deleg_cur->stateid, fmode);
1006         ret = 1;
1007 no_delegation_unlock:
1008         spin_unlock(&deleg_cur->lock);
1009 no_delegation:
1010         rcu_read_unlock();
1011
1012         if (!ret && open_stateid != NULL) {
1013                 __update_open_stateid(state, open_stateid, NULL, fmode);
1014                 ret = 1;
1015         }
1016
1017         return ret;
1018 }
1019
1020
1021 static void nfs4_return_incompatible_delegation(struct inode *inode, fmode_t fmode)
1022 {
1023         struct nfs_delegation *delegation;
1024
1025         rcu_read_lock();
1026         delegation = rcu_dereference(NFS_I(inode)->delegation);
1027         if (delegation == NULL || (delegation->type & fmode) == fmode) {
1028                 rcu_read_unlock();
1029                 return;
1030         }
1031         rcu_read_unlock();
1032         nfs_inode_return_delegation(inode);
1033 }
1034
1035 static struct nfs4_state *nfs4_try_open_cached(struct nfs4_opendata *opendata)
1036 {
1037         struct nfs4_state *state = opendata->state;
1038         struct nfs_inode *nfsi = NFS_I(state->inode);
1039         struct nfs_delegation *delegation;
1040         int open_mode = opendata->o_arg.open_flags & O_EXCL;
1041         fmode_t fmode = opendata->o_arg.fmode;
1042         nfs4_stateid stateid;
1043         int ret = -EAGAIN;
1044
1045         for (;;) {
1046                 if (can_open_cached(state, fmode, open_mode)) {
1047                         spin_lock(&state->owner->so_lock);
1048                         if (can_open_cached(state, fmode, open_mode)) {
1049                                 update_open_stateflags(state, fmode);
1050                                 spin_unlock(&state->owner->so_lock);
1051                                 goto out_return_state;
1052                         }
1053                         spin_unlock(&state->owner->so_lock);
1054                 }
1055                 rcu_read_lock();
1056                 delegation = rcu_dereference(nfsi->delegation);
1057                 if (!can_open_delegated(delegation, fmode)) {
1058                         rcu_read_unlock();
1059                         break;
1060                 }
1061                 /* Save the delegation */
1062                 memcpy(stateid.data, delegation->stateid.data, sizeof(stateid.data));
1063                 rcu_read_unlock();
1064                 ret = nfs_may_open(state->inode, state->owner->so_cred, open_mode);
1065                 if (ret != 0)
1066                         goto out;
1067                 ret = -EAGAIN;
1068
1069                 /* Try to update the stateid using the delegation */
1070                 if (update_open_stateid(state, NULL, &stateid, fmode))
1071                         goto out_return_state;
1072         }
1073 out:
1074         return ERR_PTR(ret);
1075 out_return_state:
1076         atomic_inc(&state->count);
1077         return state;
1078 }
1079
1080 static struct nfs4_state *nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data)
1081 {
1082         struct inode *inode;
1083         struct nfs4_state *state = NULL;
1084         struct nfs_delegation *delegation;
1085         int ret;
1086
1087         if (!data->rpc_done) {
1088                 state = nfs4_try_open_cached(data);
1089                 goto out;
1090         }
1091
1092         ret = -EAGAIN;
1093         if (!(data->f_attr.valid & NFS_ATTR_FATTR))
1094                 goto err;
1095         inode = nfs_fhget(data->dir->d_sb, &data->o_res.fh, &data->f_attr);
1096         ret = PTR_ERR(inode);
1097         if (IS_ERR(inode))
1098                 goto err;
1099         ret = -ENOMEM;
1100         state = nfs4_get_open_state(inode, data->owner);
1101         if (state == NULL)
1102                 goto err_put_inode;
1103         if (data->o_res.delegation_type != 0) {
1104                 int delegation_flags = 0;
1105
1106                 rcu_read_lock();
1107                 delegation = rcu_dereference(NFS_I(inode)->delegation);
1108                 if (delegation)
1109                         delegation_flags = delegation->flags;
1110                 rcu_read_unlock();
1111                 if (data->o_arg.claim == NFS4_OPEN_CLAIM_DELEGATE_CUR) {
1112                         pr_err_ratelimited("NFS: Broken NFSv4 server %s is "
1113                                         "returning a delegation for "
1114                                         "OPEN(CLAIM_DELEGATE_CUR)\n",
1115                                         NFS_CLIENT(inode)->cl_server);
1116                 } else if ((delegation_flags & 1UL<<NFS_DELEGATION_NEED_RECLAIM) == 0)
1117                         nfs_inode_set_delegation(state->inode,
1118                                         data->owner->so_cred,
1119                                         &data->o_res);
1120                 else
1121                         nfs_inode_reclaim_delegation(state->inode,
1122                                         data->owner->so_cred,
1123                                         &data->o_res);
1124         }
1125
1126         update_open_stateid(state, &data->o_res.stateid, NULL,
1127                         data->o_arg.fmode);
1128         iput(inode);
1129 out:
1130         return state;
1131 err_put_inode:
1132         iput(inode);
1133 err:
1134         return ERR_PTR(ret);
1135 }
1136
1137 static struct nfs_open_context *nfs4_state_find_open_context(struct nfs4_state *state)
1138 {
1139         struct nfs_inode *nfsi = NFS_I(state->inode);
1140         struct nfs_open_context *ctx;
1141
1142         spin_lock(&state->inode->i_lock);
1143         list_for_each_entry(ctx, &nfsi->open_files, list) {
1144                 if (ctx->state != state)
1145                         continue;
1146                 get_nfs_open_context(ctx);
1147                 spin_unlock(&state->inode->i_lock);
1148                 return ctx;
1149         }
1150         spin_unlock(&state->inode->i_lock);
1151         return ERR_PTR(-ENOENT);
1152 }
1153
1154 static struct nfs4_opendata *nfs4_open_recoverdata_alloc(struct nfs_open_context *ctx, struct nfs4_state *state)
1155 {
1156         struct nfs4_opendata *opendata;
1157
1158         opendata = nfs4_opendata_alloc(ctx->dentry, state->owner, 0, 0, NULL, GFP_NOFS);
1159         if (opendata == NULL)
1160                 return ERR_PTR(-ENOMEM);
1161         opendata->state = state;
1162         atomic_inc(&state->count);
1163         return opendata;
1164 }
1165
1166 static int nfs4_open_recover_helper(struct nfs4_opendata *opendata, fmode_t fmode, struct nfs4_state **res)
1167 {
1168         struct nfs4_state *newstate;
1169         int ret;
1170
1171         opendata->o_arg.open_flags = 0;
1172         opendata->o_arg.fmode = fmode;
1173         memset(&opendata->o_res, 0, sizeof(opendata->o_res));
1174         memset(&opendata->c_res, 0, sizeof(opendata->c_res));
1175         nfs4_init_opendata_res(opendata);
1176         ret = _nfs4_recover_proc_open(opendata);
1177         if (ret != 0)
1178                 return ret;
1179         newstate = nfs4_opendata_to_nfs4_state(opendata);
1180         if (IS_ERR(newstate))
1181                 return PTR_ERR(newstate);
1182         nfs4_close_state(newstate, fmode);
1183         *res = newstate;
1184         return 0;
1185 }
1186
1187 static int nfs4_open_recover(struct nfs4_opendata *opendata, struct nfs4_state *state)
1188 {
1189         struct nfs4_state *newstate;
1190         int ret;
1191
1192         /* memory barrier prior to reading state->n_* */
1193         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1194         smp_rmb();
1195         if (state->n_rdwr != 0) {
1196                 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1197                 ret = nfs4_open_recover_helper(opendata, FMODE_READ|FMODE_WRITE, &newstate);
1198                 if (ret != 0)
1199                         return ret;
1200                 if (newstate != state)
1201                         return -ESTALE;
1202         }
1203         if (state->n_wronly != 0) {
1204                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1205                 ret = nfs4_open_recover_helper(opendata, FMODE_WRITE, &newstate);
1206                 if (ret != 0)
1207                         return ret;
1208                 if (newstate != state)
1209                         return -ESTALE;
1210         }
1211         if (state->n_rdonly != 0) {
1212                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1213                 ret = nfs4_open_recover_helper(opendata, FMODE_READ, &newstate);
1214                 if (ret != 0)
1215                         return ret;
1216                 if (newstate != state)
1217                         return -ESTALE;
1218         }
1219         /*
1220          * We may have performed cached opens for all three recoveries.
1221          * Check if we need to update the current stateid.
1222          */
1223         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0 &&
1224             memcmp(state->stateid.data, state->open_stateid.data, sizeof(state->stateid.data)) != 0) {
1225                 write_seqlock(&state->seqlock);
1226                 if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1227                         memcpy(state->stateid.data, state->open_stateid.data, sizeof(state->stateid.data));
1228                 write_sequnlock(&state->seqlock);
1229         }
1230         return 0;
1231 }
1232
1233 /*
1234  * OPEN_RECLAIM:
1235  *      reclaim state on the server after a reboot.
1236  */
1237 static int _nfs4_do_open_reclaim(struct nfs_open_context *ctx, struct nfs4_state *state)
1238 {
1239         struct nfs_delegation *delegation;
1240         struct nfs4_opendata *opendata;
1241         fmode_t delegation_type = 0;
1242         int status;
1243
1244         opendata = nfs4_open_recoverdata_alloc(ctx, state);
1245         if (IS_ERR(opendata))
1246                 return PTR_ERR(opendata);
1247         opendata->o_arg.claim = NFS4_OPEN_CLAIM_PREVIOUS;
1248         opendata->o_arg.fh = NFS_FH(state->inode);
1249         rcu_read_lock();
1250         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
1251         if (delegation != NULL && test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) != 0)
1252                 delegation_type = delegation->type;
1253         rcu_read_unlock();
1254         opendata->o_arg.u.delegation_type = delegation_type;
1255         status = nfs4_open_recover(opendata, state);
1256         nfs4_opendata_put(opendata);
1257         return status;
1258 }
1259
1260 static int nfs4_do_open_reclaim(struct nfs_open_context *ctx, struct nfs4_state *state)
1261 {
1262         struct nfs_server *server = NFS_SERVER(state->inode);
1263         struct nfs4_exception exception = { };
1264         int err;
1265         do {
1266                 err = _nfs4_do_open_reclaim(ctx, state);
1267                 if (err != -NFS4ERR_DELAY)
1268                         break;
1269                 nfs4_handle_exception(server, err, &exception);
1270         } while (exception.retry);
1271         return err;
1272 }
1273
1274 static int nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *state)
1275 {
1276         struct nfs_open_context *ctx;
1277         int ret;
1278
1279         ctx = nfs4_state_find_open_context(state);
1280         if (IS_ERR(ctx))
1281                 return PTR_ERR(ctx);
1282         ret = nfs4_do_open_reclaim(ctx, state);
1283         put_nfs_open_context(ctx);
1284         return ret;
1285 }
1286
1287 static int _nfs4_open_delegation_recall(struct nfs_open_context *ctx, struct nfs4_state *state, const nfs4_stateid *stateid)
1288 {
1289         struct nfs4_opendata *opendata;
1290         int ret;
1291
1292         opendata = nfs4_open_recoverdata_alloc(ctx, state);
1293         if (IS_ERR(opendata))
1294                 return PTR_ERR(opendata);
1295         opendata->o_arg.claim = NFS4_OPEN_CLAIM_DELEGATE_CUR;
1296         memcpy(opendata->o_arg.u.delegation.data, stateid->data,
1297                         sizeof(opendata->o_arg.u.delegation.data));
1298         ret = nfs4_open_recover(opendata, state);
1299         nfs4_opendata_put(opendata);
1300         return ret;
1301 }
1302
1303 int nfs4_open_delegation_recall(struct nfs_open_context *ctx, struct nfs4_state *state, const nfs4_stateid *stateid)
1304 {
1305         struct nfs4_exception exception = { };
1306         struct nfs_server *server = NFS_SERVER(state->inode);
1307         int err;
1308         do {
1309                 err = _nfs4_open_delegation_recall(ctx, state, stateid);
1310                 switch (err) {
1311                         case 0:
1312                         case -ENOENT:
1313                         case -ESTALE:
1314                                 goto out;
1315                         case -NFS4ERR_BADSESSION:
1316                         case -NFS4ERR_BADSLOT:
1317                         case -NFS4ERR_BAD_HIGH_SLOT:
1318                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1319                         case -NFS4ERR_DEADSESSION:
1320                                 nfs4_schedule_session_recovery(server->nfs_client->cl_session);
1321                                 goto out;
1322                         case -NFS4ERR_STALE_CLIENTID:
1323                         case -NFS4ERR_STALE_STATEID:
1324                         case -NFS4ERR_EXPIRED:
1325                                 /* Don't recall a delegation if it was lost */
1326                                 nfs4_schedule_lease_recovery(server->nfs_client);
1327                                 goto out;
1328                         case -ERESTARTSYS:
1329                                 /*
1330                                  * The show must go on: exit, but mark the
1331                                  * stateid as needing recovery.
1332                                  */
1333                         case -NFS4ERR_DELEG_REVOKED:
1334                         case -NFS4ERR_ADMIN_REVOKED:
1335                         case -NFS4ERR_BAD_STATEID:
1336                                 nfs_inode_find_state_and_recover(state->inode,
1337                                                 stateid);
1338                                 nfs4_schedule_stateid_recovery(server, state);
1339                         case -EKEYEXPIRED:
1340                                 /*
1341                                  * User RPCSEC_GSS context has expired.
1342                                  * We cannot recover this stateid now, so
1343                                  * skip it and allow recovery thread to
1344                                  * proceed.
1345                                  */
1346                         case -ENOMEM:
1347                                 err = 0;
1348                                 goto out;
1349                 }
1350                 err = nfs4_handle_exception(server, err, &exception);
1351         } while (exception.retry);
1352 out:
1353         return err;
1354 }
1355
1356 static void nfs4_open_confirm_done(struct rpc_task *task, void *calldata)
1357 {
1358         struct nfs4_opendata *data = calldata;
1359
1360         data->rpc_status = task->tk_status;
1361         if (data->rpc_status == 0) {
1362                 memcpy(data->o_res.stateid.data, data->c_res.stateid.data,
1363                                 sizeof(data->o_res.stateid.data));
1364                 nfs_confirm_seqid(&data->owner->so_seqid, 0);
1365                 renew_lease(data->o_res.server, data->timestamp);
1366                 data->rpc_done = 1;
1367         }
1368 }
1369
1370 static void nfs4_open_confirm_release(void *calldata)
1371 {
1372         struct nfs4_opendata *data = calldata;
1373         struct nfs4_state *state = NULL;
1374
1375         /* If this request hasn't been cancelled, do nothing */
1376         if (data->cancelled == 0)
1377                 goto out_free;
1378         /* In case of error, no cleanup! */
1379         if (!data->rpc_done)
1380                 goto out_free;
1381         state = nfs4_opendata_to_nfs4_state(data);
1382         if (!IS_ERR(state))
1383                 nfs4_close_state(state, data->o_arg.fmode);
1384 out_free:
1385         nfs4_opendata_put(data);
1386 }
1387
1388 static const struct rpc_call_ops nfs4_open_confirm_ops = {
1389         .rpc_call_done = nfs4_open_confirm_done,
1390         .rpc_release = nfs4_open_confirm_release,
1391 };
1392
1393 /*
1394  * Note: On error, nfs4_proc_open_confirm will free the struct nfs4_opendata
1395  */
1396 static int _nfs4_proc_open_confirm(struct nfs4_opendata *data)
1397 {
1398         struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
1399         struct rpc_task *task;
1400         struct  rpc_message msg = {
1401                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_CONFIRM],
1402                 .rpc_argp = &data->c_arg,
1403                 .rpc_resp = &data->c_res,
1404                 .rpc_cred = data->owner->so_cred,
1405         };
1406         struct rpc_task_setup task_setup_data = {
1407                 .rpc_client = server->client,
1408                 .rpc_message = &msg,
1409                 .callback_ops = &nfs4_open_confirm_ops,
1410                 .callback_data = data,
1411                 .workqueue = nfsiod_workqueue,
1412                 .flags = RPC_TASK_ASYNC,
1413         };
1414         int status;
1415
1416         kref_get(&data->kref);
1417         data->rpc_done = 0;
1418         data->rpc_status = 0;
1419         data->timestamp = jiffies;
1420         task = rpc_run_task(&task_setup_data);
1421         if (IS_ERR(task))
1422                 return PTR_ERR(task);
1423         status = nfs4_wait_for_completion_rpc_task(task);
1424         if (status != 0) {
1425                 data->cancelled = 1;
1426                 smp_wmb();
1427         } else
1428                 status = data->rpc_status;
1429         rpc_put_task(task);
1430         return status;
1431 }
1432
1433 static void nfs4_open_prepare(struct rpc_task *task, void *calldata)
1434 {
1435         struct nfs4_opendata *data = calldata;
1436         struct nfs4_state_owner *sp = data->owner;
1437
1438         if (nfs_wait_on_sequence(data->o_arg.seqid, task) != 0)
1439                 return;
1440         /*
1441          * Check if we still need to send an OPEN call, or if we can use
1442          * a delegation instead.
1443          */
1444         if (data->state != NULL) {
1445                 struct nfs_delegation *delegation;
1446
1447                 if (can_open_cached(data->state, data->o_arg.fmode, data->o_arg.open_flags))
1448                         goto out_no_action;
1449                 rcu_read_lock();
1450                 delegation = rcu_dereference(NFS_I(data->state->inode)->delegation);
1451                 if (data->o_arg.claim != NFS4_OPEN_CLAIM_DELEGATE_CUR &&
1452                     can_open_delegated(delegation, data->o_arg.fmode))
1453                         goto unlock_no_action;
1454                 rcu_read_unlock();
1455         }
1456         /* Update sequence id. */
1457         data->o_arg.id = sp->so_owner_id.id;
1458         data->o_arg.clientid = sp->so_server->nfs_client->cl_clientid;
1459         if (data->o_arg.claim == NFS4_OPEN_CLAIM_PREVIOUS) {
1460                 task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_NOATTR];
1461                 nfs_copy_fh(&data->o_res.fh, data->o_arg.fh);
1462         }
1463         data->timestamp = jiffies;
1464         if (nfs4_setup_sequence(data->o_arg.server,
1465                                 &data->o_arg.seq_args,
1466                                 &data->o_res.seq_res,
1467                                 1, task) != 0)
1468                 nfs_release_seqid(data->o_arg.seqid);
1469         else
1470                 rpc_call_start(task);
1471         return;
1472 unlock_no_action:
1473         rcu_read_unlock();
1474 out_no_action:
1475         task->tk_action = NULL;
1476
1477 }
1478
1479 static void nfs4_recover_open_prepare(struct rpc_task *task, void *calldata)
1480 {
1481         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
1482         nfs4_open_prepare(task, calldata);
1483 }
1484
1485 static void nfs4_open_done(struct rpc_task *task, void *calldata)
1486 {
1487         struct nfs4_opendata *data = calldata;
1488
1489         data->rpc_status = task->tk_status;
1490
1491         if (!nfs4_sequence_done(task, &data->o_res.seq_res))
1492                 return;
1493
1494         if (task->tk_status == 0) {
1495                 switch (data->o_res.f_attr->mode & S_IFMT) {
1496                         case S_IFREG:
1497                                 break;
1498                         case S_IFLNK:
1499                                 data->rpc_status = -ELOOP;
1500                                 break;
1501                         case S_IFDIR:
1502                                 data->rpc_status = -EISDIR;
1503                                 break;
1504                         default:
1505                                 data->rpc_status = -ENOTDIR;
1506                 }
1507                 renew_lease(data->o_res.server, data->timestamp);
1508                 if (!(data->o_res.rflags & NFS4_OPEN_RESULT_CONFIRM))
1509                         nfs_confirm_seqid(&data->owner->so_seqid, 0);
1510         }
1511         data->rpc_done = 1;
1512 }
1513
1514 static void nfs4_open_release(void *calldata)
1515 {
1516         struct nfs4_opendata *data = calldata;
1517         struct nfs4_state *state = NULL;
1518
1519         /* If this request hasn't been cancelled, do nothing */
1520         if (data->cancelled == 0)
1521                 goto out_free;
1522         /* In case of error, no cleanup! */
1523         if (data->rpc_status != 0 || !data->rpc_done)
1524                 goto out_free;
1525         /* In case we need an open_confirm, no cleanup! */
1526         if (data->o_res.rflags & NFS4_OPEN_RESULT_CONFIRM)
1527                 goto out_free;
1528         state = nfs4_opendata_to_nfs4_state(data);
1529         if (!IS_ERR(state))
1530                 nfs4_close_state(state, data->o_arg.fmode);
1531 out_free:
1532         nfs4_opendata_put(data);
1533 }
1534
1535 static const struct rpc_call_ops nfs4_open_ops = {
1536         .rpc_call_prepare = nfs4_open_prepare,
1537         .rpc_call_done = nfs4_open_done,
1538         .rpc_release = nfs4_open_release,
1539 };
1540
1541 static const struct rpc_call_ops nfs4_recover_open_ops = {
1542         .rpc_call_prepare = nfs4_recover_open_prepare,
1543         .rpc_call_done = nfs4_open_done,
1544         .rpc_release = nfs4_open_release,
1545 };
1546
1547 static int nfs4_run_open_task(struct nfs4_opendata *data, int isrecover)
1548 {
1549         struct inode *dir = data->dir->d_inode;
1550         struct nfs_server *server = NFS_SERVER(dir);
1551         struct nfs_openargs *o_arg = &data->o_arg;
1552         struct nfs_openres *o_res = &data->o_res;
1553         struct rpc_task *task;
1554         struct rpc_message msg = {
1555                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN],
1556                 .rpc_argp = o_arg,
1557                 .rpc_resp = o_res,
1558                 .rpc_cred = data->owner->so_cred,
1559         };
1560         struct rpc_task_setup task_setup_data = {
1561                 .rpc_client = server->client,
1562                 .rpc_message = &msg,
1563                 .callback_ops = &nfs4_open_ops,
1564                 .callback_data = data,
1565                 .workqueue = nfsiod_workqueue,
1566                 .flags = RPC_TASK_ASYNC,
1567         };
1568         int status;
1569
1570         kref_get(&data->kref);
1571         data->rpc_done = 0;
1572         data->rpc_status = 0;
1573         data->cancelled = 0;
1574         if (isrecover)
1575                 task_setup_data.callback_ops = &nfs4_recover_open_ops;
1576         task = rpc_run_task(&task_setup_data);
1577         if (IS_ERR(task))
1578                 return PTR_ERR(task);
1579         status = nfs4_wait_for_completion_rpc_task(task);
1580         if (status != 0) {
1581                 data->cancelled = 1;
1582                 smp_wmb();
1583         } else
1584                 status = data->rpc_status;
1585         rpc_put_task(task);
1586
1587         return status;
1588 }
1589
1590 static int _nfs4_recover_proc_open(struct nfs4_opendata *data)
1591 {
1592         struct inode *dir = data->dir->d_inode;
1593         struct nfs_openres *o_res = &data->o_res;
1594         int status;
1595
1596         status = nfs4_run_open_task(data, 1);
1597         if (status != 0 || !data->rpc_done)
1598                 return status;
1599
1600         nfs_refresh_inode(dir, o_res->dir_attr);
1601
1602         if (o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
1603                 status = _nfs4_proc_open_confirm(data);
1604                 if (status != 0)
1605                         return status;
1606         }
1607
1608         return status;
1609 }
1610
1611 /*
1612  * Note: On error, nfs4_proc_open will free the struct nfs4_opendata
1613  */
1614 static int _nfs4_proc_open(struct nfs4_opendata *data)
1615 {
1616         struct inode *dir = data->dir->d_inode;
1617         struct nfs_server *server = NFS_SERVER(dir);
1618         struct nfs_openargs *o_arg = &data->o_arg;
1619         struct nfs_openres *o_res = &data->o_res;
1620         int status;
1621
1622         status = nfs4_run_open_task(data, 0);
1623         if (!data->rpc_done)
1624                 return status;
1625         if (status != 0) {
1626                 if (status == -NFS4ERR_BADNAME &&
1627                                 !(o_arg->open_flags & O_CREAT))
1628                         return -ENOENT;
1629                 return status;
1630         }
1631
1632         if (o_arg->open_flags & O_CREAT) {
1633                 update_changeattr(dir, &o_res->cinfo);
1634                 nfs_post_op_update_inode(dir, o_res->dir_attr);
1635         } else
1636                 nfs_refresh_inode(dir, o_res->dir_attr);
1637         if ((o_res->rflags & NFS4_OPEN_RESULT_LOCKTYPE_POSIX) == 0)
1638                 server->caps &= ~NFS_CAP_POSIX_LOCK;
1639         if(o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
1640                 status = _nfs4_proc_open_confirm(data);
1641                 if (status != 0)
1642                         return status;
1643         }
1644         if (!(o_res->f_attr->valid & NFS_ATTR_FATTR))
1645                 _nfs4_proc_getattr(server, &o_res->fh, o_res->f_attr);
1646         return 0;
1647 }
1648
1649 static int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1650 {
1651         unsigned int loop;
1652         int ret;
1653
1654         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1655                 ret = nfs4_wait_clnt_recover(clp);
1656                 if (ret != 0)
1657                         break;
1658                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1659                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1660                         break;
1661                 nfs4_schedule_state_manager(clp);
1662                 ret = -EIO;
1663         }
1664         return ret;
1665 }
1666
1667 static int nfs4_recover_expired_lease(struct nfs_server *server)
1668 {
1669         return nfs4_client_recover_expired_lease(server->nfs_client);
1670 }
1671
1672 /*
1673  * OPEN_EXPIRED:
1674  *      reclaim state on the server after a network partition.
1675  *      Assumes caller holds the appropriate lock
1676  */
1677 static int _nfs4_open_expired(struct nfs_open_context *ctx, struct nfs4_state *state)
1678 {
1679         struct nfs4_opendata *opendata;
1680         int ret;
1681
1682         opendata = nfs4_open_recoverdata_alloc(ctx, state);
1683         if (IS_ERR(opendata))
1684                 return PTR_ERR(opendata);
1685         ret = nfs4_open_recover(opendata, state);
1686         if (ret == -ESTALE)
1687                 d_drop(ctx->dentry);
1688         nfs4_opendata_put(opendata);
1689         return ret;
1690 }
1691
1692 static int nfs4_do_open_expired(struct nfs_open_context *ctx, struct nfs4_state *state)
1693 {
1694         struct nfs_server *server = NFS_SERVER(state->inode);
1695         struct nfs4_exception exception = { };
1696         int err;
1697
1698         do {
1699                 err = _nfs4_open_expired(ctx, state);
1700                 switch (err) {
1701                 default:
1702                         goto out;
1703                 case -NFS4ERR_GRACE:
1704                 case -NFS4ERR_DELAY:
1705                         nfs4_handle_exception(server, err, &exception);
1706                         err = 0;
1707                 }
1708         } while (exception.retry);
1709 out:
1710         return err;
1711 }
1712
1713 static int nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
1714 {
1715         struct nfs_open_context *ctx;
1716         int ret;
1717
1718         ctx = nfs4_state_find_open_context(state);
1719         if (IS_ERR(ctx))
1720                 return PTR_ERR(ctx);
1721         ret = nfs4_do_open_expired(ctx, state);
1722         put_nfs_open_context(ctx);
1723         return ret;
1724 }
1725
1726 #if defined(CONFIG_NFS_V4_1)
1727 static int nfs41_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
1728 {
1729         int status;
1730         struct nfs_server *server = NFS_SERVER(state->inode);
1731
1732         status = nfs41_test_stateid(server, state);
1733         if (status == NFS_OK)
1734                 return 0;
1735         nfs41_free_stateid(server, state);
1736         return nfs4_open_expired(sp, state);
1737 }
1738 #endif
1739
1740 /*
1741  * on an EXCLUSIVE create, the server should send back a bitmask with FATTR4-*
1742  * fields corresponding to attributes that were used to store the verifier.
1743  * Make sure we clobber those fields in the later setattr call
1744  */
1745 static inline void nfs4_exclusive_attrset(struct nfs4_opendata *opendata, struct iattr *sattr)
1746 {
1747         if ((opendata->o_res.attrset[1] & FATTR4_WORD1_TIME_ACCESS) &&
1748             !(sattr->ia_valid & ATTR_ATIME_SET))
1749                 sattr->ia_valid |= ATTR_ATIME;
1750
1751         if ((opendata->o_res.attrset[1] & FATTR4_WORD1_TIME_MODIFY) &&
1752             !(sattr->ia_valid & ATTR_MTIME_SET))
1753                 sattr->ia_valid |= ATTR_MTIME;
1754 }
1755
1756 /*
1757  * Returns a referenced nfs4_state
1758  */
1759 static int _nfs4_do_open(struct inode *dir, struct dentry *dentry, fmode_t fmode, int flags, struct iattr *sattr, struct rpc_cred *cred, struct nfs4_state **res)
1760 {
1761         struct nfs4_state_owner  *sp;
1762         struct nfs4_state     *state = NULL;
1763         struct nfs_server       *server = NFS_SERVER(dir);
1764         struct nfs4_opendata *opendata;
1765         int status;
1766
1767         /* Protect against reboot recovery conflicts */
1768         status = -ENOMEM;
1769         if (!(sp = nfs4_get_state_owner(server, cred))) {
1770                 dprintk("nfs4_do_open: nfs4_get_state_owner failed!\n");
1771                 goto out_err;
1772         }
1773         status = nfs4_recover_expired_lease(server);
1774         if (status != 0)
1775                 goto err_put_state_owner;
1776         if (dentry->d_inode != NULL)
1777                 nfs4_return_incompatible_delegation(dentry->d_inode, fmode);
1778         status = -ENOMEM;
1779         opendata = nfs4_opendata_alloc(dentry, sp, fmode, flags, sattr, GFP_KERNEL);
1780         if (opendata == NULL)
1781                 goto err_put_state_owner;
1782
1783         if (dentry->d_inode != NULL)
1784                 opendata->state = nfs4_get_open_state(dentry->d_inode, sp);
1785
1786         status = _nfs4_proc_open(opendata);
1787         if (status != 0)
1788                 goto err_opendata_put;
1789
1790         state = nfs4_opendata_to_nfs4_state(opendata);
1791         status = PTR_ERR(state);
1792         if (IS_ERR(state))
1793                 goto err_opendata_put;
1794         if (server->caps & NFS_CAP_POSIX_LOCK)
1795                 set_bit(NFS_STATE_POSIX_LOCKS, &state->flags);
1796
1797         if (opendata->o_arg.open_flags & O_EXCL) {
1798                 nfs4_exclusive_attrset(opendata, sattr);
1799
1800                 nfs_fattr_init(opendata->o_res.f_attr);
1801                 status = nfs4_do_setattr(state->inode, cred,
1802                                 opendata->o_res.f_attr, sattr,
1803                                 state);
1804                 if (status == 0)
1805                         nfs_setattr_update_inode(state->inode, sattr);
1806                 nfs_post_op_update_inode(state->inode, opendata->o_res.f_attr);
1807         }
1808         nfs_revalidate_inode(server, state->inode);
1809         nfs4_opendata_put(opendata);
1810         nfs4_put_state_owner(sp);
1811         *res = state;
1812         return 0;
1813 err_opendata_put:
1814         nfs4_opendata_put(opendata);
1815 err_put_state_owner:
1816         nfs4_put_state_owner(sp);
1817 out_err:
1818         *res = NULL;
1819         return status;
1820 }
1821
1822
1823 static struct nfs4_state *nfs4_do_open(struct inode *dir, struct dentry *dentry, fmode_t fmode, int flags, struct iattr *sattr, struct rpc_cred *cred)
1824 {
1825         struct nfs4_exception exception = { };
1826         struct nfs4_state *res;
1827         int status;
1828
1829         fmode &= FMODE_READ|FMODE_WRITE;
1830         do {
1831                 status = _nfs4_do_open(dir, dentry, fmode, flags, sattr, cred, &res);
1832                 if (status == 0)
1833                         break;
1834                 /* NOTE: BAD_SEQID means the server and client disagree about the
1835                  * book-keeping w.r.t. state-changing operations
1836                  * (OPEN/CLOSE/LOCK/LOCKU...)
1837                  * It is actually a sign of a bug on the client or on the server.
1838                  *
1839                  * If we receive a BAD_SEQID error in the particular case of
1840                  * doing an OPEN, we assume that nfs_increment_open_seqid() will
1841                  * have unhashed the old state_owner for us, and that we can
1842                  * therefore safely retry using a new one. We should still warn
1843                  * the user though...
1844                  */
1845                 if (status == -NFS4ERR_BAD_SEQID) {
1846                         pr_warn_ratelimited("NFS: v4 server %s "
1847                                         " returned a bad sequence-id error!\n",
1848                                         NFS_SERVER(dir)->nfs_client->cl_hostname);
1849                         exception.retry = 1;
1850                         continue;
1851                 }
1852                 /*
1853                  * BAD_STATEID on OPEN means that the server cancelled our
1854                  * state before it received the OPEN_CONFIRM.
1855                  * Recover by retrying the request as per the discussion
1856                  * on Page 181 of RFC3530.
1857                  */
1858                 if (status == -NFS4ERR_BAD_STATEID) {
1859                         exception.retry = 1;
1860                         continue;
1861                 }
1862                 if (status == -EAGAIN) {
1863                         /* We must have found a delegation */
1864                         exception.retry = 1;
1865                         continue;
1866                 }
1867                 res = ERR_PTR(nfs4_handle_exception(NFS_SERVER(dir),
1868                                         status, &exception));
1869         } while (exception.retry);
1870         return res;
1871 }
1872
1873 static int _nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
1874                             struct nfs_fattr *fattr, struct iattr *sattr,
1875                             struct nfs4_state *state)
1876 {
1877         struct nfs_server *server = NFS_SERVER(inode);
1878         struct nfs_setattrargs  arg = {
1879                 .fh             = NFS_FH(inode),
1880                 .iap            = sattr,
1881                 .server         = server,
1882                 .bitmask = server->attr_bitmask,
1883         };
1884         struct nfs_setattrres  res = {
1885                 .fattr          = fattr,
1886                 .server         = server,
1887         };
1888         struct rpc_message msg = {
1889                 .rpc_proc       = &nfs4_procedures[NFSPROC4_CLNT_SETATTR],
1890                 .rpc_argp       = &arg,
1891                 .rpc_resp       = &res,
1892                 .rpc_cred       = cred,
1893         };
1894         unsigned long timestamp = jiffies;
1895         int status;
1896
1897         nfs_fattr_init(fattr);
1898
1899         if (nfs4_copy_delegation_stateid(&arg.stateid, inode)) {
1900                 /* Use that stateid */
1901         } else if (state != NULL) {
1902                 nfs4_copy_stateid(&arg.stateid, state, current->files, current->tgid);
1903         } else
1904                 memcpy(&arg.stateid, &zero_stateid, sizeof(arg.stateid));
1905
1906         status = nfs4_call_sync(server->client, server, &msg, &arg.seq_args, &res.seq_res, 1);
1907         if (status == 0 && state != NULL)
1908                 renew_lease(server, timestamp);
1909         return status;
1910 }
1911
1912 static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
1913                            struct nfs_fattr *fattr, struct iattr *sattr,
1914                            struct nfs4_state *state)
1915 {
1916         struct nfs_server *server = NFS_SERVER(inode);
1917         struct nfs4_exception exception = {
1918                 .state = state,
1919                 .inode = inode,
1920         };
1921         int err;
1922         do {
1923                 err = nfs4_handle_exception(server,
1924                                 _nfs4_do_setattr(inode, cred, fattr, sattr, state),
1925                                 &exception);
1926         } while (exception.retry);
1927         return err;
1928 }
1929
1930 struct nfs4_closedata {
1931         struct inode *inode;
1932         struct nfs4_state *state;
1933         struct nfs_closeargs arg;
1934         struct nfs_closeres res;
1935         struct nfs_fattr fattr;
1936         unsigned long timestamp;
1937         bool roc;
1938         u32 roc_barrier;
1939 };
1940
1941 static void nfs4_free_closedata(void *data)
1942 {
1943         struct nfs4_closedata *calldata = data;
1944         struct nfs4_state_owner *sp = calldata->state->owner;
1945         struct super_block *sb = calldata->state->inode->i_sb;
1946
1947         if (calldata->roc)
1948                 pnfs_roc_release(calldata->state->inode);
1949         nfs4_put_open_state(calldata->state);
1950         nfs_free_seqid(calldata->arg.seqid);
1951         nfs4_put_state_owner(sp);
1952         nfs_sb_deactive(sb);
1953         kfree(calldata);
1954 }
1955
1956 static void nfs4_close_clear_stateid_flags(struct nfs4_state *state,
1957                 fmode_t fmode)
1958 {
1959         spin_lock(&state->owner->so_lock);
1960         if (!(fmode & FMODE_READ))
1961                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1962         if (!(fmode & FMODE_WRITE))
1963                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1964         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1965         spin_unlock(&state->owner->so_lock);
1966 }
1967
1968 static void nfs4_close_done(struct rpc_task *task, void *data)
1969 {
1970         struct nfs4_closedata *calldata = data;
1971         struct nfs4_state *state = calldata->state;
1972         struct nfs_server *server = NFS_SERVER(calldata->inode);
1973
1974         if (!nfs4_sequence_done(task, &calldata->res.seq_res))
1975                 return;
1976         /* hmm. we are done with the inode, and in the process of freeing
1977          * the state_owner. we keep this around to process errors
1978          */
1979         switch (task->tk_status) {
1980                 case 0:
1981                         if (calldata->roc)
1982                                 pnfs_roc_set_barrier(state->inode,
1983                                                      calldata->roc_barrier);
1984                         nfs_set_open_stateid(state, &calldata->res.stateid, 0);
1985                         renew_lease(server, calldata->timestamp);
1986                         nfs4_close_clear_stateid_flags(state,
1987                                         calldata->arg.fmode);
1988                         break;
1989                 case -NFS4ERR_STALE_STATEID:
1990                 case -NFS4ERR_OLD_STATEID:
1991                 case -NFS4ERR_BAD_STATEID:
1992                 case -NFS4ERR_EXPIRED:
1993                         if (calldata->arg.fmode == 0)
1994                                 break;
1995                 default:
1996                         if (nfs4_async_handle_error(task, server, state) == -EAGAIN)
1997                                 rpc_restart_call_prepare(task);
1998         }
1999         nfs_release_seqid(calldata->arg.seqid);
2000         nfs_refresh_inode(calldata->inode, calldata->res.fattr);
2001 }
2002
2003 static void nfs4_close_prepare(struct rpc_task *task, void *data)
2004 {
2005         struct nfs4_closedata *calldata = data;
2006         struct nfs4_state *state = calldata->state;
2007         int call_close = 0;
2008
2009         if (nfs_wait_on_sequence(calldata->arg.seqid, task) != 0)
2010                 return;
2011
2012         task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_DOWNGRADE];
2013         calldata->arg.fmode = FMODE_READ|FMODE_WRITE;
2014         spin_lock(&state->owner->so_lock);
2015         /* Calculate the change in open mode */
2016         if (state->n_rdwr == 0) {
2017                 if (state->n_rdonly == 0) {
2018                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
2019                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
2020                         calldata->arg.fmode &= ~FMODE_READ;
2021                 }
2022                 if (state->n_wronly == 0) {
2023                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
2024                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
2025                         calldata->arg.fmode &= ~FMODE_WRITE;
2026                 }
2027         }
2028         spin_unlock(&state->owner->so_lock);
2029
2030         if (!call_close) {
2031                 /* Note: exit _without_ calling nfs4_close_done */
2032                 task->tk_action = NULL;
2033                 return;
2034         }
2035
2036         if (calldata->arg.fmode == 0) {
2037                 task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE];
2038                 if (calldata->roc &&
2039                     pnfs_roc_drain(calldata->inode, &calldata->roc_barrier)) {
2040                         rpc_sleep_on(&NFS_SERVER(calldata->inode)->roc_rpcwaitq,
2041                                      task, NULL);
2042                         return;
2043                 }
2044         }
2045
2046         nfs_fattr_init(calldata->res.fattr);
2047         calldata->timestamp = jiffies;
2048         if (nfs4_setup_sequence(NFS_SERVER(calldata->inode),
2049                                 &calldata->arg.seq_args, &calldata->res.seq_res,
2050                                 1, task) != 0)
2051                 nfs_release_seqid(calldata->arg.seqid);
2052         else
2053                 rpc_call_start(task);
2054 }
2055
2056 static const struct rpc_call_ops nfs4_close_ops = {
2057         .rpc_call_prepare = nfs4_close_prepare,
2058         .rpc_call_done = nfs4_close_done,
2059         .rpc_release = nfs4_free_closedata,
2060 };
2061
2062 /*
2063  * It is possible for data to be read/written from a mem-mapped file
2064  * after the sys_close call (which hits the vfs layer as a flush).
2065  * This means that we can't safely call nfsv4 close on a file until
2066  * the inode is cleared. This in turn means that we are not good
2067  * NFSv4 citizens - we do not indicate to the server to update the file's
2068  * share state even when we are done with one of the three share
2069  * stateid's in the inode.
2070  *
2071  * NOTE: Caller must be holding the sp->so_owner semaphore!
2072  */
2073 int nfs4_do_close(struct nfs4_state *state, gfp_t gfp_mask, int wait, bool roc)
2074 {
2075         struct nfs_server *server = NFS_SERVER(state->inode);
2076         struct nfs4_closedata *calldata;
2077         struct nfs4_state_owner *sp = state->owner;
2078         struct rpc_task *task;
2079         struct rpc_message msg = {
2080                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE],
2081                 .rpc_cred = state->owner->so_cred,
2082         };
2083         struct rpc_task_setup task_setup_data = {
2084                 .rpc_client = server->client,
2085                 .rpc_message = &msg,
2086                 .callback_ops = &nfs4_close_ops,
2087                 .workqueue = nfsiod_workqueue,
2088                 .flags = RPC_TASK_ASYNC,
2089         };
2090         int status = -ENOMEM;
2091
2092         calldata = kzalloc(sizeof(*calldata), gfp_mask);
2093         if (calldata == NULL)
2094                 goto out;
2095         calldata->inode = state->inode;
2096         calldata->state = state;
2097         calldata->arg.fh = NFS_FH(state->inode);
2098         calldata->arg.stateid = &state->open_stateid;
2099         /* Serialization for the sequence id */
2100         calldata->arg.seqid = nfs_alloc_seqid(&state->owner->so_seqid, gfp_mask);
2101         if (calldata->arg.seqid == NULL)
2102                 goto out_free_calldata;
2103         calldata->arg.fmode = 0;
2104         calldata->arg.bitmask = server->cache_consistency_bitmask;
2105         calldata->res.fattr = &calldata->fattr;
2106         calldata->res.seqid = calldata->arg.seqid;
2107         calldata->res.server = server;
2108         calldata->roc = roc;
2109         nfs_sb_active(calldata->inode->i_sb);
2110
2111         msg.rpc_argp = &calldata->arg;
2112         msg.rpc_resp = &calldata->res;
2113         task_setup_data.callback_data = calldata;
2114         task = rpc_run_task(&task_setup_data);
2115         if (IS_ERR(task))
2116                 return PTR_ERR(task);
2117         status = 0;
2118         if (wait)
2119                 status = rpc_wait_for_completion_task(task);
2120         rpc_put_task(task);
2121         return status;
2122 out_free_calldata:
2123         kfree(calldata);
2124 out:
2125         if (roc)
2126                 pnfs_roc_release(state->inode);
2127         nfs4_put_open_state(state);
2128         nfs4_put_state_owner(sp);
2129         return status;
2130 }
2131
2132 static struct inode *
2133 nfs4_atomic_open(struct inode *dir, struct nfs_open_context *ctx, int open_flags, struct iattr *attr)
2134 {
2135         struct nfs4_state *state;
2136
2137         /* Protect against concurrent sillydeletes */
2138         state = nfs4_do_open(dir, ctx->dentry, ctx->mode, open_flags, attr, ctx->cred);
2139         if (IS_ERR(state))
2140                 return ERR_CAST(state);
2141         ctx->state = state;
2142         return igrab(state->inode);
2143 }
2144
2145 static void nfs4_close_context(struct nfs_open_context *ctx, int is_sync)
2146 {
2147         if (ctx->state == NULL)
2148                 return;
2149         if (is_sync)
2150                 nfs4_close_sync(ctx->state, ctx->mode);
2151         else
2152                 nfs4_close_state(ctx->state, ctx->mode);
2153 }
2154
2155 static int _nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
2156 {
2157         struct nfs4_server_caps_arg args = {
2158                 .fhandle = fhandle,
2159         };
2160         struct nfs4_server_caps_res res = {};
2161         struct rpc_message msg = {
2162                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_SERVER_CAPS],
2163                 .rpc_argp = &args,
2164                 .rpc_resp = &res,
2165         };
2166         int status;
2167
2168         status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
2169         if (status == 0) {
2170                 memcpy(server->attr_bitmask, res.attr_bitmask, sizeof(server->attr_bitmask));
2171                 server->caps &= ~(NFS_CAP_ACLS|NFS_CAP_HARDLINKS|
2172                                 NFS_CAP_SYMLINKS|NFS_CAP_FILEID|
2173                                 NFS_CAP_MODE|NFS_CAP_NLINK|NFS_CAP_OWNER|
2174                                 NFS_CAP_OWNER_GROUP|NFS_CAP_ATIME|
2175                                 NFS_CAP_CTIME|NFS_CAP_MTIME);
2176                 if (res.attr_bitmask[0] & FATTR4_WORD0_ACL)
2177                         server->caps |= NFS_CAP_ACLS;
2178                 if (res.has_links != 0)
2179                         server->caps |= NFS_CAP_HARDLINKS;
2180                 if (res.has_symlinks != 0)
2181                         server->caps |= NFS_CAP_SYMLINKS;
2182                 if (res.attr_bitmask[0] & FATTR4_WORD0_FILEID)
2183                         server->caps |= NFS_CAP_FILEID;
2184                 if (res.attr_bitmask[1] & FATTR4_WORD1_MODE)
2185                         server->caps |= NFS_CAP_MODE;
2186                 if (res.attr_bitmask[1] & FATTR4_WORD1_NUMLINKS)
2187                         server->caps |= NFS_CAP_NLINK;
2188                 if (res.attr_bitmask[1] & FATTR4_WORD1_OWNER)
2189                         server->caps |= NFS_CAP_OWNER;
2190                 if (res.attr_bitmask[1] & FATTR4_WORD1_OWNER_GROUP)
2191                         server->caps |= NFS_CAP_OWNER_GROUP;
2192                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_ACCESS)
2193                         server->caps |= NFS_CAP_ATIME;
2194                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_METADATA)
2195                         server->caps |= NFS_CAP_CTIME;
2196                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_MODIFY)
2197                         server->caps |= NFS_CAP_MTIME;
2198
2199                 memcpy(server->cache_consistency_bitmask, res.attr_bitmask, sizeof(server->cache_consistency_bitmask));
2200                 server->cache_consistency_bitmask[0] &= FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;
2201                 server->cache_consistency_bitmask[1] &= FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;
2202                 server->acl_bitmask = res.acl_bitmask;
2203         }
2204
2205         return status;
2206 }
2207
2208 int nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
2209 {
2210         struct nfs4_exception exception = { };
2211         int err;
2212         do {
2213                 err = nfs4_handle_exception(server,
2214                                 _nfs4_server_capabilities(server, fhandle),
2215                                 &exception);
2216         } while (exception.retry);
2217         return err;
2218 }
2219
2220 static int _nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
2221                 struct nfs_fsinfo *info)
2222 {
2223         struct nfs4_lookup_root_arg args = {
2224                 .bitmask = nfs4_fattr_bitmap,
2225         };
2226         struct nfs4_lookup_res res = {
2227                 .server = server,
2228                 .fattr = info->fattr,
2229                 .fh = fhandle,
2230         };
2231         struct rpc_message msg = {
2232                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOOKUP_ROOT],
2233                 .rpc_argp = &args,
2234                 .rpc_resp = &res,
2235         };
2236
2237         nfs_fattr_init(info->fattr);
2238         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
2239 }
2240
2241 static int nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
2242                 struct nfs_fsinfo *info)
2243 {
2244         struct nfs4_exception exception = { };
2245         int err;
2246         do {
2247                 err = _nfs4_lookup_root(server, fhandle, info);
2248                 switch (err) {
2249                 case 0:
2250                 case -NFS4ERR_WRONGSEC:
2251                         goto out;
2252                 default:
2253                         err = nfs4_handle_exception(server, err, &exception);
2254                 }
2255         } while (exception.retry);
2256 out:
2257         return err;
2258 }
2259
2260 static int nfs4_lookup_root_sec(struct nfs_server *server, struct nfs_fh *fhandle,
2261                                 struct nfs_fsinfo *info, rpc_authflavor_t flavor)
2262 {
2263         struct rpc_auth *auth;
2264         int ret;
2265
2266         auth = rpcauth_create(flavor, server->client);
2267         if (!auth) {
2268                 ret = -EIO;
2269                 goto out;
2270         }
2271         ret = nfs4_lookup_root(server, fhandle, info);
2272 out:
2273         return ret;
2274 }
2275
2276 static int nfs4_find_root_sec(struct nfs_server *server, struct nfs_fh *fhandle,
2277                               struct nfs_fsinfo *info)
2278 {
2279         int i, len, status = 0;
2280         rpc_authflavor_t flav_array[NFS_MAX_SECFLAVORS];
2281
2282         len = gss_mech_list_pseudoflavors(&flav_array[0]);
2283         flav_array[len] = RPC_AUTH_NULL;
2284         len += 1;
2285
2286         for (i = 0; i < len; i++) {
2287                 status = nfs4_lookup_root_sec(server, fhandle, info, flav_array[i]);
2288                 if (status == -NFS4ERR_WRONGSEC || status == -EACCES)
2289                         continue;
2290                 break;
2291         }
2292         /*
2293          * -EACCESS could mean that the user doesn't have correct permissions
2294          * to access the mount.  It could also mean that we tried to mount
2295          * with a gss auth flavor, but rpc.gssd isn't running.  Either way,
2296          * existing mount programs don't handle -EACCES very well so it should
2297          * be mapped to -EPERM instead.
2298          */
2299         if (status == -EACCES)
2300                 status = -EPERM;
2301         return status;
2302 }
2303
2304 /*
2305  * get the file handle for the "/" directory on the server
2306  */
2307 static int nfs4_proc_get_root(struct nfs_server *server, struct nfs_fh *fhandle,
2308                               struct nfs_fsinfo *info)
2309 {
2310         int minor_version = server->nfs_client->cl_minorversion;
2311         int status = nfs4_lookup_root(server, fhandle, info);
2312         if ((status == -NFS4ERR_WRONGSEC) && !(server->flags & NFS_MOUNT_SECFLAVOUR))
2313                 /*
2314                  * A status of -NFS4ERR_WRONGSEC will be mapped to -EPERM
2315                  * by nfs4_map_errors() as this function exits.
2316                  */
2317                 status = nfs_v4_minor_ops[minor_version]->find_root_sec(server, fhandle, info);
2318         if (status == 0)
2319                 status = nfs4_server_capabilities(server, fhandle);
2320         if (status == 0)
2321                 status = nfs4_do_fsinfo(server, fhandle, info);
2322         return nfs4_map_errors(status);
2323 }
2324
2325 static void nfs_fixup_referral_attributes(struct nfs_fattr *fattr);
2326 /*
2327  * Get locations and (maybe) other attributes of a referral.
2328  * Note that we'll actually follow the referral later when
2329  * we detect fsid mismatch in inode revalidation
2330  */
2331 static int nfs4_get_referral(struct inode *dir, const struct qstr *name,
2332                              struct nfs_fattr *fattr, struct nfs_fh *fhandle)
2333 {
2334         int status = -ENOMEM;
2335         struct page *page = NULL;
2336         struct nfs4_fs_locations *locations = NULL;
2337
2338         page = alloc_page(GFP_KERNEL);
2339         if (page == NULL)
2340                 goto out;
2341         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2342         if (locations == NULL)
2343                 goto out;
2344
2345         status = nfs4_proc_fs_locations(dir, name, locations, page);
2346         if (status != 0)
2347                 goto out;
2348         /* Make sure server returned a different fsid for the referral */
2349         if (nfs_fsid_equal(&NFS_SERVER(dir)->fsid, &locations->fattr.fsid)) {
2350                 dprintk("%s: server did not return a different fsid for"
2351                         " a referral at %s\n", __func__, name->name);
2352                 status = -EIO;
2353                 goto out;
2354         }
2355         /* Fixup attributes for the nfs_lookup() call to nfs_fhget() */
2356         nfs_fixup_referral_attributes(&locations->fattr);
2357
2358         /* replace the lookup nfs_fattr with the locations nfs_fattr */
2359         memcpy(fattr, &locations->fattr, sizeof(struct nfs_fattr));
2360         memset(fhandle, 0, sizeof(struct nfs_fh));
2361 out:
2362         if (page)
2363                 __free_page(page);
2364         kfree(locations);
2365         return status;
2366 }
2367
2368 static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fattr *fattr)
2369 {
2370         struct nfs4_getattr_arg args = {
2371                 .fh = fhandle,
2372                 .bitmask = server->attr_bitmask,
2373         };
2374         struct nfs4_getattr_res res = {
2375                 .fattr = fattr,
2376                 .server = server,
2377         };
2378         struct rpc_message msg = {
2379                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETATTR],
2380                 .rpc_argp = &args,
2381                 .rpc_resp = &res,
2382         };
2383
2384         nfs_fattr_init(fattr);
2385         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
2386 }
2387
2388 static int nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fattr *fattr)
2389 {
2390         struct nfs4_exception exception = { };
2391         int err;
2392         do {
2393                 err = nfs4_handle_exception(server,
2394                                 _nfs4_proc_getattr(server, fhandle, fattr),
2395                                 &exception);
2396         } while (exception.retry);
2397         return err;
2398 }
2399
2400 /*
2401  * The file is not closed if it is opened due to the a request to change
2402  * the size of the file. The open call will not be needed once the
2403  * VFS layer lookup-intents are implemented.
2404  *
2405  * Close is called when the inode is destroyed.
2406  * If we haven't opened the file for O_WRONLY, we
2407  * need to in the size_change case to obtain a stateid.
2408  *
2409  * Got race?
2410  * Because OPEN is always done by name in nfsv4, it is
2411  * possible that we opened a different file by the same
2412  * name.  We can recognize this race condition, but we
2413  * can't do anything about it besides returning an error.
2414  *
2415  * This will be fixed with VFS changes (lookup-intent).
2416  */
2417 static int
2418 nfs4_proc_setattr(struct dentry *dentry, struct nfs_fattr *fattr,
2419                   struct iattr *sattr)
2420 {
2421         struct inode *inode = dentry->d_inode;
2422         struct rpc_cred *cred = NULL;
2423         struct nfs4_state *state = NULL;
2424         int status;
2425
2426         if (pnfs_ld_layoutret_on_setattr(inode))
2427                 pnfs_return_layout(inode);
2428
2429         nfs_fattr_init(fattr);
2430
2431         /* Search for an existing open(O_WRITE) file */
2432         if (sattr->ia_valid & ATTR_FILE) {
2433                 struct nfs_open_context *ctx;
2434
2435                 ctx = nfs_file_open_context(sattr->ia_file);
2436                 if (ctx) {
2437                         cred = ctx->cred;
2438                         state = ctx->state;
2439                 }
2440         }
2441
2442         status = nfs4_do_setattr(inode, cred, fattr, sattr, state);
2443         if (status == 0)
2444                 nfs_setattr_update_inode(inode, sattr);
2445         return status;
2446 }
2447
2448 static int _nfs4_proc_lookup(struct rpc_clnt *clnt, struct inode *dir,
2449                 const struct qstr *name, struct nfs_fh *fhandle,
2450                 struct nfs_fattr *fattr)
2451 {
2452         struct nfs_server *server = NFS_SERVER(dir);
2453         int                    status;
2454         struct nfs4_lookup_arg args = {
2455                 .bitmask = server->attr_bitmask,
2456                 .dir_fh = NFS_FH(dir),
2457                 .name = name,
2458         };
2459         struct nfs4_lookup_res res = {
2460                 .server = server,
2461                 .fattr = fattr,
2462                 .fh = fhandle,
2463         };
2464         struct rpc_message msg = {
2465                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOOKUP],
2466                 .rpc_argp = &args,
2467                 .rpc_resp = &res,
2468         };
2469
2470         nfs_fattr_init(fattr);
2471
2472         dprintk("NFS call  lookup %s\n", name->name);
2473         status = nfs4_call_sync(clnt, server, &msg, &args.seq_args, &res.seq_res, 0);
2474         dprintk("NFS reply lookup: %d\n", status);
2475         return status;
2476 }
2477
2478 void nfs_fixup_secinfo_attributes(struct nfs_fattr *fattr, struct nfs_fh *fh)
2479 {
2480         memset(fh, 0, sizeof(struct nfs_fh));
2481         fattr->fsid.major = 1;
2482         fattr->valid |= NFS_ATTR_FATTR_TYPE | NFS_ATTR_FATTR_MODE |
2483                 NFS_ATTR_FATTR_NLINK | NFS_ATTR_FATTR_FSID | NFS_ATTR_FATTR_MOUNTPOINT;
2484         fattr->mode = S_IFDIR | S_IRUGO | S_IXUGO;
2485         fattr->nlink = 2;
2486 }
2487
2488 static int nfs4_proc_lookup(struct rpc_clnt *clnt, struct inode *dir, struct qstr *name,
2489                             struct nfs_fh *fhandle, struct nfs_fattr *fattr)
2490 {
2491         struct nfs4_exception exception = { };
2492         int err;
2493         do {
2494                 int status;
2495
2496                 status = _nfs4_proc_lookup(clnt, dir, name, fhandle, fattr);
2497                 switch (status) {
2498                 case -NFS4ERR_BADNAME:
2499                         return -ENOENT;
2500                 case -NFS4ERR_MOVED:
2501                         return nfs4_get_referral(dir, name, fattr, fhandle);
2502                 case -NFS4ERR_WRONGSEC:
2503                         nfs_fixup_secinfo_attributes(fattr, fhandle);
2504                 }
2505                 err = nfs4_handle_exception(NFS_SERVER(dir),
2506                                 status, &exception);
2507         } while (exception.retry);
2508         return err;
2509 }
2510
2511 static int _nfs4_proc_access(struct inode *inode, struct nfs_access_entry *entry)
2512 {
2513         struct nfs_server *server = NFS_SERVER(inode);
2514         struct nfs4_accessargs args = {
2515                 .fh = NFS_FH(inode),
2516                 .bitmask = server->attr_bitmask,
2517         };
2518         struct nfs4_accessres res = {
2519                 .server = server,
2520         };
2521         struct rpc_message msg = {
2522                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_ACCESS],
2523                 .rpc_argp = &args,
2524                 .rpc_resp = &res,
2525                 .rpc_cred = entry->cred,
2526         };
2527         int mode = entry->mask;
2528         int status;
2529
2530         /*
2531          * Determine which access bits we want to ask for...
2532          */
2533         if (mode & MAY_READ)
2534                 args.access |= NFS4_ACCESS_READ;
2535         if (S_ISDIR(inode->i_mode)) {
2536                 if (mode & MAY_WRITE)
2537                         args.access |= NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE;
2538                 if (mode & MAY_EXEC)
2539                         args.access |= NFS4_ACCESS_LOOKUP;
2540         } else {
2541                 if (mode & MAY_WRITE)
2542                         args.access |= NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND;
2543                 if (mode & MAY_EXEC)
2544                         args.access |= NFS4_ACCESS_EXECUTE;
2545         }
2546
2547         res.fattr = nfs_alloc_fattr();
2548         if (res.fattr == NULL)
2549                 return -ENOMEM;
2550
2551         status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
2552         if (!status) {
2553                 entry->mask = 0;
2554                 if (res.access & NFS4_ACCESS_READ)
2555                         entry->mask |= MAY_READ;
2556                 if (res.access & (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2557                         entry->mask |= MAY_WRITE;
2558                 if (res.access & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2559                         entry->mask |= MAY_EXEC;
2560                 nfs_refresh_inode(inode, res.fattr);
2561         }
2562         nfs_free_fattr(res.fattr);
2563         return status;
2564 }
2565
2566 static int nfs4_proc_access(struct inode *inode, struct nfs_access_entry *entry)
2567 {
2568         struct nfs4_exception exception = { };
2569         int err;
2570         do {
2571                 err = nfs4_handle_exception(NFS_SERVER(inode),
2572                                 _nfs4_proc_access(inode, entry),
2573                                 &exception);
2574         } while (exception.retry);
2575         return err;
2576 }
2577
2578 /*
2579  * TODO: For the time being, we don't try to get any attributes
2580  * along with any of the zero-copy operations READ, READDIR,
2581  * READLINK, WRITE.
2582  *
2583  * In the case of the first three, we want to put the GETATTR
2584  * after the read-type operation -- this is because it is hard
2585  * to predict the length of a GETATTR response in v4, and thus
2586  * align the READ data correctly.  This means that the GETATTR
2587  * may end up partially falling into the page cache, and we should
2588  * shift it into the 'tail' of the xdr_buf before processing.
2589  * To do this efficiently, we need to know the total length
2590  * of data received, which doesn't seem to be available outside
2591  * of the RPC layer.
2592  *
2593  * In the case of WRITE, we also want to put the GETATTR after
2594  * the operation -- in this case because we want to make sure
2595  * we get the post-operation mtime and size.  This means that
2596  * we can't use xdr_encode_pages() as written: we need a variant
2597  * of it which would leave room in the 'tail' iovec.
2598  *
2599  * Both of these changes to the XDR layer would in fact be quite
2600  * minor, but I decided to leave them for a subsequent patch.
2601  */
2602 static int _nfs4_proc_readlink(struct inode *inode, struct page *page,
2603                 unsigned int pgbase, unsigned int pglen)
2604 {
2605         struct nfs4_readlink args = {
2606                 .fh       = NFS_FH(inode),
2607                 .pgbase   = pgbase,
2608                 .pglen    = pglen,
2609                 .pages    = &page,
2610         };
2611         struct nfs4_readlink_res res;
2612         struct rpc_message msg = {
2613                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_READLINK],
2614                 .rpc_argp = &args,
2615                 .rpc_resp = &res,
2616         };
2617
2618         return nfs4_call_sync(NFS_SERVER(inode)->client, NFS_SERVER(inode), &msg, &args.seq_args, &res.seq_res, 0);
2619 }
2620
2621 static int nfs4_proc_readlink(struct inode *inode, struct page *page,
2622                 unsigned int pgbase, unsigned int pglen)
2623 {
2624         struct nfs4_exception exception = { };
2625         int err;
2626         do {
2627                 err = nfs4_handle_exception(NFS_SERVER(inode),
2628                                 _nfs4_proc_readlink(inode, page, pgbase, pglen),
2629                                 &exception);
2630         } while (exception.retry);
2631         return err;
2632 }
2633
2634 /*
2635  * Got race?
2636  * We will need to arrange for the VFS layer to provide an atomic open.
2637  * Until then, this create/open method is prone to inefficiency and race
2638  * conditions due to the lookup, create, and open VFS calls from sys_open()
2639  * placed on the wire.
2640  *
2641  * Given the above sorry state of affairs, I'm simply sending an OPEN.
2642  * The file will be opened again in the subsequent VFS open call
2643  * (nfs4_proc_file_open).
2644  *
2645  * The open for read will just hang around to be used by any process that
2646  * opens the file O_RDONLY. This will all be resolved with the VFS changes.
2647  */
2648
2649 static int
2650 nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
2651                  int flags, struct nfs_open_context *ctx)
2652 {
2653         struct dentry *de = dentry;
2654         struct nfs4_state *state;
2655         struct rpc_cred *cred = NULL;
2656         fmode_t fmode = 0;
2657         int status = 0;
2658
2659         if (ctx != NULL) {
2660                 cred = ctx->cred;
2661                 de = ctx->dentry;
2662                 fmode = ctx->mode;
2663         }
2664         sattr->ia_mode &= ~current_umask();
2665         state = nfs4_do_open(dir, de, fmode, flags, sattr, cred);
2666         d_drop(dentry);
2667         if (IS_ERR(state)) {
2668                 status = PTR_ERR(state);
2669                 goto out;
2670         }
2671         d_add(dentry, igrab(state->inode));
2672         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2673         if (ctx != NULL)
2674                 ctx->state = state;
2675         else
2676                 nfs4_close_sync(state, fmode);
2677 out:
2678         return status;
2679 }
2680
2681 static int _nfs4_proc_remove(struct inode *dir, struct qstr *name)
2682 {
2683         struct nfs_server *server = NFS_SERVER(dir);
2684         struct nfs_removeargs args = {
2685                 .fh = NFS_FH(dir),
2686                 .name.len = name->len,
2687                 .name.name = name->name,
2688                 .bitmask = server->attr_bitmask,
2689         };
2690         struct nfs_removeres res = {
2691                 .server = server,
2692         };
2693         struct rpc_message msg = {
2694                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_REMOVE],
2695                 .rpc_argp = &args,
2696                 .rpc_resp = &res,
2697         };
2698         int status = -ENOMEM;
2699
2700         res.dir_attr = nfs_alloc_fattr();
2701         if (res.dir_attr == NULL)
2702                 goto out;
2703
2704         status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 1);
2705         if (status == 0) {
2706                 update_changeattr(dir, &res.cinfo);
2707                 nfs_post_op_update_inode(dir, res.dir_attr);
2708         }
2709         nfs_free_fattr(res.dir_attr);
2710 out:
2711         return status;
2712 }
2713
2714 static int nfs4_proc_remove(struct inode *dir, struct qstr *name)
2715 {
2716         struct nfs4_exception exception = { };
2717         int err;
2718         do {
2719                 err = nfs4_handle_exception(NFS_SERVER(dir),
2720                                 _nfs4_proc_remove(dir, name),
2721                                 &exception);
2722         } while (exception.retry);
2723         return err;
2724 }
2725
2726 static void nfs4_proc_unlink_setup(struct rpc_message *msg, struct inode *dir)
2727 {
2728         struct nfs_server *server = NFS_SERVER(dir);
2729         struct nfs_removeargs *args = msg->rpc_argp;
2730         struct nfs_removeres *res = msg->rpc_resp;
2731
2732         args->bitmask = server->cache_consistency_bitmask;
2733         res->server = server;
2734         res->seq_res.sr_slot = NULL;
2735         msg->rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_REMOVE];
2736 }
2737
2738 static int nfs4_proc_unlink_done(struct rpc_task *task, struct inode *dir)
2739 {
2740         struct nfs_removeres *res = task->tk_msg.rpc_resp;
2741
2742         if (!nfs4_sequence_done(task, &res->seq_res))
2743                 return 0;
2744         if (nfs4_async_handle_error(task, res->server, NULL) == -EAGAIN)
2745                 return 0;
2746         update_changeattr(dir, &res->cinfo);
2747         nfs_post_op_update_inode(dir, res->dir_attr);
2748         return 1;
2749 }
2750
2751 static void nfs4_proc_rename_setup(struct rpc_message *msg, struct inode *dir)
2752 {
2753         struct nfs_server *server = NFS_SERVER(dir);
2754         struct nfs_renameargs *arg = msg->rpc_argp;
2755         struct nfs_renameres *res = msg->rpc_resp;
2756
2757         msg->rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_RENAME];
2758         arg->bitmask = server->attr_bitmask;
2759         res->server = server;
2760 }
2761
2762 static int nfs4_proc_rename_done(struct rpc_task *task, struct inode *old_dir,
2763                                  struct inode *new_dir)
2764 {
2765         struct nfs_renameres *res = task->tk_msg.rpc_resp;
2766
2767         if (!nfs4_sequence_done(task, &res->seq_res))
2768                 return 0;
2769         if (nfs4_async_handle_error(task, res->server, NULL) == -EAGAIN)
2770                 return 0;
2771
2772         update_changeattr(old_dir, &res->old_cinfo);
2773         nfs_post_op_update_inode(old_dir, res->old_fattr);
2774         update_changeattr(new_dir, &res->new_cinfo);
2775         nfs_post_op_update_inode(new_dir, res->new_fattr);
2776         return 1;
2777 }
2778
2779 static int _nfs4_proc_rename(struct inode *old_dir, struct qstr *old_name,
2780                 struct inode *new_dir, struct qstr *new_name)
2781 {
2782         struct nfs_server *server = NFS_SERVER(old_dir);
2783         struct nfs_renameargs arg = {
2784                 .old_dir = NFS_FH(old_dir),
2785                 .new_dir = NFS_FH(new_dir),
2786                 .old_name = old_name,
2787                 .new_name = new_name,
2788                 .bitmask = server->attr_bitmask,
2789         };
2790         struct nfs_renameres res = {
2791                 .server = server,
2792         };
2793         struct rpc_message msg = {
2794                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_RENAME],
2795                 .rpc_argp = &arg,
2796                 .rpc_resp = &res,
2797         };
2798         int status = -ENOMEM;
2799
2800         res.old_fattr = nfs_alloc_fattr();
2801         res.new_fattr = nfs_alloc_fattr();
2802         if (res.old_fattr == NULL || res.new_fattr == NULL)
2803                 goto out;
2804
2805         status = nfs4_call_sync(server->client, server, &msg, &arg.seq_args, &res.seq_res, 1);
2806         if (!status) {
2807                 update_changeattr(old_dir, &res.old_cinfo);
2808                 nfs_post_op_update_inode(old_dir, res.old_fattr);
2809                 update_changeattr(new_dir, &res.new_cinfo);
2810                 nfs_post_op_update_inode(new_dir, res.new_fattr);
2811         }
2812 out:
2813         nfs_free_fattr(res.new_fattr);
2814         nfs_free_fattr(res.old_fattr);
2815         return status;
2816 }
2817
2818 static int nfs4_proc_rename(struct inode *old_dir, struct qstr *old_name,
2819                 struct inode *new_dir, struct qstr *new_name)
2820 {
2821         struct nfs4_exception exception = { };
2822         int err;
2823         do {
2824                 err = nfs4_handle_exception(NFS_SERVER(old_dir),
2825                                 _nfs4_proc_rename(old_dir, old_name,
2826                                         new_dir, new_name),
2827                                 &exception);
2828         } while (exception.retry);
2829         return err;
2830 }
2831
2832 static int _nfs4_proc_link(struct inode *inode, struct inode *dir, struct qstr *name)
2833 {
2834         struct nfs_server *server = NFS_SERVER(inode);
2835         struct nfs4_link_arg arg = {
2836                 .fh     = NFS_FH(inode),
2837                 .dir_fh = NFS_FH(dir),
2838                 .name   = name,
2839                 .bitmask = server->attr_bitmask,
2840         };
2841         struct nfs4_link_res res = {
2842                 .server = server,
2843         };
2844         struct rpc_message msg = {
2845                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LINK],
2846                 .rpc_argp = &arg,
2847                 .rpc_resp = &res,
2848         };
2849         int status = -ENOMEM;
2850
2851         res.fattr = nfs_alloc_fattr();
2852         res.dir_attr = nfs_alloc_fattr();
2853         if (res.fattr == NULL || res.dir_attr == NULL)
2854                 goto out;
2855
2856         status = nfs4_call_sync(server->client, server, &msg, &arg.seq_args, &res.seq_res, 1);
2857         if (!status) {
2858                 update_changeattr(dir, &res.cinfo);
2859                 nfs_post_op_update_inode(dir, res.dir_attr);
2860                 nfs_post_op_update_inode(inode, res.fattr);
2861         }
2862 out:
2863         nfs_free_fattr(res.dir_attr);
2864         nfs_free_fattr(res.fattr);
2865         return status;
2866 }
2867
2868 static int nfs4_proc_link(struct inode *inode, struct inode *dir, struct qstr *name)
2869 {
2870         struct nfs4_exception exception = { };
2871         int err;
2872         do {
2873                 err = nfs4_handle_exception(NFS_SERVER(inode),
2874                                 _nfs4_proc_link(inode, dir, name),
2875                                 &exception);
2876         } while (exception.retry);
2877         return err;
2878 }
2879
2880 struct nfs4_createdata {
2881         struct rpc_message msg;
2882         struct nfs4_create_arg arg;
2883         struct nfs4_create_res res;
2884         struct nfs_fh fh;
2885         struct nfs_fattr fattr;
2886         struct nfs_fattr dir_fattr;
2887 };
2888
2889 static struct nfs4_createdata *nfs4_alloc_createdata(struct inode *dir,
2890                 struct qstr *name, struct iattr *sattr, u32 ftype)
2891 {
2892         struct nfs4_createdata *data;
2893
2894         data = kzalloc(sizeof(*data), GFP_KERNEL);
2895         if (data != NULL) {
2896                 struct nfs_server *server = NFS_SERVER(dir);
2897
2898                 data->msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CREATE];
2899                 data->msg.rpc_argp = &data->arg;
2900                 data->msg.rpc_resp = &data->res;
2901                 data->arg.dir_fh = NFS_FH(dir);
2902                 data->arg.server = server;
2903                 data->arg.name = name;
2904                 data->arg.attrs = sattr;
2905                 data->arg.ftype = ftype;
2906                 data->arg.bitmask = server->attr_bitmask;
2907                 data->res.server = server;
2908                 data->res.fh = &data->fh;
2909                 data->res.fattr = &data->fattr;
2910                 data->res.dir_fattr = &data->dir_fattr;
2911                 nfs_fattr_init(data->res.fattr);
2912                 nfs_fattr_init(data->res.dir_fattr);
2913         }
2914         return data;
2915 }
2916
2917 static int nfs4_do_create(struct inode *dir, struct dentry *dentry, struct nfs4_createdata *data)
2918 {
2919         int status = nfs4_call_sync(NFS_SERVER(dir)->client, NFS_SERVER(dir), &data->msg,
2920                                     &data->arg.seq_args, &data->res.seq_res, 1);
2921         if (status == 0) {
2922                 update_changeattr(dir, &data->res.dir_cinfo);
2923                 nfs_post_op_update_inode(dir, data->res.dir_fattr);
2924                 status = nfs_instantiate(dentry, data->res.fh, data->res.fattr);
2925         }
2926         return status;
2927 }
2928
2929 static void nfs4_free_createdata(struct nfs4_createdata *data)
2930 {
2931         kfree(data);
2932 }
2933
2934 static int _nfs4_proc_symlink(struct inode *dir, struct dentry *dentry,
2935                 struct page *page, unsigned int len, struct iattr *sattr)
2936 {
2937         struct nfs4_createdata *data;
2938         int status = -ENAMETOOLONG;
2939
2940         if (len > NFS4_MAXPATHLEN)
2941                 goto out;
2942
2943         status = -ENOMEM;
2944         data = nfs4_alloc_createdata(dir, &dentry->d_name, sattr, NF4LNK);
2945         if (data == NULL)
2946                 goto out;
2947
2948         data->msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_SYMLINK];
2949         data->arg.u.symlink.pages = &page;
2950         data->arg.u.symlink.len = len;
2951
2952         status = nfs4_do_create(dir, dentry, data);
2953
2954         nfs4_free_createdata(data);
2955 out:
2956         return status;
2957 }
2958
2959 static int nfs4_proc_symlink(struct inode *dir, struct dentry *dentry,
2960                 struct page *page, unsigned int len, struct iattr *sattr)
2961 {
2962         struct nfs4_exception exception = { };
2963         int err;
2964         do {
2965                 err = nfs4_handle_exception(NFS_SERVER(dir),
2966                                 _nfs4_proc_symlink(dir, dentry, page,
2967                                                         len, sattr),
2968                                 &exception);
2969         } while (exception.retry);
2970         return err;
2971 }
2972
2973 static int _nfs4_proc_mkdir(struct inode *dir, struct dentry *dentry,
2974                 struct iattr *sattr)
2975 {
2976         struct nfs4_createdata *data;
2977         int status = -ENOMEM;
2978
2979         data = nfs4_alloc_createdata(dir, &dentry->d_name, sattr, NF4DIR);
2980         if (data == NULL)
2981                 goto out;
2982
2983         status = nfs4_do_create(dir, dentry, data);
2984
2985         nfs4_free_createdata(data);
2986 out:
2987         return status;
2988 }
2989
2990 static int nfs4_proc_mkdir(struct inode *dir, struct dentry *dentry,
2991                 struct iattr *sattr)
2992 {
2993         struct nfs4_exception exception = { };
2994         int err;
2995
2996         sattr->ia_mode &= ~current_umask();
2997         do {
2998                 err = nfs4_handle_exception(NFS_SERVER(dir),
2999                                 _nfs4_proc_mkdir(dir, dentry, sattr),
3000                                 &exception);
3001         } while (exception.retry);
3002         return err;
3003 }
3004
3005 static int _nfs4_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
3006                 u64 cookie, struct page **pages, unsigned int count, int plus)
3007 {
3008         struct inode            *dir = dentry->d_inode;
3009         struct nfs4_readdir_arg args = {
3010                 .fh = NFS_FH(dir),
3011                 .pages = pages,
3012                 .pgbase = 0,
3013                 .count = count,
3014                 .bitmask = NFS_SERVER(dentry->d_inode)->attr_bitmask,
3015                 .plus = plus,
3016         };
3017         struct nfs4_readdir_res res;
3018         struct rpc_message msg = {
3019                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_READDIR],
3020                 .rpc_argp = &args,
3021                 .rpc_resp = &res,
3022                 .rpc_cred = cred,
3023         };
3024         int                     status;
3025
3026         dprintk("%s: dentry = %s/%s, cookie = %Lu\n", __func__,
3027                         dentry->d_parent->d_name.name,
3028                         dentry->d_name.name,
3029                         (unsigned long long)cookie);
3030         nfs4_setup_readdir(cookie, NFS_I(dir)->cookieverf, dentry, &args);
3031         res.pgbase = args.pgbase;
3032         status = nfs4_call_sync(NFS_SERVER(dir)->client, NFS_SERVER(dir), &msg, &args.seq_args, &res.seq_res, 0);
3033         if (status >= 0) {
3034                 memcpy(NFS_I(dir)->cookieverf, res.verifier.data, NFS4_VERIFIER_SIZE);
3035                 status += args.pgbase;
3036         }
3037
3038         nfs_invalidate_atime(dir);
3039
3040         dprintk("%s: returns %d\n", __func__, status);
3041         return status;
3042 }
3043
3044 static int nfs4_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
3045                 u64 cookie, struct page **pages, unsigned int count, int plus)
3046 {
3047         struct nfs4_exception exception = { };
3048         int err;
3049         do {
3050                 err = nfs4_handle_exception(NFS_SERVER(dentry->d_inode),
3051                                 _nfs4_proc_readdir(dentry, cred, cookie,
3052                                         pages, count, plus),
3053                                 &exception);
3054         } while (exception.retry);
3055         return err;
3056 }
3057
3058 static int _nfs4_proc_mknod(struct inode *dir, struct dentry *dentry,
3059                 struct iattr *sattr, dev_t rdev)
3060 {
3061         struct nfs4_createdata *data;
3062         int mode = sattr->ia_mode;
3063         int status = -ENOMEM;
3064
3065         BUG_ON(!(sattr->ia_valid & ATTR_MODE));
3066         BUG_ON(!S_ISFIFO(mode) && !S_ISBLK(mode) && !S_ISCHR(mode) && !S_ISSOCK(mode));
3067
3068         data = nfs4_alloc_createdata(dir, &dentry->d_name, sattr, NF4SOCK);
3069         if (data == NULL)
3070                 goto out;
3071
3072         if (S_ISFIFO(mode))
3073                 data->arg.ftype = NF4FIFO;
3074         else if (S_ISBLK(mode)) {
3075                 data->arg.ftype = NF4BLK;
3076                 data->arg.u.device.specdata1 = MAJOR(rdev);
3077                 data->arg.u.device.specdata2 = MINOR(rdev);
3078         }
3079         else if (S_ISCHR(mode)) {
3080                 data->arg.ftype = NF4CHR;
3081                 data->arg.u.device.specdata1 = MAJOR(rdev);
3082                 data->arg.u.device.specdata2 = MINOR(rdev);
3083         }
3084
3085         status = nfs4_do_create(dir, dentry, data);
3086
3087         nfs4_free_createdata(data);
3088 out:
3089         return status;
3090 }
3091
3092 static int nfs4_proc_mknod(struct inode *dir, struct dentry *dentry,
3093                 struct iattr *sattr, dev_t rdev)
3094 {
3095         struct nfs4_exception exception = { };
3096         int err;
3097
3098         sattr->ia_mode &= ~current_umask();
3099         do {
3100                 err = nfs4_handle_exception(NFS_SERVER(dir),
3101                                 _nfs4_proc_mknod(dir, dentry, sattr, rdev),
3102                                 &exception);
3103         } while (exception.retry);
3104         return err;
3105 }
3106
3107 static int _nfs4_proc_statfs(struct nfs_server *server, struct nfs_fh *fhandle,
3108                  struct nfs_fsstat *fsstat)
3109 {
3110         struct nfs4_statfs_arg args = {
3111                 .fh = fhandle,
3112                 .bitmask = server->attr_bitmask,
3113         };
3114         struct nfs4_statfs_res res = {
3115                 .fsstat = fsstat,
3116         };
3117         struct rpc_message msg = {
3118                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_STATFS],
3119                 .rpc_argp = &args,
3120                 .rpc_resp = &res,
3121         };
3122
3123         nfs_fattr_init(fsstat->fattr);
3124         return  nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3125 }
3126
3127 static int nfs4_proc_statfs(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fsstat *fsstat)
3128 {
3129         struct nfs4_exception exception = { };
3130         int err;
3131         do {
3132                 err = nfs4_handle_exception(server,
3133                                 _nfs4_proc_statfs(server, fhandle, fsstat),
3134                                 &exception);
3135         } while (exception.retry);
3136         return err;
3137 }
3138
3139 static int _nfs4_do_fsinfo(struct nfs_server *server, struct nfs_fh *fhandle,
3140                 struct nfs_fsinfo *fsinfo)
3141 {
3142         struct nfs4_fsinfo_arg args = {
3143                 .fh = fhandle,
3144                 .bitmask = server->attr_bitmask,
3145         };
3146         struct nfs4_fsinfo_res res = {
3147                 .fsinfo = fsinfo,
3148         };
3149         struct rpc_message msg = {
3150                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_FSINFO],
3151                 .rpc_argp = &args,
3152                 .rpc_resp = &res,
3153         };
3154
3155         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3156 }
3157
3158 static int nfs4_do_fsinfo(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fsinfo *fsinfo)
3159 {
3160         struct nfs4_exception exception = { };
3161         int err;
3162
3163         do {
3164                 err = nfs4_handle_exception(server,
3165                                 _nfs4_do_fsinfo(server, fhandle, fsinfo),
3166                                 &exception);
3167         } while (exception.retry);
3168         return err;
3169 }
3170
3171 static int nfs4_proc_fsinfo(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fsinfo *fsinfo)
3172 {
3173         nfs_fattr_init(fsinfo->fattr);
3174         return nfs4_do_fsinfo(server, fhandle, fsinfo);
3175 }
3176
3177 static int _nfs4_proc_pathconf(struct nfs_server *server, struct nfs_fh *fhandle,
3178                 struct nfs_pathconf *pathconf)
3179 {
3180         struct nfs4_pathconf_arg args = {
3181                 .fh = fhandle,
3182                 .bitmask = server->attr_bitmask,
3183         };
3184         struct nfs4_pathconf_res res = {
3185                 .pathconf = pathconf,
3186         };
3187         struct rpc_message msg = {
3188                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_PATHCONF],
3189                 .rpc_argp = &args,
3190                 .rpc_resp = &res,
3191         };
3192
3193         /* None of the pathconf attributes are mandatory to implement */
3194         if ((args.bitmask[0] & nfs4_pathconf_bitmap[0]) == 0) {
3195                 memset(pathconf, 0, sizeof(*pathconf));
3196                 return 0;
3197         }
3198
3199         nfs_fattr_init(pathconf->fattr);
3200         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3201 }
3202
3203 static int nfs4_proc_pathconf(struct nfs_server *server, struct nfs_fh *fhandle,
3204                 struct nfs_pathconf *pathconf)
3205 {
3206         struct nfs4_exception exception = { };
3207         int err;
3208
3209         do {
3210                 err = nfs4_handle_exception(server,
3211                                 _nfs4_proc_pathconf(server, fhandle, pathconf),
3212                                 &exception);
3213         } while (exception.retry);
3214         return err;
3215 }
3216
3217 void __nfs4_read_done_cb(struct nfs_read_data *data)
3218 {
3219         nfs_invalidate_atime(data->inode);
3220 }
3221
3222 static int nfs4_read_done_cb(struct rpc_task *task, struct nfs_read_data *data)
3223 {
3224         struct nfs_server *server = NFS_SERVER(data->inode);
3225
3226         if (nfs4_async_handle_error(task, server, data->args.context->state) == -EAGAIN) {
3227                 rpc_restart_call_prepare(task);
3228                 return -EAGAIN;
3229         }
3230
3231         __nfs4_read_done_cb(data);
3232         if (task->tk_status > 0)
3233                 renew_lease(server, data->timestamp);
3234         return 0;
3235 }
3236
3237 static int nfs4_read_done(struct rpc_task *task, struct nfs_read_data *data)
3238 {
3239
3240         dprintk("--> %s\n", __func__);
3241
3242         if (!nfs4_sequence_done(task, &data->res.seq_res))
3243                 return -EAGAIN;
3244
3245         return data->read_done_cb ? data->read_done_cb(task, data) :
3246                                     nfs4_read_done_cb(task, data);
3247 }
3248
3249 static void nfs4_proc_read_setup(struct nfs_read_data *data, struct rpc_message *msg)
3250 {
3251         data->timestamp   = jiffies;
3252         data->read_done_cb = nfs4_read_done_cb;
3253         msg->rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_READ];
3254 }
3255
3256 /* Reset the the nfs_read_data to send the read to the MDS. */
3257 void nfs4_reset_read(struct rpc_task *task, struct nfs_read_data *data)
3258 {
3259         dprintk("%s Reset task for i/o through\n", __func__);
3260         put_lseg(data->lseg);
3261         data->lseg = NULL;
3262         /* offsets will differ in the dense stripe case */
3263         data->args.offset = data->mds_offset;
3264         data->ds_clp = NULL;
3265         data->args.fh     = NFS_FH(data->inode);
3266         data->read_done_cb = nfs4_read_done_cb;
3267         task->tk_ops = data->mds_ops;
3268         rpc_task_reset_client(task, NFS_CLIENT(data->inode));
3269 }
3270 EXPORT_SYMBOL_GPL(nfs4_reset_read);
3271
3272 static int nfs4_write_done_cb(struct rpc_task *task, struct nfs_write_data *data)
3273 {
3274         struct inode *inode = data->inode;
3275
3276         if (nfs4_async_handle_error(task, NFS_SERVER(inode), data->args.context->state) == -EAGAIN) {
3277                 rpc_restart_call_prepare(task);
3278                 return -EAGAIN;
3279         }
3280         if (task->tk_status >= 0) {
3281                 renew_lease(NFS_SERVER(inode), data->timestamp);
3282                 nfs_post_op_update_inode_force_wcc(inode, data->res.fattr);
3283         }
3284         return 0;
3285 }
3286
3287 static int nfs4_write_done(struct rpc_task *task, struct nfs_write_data *data)
3288 {
3289         if (!nfs4_sequence_done(task, &data->res.seq_res))
3290                 return -EAGAIN;
3291         return data->write_done_cb ? data->write_done_cb(task, data) :
3292                 nfs4_write_done_cb(task, data);
3293 }
3294
3295 /* Reset the the nfs_write_data to send the write to the MDS. */
3296 void nfs4_reset_write(struct rpc_task *task, struct nfs_write_data *data)
3297 {
3298         dprintk("%s Reset task for i/o through\n", __func__);
3299         put_lseg(data->lseg);
3300         data->lseg          = NULL;
3301         data->ds_clp        = NULL;
3302         data->write_done_cb = nfs4_write_done_cb;
3303         data->args.fh       = NFS_FH(data->inode);
3304         data->args.bitmask  = data->res.server->cache_consistency_bitmask;
3305         data->args.offset   = data->mds_offset;
3306         data->res.fattr     = &data->fattr;
3307         task->tk_ops        = data->mds_ops;
3308         rpc_task_reset_client(task, NFS_CLIENT(data->inode));
3309 }
3310 EXPORT_SYMBOL_GPL(nfs4_reset_write);
3311
3312 static void nfs4_proc_write_setup(struct nfs_write_data *data, struct rpc_message *msg)
3313 {
3314         struct nfs_server *server = NFS_SERVER(data->inode);
3315
3316         if (data->lseg) {
3317                 data->args.bitmask = NULL;
3318                 data->res.fattr = NULL;
3319         } else
3320                 data->args.bitmask = server->cache_consistency_bitmask;
3321         if (!data->write_done_cb)
3322                 data->write_done_cb = nfs4_write_done_cb;
3323         data->res.server = server;
3324         data->timestamp   = jiffies;
3325
3326         msg->rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_WRITE];
3327 }
3328
3329 static int nfs4_commit_done_cb(struct rpc_task *task, struct nfs_write_data *data)
3330 {
3331         struct inode *inode = data->inode;
3332
3333         if (nfs4_async_handle_error(task, NFS_SERVER(inode), NULL) == -EAGAIN) {
3334                 rpc_restart_call_prepare(task);
3335                 return -EAGAIN;
3336         }
3337         nfs_refresh_inode(inode, data->res.fattr);
3338         return 0;
3339 }
3340
3341 static int nfs4_commit_done(struct rpc_task *task, struct nfs_write_data *data)
3342 {
3343         if (!nfs4_sequence_done(task, &data->res.seq_res))
3344                 return -EAGAIN;
3345         return data->write_done_cb(task, data);
3346 }
3347
3348 static void nfs4_proc_commit_setup(struct nfs_write_data *data, struct rpc_message *msg)
3349 {
3350         struct nfs_server *server = NFS_SERVER(data->inode);
3351
3352         if (data->lseg) {
3353                 data->args.bitmask = NULL;
3354                 data->res.fattr = NULL;
3355         } else
3356                 data->args.bitmask = server->cache_consistency_bitmask;
3357         if (!data->write_done_cb)
3358                 data->write_done_cb = nfs4_commit_done_cb;
3359         data->res.server = server;
3360         msg->rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_COMMIT];
3361 }
3362
3363 struct nfs4_renewdata {
3364         struct nfs_client       *client;
3365         unsigned long           timestamp;
3366 };
3367
3368 /*
3369  * nfs4_proc_async_renew(): This is not one of the nfs_rpc_ops; it is a special
3370  * standalone procedure for queueing an asynchronous RENEW.
3371  */
3372 static void nfs4_renew_release(void *calldata)
3373 {
3374         struct nfs4_renewdata *data = calldata;
3375         struct nfs_client *clp = data->client;
3376
3377         if (atomic_read(&clp->cl_count) > 1)
3378                 nfs4_schedule_state_renewal(clp);
3379         nfs_put_client(clp);
3380         kfree(data);
3381 }
3382
3383 static void nfs4_renew_done(struct rpc_task *task, void *calldata)
3384 {
3385         struct nfs4_renewdata *data = calldata;
3386         struct nfs_client *clp = data->client;
3387         unsigned long timestamp = data->timestamp;
3388
3389         if (task->tk_status < 0) {
3390                 /* Unless we're shutting down, schedule state recovery! */
3391                 if (test_bit(NFS_CS_RENEWD, &clp->cl_res_state) == 0)
3392                         return;
3393                 if (task->tk_status != NFS4ERR_CB_PATH_DOWN) {
3394                         nfs4_schedule_lease_recovery(clp);
3395                         return;
3396                 }
3397                 nfs4_schedule_path_down_recovery(clp);
3398         }
3399         do_renew_lease(clp, timestamp);
3400 }
3401
3402 static const struct rpc_call_ops nfs4_renew_ops = {
3403         .rpc_call_done = nfs4_renew_done,
3404         .rpc_release = nfs4_renew_release,
3405 };
3406
3407 static int nfs4_proc_async_renew(struct nfs_client *clp, struct rpc_cred *cred, unsigned renew_flags)
3408 {
3409         struct rpc_message msg = {
3410                 .rpc_proc       = &nfs4_procedures[NFSPROC4_CLNT_RENEW],
3411                 .rpc_argp       = clp,
3412                 .rpc_cred       = cred,
3413         };
3414         struct nfs4_renewdata *data;
3415
3416         if (renew_flags == 0)
3417                 return 0;
3418         if (!atomic_inc_not_zero(&clp->cl_count))
3419                 return -EIO;
3420         data = kmalloc(sizeof(*data), GFP_NOFS);
3421         if (data == NULL)
3422                 return -ENOMEM;
3423         data->client = clp;
3424         data->timestamp = jiffies;
3425         return rpc_call_async(clp->cl_rpcclient, &msg, RPC_TASK_SOFT,
3426                         &nfs4_renew_ops, data);
3427 }
3428
3429 static int nfs4_proc_renew(struct nfs_client *clp, struct rpc_cred *cred)
3430 {
3431         struct rpc_message msg = {
3432                 .rpc_proc       = &nfs4_procedures[NFSPROC4_CLNT_RENEW],
3433                 .rpc_argp       = clp,
3434                 .rpc_cred       = cred,
3435         };
3436         unsigned long now = jiffies;
3437         int status;
3438
3439         status = rpc_call_sync(clp->cl_rpcclient, &msg, 0);
3440         if (status < 0)
3441                 return status;
3442         do_renew_lease(clp, now);
3443         return 0;
3444 }
3445
3446 static inline int nfs4_server_supports_acls(struct nfs_server *server)
3447 {
3448         return (server->caps & NFS_CAP_ACLS)
3449                 && (server->acl_bitmask & ACL4_SUPPORT_ALLOW_ACL)
3450                 && (server->acl_bitmask & ACL4_SUPPORT_DENY_ACL);
3451 }
3452
3453 /* Assuming that XATTR_SIZE_MAX is a multiple of PAGE_CACHE_SIZE, and that
3454  * it's OK to put sizeof(void) * (XATTR_SIZE_MAX/PAGE_CACHE_SIZE) bytes on
3455  * the stack.
3456  */
3457 #define NFS4ACL_MAXPAGES (XATTR_SIZE_MAX >> PAGE_CACHE_SHIFT)
3458
3459 static int buf_to_pages_noslab(const void *buf, size_t buflen,
3460                 struct page **pages, unsigned int *pgbase)
3461 {
3462         struct page *newpage, **spages;
3463         int rc = 0;
3464         size_t len;
3465         spages = pages;
3466
3467         do {
3468                 len = min_t(size_t, PAGE_CACHE_SIZE, buflen);
3469                 newpage = alloc_page(GFP_KERNEL);
3470
3471                 if (newpage == NULL)
3472                         goto unwind;
3473                 memcpy(page_address(newpage), buf, len);
3474                 buf += len;
3475                 buflen -= len;
3476                 *pages++ = newpage;
3477                 rc++;
3478         } while (buflen != 0);
3479
3480         return rc;
3481
3482 unwind:
3483         for(; rc > 0; rc--)
3484                 __free_page(spages[rc-1]);
3485         return -ENOMEM;
3486 }
3487
3488 struct nfs4_cached_acl {
3489         int cached;
3490         size_t len;
3491         char data[0];
3492 };
3493
3494 static void nfs4_set_cached_acl(struct inode *inode, struct nfs4_cached_acl *acl)
3495 {
3496         struct nfs_inode *nfsi = NFS_I(inode);
3497
3498         spin_lock(&inode->i_lock);
3499         kfree(nfsi->nfs4_acl);
3500         nfsi->nfs4_acl = acl;
3501         spin_unlock(&inode->i_lock);
3502 }
3503
3504 static void nfs4_zap_acl_attr(struct inode *inode)
3505 {
3506         nfs4_set_cached_acl(inode, NULL);
3507 }
3508
3509 static inline ssize_t nfs4_read_cached_acl(struct inode *inode, char *buf, size_t buflen)
3510 {
3511         struct nfs_inode *nfsi = NFS_I(inode);
3512         struct nfs4_cached_acl *acl;
3513         int ret = -ENOENT;
3514
3515         spin_lock(&inode->i_lock);
3516         acl = nfsi->nfs4_acl;
3517         if (acl == NULL)
3518                 goto out;
3519         if (buf == NULL) /* user is just asking for length */
3520                 goto out_len;
3521         if (acl->cached == 0)
3522                 goto out;
3523         ret = -ERANGE; /* see getxattr(2) man page */
3524         if (acl->len > buflen)
3525                 goto out;
3526         memcpy(buf, acl->data, acl->len);
3527 out_len:
3528         ret = acl->len;
3529 out:
3530         spin_unlock(&inode->i_lock);
3531         return ret;
3532 }
3533
3534 static void nfs4_write_cached_acl(struct inode *inode, struct page **pages, size_t pgbase, size_t acl_len)
3535 {
3536         struct nfs4_cached_acl *acl;
3537
3538         if (pages && acl_len <= PAGE_SIZE) {
3539                 acl = kmalloc(sizeof(*acl) + acl_len, GFP_KERNEL);
3540                 if (acl == NULL)
3541                         goto out;
3542                 acl->cached = 1;
3543                 _copy_from_pages(acl->data, pages, pgbase, acl_len);
3544         } else {
3545                 acl = kmalloc(sizeof(*acl), GFP_KERNEL);
3546                 if (acl == NULL)
3547                         goto out;
3548                 acl->cached = 0;
3549         }
3550         acl->len = acl_len;
3551 out:
3552         nfs4_set_cached_acl(inode, acl);
3553 }
3554
3555 /*
3556  * The getxattr API returns the required buffer length when called with a
3557  * NULL buf. The NFSv4 acl tool then calls getxattr again after allocating
3558  * the required buf.  On a NULL buf, we send a page of data to the server
3559  * guessing that the ACL request can be serviced by a page. If so, we cache
3560  * up to the page of ACL data, and the 2nd call to getxattr is serviced by
3561  * the cache. If not so, we throw away the page, and cache the required
3562  * length. The next getxattr call will then produce another round trip to
3563  * the server, this time with the input buf of the required size.
3564  */
3565 static ssize_t __nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
3566 {
3567         struct page *pages[NFS4ACL_MAXPAGES] = {NULL, };
3568         struct nfs_getaclargs args = {
3569                 .fh = NFS_FH(inode),
3570                 .acl_pages = pages,
3571                 .acl_len = buflen,
3572         };
3573         struct nfs_getaclres res = {
3574                 .acl_len = buflen,
3575         };
3576         struct rpc_message msg = {
3577                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETACL],
3578                 .rpc_argp = &args,
3579                 .rpc_resp = &res,
3580         };
3581         int ret = -ENOMEM, npages, i;
3582         size_t acl_len = 0;
3583
3584         npages = (buflen + PAGE_SIZE - 1) >> PAGE_SHIFT;
3585         /* As long as we're doing a round trip to the server anyway,
3586          * let's be prepared for a page of acl data. */
3587         if (npages == 0)
3588                 npages = 1;
3589
3590         /* Add an extra page to handle the bitmap returned */
3591         npages++;
3592
3593         for (i = 0; i < npages; i++) {
3594                 pages[i] = alloc_page(GFP_KERNEL);
3595                 if (!pages[i])
3596                         goto out_free;
3597         }
3598
3599         /* for decoding across pages */
3600         res.acl_scratch = alloc_page(GFP_KERNEL);
3601         if (!res.acl_scratch)
3602                 goto out_free;
3603
3604         args.acl_len = npages * PAGE_SIZE;
3605         args.acl_pgbase = 0;
3606
3607         /* Let decode_getfacl know not to fail if the ACL data is larger than
3608          * the page we send as a guess */
3609         if (buf == NULL)
3610                 res.acl_flags |= NFS4_ACL_LEN_REQUEST;
3611
3612         dprintk("%s  buf %p buflen %zu npages %d args.acl_len %zu\n",
3613                 __func__, buf, buflen, npages, args.acl_len);
3614         ret = nfs4_call_sync(NFS_SERVER(inode)->client, NFS_SERVER(inode),
3615                              &msg, &args.seq_args, &res.seq_res, 0);
3616         if (ret)
3617                 goto out_free;
3618
3619         acl_len = res.acl_len - res.acl_data_offset;
3620         if (acl_len > args.acl_len)
3621                 nfs4_write_cached_acl(inode, NULL, 0, acl_len);
3622         else
3623                 nfs4_write_cached_acl(inode, pages, res.acl_data_offset,
3624                                       acl_len);
3625         if (buf) {
3626                 ret = -ERANGE;
3627                 if (acl_len > buflen)
3628                         goto out_free;
3629                 _copy_from_pages(buf, pages, res.acl_data_offset,
3630                                 acl_len);
3631         }
3632         ret = acl_len;
3633 out_free:
3634         for (i = 0; i < npages; i++)
3635                 if (pages[i])
3636                         __free_page(pages[i]);
3637         if (res.acl_scratch)
3638                 __free_page(res.acl_scratch);
3639         return ret;
3640 }
3641
3642 static ssize_t nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
3643 {
3644         struct nfs4_exception exception = { };
3645         ssize_t ret;
3646         do {
3647                 ret = __nfs4_get_acl_uncached(inode, buf, buflen);
3648                 if (ret >= 0)
3649                         break;
3650                 ret = nfs4_handle_exception(NFS_SERVER(inode), ret, &exception);
3651         } while (exception.retry);
3652         return ret;
3653 }
3654
3655 static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
3656 {
3657         struct nfs_server *server = NFS_SERVER(inode);
3658         int ret;
3659
3660         if (!nfs4_server_supports_acls(server))
3661                 return -EOPNOTSUPP;
3662         ret = nfs_revalidate_inode(server, inode);
3663         if (ret < 0)
3664                 return ret;
3665         if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_ACL)
3666                 nfs_zap_acl_cache(inode);
3667         ret = nfs4_read_cached_acl(inode, buf, buflen);
3668         if (ret != -ENOENT)
3669                 /* -ENOENT is returned if there is no ACL or if there is an ACL
3670                  * but no cached acl data, just the acl length */
3671                 return ret;
3672         return nfs4_get_acl_uncached(inode, buf, buflen);
3673 }
3674
3675 static int __nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen)
3676 {
3677         struct nfs_server *server = NFS_SERVER(inode);
3678         struct page *pages[NFS4ACL_MAXPAGES];
3679         struct nfs_setaclargs arg = {
3680                 .fh             = NFS_FH(inode),
3681                 .acl_pages      = pages,
3682                 .acl_len        = buflen,
3683         };
3684         struct nfs_setaclres res;
3685         struct rpc_message msg = {
3686                 .rpc_proc       = &nfs4_procedures[NFSPROC4_CLNT_SETACL],
3687                 .rpc_argp       = &arg,
3688                 .rpc_resp       = &res,
3689         };
3690         int ret, i;
3691
3692         if (!nfs4_server_supports_acls(server))
3693                 return -EOPNOTSUPP;
3694         i = buf_to_pages_noslab(buf, buflen, arg.acl_pages, &arg.acl_pgbase);
3695         if (i < 0)
3696                 return i;
3697         nfs_inode_return_delegation(inode);
3698         ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args, &res.seq_res, 1);
3699
3700         /*
3701          * Free each page after tx, so the only ref left is
3702          * held by the network stack
3703          */
3704         for (; i > 0; i--)
3705                 put_page(pages[i-1]);
3706
3707         /*
3708          * Acl update can result in inode attribute update.
3709          * so mark the attribute cache invalid.
3710          */
3711         spin_lock(&inode->i_lock);
3712         NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
3713         spin_unlock(&inode->i_lock);
3714         nfs_access_zap_cache(inode);
3715         nfs_zap_acl_cache(inode);
3716         return ret;
3717 }
3718
3719 static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen)
3720 {
3721         struct nfs4_exception exception = { };
3722         int err;
3723         do {
3724                 err = nfs4_handle_exception(NFS_SERVER(inode),
3725                                 __nfs4_proc_set_acl(inode, buf, buflen),
3726                                 &exception);
3727         } while (exception.retry);
3728         return err;
3729 }
3730
3731 static int
3732 nfs4_async_handle_error(struct rpc_task *task, const struct nfs_server *server, struct nfs4_state *state)
3733 {
3734         struct nfs_client *clp = server->nfs_client;
3735
3736         if (task->tk_status >= 0)
3737                 return 0;
3738         switch(task->tk_status) {
3739                 case -NFS4ERR_DELEG_REVOKED:
3740                 case -NFS4ERR_ADMIN_REVOKED:
3741                 case -NFS4ERR_BAD_STATEID:
3742                         if (state != NULL)
3743                                 nfs_remove_bad_delegation(state->inode);
3744                 case -NFS4ERR_OPENMODE:
3745                         if (state == NULL)
3746                                 break;
3747                         nfs4_schedule_stateid_recovery(server, state);
3748                      &nbs