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/nfs.h>
  47 #include <linux/nfs4.h>
  48 #include <linux/nfs_fs.h>
  49 #include <linux/nfs_page.h>
  50 #include <linux/nfs_mount.h>
  51 #include <linux/namei.h>
  52 #include <linux/mount.h>
  53 #include <linux/module.h>
  54 #include <linux/nfs_idmap.h>
  55 #include <linux/xattr.h>
  56 #include <linux/utsname.h>
  57 #include <linux/freezer.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 #include "netns.h"
  66 #include "nfs4session.h"
  67 #include "fscache.h"
  68
  69 #include "nfs4trace.h"
  70
  71 #define NFSDBG_FACILITY         NFSDBG_PROC
  72
  73 #define NFS4_POLL_RETRY_MIN     (HZ/10)
  74 #define NFS4_POLL_RETRY_MAX     (15*HZ)
  75
  76 struct nfs4_opendata;
  77 static int _nfs4_proc_open(struct nfs4_opendata *data);
  78 static int _nfs4_recover_proc_open(struct nfs4_opendata *data);
  79 static int nfs4_do_fsinfo(struct nfs_server *, struct nfs_fh *, struct nfs_fsinfo *);
  80 static int nfs4_async_handle_error(struct rpc_task *, const struct nfs_server *, struct nfs4_state *);
  81 static void nfs_fixup_referral_attributes(struct nfs_fattr *fattr);
  82 static int nfs4_proc_getattr(struct nfs_server *, struct nfs_fh *, struct nfs_fattr *, struct nfs4_label *label);
  83 static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fattr *fattr, struct nfs4_label *label);
  84 static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
  85                             struct nfs_fattr *fattr, struct iattr *sattr,
  86                             struct nfs4_state *state, struct nfs4_label *ilabel,
  87                             struct nfs4_label *olabel);
  88 #ifdef CONFIG_NFS_V4_1
  89 static int nfs41_test_stateid(struct nfs_server *, nfs4_stateid *,
  90                 struct rpc_cred *);
  91 static int nfs41_free_stateid(struct nfs_server *, nfs4_stateid *,
  92                 struct rpc_cred *);
  93 #endif
  94
  95 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
  96 static inline struct nfs4_label *
  97 nfs4_label_init_security(struct inode *dir, struct dentry *dentry,
  98         struct iattr *sattr, struct nfs4_label *label)
  99 {
 100         int err;
 101
 102         if (label == NULL)
 103                 return NULL;
 104
 105         if (nfs_server_capable(dir, NFS_CAP_SECURITY_LABEL) == 0)
 106                 return NULL;
 107
 108         err = security_dentry_init_security(dentry, sattr->ia_mode,
 109                                 &dentry->d_name, (void **)&label->label, &label->len);
 110         if (err == 0)
 111                 return label;
 112
 113         return NULL;
 114 }
 115 static inline void
 116 nfs4_label_release_security(struct nfs4_label *label)
 117 {
 118         if (label)
 119                 security_release_secctx(label->label, label->len);
 120 }
 121 static inline u32 *nfs4_bitmask(struct nfs_server *server, struct nfs4_label *label)
 122 {
 123         if (label)
 124                 return server->attr_bitmask;
 125
 126         return server->attr_bitmask_nl;
 127 }
 128 #else
 129 static inline struct nfs4_label *
 130 nfs4_label_init_security(struct inode *dir, struct dentry *dentry,
 131         struct iattr *sattr, struct nfs4_label *l)
 132 { return NULL; }
 133 static inline void
 134 nfs4_label_release_security(struct nfs4_label *label)
 135 { return; }
 136 static inline u32 *
 137 nfs4_bitmask(struct nfs_server *server, struct nfs4_label *label)
 138 { return server->attr_bitmask; }
 139 #endif
 140
 141 /* Prevent leaks of NFSv4 errors into userland */
 142 static int nfs4_map_errors(int err)
 143 {
 144         if (err >= -1000)
 145                 return err;
 146         switch (err) {
 147         case -NFS4ERR_RESOURCE:
 148         case -NFS4ERR_LAYOUTTRYLATER:
 149         case -NFS4ERR_RECALLCONFLICT:
 150                 return -EREMOTEIO;
 151         case -NFS4ERR_WRONGSEC:
 152         case -NFS4ERR_WRONG_CRED:
 153                 return -EPERM;
 154         case -NFS4ERR_BADOWNER:
 155         case -NFS4ERR_BADNAME:
 156                 return -EINVAL;
 157         case -NFS4ERR_SHARE_DENIED:
 158                 return -EACCES;
 159         case -NFS4ERR_MINOR_VERS_MISMATCH:
 160                 return -EPROTONOSUPPORT;
 161         case -NFS4ERR_ACCESS:
 162                 return -EACCES;
 163         case -NFS4ERR_FILE_OPEN:
 164                 return -EBUSY;
 165         default:
 166                 dprintk("%s could not handle NFSv4 error %d\n",
 167                                 __func__, -err);
 168                 break;
 169         }
 170         return -EIO;
 171 }
 172
 173 /*
 174  * This is our standard bitmap for GETATTR requests.
 175  */
 176 const u32 nfs4_fattr_bitmap[3] = {
 177         FATTR4_WORD0_TYPE
 178         | FATTR4_WORD0_CHANGE
 179         | FATTR4_WORD0_SIZE
 180         | FATTR4_WORD0_FSID
 181         | FATTR4_WORD0_FILEID,
 182         FATTR4_WORD1_MODE
 183         | FATTR4_WORD1_NUMLINKS
 184         | FATTR4_WORD1_OWNER
 185         | FATTR4_WORD1_OWNER_GROUP
 186         | FATTR4_WORD1_RAWDEV
 187         | FATTR4_WORD1_SPACE_USED
 188         | FATTR4_WORD1_TIME_ACCESS
 189         | FATTR4_WORD1_TIME_METADATA
 190         | FATTR4_WORD1_TIME_MODIFY,
 191 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
 192         FATTR4_WORD2_SECURITY_LABEL
 193 #endif
 194 };
 195
 196 static const u32 nfs4_pnfs_open_bitmap[3] = {
 197         FATTR4_WORD0_TYPE
 198         | FATTR4_WORD0_CHANGE
 199         | FATTR4_WORD0_SIZE
 200         | FATTR4_WORD0_FSID
 201         | FATTR4_WORD0_FILEID,
 202         FATTR4_WORD1_MODE
 203         | FATTR4_WORD1_NUMLINKS
 204         | FATTR4_WORD1_OWNER
 205         | FATTR4_WORD1_OWNER_GROUP
 206         | FATTR4_WORD1_RAWDEV
 207         | FATTR4_WORD1_SPACE_USED
 208         | FATTR4_WORD1_TIME_ACCESS
 209         | FATTR4_WORD1_TIME_METADATA
 210         | FATTR4_WORD1_TIME_MODIFY,
 211         FATTR4_WORD2_MDSTHRESHOLD
 212 };
 213
 214 static const u32 nfs4_open_noattr_bitmap[3] = {
 215         FATTR4_WORD0_TYPE
 216         | FATTR4_WORD0_CHANGE
 217         | FATTR4_WORD0_FILEID,
 218 };
 219
 220 const u32 nfs4_statfs_bitmap[3] = {
 221         FATTR4_WORD0_FILES_AVAIL
 222         | FATTR4_WORD0_FILES_FREE
 223         | FATTR4_WORD0_FILES_TOTAL,
 224         FATTR4_WORD1_SPACE_AVAIL
 225         | FATTR4_WORD1_SPACE_FREE
 226         | FATTR4_WORD1_SPACE_TOTAL
 227 };
 228
 229 const u32 nfs4_pathconf_bitmap[3] = {
 230         FATTR4_WORD0_MAXLINK
 231         | FATTR4_WORD0_MAXNAME,
 232         0
 233 };
 234
 235 const u32 nfs4_fsinfo_bitmap[3] = { FATTR4_WORD0_MAXFILESIZE
 236                         | FATTR4_WORD0_MAXREAD
 237                         | FATTR4_WORD0_MAXWRITE
 238                         | FATTR4_WORD0_LEASE_TIME,
 239                         FATTR4_WORD1_TIME_DELTA
 240                         | FATTR4_WORD1_FS_LAYOUT_TYPES,
 241                         FATTR4_WORD2_LAYOUT_BLKSIZE
 242 };
 243
 244 const u32 nfs4_fs_locations_bitmap[3] = {
 245         FATTR4_WORD0_TYPE
 246         | FATTR4_WORD0_CHANGE
 247         | FATTR4_WORD0_SIZE
 248         | FATTR4_WORD0_FSID
 249         | FATTR4_WORD0_FILEID
 250         | FATTR4_WORD0_FS_LOCATIONS,
 251         FATTR4_WORD1_MODE
 252         | FATTR4_WORD1_NUMLINKS
 253         | FATTR4_WORD1_OWNER
 254         | FATTR4_WORD1_OWNER_GROUP
 255         | FATTR4_WORD1_RAWDEV
 256         | FATTR4_WORD1_SPACE_USED
 257         | FATTR4_WORD1_TIME_ACCESS
 258         | FATTR4_WORD1_TIME_METADATA
 259         | FATTR4_WORD1_TIME_MODIFY
 260         | FATTR4_WORD1_MOUNTED_ON_FILEID,
 261 };
 262
 263 static void nfs4_setup_readdir(u64 cookie, __be32 *verifier, struct dentry *dentry,
 264                 struct nfs4_readdir_arg *readdir)
 265 {
 266         __be32 *start, *p;
 267
 268         if (cookie > 2) {
 269                 readdir->cookie = cookie;
 270                 memcpy(&readdir->verifier, verifier, sizeof(readdir->verifier));
 271                 return;
 272         }
 273
 274         readdir->cookie = 0;
 275         memset(&readdir->verifier, 0, sizeof(readdir->verifier));
 276         if (cookie == 2)
 277                 return;
 278
 279         /*
 280          * NFSv4 servers do not return entries for '.' and '..'
 281          * Therefore, we fake these entries here.  We let '.'
 282          * have cookie 0 and '..' have cookie 1.  Note that
 283          * when talking to the server, we always send cookie 0
 284          * instead of 1 or 2.
 285          */
 286         start = p = kmap_atomic(*readdir->pages);
 287
 288         if (cookie == 0) {
 289                 *p++ = xdr_one;                                  /* next */
 290                 *p++ = xdr_zero;                   /* cookie, first word */
 291                 *p++ = xdr_one;                   /* cookie, second word */
 292                 *p++ = xdr_one;                             /* entry len */
 293                 memcpy(p, ".\0\0\0", 4);                        /* entry */
 294                 p++;
 295                 *p++ = xdr_one;                         /* bitmap length */
 296                 *p++ = htonl(FATTR4_WORD0_FILEID);             /* bitmap */
 297                 *p++ = htonl(8);              /* attribute buffer length */
 298                 p = xdr_encode_hyper(p, NFS_FILEID(dentry->d_inode));
 299         }
 300
 301         *p++ = xdr_one;                                  /* next */
 302         *p++ = xdr_zero;                   /* cookie, first word */
 303         *p++ = xdr_two;                   /* cookie, second word */
 304         *p++ = xdr_two;                             /* entry len */
 305         memcpy(p, "..\0\0", 4);                         /* entry */
 306         p++;
 307         *p++ = xdr_one;                         /* bitmap length */
 308         *p++ = htonl(FATTR4_WORD0_FILEID);             /* bitmap */
 309         *p++ = htonl(8);              /* attribute buffer length */
 310         p = xdr_encode_hyper(p, NFS_FILEID(dentry->d_parent->d_inode));
 311
 312         readdir->pgbase = (char *)p - (char *)start;
 313         readdir->count -= readdir->pgbase;
 314         kunmap_atomic(start);
 315 }
 316
 317 static int nfs4_delay(struct rpc_clnt *clnt, long *timeout)
 318 {
 319         int res = 0;
 320
 321         might_sleep();
 322
 323         if (*timeout <= 0)
 324                 *timeout = NFS4_POLL_RETRY_MIN;
 325         if (*timeout > NFS4_POLL_RETRY_MAX)
 326                 *timeout = NFS4_POLL_RETRY_MAX;
 327         freezable_schedule_timeout_killable_unsafe(*timeout);
 328         if (fatal_signal_pending(current))
 329                 res = -ERESTARTSYS;
 330         *timeout <<= 1;
 331         return res;
 332 }
 333
 334 /* This is the error handling routine for processes that are allowed
 335  * to sleep.
 336  */
 337 static int nfs4_handle_exception(struct nfs_server *server, int errorcode, struct nfs4_exception *exception)
 338 {
 339         struct nfs_client *clp = server->nfs_client;
 340         struct nfs4_state *state = exception->state;
 341         struct inode *inode = exception->inode;
 342         int ret = errorcode;
 343
 344         exception->retry = 0;
 345         switch(errorcode) {
 346                 case 0:
 347                         return 0;
 348                 case -NFS4ERR_OPENMODE:
 349                         if (inode && nfs4_have_delegation(inode, FMODE_READ)) {
 350                                 nfs4_inode_return_delegation(inode);
 351                                 exception->retry = 1;
 352                                 return 0;
 353                         }
 354                         if (state == NULL)
 355                                 break;
 356                         ret = nfs4_schedule_stateid_recovery(server, state);
 357                         if (ret < 0)
 358                                 break;
 359                         goto wait_on_recovery;
 360                 case -NFS4ERR_DELEG_REVOKED:
 361                 case -NFS4ERR_ADMIN_REVOKED:
 362                 case -NFS4ERR_BAD_STATEID:
 363                         if (inode != NULL && nfs4_have_delegation(inode, FMODE_READ)) {
 364                                 nfs_remove_bad_delegation(inode);
 365                                 exception->retry = 1;
 366                                 break;
 367                         }
 368                         if (state == NULL)
 369                                 break;
 370                         ret = nfs4_schedule_stateid_recovery(server, state);
 371                         if (ret < 0)
 372                                 break;
 373                         goto wait_on_recovery;
 374                 case -NFS4ERR_EXPIRED:
 375                         if (state != NULL) {
 376                                 ret = nfs4_schedule_stateid_recovery(server, state);
 377                                 if (ret < 0)
 378                                         break;
 379                         }
 380                 case -NFS4ERR_STALE_STATEID:
 381                 case -NFS4ERR_STALE_CLIENTID:
 382                         nfs4_schedule_lease_recovery(clp);
 383                         goto wait_on_recovery;
 384                 case -NFS4ERR_MOVED:
 385                         ret = nfs4_schedule_migration_recovery(server);
 386                         if (ret < 0)
 387                                 break;
 388                         goto wait_on_recovery;
 389                 case -NFS4ERR_LEASE_MOVED:
 390                         nfs4_schedule_lease_moved_recovery(clp);
 391                         goto wait_on_recovery;
 392 #if defined(CONFIG_NFS_V4_1)
 393                 case -NFS4ERR_BADSESSION:
 394                 case -NFS4ERR_BADSLOT:
 395                 case -NFS4ERR_BAD_HIGH_SLOT:
 396                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 397                 case -NFS4ERR_DEADSESSION:
 398                 case -NFS4ERR_SEQ_FALSE_RETRY:
 399                 case -NFS4ERR_SEQ_MISORDERED:
 400                         dprintk("%s ERROR: %d Reset session\n", __func__,
 401                                 errorcode);
 402                         nfs4_schedule_session_recovery(clp->cl_session, errorcode);
 403                         goto wait_on_recovery;
 404 #endif /* defined(CONFIG_NFS_V4_1) */
 405                 case -NFS4ERR_FILE_OPEN:
 406                         if (exception->timeout > HZ) {
 407                                 /* We have retried a decent amount, time to
 408                                  * fail
 409                                  */
 410                                 ret = -EBUSY;
 411                                 break;
 412                         }
 413                 case -NFS4ERR_GRACE:
 414                 case -NFS4ERR_DELAY:
 415                         ret = nfs4_delay(server->client, &exception->timeout);
 416                         if (ret != 0)
 417                                 break;
 418                 case -NFS4ERR_RETRY_UNCACHED_REP:
 419                 case -NFS4ERR_OLD_STATEID:
 420                         exception->retry = 1;
 421                         break;
 422                 case -NFS4ERR_BADOWNER:
 423                         /* The following works around a Linux server bug! */
 424                 case -NFS4ERR_BADNAME:
 425                         if (server->caps & NFS_CAP_UIDGID_NOMAP) {
 426                                 server->caps &= ~NFS_CAP_UIDGID_NOMAP;
 427                                 exception->retry = 1;
 428                                 printk(KERN_WARNING "NFS: v4 server %s "
 429                                                 "does not accept raw "
 430                                                 "uid/gids. "
 431                                                 "Reenabling the idmapper.\n",
 432                                                 server->nfs_client->cl_hostname);
 433                         }
 434         }
 435         /* We failed to handle the error */
 436         return nfs4_map_errors(ret);
 437 wait_on_recovery:
 438         ret = nfs4_wait_clnt_recover(clp);
 439         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
 440                 return -EIO;
 441         if (ret == 0)
 442                 exception->retry = 1;
 443         return ret;
 444 }
 445
 446 /*
 447  * Return 'true' if 'clp' is using an rpc_client that is integrity protected
 448  * or 'false' otherwise.
 449  */
 450 static bool _nfs4_is_integrity_protected(struct nfs_client *clp)
 451 {
 452         rpc_authflavor_t flavor = clp->cl_rpcclient->cl_auth->au_flavor;
 453
 454         if (flavor == RPC_AUTH_GSS_KRB5I ||
 455             flavor == RPC_AUTH_GSS_KRB5P)
 456                 return true;
 457
 458         return false;
 459 }
 460
 461 static void do_renew_lease(struct nfs_client *clp, unsigned long timestamp)
 462 {
 463         spin_lock(&clp->cl_lock);
 464         if (time_before(clp->cl_last_renewal,timestamp))
 465                 clp->cl_last_renewal = timestamp;
 466         spin_unlock(&clp->cl_lock);
 467 }
 468
 469 static void renew_lease(const struct nfs_server *server, unsigned long timestamp)
 470 {
 471         do_renew_lease(server->nfs_client, timestamp);
 472 }
 473
 474 struct nfs4_call_sync_data {
 475         const struct nfs_server *seq_server;
 476         struct nfs4_sequence_args *seq_args;
 477         struct nfs4_sequence_res *seq_res;
 478 };
 479
 480 static void nfs4_init_sequence(struct nfs4_sequence_args *args,
 481                                struct nfs4_sequence_res *res, int cache_reply)
 482 {
 483         args->sa_slot = NULL;
 484         args->sa_cache_this = cache_reply;
 485         args->sa_privileged = 0;
 486
 487         res->sr_slot = NULL;
 488 }
 489
 490 static void nfs4_set_sequence_privileged(struct nfs4_sequence_args *args)
 491 {
 492         args->sa_privileged = 1;
 493 }
 494
 495 static int nfs40_setup_sequence(const struct nfs_server *server,
 496                                 struct nfs4_sequence_args *args,
 497                                 struct nfs4_sequence_res *res,
 498                                 struct rpc_task *task)
 499 {
 500         struct nfs4_slot_table *tbl = server->nfs_client->cl_slot_tbl;
 501         struct nfs4_slot *slot;
 502
 503         /* slot already allocated? */
 504         if (res->sr_slot != NULL)
 505                 goto out_start;
 506
 507         spin_lock(&tbl->slot_tbl_lock);
 508         if (nfs4_slot_tbl_draining(tbl) && !args->sa_privileged)
 509                 goto out_sleep;
 510
 511         slot = nfs4_alloc_slot(tbl);
 512         if (IS_ERR(slot)) {
 513                 if (slot == ERR_PTR(-ENOMEM))
 514                         task->tk_timeout = HZ >> 2;
 515                 goto out_sleep;
 516         }
 517         spin_unlock(&tbl->slot_tbl_lock);
 518
 519         args->sa_slot = slot;
 520         res->sr_slot = slot;
 521
 522 out_start:
 523         rpc_call_start(task);
 524         return 0;
 525
 526 out_sleep:
 527         if (args->sa_privileged)
 528                 rpc_sleep_on_priority(&tbl->slot_tbl_waitq, task,
 529                                 NULL, RPC_PRIORITY_PRIVILEGED);
 530         else
 531                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
 532         spin_unlock(&tbl->slot_tbl_lock);
 533         return -EAGAIN;
 534 }
 535
 536 static int nfs40_sequence_done(struct rpc_task *task,
 537                                struct nfs4_sequence_res *res)
 538 {
 539         struct nfs4_slot *slot = res->sr_slot;
 540         struct nfs4_slot_table *tbl;
 541
 542         if (slot == NULL)
 543                 goto out;
 544
 545         tbl = slot->table;
 546         spin_lock(&tbl->slot_tbl_lock);
 547         if (!nfs41_wake_and_assign_slot(tbl, slot))
 548                 nfs4_free_slot(tbl, slot);
 549         spin_unlock(&tbl->slot_tbl_lock);
 550
 551         res->sr_slot = NULL;
 552 out:
 553         return 1;
 554 }
 555
 556 #if defined(CONFIG_NFS_V4_1)
 557
 558 static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 559 {
 560         struct nfs4_session *session;
 561         struct nfs4_slot_table *tbl;
 562         struct nfs4_slot *slot = res->sr_slot;
 563         bool send_new_highest_used_slotid = false;
 564
 565         tbl = slot->table;
 566         session = tbl->session;
 567
 568         spin_lock(&tbl->slot_tbl_lock);
 569         /* Be nice to the server: try to ensure that the last transmitted
 570          * value for highest_user_slotid <= target_highest_slotid
 571          */
 572         if (tbl->highest_used_slotid > tbl->target_highest_slotid)
 573                 send_new_highest_used_slotid = true;
 574
 575         if (nfs41_wake_and_assign_slot(tbl, slot)) {
 576                 send_new_highest_used_slotid = false;
 577                 goto out_unlock;
 578         }
 579         nfs4_free_slot(tbl, slot);
 580
 581         if (tbl->highest_used_slotid != NFS4_NO_SLOT)
 582                 send_new_highest_used_slotid = false;
 583 out_unlock:
 584         spin_unlock(&tbl->slot_tbl_lock);
 585         res->sr_slot = NULL;
 586         if (send_new_highest_used_slotid)
 587                 nfs41_server_notify_highest_slotid_update(session->clp);
 588 }
 589
 590 int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
 591 {
 592         struct nfs4_session *session;
 593         struct nfs4_slot *slot = res->sr_slot;
 594         struct nfs_client *clp;
 595         bool interrupted = false;
 596         int ret = 1;
 597
 598         if (slot == NULL)
 599                 goto out_noaction;
 600         /* don't increment the sequence number if the task wasn't sent */
 601         if (!RPC_WAS_SENT(task))
 602                 goto out;
 603
 604         session = slot->table->session;
 605
 606         if (slot->interrupted) {
 607                 slot->interrupted = 0;
 608                 interrupted = true;
 609         }
 610
 611         trace_nfs4_sequence_done(session, res);
 612         /* Check the SEQUENCE operation status */
 613         switch (res->sr_status) {
 614         case 0:
 615                 /* Update the slot's sequence and clientid lease timer */
 616                 ++slot->seq_nr;
 617                 clp = session->clp;
 618                 do_renew_lease(clp, res->sr_timestamp);
 619                 /* Check sequence flags */
 620                 if (res->sr_status_flags != 0)
 621                         nfs4_schedule_lease_recovery(clp);
 622                 nfs41_update_target_slotid(slot->table, slot, res);
 623                 break;
 624         case 1:
 625                 /*
 626                  * sr_status remains 1 if an RPC level error occurred.
 627                  * The server may or may not have processed the sequence
 628                  * operation..
 629                  * Mark the slot as having hosted an interrupted RPC call.
 630                  */
 631                 slot->interrupted = 1;
 632                 goto out;
 633         case -NFS4ERR_DELAY:
 634                 /* The server detected a resend of the RPC call and
 635                  * returned NFS4ERR_DELAY as per Section 2.10.6.2
 636                  * of RFC5661.
 637                  */
 638                 dprintk("%s: slot=%u seq=%u: Operation in progress\n",
 639                         __func__,
 640                         slot->slot_nr,
 641                         slot->seq_nr);
 642                 goto out_retry;
 643         case -NFS4ERR_BADSLOT:
 644                 /*
 645                  * The slot id we used was probably retired. Try again
 646                  * using a different slot id.
 647                  */
 648                 goto retry_nowait;
 649         case -NFS4ERR_SEQ_MISORDERED:
 650                 /*
 651                  * Was the last operation on this sequence interrupted?
 652                  * If so, retry after bumping the sequence number.
 653                  */
 654                 if (interrupted) {
 655                         ++slot->seq_nr;
 656                         goto retry_nowait;
 657                 }
 658                 /*
 659                  * Could this slot have been previously retired?
 660                  * If so, then the server may be expecting seq_nr = 1!
 661                  */
 662                 if (slot->seq_nr != 1) {
 663                         slot->seq_nr = 1;
 664                         goto retry_nowait;
 665                 }
 666                 break;
 667         case -NFS4ERR_SEQ_FALSE_RETRY:
 668                 ++slot->seq_nr;
 669                 goto retry_nowait;
 670         default:
 671                 /* Just update the slot sequence no. */
 672                 ++slot->seq_nr;
 673         }
 674 out:
 675         /* The session may be reset by one of the error handlers. */
 676         dprintk("%s: Error %d free the slot \n", __func__, res->sr_status);
 677         nfs41_sequence_free_slot(res);
 678 out_noaction:
 679         return ret;
 680 retry_nowait:
 681         if (rpc_restart_call_prepare(task)) {
 682                 task->tk_status = 0;
 683                 ret = 0;
 684         }
 685         goto out;
 686 out_retry:
 687         if (!rpc_restart_call(task))
 688                 goto out;
 689         rpc_delay(task, NFS4_POLL_RETRY_MAX);
 690         return 0;
 691 }
 692 EXPORT_SYMBOL_GPL(nfs41_sequence_done);
 693
 694 static int nfs4_sequence_done(struct rpc_task *task,
 695                                struct nfs4_sequence_res *res)
 696 {
 697         if (res->sr_slot == NULL)
 698                 return 1;
 699         if (!res->sr_slot->table->session)
 700                 return nfs40_sequence_done(task, res);
 701         return nfs41_sequence_done(task, res);
 702 }
 703
 704 int nfs41_setup_sequence(struct nfs4_session *session,
 705                                 struct nfs4_sequence_args *args,
 706                                 struct nfs4_sequence_res *res,
 707                                 struct rpc_task *task)
 708 {
 709         struct nfs4_slot *slot;
 710         struct nfs4_slot_table *tbl;
 711
 712         dprintk("--> %s\n", __func__);
 713         /* slot already allocated? */
 714         if (res->sr_slot != NULL)
 715                 goto out_success;
 716
 717         tbl = &session->fc_slot_table;
 718
 719         task->tk_timeout = 0;
 720
 721         spin_lock(&tbl->slot_tbl_lock);
 722         if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state) &&
 723             !args->sa_privileged) {
 724                 /* The state manager will wait until the slot table is empty */
 725                 dprintk("%s session is draining\n", __func__);
 726                 goto out_sleep;
 727         }
 728
 729         slot = nfs4_alloc_slot(tbl);
 730         if (IS_ERR(slot)) {
 731                 /* If out of memory, try again in 1/4 second */
 732                 if (slot == ERR_PTR(-ENOMEM))
 733                         task->tk_timeout = HZ >> 2;
 734                 dprintk("<-- %s: no free slots\n", __func__);
 735                 goto out_sleep;
 736         }
 737         spin_unlock(&tbl->slot_tbl_lock);
 738
 739         args->sa_slot = slot;
 740
 741         dprintk("<-- %s slotid=%u seqid=%u\n", __func__,
 742                         slot->slot_nr, slot->seq_nr);
 743
 744         res->sr_slot = slot;
 745         res->sr_timestamp = jiffies;
 746         res->sr_status_flags = 0;
 747         /*
 748          * sr_status is only set in decode_sequence, and so will remain
 749          * set to 1 if an rpc level failure occurs.
 750          */
 751         res->sr_status = 1;
 752         trace_nfs4_setup_sequence(session, args);
 753 out_success:
 754         rpc_call_start(task);
 755         return 0;
 756 out_sleep:
 757         /* Privileged tasks are queued with top priority */
 758         if (args->sa_privileged)
 759                 rpc_sleep_on_priority(&tbl->slot_tbl_waitq, task,
 760                                 NULL, RPC_PRIORITY_PRIVILEGED);
 761         else
 762                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
 763         spin_unlock(&tbl->slot_tbl_lock);
 764         return -EAGAIN;
 765 }
 766 EXPORT_SYMBOL_GPL(nfs41_setup_sequence);
 767
 768 static int nfs4_setup_sequence(const struct nfs_server *server,
 769                                struct nfs4_sequence_args *args,
 770                                struct nfs4_sequence_res *res,
 771                                struct rpc_task *task)
 772 {
 773         struct nfs4_session *session = nfs4_get_session(server);
 774         int ret = 0;
 775
 776         if (!session)
 777                 return nfs40_setup_sequence(server, args, res, task);
 778
 779         dprintk("--> %s clp %p session %p sr_slot %u\n",
 780                 __func__, session->clp, session, res->sr_slot ?
 781                         res->sr_slot->slot_nr : NFS4_NO_SLOT);
 782
 783         ret = nfs41_setup_sequence(session, args, res, task);
 784
 785         dprintk("<-- %s status=%d\n", __func__, ret);
 786         return ret;
 787 }
 788
 789 static void nfs41_call_sync_prepare(struct rpc_task *task, void *calldata)
 790 {
 791         struct nfs4_call_sync_data *data = calldata;
 792         struct nfs4_session *session = nfs4_get_session(data->seq_server);
 793
 794         dprintk("--> %s data->seq_server %p\n", __func__, data->seq_server);
 795
 796         nfs41_setup_sequence(session, data->seq_args, data->seq_res, task);
 797 }
 798
 799 static void nfs41_call_sync_done(struct rpc_task *task, void *calldata)
 800 {
 801         struct nfs4_call_sync_data *data = calldata;
 802
 803         nfs41_sequence_done(task, data->seq_res);
 804 }
 805
 806 static const struct rpc_call_ops nfs41_call_sync_ops = {
 807         .rpc_call_prepare = nfs41_call_sync_prepare,
 808         .rpc_call_done = nfs41_call_sync_done,
 809 };
 810
 811 #else   /* !CONFIG_NFS_V4_1 */
 812
 813 static int nfs4_setup_sequence(const struct nfs_server *server,
 814                                struct nfs4_sequence_args *args,
 815                                struct nfs4_sequence_res *res,
 816                                struct rpc_task *task)
 817 {
 818         return nfs40_setup_sequence(server, args, res, task);
 819 }
 820
 821 static int nfs4_sequence_done(struct rpc_task *task,
 822                                struct nfs4_sequence_res *res)
 823 {
 824         return nfs40_sequence_done(task, res);
 825 }
 826
 827 #endif  /* !CONFIG_NFS_V4_1 */
 828
 829 static void nfs40_call_sync_prepare(struct rpc_task *task, void *calldata)
 830 {
 831         struct nfs4_call_sync_data *data = calldata;
 832         nfs4_setup_sequence(data->seq_server,
 833                                 data->seq_args, data->seq_res, task);
 834 }
 835
 836 static void nfs40_call_sync_done(struct rpc_task *task, void *calldata)
 837 {
 838         struct nfs4_call_sync_data *data = calldata;
 839         nfs4_sequence_done(task, data->seq_res);
 840 }
 841
 842 static const struct rpc_call_ops nfs40_call_sync_ops = {
 843         .rpc_call_prepare = nfs40_call_sync_prepare,
 844         .rpc_call_done = nfs40_call_sync_done,
 845 };
 846
 847 static int nfs4_call_sync_sequence(struct rpc_clnt *clnt,
 848                                    struct nfs_server *server,
 849                                    struct rpc_message *msg,
 850                                    struct nfs4_sequence_args *args,
 851                                    struct nfs4_sequence_res *res)
 852 {
 853         int ret;
 854         struct rpc_task *task;
 855         struct nfs_client *clp = server->nfs_client;
 856         struct nfs4_call_sync_data data = {
 857                 .seq_server = server,
 858                 .seq_args = args,
 859                 .seq_res = res,
 860         };
 861         struct rpc_task_setup task_setup = {
 862                 .rpc_client = clnt,
 863                 .rpc_message = msg,
 864                 .callback_ops = clp->cl_mvops->call_sync_ops,
 865                 .callback_data = &data
 866         };
 867
 868         task = rpc_run_task(&task_setup);
 869         if (IS_ERR(task))
 870                 ret = PTR_ERR(task);
 871         else {
 872                 ret = task->tk_status;
 873                 rpc_put_task(task);
 874         }
 875         return ret;
 876 }
 877
 878 static
 879 int nfs4_call_sync(struct rpc_clnt *clnt,
 880                    struct nfs_server *server,
 881                    struct rpc_message *msg,
 882                    struct nfs4_sequence_args *args,
 883                    struct nfs4_sequence_res *res,
 884                    int cache_reply)
 885 {
 886         nfs4_init_sequence(args, res, cache_reply);
 887         return nfs4_call_sync_sequence(clnt, server, msg, args, res);
 888 }
 889
 890 static void update_changeattr(struct inode *dir, struct nfs4_change_info *cinfo)
 891 {
 892         struct nfs_inode *nfsi = NFS_I(dir);
 893
 894         spin_lock(&dir->i_lock);
 895         nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
 896         if (!cinfo->atomic || cinfo->before != dir->i_version)
 897                 nfs_force_lookup_revalidate(dir);
 898         dir->i_version = cinfo->after;
 899         nfs_fscache_invalidate(dir);
 900         spin_unlock(&dir->i_lock);
 901 }
 902
 903 struct nfs4_opendata {
 904         struct kref kref;
 905         struct nfs_openargs o_arg;
 906         struct nfs_openres o_res;
 907         struct nfs_open_confirmargs c_arg;
 908         struct nfs_open_confirmres c_res;
 909         struct nfs4_string owner_name;
 910         struct nfs4_string group_name;
 911         struct nfs_fattr f_attr;
 912         struct nfs4_label *f_label;
 913         struct dentry *dir;
 914         struct dentry *dentry;
 915         struct nfs4_state_owner *owner;
 916         struct nfs4_state *state;
 917         struct iattr attrs;
 918         unsigned long timestamp;
 919         unsigned int rpc_done : 1;
 920         unsigned int file_created : 1;
 921         unsigned int is_recover : 1;
 922         int rpc_status;
 923         int cancelled;
 924 };
 925
 926 static bool nfs4_clear_cap_atomic_open_v1(struct nfs_server *server,
 927                 int err, struct nfs4_exception *exception)
 928 {
 929         if (err != -EINVAL)
 930                 return false;
 931         if (!(server->caps & NFS_CAP_ATOMIC_OPEN_V1))
 932                 return false;
 933         server->caps &= ~NFS_CAP_ATOMIC_OPEN_V1;
 934         exception->retry = 1;
 935         return true;
 936 }
 937
 938 static enum open_claim_type4
 939 nfs4_map_atomic_open_claim(struct nfs_server *server,
 940                 enum open_claim_type4 claim)
 941 {
 942         if (server->caps & NFS_CAP_ATOMIC_OPEN_V1)
 943                 return claim;
 944         switch (claim) {
 945         default:
 946                 return claim;
 947         case NFS4_OPEN_CLAIM_FH:
 948                 return NFS4_OPEN_CLAIM_NULL;
 949         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
 950                 return NFS4_OPEN_CLAIM_DELEGATE_CUR;
 951         case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
 952                 return NFS4_OPEN_CLAIM_DELEGATE_PREV;
 953         }
 954 }
 955
 956 static void nfs4_init_opendata_res(struct nfs4_opendata *p)
 957 {
 958         p->o_res.f_attr = &p->f_attr;
 959         p->o_res.f_label = p->f_label;
 960         p->o_res.seqid = p->o_arg.seqid;
 961         p->c_res.seqid = p->c_arg.seqid;
 962         p->o_res.server = p->o_arg.server;
 963         p->o_res.access_request = p->o_arg.access;
 964         nfs_fattr_init(&p->f_attr);
 965         nfs_fattr_init_names(&p->f_attr, &p->owner_name, &p->group_name);
 966 }
 967
 968 static struct nfs4_opendata *nfs4_opendata_alloc(struct dentry *dentry,
 969                 struct nfs4_state_owner *sp, fmode_t fmode, int flags,
 970                 const struct iattr *attrs,
 971                 struct nfs4_label *label,
 972                 enum open_claim_type4 claim,
 973                 gfp_t gfp_mask)
 974 {
 975         struct dentry *parent = dget_parent(dentry);
 976         struct inode *dir = parent->d_inode;
 977         struct nfs_server *server = NFS_SERVER(dir);
 978         struct nfs4_opendata *p;
 979
 980         p = kzalloc(sizeof(*p), gfp_mask);
 981         if (p == NULL)
 982                 goto err;
 983
 984         p->f_label = nfs4_label_alloc(server, gfp_mask);
 985         if (IS_ERR(p->f_label))
 986                 goto err_free_p;
 987
 988         p->o_arg.seqid = nfs_alloc_seqid(&sp->so_seqid, gfp_mask);
 989         if (p->o_arg.seqid == NULL)
 990                 goto err_free_label;
 991         nfs_sb_active(dentry->d_sb);
 992         p->dentry = dget(dentry);
 993         p->dir = parent;
 994         p->owner = sp;
 995         atomic_inc(&sp->so_count);
 996         p->o_arg.open_flags = flags;
 997         p->o_arg.fmode = fmode & (FMODE_READ|FMODE_WRITE);
 998         /* don't put an ACCESS op in OPEN compound if O_EXCL, because ACCESS
 999          * will return permission denied for all bits until close */
1000         if (!(flags & O_EXCL)) {
1001                 /* ask server to check for all possible rights as results
1002                  * are cached */
1003                 p->o_arg.access = NFS4_ACCESS_READ | NFS4_ACCESS_MODIFY |
1004                                   NFS4_ACCESS_EXTEND | NFS4_ACCESS_EXECUTE;
1005         }
1006         p->o_arg.clientid = server->nfs_client->cl_clientid;
1007         p->o_arg.id.create_time = ktime_to_ns(sp->so_seqid.create_time);
1008         p->o_arg.id.uniquifier = sp->so_seqid.owner_id;
1009         p->o_arg.name = &dentry->d_name;
1010         p->o_arg.server = server;
1011         p->o_arg.bitmask = nfs4_bitmask(server, label);
1012         p->o_arg.open_bitmap = &nfs4_fattr_bitmap[0];
1013         p->o_arg.label = label;
1014         p->o_arg.claim = nfs4_map_atomic_open_claim(server, claim);
1015         switch (p->o_arg.claim) {
1016         case NFS4_OPEN_CLAIM_NULL:
1017         case NFS4_OPEN_CLAIM_DELEGATE_CUR:
1018         case NFS4_OPEN_CLAIM_DELEGATE_PREV:
1019                 p->o_arg.fh = NFS_FH(dir);
1020                 break;
1021         case NFS4_OPEN_CLAIM_PREVIOUS:
1022         case NFS4_OPEN_CLAIM_FH:
1023         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
1024         case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
1025                 p->o_arg.fh = NFS_FH(dentry->d_inode);
1026         }
1027         if (attrs != NULL && attrs->ia_valid != 0) {
1028                 __u32 verf[2];
1029
1030                 p->o_arg.u.attrs = &p->attrs;
1031                 memcpy(&p->attrs, attrs, sizeof(p->attrs));
1032
1033                 verf[0] = jiffies;
1034                 verf[1] = current->pid;
1035                 memcpy(p->o_arg.u.verifier.data, verf,
1036                                 sizeof(p->o_arg.u.verifier.data));
1037         }
1038         p->c_arg.fh = &p->o_res.fh;
1039         p->c_arg.stateid = &p->o_res.stateid;
1040         p->c_arg.seqid = p->o_arg.seqid;
1041         nfs4_init_opendata_res(p);
1042         kref_init(&p->kref);
1043         return p;
1044
1045 err_free_label:
1046         nfs4_label_free(p->f_label);
1047 err_free_p:
1048         kfree(p);
1049 err:
1050         dput(parent);
1051         return NULL;
1052 }
1053
1054 static void nfs4_opendata_free(struct kref *kref)
1055 {
1056         struct nfs4_opendata *p = container_of(kref,
1057                         struct nfs4_opendata, kref);
1058         struct super_block *sb = p->dentry->d_sb;
1059
1060         nfs_free_seqid(p->o_arg.seqid);
1061         if (p->state != NULL)
1062                 nfs4_put_open_state(p->state);
1063         nfs4_put_state_owner(p->owner);
1064
1065         nfs4_label_free(p->f_label);
1066
1067         dput(p->dir);
1068         dput(p->dentry);
1069         nfs_sb_deactive(sb);
1070         nfs_fattr_free_names(&p->f_attr);
1071         kfree(p->f_attr.mdsthreshold);
1072         kfree(p);
1073 }
1074
1075 static void nfs4_opendata_put(struct nfs4_opendata *p)
1076 {
1077         if (p != NULL)
1078                 kref_put(&p->kref, nfs4_opendata_free);
1079 }
1080
1081 static int nfs4_wait_for_completion_rpc_task(struct rpc_task *task)
1082 {
1083         int ret;
1084
1085         ret = rpc_wait_for_completion_task(task);
1086         return ret;
1087 }
1088
1089 static int can_open_cached(struct nfs4_state *state, fmode_t mode, int open_mode)
1090 {
1091         int ret = 0;
1092
1093         if (open_mode & (O_EXCL|O_TRUNC))
1094                 goto out;
1095         switch (mode & (FMODE_READ|FMODE_WRITE)) {
1096                 case FMODE_READ:
1097                         ret |= test_bit(NFS_O_RDONLY_STATE, &state->flags) != 0
1098                                 && state->n_rdonly != 0;
1099                         break;
1100                 case FMODE_WRITE:
1101                         ret |= test_bit(NFS_O_WRONLY_STATE, &state->flags) != 0
1102                                 && state->n_wronly != 0;
1103                         break;
1104                 case FMODE_READ|FMODE_WRITE:
1105                         ret |= test_bit(NFS_O_RDWR_STATE, &state->flags) != 0
1106                                 && state->n_rdwr != 0;
1107         }
1108 out:
1109         return ret;
1110 }
1111
1112 static int can_open_delegated(struct nfs_delegation *delegation, fmode_t fmode)
1113 {
1114         if (delegation == NULL)
1115                 return 0;
1116         if ((delegation->type & fmode) != fmode)
1117                 return 0;
1118         if (test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags))
1119                 return 0;
1120         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
1121                 return 0;
1122         nfs_mark_delegation_referenced(delegation);
1123         return 1;
1124 }
1125
1126 static void update_open_stateflags(struct nfs4_state *state, fmode_t fmode)
1127 {
1128         switch (fmode) {
1129                 case FMODE_WRITE:
1130                         state->n_wronly++;
1131                         break;
1132                 case FMODE_READ:
1133                         state->n_rdonly++;
1134                         break;
1135                 case FMODE_READ|FMODE_WRITE:
1136                         state->n_rdwr++;
1137         }
1138         nfs4_state_set_mode_locked(state, state->state | fmode);
1139 }
1140
1141 static void nfs_test_and_clear_all_open_stateid(struct nfs4_state *state)
1142 {
1143         struct nfs_client *clp = state->owner->so_server->nfs_client;
1144         bool need_recover = false;
1145
1146         if (test_and_clear_bit(NFS_O_RDONLY_STATE, &state->flags) && state->n_rdonly)
1147                 need_recover = true;
1148         if (test_and_clear_bit(NFS_O_WRONLY_STATE, &state->flags) && state->n_wronly)
1149                 need_recover = true;
1150         if (test_and_clear_bit(NFS_O_RDWR_STATE, &state->flags) && state->n_rdwr)
1151                 need_recover = true;
1152         if (need_recover)
1153                 nfs4_state_mark_reclaim_nograce(clp, state);
1154 }
1155
1156 static bool nfs_need_update_open_stateid(struct nfs4_state *state,
1157                 nfs4_stateid *stateid)
1158 {
1159         if (test_and_set_bit(NFS_OPEN_STATE, &state->flags) == 0)
1160                 return true;
1161         if (!nfs4_stateid_match_other(stateid, &state->open_stateid)) {
1162                 nfs_test_and_clear_all_open_stateid(state);
1163                 return true;
1164         }
1165         if (nfs4_stateid_is_newer(stateid, &state->open_stateid))
1166                 return true;
1167         return false;
1168 }
1169
1170 static void nfs_clear_open_stateid_locked(struct nfs4_state *state,
1171                 nfs4_stateid *stateid, fmode_t fmode)
1172 {
1173         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1174         switch (fmode & (FMODE_READ|FMODE_WRITE)) {
1175         case FMODE_WRITE:
1176                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1177                 break;
1178         case FMODE_READ:
1179                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1180                 break;
1181         case 0:
1182                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1183                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1184                 clear_bit(NFS_OPEN_STATE, &state->flags);
1185         }
1186         if (stateid == NULL)
1187                 return;
1188         if (!nfs_need_update_open_stateid(state, stateid))
1189                 return;
1190         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1191                 nfs4_stateid_copy(&state->stateid, stateid);
1192         nfs4_stateid_copy(&state->open_stateid, stateid);
1193 }
1194
1195 static void nfs_clear_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, fmode_t fmode)
1196 {
1197         write_seqlock(&state->seqlock);
1198         nfs_clear_open_stateid_locked(state, stateid, fmode);
1199         write_sequnlock(&state->seqlock);
1200         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags))
1201                 nfs4_schedule_state_manager(state->owner->so_server->nfs_client);
1202 }
1203
1204 static void nfs_set_open_stateid_locked(struct nfs4_state *state, nfs4_stateid *stateid, fmode_t fmode)
1205 {
1206         switch (fmode) {
1207                 case FMODE_READ:
1208                         set_bit(NFS_O_RDONLY_STATE, &state->flags);
1209                         break;
1210                 case FMODE_WRITE:
1211                         set_bit(NFS_O_WRONLY_STATE, &state->flags);
1212                         break;
1213                 case FMODE_READ|FMODE_WRITE:
1214                         set_bit(NFS_O_RDWR_STATE, &state->flags);
1215         }
1216         if (!nfs_need_update_open_stateid(state, stateid))
1217                 return;
1218         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1219                 nfs4_stateid_copy(&state->stateid, stateid);
1220         nfs4_stateid_copy(&state->open_stateid, stateid);
1221 }
1222
1223 static void __update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, const nfs4_stateid *deleg_stateid, fmode_t fmode)
1224 {
1225         /*
1226          * Protect the call to nfs4_state_set_mode_locked and
1227          * serialise the stateid update
1228          */
1229         write_seqlock(&state->seqlock);
1230         if (deleg_stateid != NULL) {
1231                 nfs4_stateid_copy(&state->stateid, deleg_stateid);
1232                 set_bit(NFS_DELEGATED_STATE, &state->flags);
1233         }
1234         if (open_stateid != NULL)
1235                 nfs_set_open_stateid_locked(state, open_stateid, fmode);
1236         write_sequnlock(&state->seqlock);
1237         spin_lock(&state->owner->so_lock);
1238         update_open_stateflags(state, fmode);
1239         spin_unlock(&state->owner->so_lock);
1240 }
1241
1242 static int update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, nfs4_stateid *delegation, fmode_t fmode)
1243 {
1244         struct nfs_inode *nfsi = NFS_I(state->inode);
1245         struct nfs_delegation *deleg_cur;
1246         int ret = 0;
1247
1248         fmode &= (FMODE_READ|FMODE_WRITE);
1249
1250         rcu_read_lock();
1251         deleg_cur = rcu_dereference(nfsi->delegation);
1252         if (deleg_cur == NULL)
1253                 goto no_delegation;
1254
1255         spin_lock(&deleg_cur->lock);
1256         if (rcu_dereference(nfsi->delegation) != deleg_cur ||
1257            test_bit(NFS_DELEGATION_RETURNING, &deleg_cur->flags) ||
1258             (deleg_cur->type & fmode) != fmode)
1259                 goto no_delegation_unlock;
1260
1261         if (delegation == NULL)
1262                 delegation = &deleg_cur->stateid;
1263         else if (!nfs4_stateid_match(&deleg_cur->stateid, delegation))
1264                 goto no_delegation_unlock;
1265
1266         nfs_mark_delegation_referenced(deleg_cur);
1267         __update_open_stateid(state, open_stateid, &deleg_cur->stateid, fmode);
1268         ret = 1;
1269 no_delegation_unlock:
1270         spin_unlock(&deleg_cur->lock);
1271 no_delegation:
1272         rcu_read_unlock();
1273
1274         if (!ret && open_stateid != NULL) {
1275                 __update_open_stateid(state, open_stateid, NULL, fmode);
1276                 ret = 1;
1277         }
1278         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags))
1279                 nfs4_schedule_state_manager(state->owner->so_server->nfs_client);
1280
1281         return ret;
1282 }
1283
1284
1285 static void nfs4_return_incompatible_delegation(struct inode *inode, fmode_t fmode)
1286 {
1287         struct nfs_delegation *delegation;
1288
1289         rcu_read_lock();
1290         delegation = rcu_dereference(NFS_I(inode)->delegation);
1291         if (delegation == NULL || (delegation->type & fmode) == fmode) {
1292                 rcu_read_unlock();
1293                 return;
1294         }
1295         rcu_read_unlock();
1296         nfs4_inode_return_delegation(inode);
1297 }
1298
1299 static struct nfs4_state *nfs4_try_open_cached(struct nfs4_opendata *opendata)
1300 {
1301         struct nfs4_state *state = opendata->state;
1302         struct nfs_inode *nfsi = NFS_I(state->inode);
1303         struct nfs_delegation *delegation;
1304         int open_mode = opendata->o_arg.open_flags;
1305         fmode_t fmode = opendata->o_arg.fmode;
1306         nfs4_stateid stateid;
1307         int ret = -EAGAIN;
1308
1309         for (;;) {
1310                 if (can_open_cached(state, fmode, open_mode)) {
1311                         spin_lock(&state->owner->so_lock);
1312                         if (can_open_cached(state, fmode, open_mode)) {
1313                                 update_open_stateflags(state, fmode);
1314                                 spin_unlock(&state->owner->so_lock);
1315                                 goto out_return_state;
1316                         }
1317                         spin_unlock(&state->owner->so_lock);
1318                 }
1319                 rcu_read_lock();
1320                 delegation = rcu_dereference(nfsi->delegation);
1321                 if (!can_open_delegated(delegation, fmode)) {
1322                         rcu_read_unlock();
1323                         break;
1324                 }
1325                 /* Save the delegation */
1326                 nfs4_stateid_copy(&stateid, &delegation->stateid);
1327                 rcu_read_unlock();
1328                 nfs_release_seqid(opendata->o_arg.seqid);
1329                 if (!opendata->is_recover) {
1330                         ret = nfs_may_open(state->inode, state->owner->so_cred, open_mode);
1331                         if (ret != 0)
1332                                 goto out;
1333                 }
1334                 ret = -EAGAIN;
1335
1336                 /* Try to update the stateid using the delegation */
1337                 if (update_open_stateid(state, NULL, &stateid, fmode))
1338                         goto out_return_state;
1339         }
1340 out:
1341         return ERR_PTR(ret);
1342 out_return_state:
1343         atomic_inc(&state->count);
1344         return state;
1345 }
1346
1347 static void
1348 nfs4_opendata_check_deleg(struct nfs4_opendata *data, struct nfs4_state *state)
1349 {
1350         struct nfs_client *clp = NFS_SERVER(state->inode)->nfs_client;
1351         struct nfs_delegation *delegation;
1352         int delegation_flags = 0;
1353
1354         rcu_read_lock();
1355         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
1356         if (delegation)
1357                 delegation_flags = delegation->flags;
1358         rcu_read_unlock();
1359         if (data->o_arg.claim == NFS4_OPEN_CLAIM_DELEGATE_CUR) {
1360                 pr_err_ratelimited("NFS: Broken NFSv4 server %s is "
1361                                    "returning a delegation for "
1362                                    "OPEN(CLAIM_DELEGATE_CUR)\n",
1363                                    clp->cl_hostname);
1364         } else if ((delegation_flags & 1UL<<NFS_DELEGATION_NEED_RECLAIM) == 0)
1365                 nfs_inode_set_delegation(state->inode,
1366                                          data->owner->so_cred,
1367                                          &data->o_res);
1368         else
1369                 nfs_inode_reclaim_delegation(state->inode,
1370                                              data->owner->so_cred,
1371                                              &data->o_res);
1372 }
1373
1374 /*
1375  * Check the inode attributes against the CLAIM_PREVIOUS returned attributes
1376  * and update the nfs4_state.
1377  */
1378 static struct nfs4_state *
1379 _nfs4_opendata_reclaim_to_nfs4_state(struct nfs4_opendata *data)
1380 {
1381         struct inode *inode = data->state->inode;
1382         struct nfs4_state *state = data->state;
1383         int ret;
1384
1385         if (!data->rpc_done) {
1386                 if (data->rpc_status) {
1387                         ret = data->rpc_status;
1388                         goto err;
1389                 }
1390                 /* cached opens have already been processed */
1391                 goto update;
1392         }
1393
1394         ret = nfs_refresh_inode(inode, &data->f_attr);
1395         if (ret)
1396                 goto err;
1397
1398         if (data->o_res.delegation_type != 0)
1399                 nfs4_opendata_check_deleg(data, state);
1400 update:
1401         update_open_stateid(state, &data->o_res.stateid, NULL,
1402                             data->o_arg.fmode);
1403         atomic_inc(&state->count);
1404
1405         return state;
1406 err:
1407         return ERR_PTR(ret);
1408
1409 }
1410
1411 static struct nfs4_state *
1412 _nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data)
1413 {
1414         struct inode *inode;
1415         struct nfs4_state *state = NULL;
1416         int ret;
1417
1418         if (!data->rpc_done) {
1419                 state = nfs4_try_open_cached(data);
1420                 goto out;
1421         }
1422
1423         ret = -EAGAIN;
1424         if (!(data->f_attr.valid & NFS_ATTR_FATTR))
1425                 goto err;
1426         inode = nfs_fhget(data->dir->d_sb, &data->o_res.fh, &data->f_attr, data->f_label);
1427         ret = PTR_ERR(inode);
1428         if (IS_ERR(inode))
1429                 goto err;
1430         ret = -ENOMEM;
1431         state = nfs4_get_open_state(inode, data->owner);
1432         if (state == NULL)
1433                 goto err_put_inode;
1434         if (data->o_res.delegation_type != 0)
1435                 nfs4_opendata_check_deleg(data, state);
1436         update_open_stateid(state, &data->o_res.stateid, NULL,
1437                         data->o_arg.fmode);
1438         iput(inode);
1439 out:
1440         nfs_release_seqid(data->o_arg.seqid);
1441         return state;
1442 err_put_inode:
1443         iput(inode);
1444 err:
1445         return ERR_PTR(ret);
1446 }
1447
1448 static struct nfs4_state *
1449 nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data)
1450 {
1451         if (data->o_arg.claim == NFS4_OPEN_CLAIM_PREVIOUS)
1452                 return _nfs4_opendata_reclaim_to_nfs4_state(data);
1453         return _nfs4_opendata_to_nfs4_state(data);
1454 }
1455
1456 static struct nfs_open_context *nfs4_state_find_open_context(struct nfs4_state *state)
1457 {
1458         struct nfs_inode *nfsi = NFS_I(state->inode);
1459         struct nfs_open_context *ctx;
1460
1461         spin_lock(&state->inode->i_lock);
1462         list_for_each_entry(ctx, &nfsi->open_files, list) {
1463                 if (ctx->state != state)
1464                         continue;
1465                 get_nfs_open_context(ctx);
1466                 spin_unlock(&state->inode->i_lock);
1467                 return ctx;
1468         }
1469         spin_unlock(&state->inode->i_lock);
1470         return ERR_PTR(-ENOENT);
1471 }
1472
1473 static struct nfs4_opendata *nfs4_open_recoverdata_alloc(struct nfs_open_context *ctx,
1474                 struct nfs4_state *state, enum open_claim_type4 claim)
1475 {
1476         struct nfs4_opendata *opendata;
1477
1478         opendata = nfs4_opendata_alloc(ctx->dentry, state->owner, 0, 0,
1479                         NULL, NULL, claim, GFP_NOFS);
1480         if (opendata == NULL)
1481                 return ERR_PTR(-ENOMEM);
1482         opendata->state = state;
1483         atomic_inc(&state->count);
1484         return opendata;
1485 }
1486
1487 static int nfs4_open_recover_helper(struct nfs4_opendata *opendata, fmode_t fmode, struct nfs4_state **res)
1488 {
1489         struct nfs4_state *newstate;
1490         int ret;
1491
1492         opendata->o_arg.open_flags = 0;
1493         opendata->o_arg.fmode = fmode;
1494         memset(&opendata->o_res, 0, sizeof(opendata->o_res));
1495         memset(&opendata->c_res, 0, sizeof(opendata->c_res));
1496         nfs4_init_opendata_res(opendata);
1497         ret = _nfs4_recover_proc_open(opendata);
1498         if (ret != 0)
1499                 return ret;
1500         newstate = nfs4_opendata_to_nfs4_state(opendata);
1501         if (IS_ERR(newstate))
1502                 return PTR_ERR(newstate);
1503         nfs4_close_state(newstate, fmode);
1504         *res = newstate;
1505         return 0;
1506 }
1507
1508 static int nfs4_open_recover(struct nfs4_opendata *opendata, struct nfs4_state *state)
1509 {
1510         struct nfs4_state *newstate;
1511         int ret;
1512
1513         /* Don't trigger recovery in nfs_test_and_clear_all_open_stateid */
1514         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1515         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1516         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1517         /* memory barrier prior to reading state->n_* */
1518         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1519         clear_bit(NFS_OPEN_STATE, &state->flags);
1520         smp_rmb();
1521         if (state->n_rdwr != 0) {
1522                 ret = nfs4_open_recover_helper(opendata, FMODE_READ|FMODE_WRITE, &newstate);
1523                 if (ret != 0)
1524                         return ret;
1525                 if (newstate != state)
1526                         return -ESTALE;
1527         }
1528         if (state->n_wronly != 0) {
1529                 ret = nfs4_open_recover_helper(opendata, FMODE_WRITE, &newstate);
1530                 if (ret != 0)
1531                         return ret;
1532                 if (newstate != state)
1533                         return -ESTALE;
1534         }
1535         if (state->n_rdonly != 0) {
1536                 ret = nfs4_open_recover_helper(opendata, FMODE_READ, &newstate);
1537                 if (ret != 0)
1538                         return ret;
1539                 if (newstate != state)
1540                         return -ESTALE;
1541         }
1542         /*
1543          * We may have performed cached opens for all three recoveries.
1544          * Check if we need to update the current stateid.
1545          */
1546         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0 &&
1547             !nfs4_stateid_match(&state->stateid, &state->open_stateid)) {
1548                 write_seqlock(&state->seqlock);
1549                 if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1550                         nfs4_stateid_copy(&state->stateid, &state->open_stateid);
1551                 write_sequnlock(&state->seqlock);
1552         }
1553         return 0;
1554 }
1555
1556 /*
1557  * OPEN_RECLAIM:
1558  *      reclaim state on the server after a reboot.
1559  */
1560 static int _nfs4_do_open_reclaim(struct nfs_open_context *ctx, struct nfs4_state *state)
1561 {
1562         struct nfs_delegation *delegation;
1563         struct nfs4_opendata *opendata;
1564         fmode_t delegation_type = 0;
1565         int status;
1566
1567         opendata = nfs4_open_recoverdata_alloc(ctx, state,
1568                         NFS4_OPEN_CLAIM_PREVIOUS);
1569         if (IS_ERR(opendata))
1570                 return PTR_ERR(opendata);
1571         rcu_read_lock();
1572         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
1573         if (delegation != NULL && test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) != 0)
1574                 delegation_type = delegation->type;
1575         rcu_read_unlock();
1576         opendata->o_arg.u.delegation_type = delegation_type;
1577         status = nfs4_open_recover(opendata, state);
1578         nfs4_opendata_put(opendata);
1579         return status;
1580 }
1581
1582 static int nfs4_do_open_reclaim(struct nfs_open_context *ctx, struct nfs4_state *state)
1583 {
1584         struct nfs_server *server = NFS_SERVER(state->inode);
1585         struct nfs4_exception exception = { };
1586         int err;
1587         do {
1588                 err = _nfs4_do_open_reclaim(ctx, state);
1589                 trace_nfs4_open_reclaim(ctx, 0, err);
1590                 if (nfs4_clear_cap_atomic_open_v1(server, err, &exception))
1591                         continue;
1592                 if (err != -NFS4ERR_DELAY)
1593                         break;
1594                 nfs4_handle_exception(server, err, &exception);
1595         } while (exception.retry);
1596         return err;
1597 }
1598
1599 static int nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *state)
1600 {
1601         struct nfs_open_context *ctx;
1602         int ret;
1603
1604         ctx = nfs4_state_find_open_context(state);
1605         if (IS_ERR(ctx))
1606                 return -EAGAIN;
1607         ret = nfs4_do_open_reclaim(ctx, state);
1608         put_nfs_open_context(ctx);
1609         return ret;
1610 }
1611
1612 static int nfs4_handle_delegation_recall_error(struct nfs_server *server, struct nfs4_state *state, const nfs4_stateid *stateid, int err)
1613 {
1614         switch (err) {
1615                 default:
1616                         printk(KERN_ERR "NFS: %s: unhandled error "
1617                                         "%d.\n", __func__, err);
1618                 case 0:
1619                 case -ENOENT:
1620                 case -ESTALE:
1621                         break;
1622                 case -NFS4ERR_BADSESSION:
1623                 case -NFS4ERR_BADSLOT:
1624                 case -NFS4ERR_BAD_HIGH_SLOT:
1625                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1626                 case -NFS4ERR_DEADSESSION:
1627                         set_bit(NFS_DELEGATED_STATE, &state->flags);
1628                         nfs4_schedule_session_recovery(server->nfs_client->cl_session, err);
1629                         return -EAGAIN;
1630                 case -NFS4ERR_STALE_CLIENTID:
1631                 case -NFS4ERR_STALE_STATEID:
1632                         set_bit(NFS_DELEGATED_STATE, &state->flags);
1633                 case -NFS4ERR_EXPIRED:
1634                         /* Don't recall a delegation if it was lost */
1635                         nfs4_schedule_lease_recovery(server->nfs_client);
1636                         return -EAGAIN;
1637                 case -NFS4ERR_MOVED:
1638                         nfs4_schedule_migration_recovery(server);
1639                         return -EAGAIN;
1640                 case -NFS4ERR_LEASE_MOVED:
1641                         nfs4_schedule_lease_moved_recovery(server->nfs_client);
1642                         return -EAGAIN;
1643                 case -NFS4ERR_DELEG_REVOKED:
1644                 case -NFS4ERR_ADMIN_REVOKED:
1645                 case -NFS4ERR_BAD_STATEID:
1646                 case -NFS4ERR_OPENMODE:
1647                         nfs_inode_find_state_and_recover(state->inode,
1648                                         stateid);
1649                         nfs4_schedule_stateid_recovery(server, state);
1650                         return 0;
1651                 case -NFS4ERR_DELAY:
1652                 case -NFS4ERR_GRACE:
1653                         set_bit(NFS_DELEGATED_STATE, &state->flags);
1654                         ssleep(1);
1655                         return -EAGAIN;
1656                 case -ENOMEM:
1657                 case -NFS4ERR_DENIED:
1658                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1659                         return 0;
1660         }
1661         return err;
1662 }
1663
1664 int nfs4_open_delegation_recall(struct nfs_open_context *ctx, struct nfs4_state *state, const nfs4_stateid *stateid)
1665 {
1666         struct nfs_server *server = NFS_SERVER(state->inode);
1667         struct nfs4_opendata *opendata;
1668         int err;
1669
1670         opendata = nfs4_open_recoverdata_alloc(ctx, state,
1671                         NFS4_OPEN_CLAIM_DELEG_CUR_FH);
1672         if (IS_ERR(opendata))
1673                 return PTR_ERR(opendata);
1674         nfs4_stateid_copy(&opendata->o_arg.u.delegation, stateid);
1675         err = nfs4_open_recover(opendata, state);
1676         nfs4_opendata_put(opendata);
1677         return nfs4_handle_delegation_recall_error(server, state, stateid, err);
1678 }
1679
1680 static void nfs4_open_confirm_prepare(struct rpc_task *task, void *calldata)
1681 {
1682         struct nfs4_opendata *data = calldata;
1683
1684         nfs40_setup_sequence(data->o_arg.server, &data->c_arg.seq_args,
1685                                 &data->c_res.seq_res, task);
1686 }
1687
1688 static void nfs4_open_confirm_done(struct rpc_task *task, void *calldata)
1689 {
1690         struct nfs4_opendata *data = calldata;
1691
1692         nfs40_sequence_done(task, &data->c_res.seq_res);
1693
1694         data->rpc_status = task->tk_status;
1695         if (data->rpc_status == 0) {
1696                 nfs4_stateid_copy(&data->o_res.stateid, &data->c_res.stateid);
1697                 nfs_confirm_seqid(&data->owner->so_seqid, 0);
1698                 renew_lease(data->o_res.server, data->timestamp);
1699                 data->rpc_done = 1;
1700         }
1701 }
1702
1703 static void nfs4_open_confirm_release(void *calldata)
1704 {
1705         struct nfs4_opendata *data = calldata;
1706         struct nfs4_state *state = NULL;
1707
1708         /* If this request hasn't been cancelled, do nothing */
1709         if (data->cancelled == 0)
1710                 goto out_free;
1711         /* In case of error, no cleanup! */
1712         if (!data->rpc_done)
1713                 goto out_free;
1714         state = nfs4_opendata_to_nfs4_state(data);
1715         if (!IS_ERR(state))
1716                 nfs4_close_state(state, data->o_arg.fmode);
1717 out_free:
1718         nfs4_opendata_put(data);
1719 }
1720
1721 static const struct rpc_call_ops nfs4_open_confirm_ops = {
1722         .rpc_call_prepare = nfs4_open_confirm_prepare,
1723         .rpc_call_done = nfs4_open_confirm_done,
1724         .rpc_release = nfs4_open_confirm_release,
1725 };
1726
1727 /*
1728  * Note: On error, nfs4_proc_open_confirm will free the struct nfs4_opendata
1729  */
1730 static int _nfs4_proc_open_confirm(struct nfs4_opendata *data)
1731 {
1732         struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
1733         struct rpc_task *task;
1734         struct  rpc_message msg = {
1735                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_CONFIRM],
1736                 .rpc_argp = &data->c_arg,
1737                 .rpc_resp = &data->c_res,
1738                 .rpc_cred = data->owner->so_cred,
1739         };
1740         struct rpc_task_setup task_setup_data = {
1741                 .rpc_client = server->client,
1742                 .rpc_message = &msg,
1743                 .callback_ops = &nfs4_open_confirm_ops,
1744                 .callback_data = data,
1745                 .workqueue = nfsiod_workqueue,
1746                 .flags = RPC_TASK_ASYNC,
1747         };
1748         int status;
1749
1750         nfs4_init_sequence(&data->c_arg.seq_args, &data->c_res.seq_res, 1);
1751         kref_get(&data->kref);
1752         data->rpc_done = 0;
1753         data->rpc_status = 0;
1754         data->timestamp = jiffies;
1755         task = rpc_run_task(&task_setup_data);
1756         if (IS_ERR(task))
1757                 return PTR_ERR(task);
1758         status = nfs4_wait_for_completion_rpc_task(task);
1759         if (status != 0) {
1760                 data->cancelled = 1;
1761                 smp_wmb();
1762         } else
1763                 status = data->rpc_status;
1764         rpc_put_task(task);
1765         return status;
1766 }
1767
1768 static void nfs4_open_prepare(struct rpc_task *task, void *calldata)
1769 {
1770         struct nfs4_opendata *data = calldata;
1771         struct nfs4_state_owner *sp = data->owner;
1772         struct nfs_client *clp = sp->so_server->nfs_client;
1773
1774         if (nfs_wait_on_sequence(data->o_arg.seqid, task) != 0)
1775                 goto out_wait;
1776         /*
1777          * Check if we still need to send an OPEN call, or if we can use
1778          * a delegation instead.
1779          */
1780         if (data->state != NULL) {
1781                 struct nfs_delegation *delegation;
1782
1783                 if (can_open_cached(data->state, data->o_arg.fmode, data->o_arg.open_flags))
1784                         goto out_no_action;
1785                 rcu_read_lock();
1786                 delegation = rcu_dereference(NFS_I(data->state->inode)->delegation);
1787                 if (data->o_arg.claim != NFS4_OPEN_CLAIM_DELEGATE_CUR &&
1788                     data->o_arg.claim != NFS4_OPEN_CLAIM_DELEG_CUR_FH &&
1789                     can_open_delegated(delegation, data->o_arg.fmode))
1790                         goto unlock_no_action;
1791                 rcu_read_unlock();
1792         }
1793         /* Update client id. */
1794         data->o_arg.clientid = clp->cl_clientid;
1795         switch (data->o_arg.claim) {
1796         case NFS4_OPEN_CLAIM_PREVIOUS:
1797         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
1798         case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
1799                 data->o_arg.open_bitmap = &nfs4_open_noattr_bitmap[0];
1800         case NFS4_OPEN_CLAIM_FH:
1801                 task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_NOATTR];
1802                 nfs_copy_fh(&data->o_res.fh, data->o_arg.fh);
1803         }
1804         data->timestamp = jiffies;
1805         if (nfs4_setup_sequence(data->o_arg.server,
1806                                 &data->o_arg.seq_args,
1807                                 &data->o_res.seq_res,
1808                                 task) != 0)
1809                 nfs_release_seqid(data->o_arg.seqid);
1810
1811         /* Set the create mode (note dependency on the session type) */
1812         data->o_arg.createmode = NFS4_CREATE_UNCHECKED;
1813         if (data->o_arg.open_flags & O_EXCL) {
1814                 data->o_arg.createmode = NFS4_CREATE_EXCLUSIVE;
1815                 if (nfs4_has_persistent_session(clp))
1816                         data->o_arg.createmode = NFS4_CREATE_GUARDED;
1817                 else if (clp->cl_mvops->minor_version > 0)
1818                         data->o_arg.createmode = NFS4_CREATE_EXCLUSIVE4_1;
1819         }
1820         return;
1821 unlock_no_action:
1822         rcu_read_unlock();
1823 out_no_action:
1824         task->tk_action = NULL;
1825 out_wait:
1826         nfs4_sequence_done(task, &data->o_res.seq_res);
1827 }
1828
1829 static void nfs4_open_done(struct rpc_task *task, void *calldata)
1830 {
1831         struct nfs4_opendata *data = calldata;
1832
1833         data->rpc_status = task->tk_status;
1834
1835         if (!nfs4_sequence_done(task, &data->o_res.seq_res))
1836                 return;
1837
1838         if (task->tk_status == 0) {
1839                 if (data->o_res.f_attr->valid & NFS_ATTR_FATTR_TYPE) {
1840                         switch (data->o_res.f_attr->mode & S_IFMT) {
1841                         case S_IFREG:
1842                                 break;
1843                         case S_IFLNK:
1844                                 data->rpc_status = -ELOOP;
1845                                 break;
1846                         case S_IFDIR:
1847                                 data->rpc_status = -EISDIR;
1848                                 break;
1849                         default:
1850                                 data->rpc_status = -ENOTDIR;
1851                         }
1852                 }
1853                 renew_lease(data->o_res.server, data->timestamp);
1854                 if (!(data->o_res.rflags & NFS4_OPEN_RESULT_CONFIRM))
1855                         nfs_confirm_seqid(&data->owner->so_seqid, 0);
1856         }
1857         data->rpc_done = 1;
1858 }
1859
1860 static void nfs4_open_release(void *calldata)
1861 {
1862         struct nfs4_opendata *data = calldata;
1863         struct nfs4_state *state = NULL;
1864
1865         /* If this request hasn't been cancelled, do nothing */
1866         if (data->cancelled == 0)
1867                 goto out_free;
1868         /* In case of error, no cleanup! */
1869         if (data->rpc_status != 0 || !data->rpc_done)
1870                 goto out_free;
1871         /* In case we need an open_confirm, no cleanup! */
1872         if (data->o_res.rflags & NFS4_OPEN_RESULT_CONFIRM)
1873                 goto out_free;
1874         state = nfs4_opendata_to_nfs4_state(data);
1875         if (!IS_ERR(state))
1876                 nfs4_close_state(state, data->o_arg.fmode);
1877 out_free:
1878         nfs4_opendata_put(data);
1879 }
1880
1881 static const struct rpc_call_ops nfs4_open_ops = {
1882         .rpc_call_prepare = nfs4_open_prepare,
1883         .rpc_call_done = nfs4_open_done,
1884         .rpc_release = nfs4_open_release,
1885 };
1886
1887 static int nfs4_run_open_task(struct nfs4_opendata *data, int isrecover)
1888 {
1889         struct inode *dir = data->dir->d_inode;
1890         struct nfs_server *server = NFS_SERVER(dir);
1891         struct nfs_openargs *o_arg = &data->o_arg;
1892         struct nfs_openres *o_res = &data->o_res;
1893         struct rpc_task *task;
1894         struct rpc_message msg = {
1895                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN],
1896                 .rpc_argp = o_arg,
1897                 .rpc_resp = o_res,
1898                 .rpc_cred = data->owner->so_cred,
1899         };
1900         struct rpc_task_setup task_setup_data = {
1901                 .rpc_client = server->client,
1902                 .rpc_message = &msg,
1903                 .callback_ops = &nfs4_open_ops,
1904                 .callback_data = data,
1905                 .workqueue = nfsiod_workqueue,
1906                 .flags = RPC_TASK_ASYNC,
1907         };
1908         int status;
1909
1910         nfs4_init_sequence(&o_arg->seq_args, &o_res->seq_res, 1);
1911         kref_get(&data->kref);
1912         data->rpc_done = 0;
1913         data->rpc_status = 0;
1914         data->cancelled = 0;
1915         data->is_recover = 0;
1916         if (isrecover) {
1917                 nfs4_set_sequence_privileged(&o_arg->seq_args);
1918                 data->is_recover = 1;
1919         }
1920         task = rpc_run_task(&task_setup_data);
1921         if (IS_ERR(task))
1922                 return PTR_ERR(task);
1923         status = nfs4_wait_for_completion_rpc_task(task);
1924         if (status != 0) {
1925                 data->cancelled = 1;
1926                 smp_wmb();
1927         } else
1928                 status = data->rpc_status;
1929         rpc_put_task(task);
1930
1931         return status;
1932 }
1933
1934 static int _nfs4_recover_proc_open(struct nfs4_opendata *data)
1935 {
1936         struct inode *dir = data->dir->d_inode;
1937         struct nfs_openres *o_res = &data->o_res;
1938         int status;
1939
1940         status = nfs4_run_open_task(data, 1);
1941         if (status != 0 || !data->rpc_done)
1942                 return status;
1943
1944         nfs_fattr_map_and_free_names(NFS_SERVER(dir), &data->f_attr);
1945
1946         if (o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
1947                 status = _nfs4_proc_open_confirm(data);
1948                 if (status != 0)
1949                         return status;
1950         }
1951
1952         return status;
1953 }
1954
1955 /*
1956  * Additional permission checks in order to distinguish between an
1957  * open for read, and an open for execute. This works around the
1958  * fact that NFSv4 OPEN treats read and execute permissions as being
1959  * the same.
1960  * Note that in the non-execute case, we want to turn off permission
1961  * checking if we just created a new file (POSIX open() semantics).
1962  */
1963 static int nfs4_opendata_access(struct rpc_cred *cred,
1964                                 struct nfs4_opendata *opendata,
1965                                 struct nfs4_state *state, fmode_t fmode,
1966                                 int openflags)
1967 {
1968         struct nfs_access_entry cache;
1969         u32 mask;
1970
1971         /* access call failed or for some reason the server doesn't
1972          * support any access modes -- defer access call until later */
1973         if (opendata->o_res.access_supported == 0)
1974                 return 0;
1975
1976         mask = 0;
1977         /*
1978          * Use openflags to check for exec, because fmode won't
1979          * always have FMODE_EXEC set when file open for exec.
1980          */
1981         if (openflags & __FMODE_EXEC) {
1982                 /* ONLY check for exec rights */
1983                 mask = MAY_EXEC;
1984         } else if ((fmode & FMODE_READ) && !opendata->file_created)
1985                 mask = MAY_READ;
1986
1987         cache.cred = cred;
1988         cache.jiffies = jiffies;
1989         nfs_access_set_mask(&cache, opendata->o_res.access_result);
1990         nfs_access_add_cache(state->inode, &cache);
1991
1992         if ((mask & ~cache.mask & (MAY_READ | MAY_EXEC)) == 0)
1993                 return 0;
1994
1995         /* even though OPEN succeeded, access is denied. Close the file */
1996         nfs4_close_state(state, fmode);
1997         return -EACCES;
1998 }
1999
2000 /*
2001  * Note: On error, nfs4_proc_open will free the struct nfs4_opendata
2002  */
2003 static int _nfs4_proc_open(struct nfs4_opendata *data)
2004 {
2005         struct inode *dir = data->dir->d_inode;
2006         struct nfs_server *server = NFS_SERVER(dir);
2007         struct nfs_openargs *o_arg = &data->o_arg;
2008         struct nfs_openres *o_res = &data->o_res;
2009         int status;
2010
2011         status = nfs4_run_open_task(data, 0);
2012         if (!data->rpc_done)
2013                 return status;
2014         if (status != 0) {
2015                 if (status == -NFS4ERR_BADNAME &&
2016                                 !(o_arg->open_flags & O_CREAT))
2017                         return -ENOENT;
2018                 return status;
2019         }
2020
2021         nfs_fattr_map_and_free_names(server, &data->f_attr);
2022
2023         if (o_arg->open_flags & O_CREAT) {
2024                 update_changeattr(dir, &o_res->cinfo);
2025                 if (o_arg->open_flags & O_EXCL)
2026                         data->file_created = 1;
2027                 else if (o_res->cinfo.before != o_res->cinfo.after)
2028                         data->file_created = 1;
2029         }
2030         if ((o_res->rflags & NFS4_OPEN_RESULT_LOCKTYPE_POSIX) == 0)
2031                 server->caps &= ~NFS_CAP_POSIX_LOCK;
2032         if(o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
2033                 status = _nfs4_proc_open_confirm(data);
2034                 if (status != 0)
2035                         return status;
2036         }
2037         if (!(o_res->f_attr->valid & NFS_ATTR_FATTR))
2038                 nfs4_proc_getattr(server, &o_res->fh, o_res->f_attr, o_res->f_label);
2039         return 0;
2040 }
2041
2042 static int nfs4_recover_expired_lease(struct nfs_server *server)
2043 {
2044         return nfs4_client_recover_expired_lease(server->nfs_client);
2045 }
2046
2047 /*
2048  * OPEN_EXPIRED:
2049  *      reclaim state on the server after a network partition.
2050  *      Assumes caller holds the appropriate lock
2051  */
2052 static int _nfs4_open_expired(struct nfs_open_context *ctx, struct nfs4_state *state)
2053 {
2054         struct nfs4_opendata *opendata;
2055         int ret;
2056
2057         opendata = nfs4_open_recoverdata_alloc(ctx, state,
2058                         NFS4_OPEN_CLAIM_FH);
2059         if (IS_ERR(opendata))
2060                 return PTR_ERR(opendata);
2061         ret = nfs4_open_recover(opendata, state);
2062         if (ret == -ESTALE)
2063                 d_drop(ctx->dentry);
2064         nfs4_opendata_put(opendata);
2065         return ret;
2066 }
2067
2068 static int nfs4_do_open_expired(struct nfs_open_context *ctx, struct nfs4_state *state)
2069 {
2070         struct nfs_server *server = NFS_SERVER(state->inode);
2071         struct nfs4_exception exception = { };
2072         int err;
2073
2074         do {
2075                 err = _nfs4_open_expired(ctx, state);
2076                 trace_nfs4_open_expired(ctx, 0, err);
2077                 if (nfs4_clear_cap_atomic_open_v1(server, err, &exception))
2078                         continue;
2079                 switch (err) {
2080                 default:
2081                         goto out;
2082                 case -NFS4ERR_GRACE:
2083                 case -NFS4ERR_DELAY:
2084                         nfs4_handle_exception(server, err, &exception);
2085                         err = 0;
2086                 }
2087         } while (exception.retry);
2088 out:
2089         return err;
2090 }
2091
2092 static int nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
2093 {
2094         struct nfs_open_context *ctx;
2095         int ret;
2096
2097         ctx = nfs4_state_find_open_context(state);
2098         if (IS_ERR(ctx))
2099                 return -EAGAIN;
2100         ret = nfs4_do_open_expired(ctx, state);
2101         put_nfs_open_context(ctx);
2102         return ret;
2103 }
2104
2105 #if defined(CONFIG_NFS_V4_1)
2106 static void nfs41_clear_delegation_stateid(struct nfs4_state *state)
2107 {
2108         struct nfs_server *server = NFS_SERVER(state->inode);
2109         nfs4_stateid *stateid = &state->stateid;
2110         struct nfs_delegation *delegation;
2111         struct rpc_cred *cred = NULL;
2112         int status = -NFS4ERR_BAD_STATEID;
2113
2114         /* If a state reset has been done, test_stateid is unneeded */
2115         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
2116                 return;
2117
2118         /* Get the delegation credential for use by test/free_stateid */
2119         rcu_read_lock();
2120         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
2121         if (delegation != NULL &&
2122             nfs4_stateid_match(&delegation->stateid, stateid)) {
2123                 cred = get_rpccred(delegation->cred);
2124                 rcu_read_unlock();
2125                 status = nfs41_test_stateid(server, stateid, cred);
2126                 trace_nfs4_test_delegation_stateid(state, NULL, status);
2127         } else
2128                 rcu_read_unlock();
2129
2130         if (status != NFS_OK) {
2131                 /* Free the stateid unless the server explicitly
2132                  * informs us the stateid is unrecognized. */
2133                 if (status != -NFS4ERR_BAD_STATEID)
2134                         nfs41_free_stateid(server, stateid, cred);
2135                 nfs_remove_bad_delegation(state->inode);
2136
2137                 write_seqlock(&state->seqlock);
2138                 nfs4_stateid_copy(&state->stateid, &state->open_stateid);
2139                 write_sequnlock(&state->seqlock);
2140                 clear_bit(NFS_DELEGATED_STATE, &state->flags);
2141         }
2142
2143         if (cred != NULL)
2144                 put_rpccred(cred);
2145 }
2146
2147 /**
2148  * nfs41_check_open_stateid - possibly free an open stateid
2149  *
2150  * @state: NFSv4 state for an inode
2151  *
2152  * Returns NFS_OK if recovery for this stateid is now finished.
2153  * Otherwise a negative NFS4ERR value is returned.
2154  */
2155 static int nfs41_check_open_stateid(struct nfs4_state *state)
2156 {
2157         struct nfs_server *server = NFS_SERVER(state->inode);
2158         nfs4_stateid *stateid = &state->open_stateid;
2159         struct rpc_cred *cred = state->owner->so_cred;
2160         int status;
2161
2162         /* If a state reset has been done, test_stateid is unneeded */
2163         if ((test_bit(NFS_O_RDONLY_STATE, &state->flags) == 0) &&
2164             (test_bit(NFS_O_WRONLY_STATE, &state->flags) == 0) &&
2165             (test_bit(NFS_O_RDWR_STATE, &state->flags) == 0))
2166                 return -NFS4ERR_BAD_STATEID;
2167
2168         status = nfs41_test_stateid(server, stateid, cred);
2169         trace_nfs4_test_open_stateid(state, NULL, status);
2170         if (status != NFS_OK) {
2171                 /* Free the stateid unless the server explicitly
2172                  * informs us the stateid is unrecognized. */
2173                 if (status != -NFS4ERR_BAD_STATEID)
2174                         nfs41_free_stateid(server, stateid, cred);
2175
2176                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
2177                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
2178                 clear_bit(NFS_O_RDWR_STATE, &state->flags);
2179                 clear_bit(NFS_OPEN_STATE, &state->flags);
2180         }
2181         return status;
2182 }
2183
2184 static int nfs41_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
2185 {
2186         int status;
2187
2188         nfs41_clear_delegation_stateid(state);
2189         status = nfs41_check_open_stateid(state);
2190         if (status != NFS_OK)
2191                 status = nfs4_open_expired(sp, state);
2192         return status;
2193 }
2194 #endif
2195
2196 /*
2197  * on an EXCLUSIVE create, the server should send back a bitmask with FATTR4-*
2198  * fields corresponding to attributes that were used to store the verifier.
2199  * Make sure we clobber those fields in the later setattr call
2200  */
2201 static inline void nfs4_exclusive_attrset(struct nfs4_opendata *opendata, struct iattr *sattr)
2202 {
2203         if ((opendata->o_res.attrset[1] & FATTR4_WORD1_TIME_ACCESS) &&
2204             !(sattr->ia_valid & ATTR_ATIME_SET))
2205                 sattr->ia_valid |= ATTR_ATIME;
2206
2207         if ((opendata->o_res.attrset[1] & FATTR4_WORD1_TIME_MODIFY) &&
2208             !(sattr->ia_valid & ATTR_MTIME_SET))
2209                 sattr->ia_valid |= ATTR_MTIME;
2210 }
2211
2212 static int _nfs4_open_and_get_state(struct nfs4_opendata *opendata,
2213                 fmode_t fmode,
2214                 int flags,
2215                 struct nfs_open_context *ctx)
2216 {
2217         struct nfs4_state_owner *sp = opendata->owner;
2218         struct nfs_server *server = sp->so_server;
2219         struct dentry *dentry;
2220         struct nfs4_state *state;
2221         unsigned int seq;
2222         int ret;
2223
2224         seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
2225
2226         ret = _nfs4_proc_open(opendata);
2227         if (ret != 0) {
2228                 if (ret == -ENOENT) {
2229                         d_drop(opendata->dentry);
2230                         d_add(opendata->dentry, NULL);
2231                         nfs_set_verifier(opendata->dentry,
2232                                          nfs_save_change_attribute(opendata->dir->d_inode));
2233                 }
2234                 goto out;
2235         }
2236
2237         state = nfs4_opendata_to_nfs4_state(opendata);
2238         ret = PTR_ERR(state);
2239         if (IS_ERR(state))
2240                 goto out;
2241         if (server->caps & NFS_CAP_POSIX_LOCK)
2242                 set_bit(NFS_STATE_POSIX_LOCKS, &state->flags);
2243
2244         dentry = opendata->dentry;
2245         if (dentry->d_inode == NULL) {
2246                 /* FIXME: Is this d_drop() ever needed? */
2247                 d_drop(dentry);
2248                 dentry = d_add_unique(dentry, igrab(state->inode));
2249                 if (dentry == NULL) {
2250                         dentry = opendata->dentry;
2251                 } else if (dentry != ctx->dentry) {
2252                         dput(ctx->dentry);
2253                         ctx->dentry = dget(dentry);
2254                 }
2255                 nfs_set_verifier(dentry,
2256                                 nfs_save_change_attribute(opendata->dir->d_inode));
2257         }
2258
2259         ret = nfs4_opendata_access(sp->so_cred, opendata, state, fmode, flags);
2260         if (ret != 0)
2261                 goto out;
2262
2263         ctx->state = state;
2264         if (dentry->d_inode == state->inode) {
2265                 nfs_inode_attach_open_context(ctx);
2266                 if (read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
2267                         nfs4_schedule_stateid_recovery(server, state);
2268         }
2269 out:
2270         return ret;
2271 }
2272
2273 /*
2274  * Returns a referenced nfs4_state
2275  */
2276 static int _nfs4_do_open(struct inode *dir,
2277                         struct nfs_open_context *ctx,
2278                         int flags,
2279                         struct iattr *sattr,
2280                         struct nfs4_label *label,
2281                         int *opened)
2282 {
2283         struct nfs4_state_owner  *sp;
2284         struct nfs4_state     *state = NULL;
2285         struct nfs_server       *server = NFS_SERVER(dir);
2286         struct nfs4_opendata *opendata;
2287         struct dentry *dentry = ctx->dentry;
2288         struct rpc_cred *cred = ctx->cred;
2289         struct nfs4_threshold **ctx_th = &ctx->mdsthreshold;
2290         fmode_t fmode = ctx->mode & (FMODE_READ|FMODE_WRITE|FMODE_EXEC);
2291         enum open_claim_type4 claim = NFS4_OPEN_CLAIM_NULL;
2292         struct nfs4_label *olabel = NULL;
2293         int status;
2294
2295         /* Protect against reboot recovery conflicts */
2296         status = -ENOMEM;
2297         sp = nfs4_get_state_owner(server, cred, GFP_KERNEL);
2298         if (sp == NULL) {
2299                 dprintk("nfs4_do_open: nfs4_get_state_owner failed!\n");
2300                 goto out_err;
2301         }
2302         status = nfs4_recover_expired_lease(server);
2303         if (status != 0)
2304                 goto err_put_state_owner;
2305         if (dentry->d_inode != NULL)
2306                 nfs4_return_incompatible_delegation(dentry->d_inode, fmode);
2307         status = -ENOMEM;
2308         if (dentry->d_inode)
2309                 claim = NFS4_OPEN_CLAIM_FH;
2310         opendata = nfs4_opendata_alloc(dentry, sp, fmode, flags, sattr,
2311                         label, claim, GFP_KERNEL);
2312         if (opendata == NULL)
2313                 goto err_put_state_owner;
2314
2315         if (label) {
2316                 olabel = nfs4_label_alloc(server, GFP_KERNEL);
2317                 if (IS_ERR(olabel)) {
2318                         status = PTR_ERR(olabel);
2319                         goto err_opendata_put;
2320                 }
2321         }
2322
2323         if (server->attr_bitmask[2] & FATTR4_WORD2_MDSTHRESHOLD) {
2324                 if (!opendata->f_attr.mdsthreshold) {
2325                         opendata->f_attr.mdsthreshold = pnfs_mdsthreshold_alloc();
2326                         if (!opendata->f_attr.mdsthreshold)
2327                                 goto err_free_label;
2328                 }
2329                 opendata->o_arg.open_bitmap = &nfs4_pnfs_open_bitmap[0];
2330         }
2331         if (dentry->d_inode != NULL)
2332                 opendata->state = nfs4_get_open_state(dentry->d_inode, sp);
2333
2334         status = _nfs4_open_and_get_state(opendata, fmode, flags, ctx);
2335         if (status != 0)
2336                 goto err_free_label;
2337         state = ctx->state;
2338
2339         if ((opendata->o_arg.open_flags & O_EXCL) &&
2340             (opendata->o_arg.createmode != NFS4_CREATE_GUARDED)) {
2341                 nfs4_exclusive_attrset(opendata, sattr);
2342
2343                 nfs_fattr_init(opendata->o_res.f_attr);
2344                 status = nfs4_do_setattr(state->inode, cred,
2345                                 opendata->o_res.f_attr, sattr,
2346                                 state, label, olabel);
2347                 if (status == 0) {
2348                         nfs_setattr_update_inode(state->inode, sattr);
2349                         nfs_post_op_update_inode(state->inode, opendata->o_res.f_attr);
2350                         nfs_setsecurity(state->inode, opendata->o_res.f_attr, olabel);
2351                 }
2352         }
2353         if (opendata->file_created)
2354                 *opened |= FILE_CREATED;
2355
2356         if (pnfs_use_threshold(ctx_th, opendata->f_attr.mdsthreshold, server)) {
2357                 *ctx_th = opendata->f_attr.mdsthreshold;
2358                 opendata->f_attr.mdsthreshold = NULL;
2359         }
2360
2361         nfs4_label_free(olabel);
2362
2363         nfs4_opendata_put(opendata);
2364         nfs4_put_state_owner(sp);
2365         return 0;
2366 err_free_label:
2367         nfs4_label_free(olabel);
2368 err_opendata_put:
2369         nfs4_opendata_put(opendata);
2370 err_put_state_owner:
2371         nfs4_put_state_owner(sp);
2372 out_err:
2373         return status;
2374 }
2375
2376
2377 static struct nfs4_state *nfs4_do_open(struct inode *dir,
2378                                         struct nfs_open_context *ctx,
2379                                         int flags,
2380                                         struct iattr *sattr,
2381                                         struct nfs4_label *label,
2382                                         int *opened)
2383 {
2384         struct nfs_server *server = NFS_SERVER(dir);
2385         struct nfs4_exception exception = { };
2386         struct nfs4_state *res;
2387         int status;
2388
2389         do {
2390                 status = _nfs4_do_open(dir, ctx, flags, sattr, label, opened);
2391                 res = ctx->state;
2392                 trace_nfs4_open_file(ctx, flags, status);
2393                 if (status == 0)
2394                         break;
2395                 /* NOTE: BAD_SEQID means the server and client disagree about the
2396                  * book-keeping w.r.t. state-changing operations
2397                  * (OPEN/CLOSE/LOCK/LOCKU...)
2398                  * It is actually a sign of a bug on the client or on the server.
2399                  *
2400                  * If we receive a BAD_SEQID error in the particular case of
2401                  * doing an OPEN, we assume that nfs_increment_open_seqid() will
2402                  * have unhashed the old state_owner for us, and that we can
2403                  * therefore safely retry using a new one. We should still warn
2404                  * the user though...
2405                  */
2406                 if (status == -NFS4ERR_BAD_SEQID) {
2407                         pr_warn_ratelimited("NFS: v4 server %s "
2408                                         " returned a bad sequence-id error!\n",
2409                                         NFS_SERVER(dir)->nfs_client->cl_hostname);
2410                         exception.retry = 1;
2411                         continue;
2412                 }
2413                 /*
2414                  * BAD_STATEID on OPEN means that the server cancelled our
2415                  * state before it received the OPEN_CONFIRM.
2416                  * Recover by retrying the request as per the discussion
2417                  * on Page 181 of RFC3530.
2418                  */
2419                 if (status == -NFS4ERR_BAD_STATEID) {
2420                         exception.retry = 1;
2421                         continue;
2422                 }
2423                 if (status == -EAGAIN) {
2424                         /* We must have found a delegation */
2425                         exception.retry = 1;
2426                         continue;
2427                 }
2428                 if (nfs4_clear_cap_atomic_open_v1(server, status, &exception))
2429                         continue;
2430                 res = ERR_PTR(nfs4_handle_exception(server,
2431                                         status, &exception));
2432         } while (exception.retry);
2433         return res;
2434 }
2435
2436 static int _nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
2437                             struct nfs_fattr *fattr, struct iattr *sattr,
2438                             struct nfs4_state *state, struct nfs4_label *ilabel,
2439                             struct nfs4_label *olabel)
2440 {
2441         struct nfs_server *server = NFS_SERVER(inode);
2442         struct nfs_setattrargs  arg = {
2443                 .fh             = NFS_FH(inode),
2444                 .iap            = sattr,
2445                 .server         = server,
2446                 .bitmask = server->attr_bitmask,
2447                 .label          = ilabel,
2448         };
2449         struct nfs_setattrres  res = {
2450                 .fattr          = fattr,
2451                 .label          = olabel,
2452                 .server         = server,
2453         };
2454         struct rpc_message msg = {
2455                 .rpc_proc       = &nfs4_procedures[NFSPROC4_CLNT_SETATTR],
2456                 .rpc_argp       = &arg,
2457                 .rpc_resp       = &res,
2458                 .rpc_cred       = cred,
2459         };
2460         unsigned long timestamp = jiffies;
2461         fmode_t fmode;
2462         bool truncate;
2463         int status;
2464
2465         arg.bitmask = nfs4_bitmask(server, ilabel);
2466         if (ilabel)
2467                 arg.bitmask = nfs4_bitmask(server, olabel);
2468
2469         nfs_fattr_init(fattr);
2470
2471         /* Servers should only apply open mode checks for file size changes */
2472         truncate = (sattr->ia_valid & ATTR_SIZE) ? true : false;
2473         fmode = truncate ? FMODE_WRITE : FMODE_READ;
2474
2475         if (nfs4_copy_delegation_stateid(&arg.stateid, inode, fmode)) {
2476                 /* Use that stateid */
2477         } else if (truncate && state != NULL) {
2478                 struct nfs_lockowner lockowner = {
2479                         .l_owner = current->files,
2480                         .l_pid = current->tgid,
2481                 };
2482                 if (!nfs4_valid_open_stateid(state))
2483                         return -EBADF;
2484                 if (nfs4_select_rw_stateid(&arg.stateid, state, FMODE_WRITE,
2485                                 &lockowner) == -EIO)
2486                         return -EBADF;
2487         } else
2488                 nfs4_stateid_copy(&arg.stateid, &zero_stateid);
2489
2490         status = nfs4_call_sync(server->client, server, &msg, &arg.seq_args, &res.seq_res, 1);
2491         if (status == 0 && state != NULL)
2492                 renew_lease(server, timestamp);
2493         return status;
2494 }
2495
2496 static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
2497                            struct nfs_fattr *fattr, struct iattr *sattr,
2498                            struct nfs4_state *state, struct nfs4_label *ilabel,
2499                            struct nfs4_label *olabel)
2500 {
2501         struct nfs_server *server = NFS_SERVER(inode);
2502         struct nfs4_exception exception = {
2503                 .state = state,
2504                 .inode = inode,
2505         };
2506         int err;
2507         do {
2508                 err = _nfs4_do_setattr(inode, cred, fattr, sattr, state, ilabel, olabel);
2509                 trace_nfs4_setattr(inode, err);
2510                 switch (err) {
2511                 case -NFS4ERR_OPENMODE:
2512                         if (!(sattr->ia_valid & ATTR_SIZE)) {
2513                                 pr_warn_once("NFSv4: server %s is incorrectly "
2514                                                 "applying open mode checks to "
2515                                                 "a SETATTR that is not "
2516                                                 "changing file size.\n",
2517                                                 server->nfs_client->cl_hostname);
2518                         }
2519                         if (state && !(state->state & FMODE_WRITE)) {
2520                                 err = -EBADF;
2521                                 if (sattr->ia_valid & ATTR_OPEN)
2522                                         err = -EACCES;
2523                                 goto out;
2524                         }
2525                 }
2526                 err = nfs4_handle_exception(server, err, &exception);
2527         } while (exception.retry);
2528 out:
2529         return err;
2530 }
2531
2532 struct nfs4_closedata {
2533         struct inode *inode;
2534         struct nfs4_state *state;
2535         struct nfs_closeargs arg;
2536         struct nfs_closeres res;
2537         struct nfs_fattr fattr;
2538         unsigned long timestamp;
2539         bool roc;
2540         u32 roc_barrier;
2541 };
2542
2543 static void nfs4_free_closedata(void *data)
2544 {
2545         struct nfs4_closedata *calldata = data;
2546         struct nfs4_state_owner *sp = calldata->state->owner;
2547         struct super_block *sb = calldata->state->inode->i_sb;
2548
2549         if (calldata->roc)
2550                 pnfs_roc_release(calldata->state->inode);
2551         nfs4_put_open_state(calldata->state);
2552         nfs_free_seqid(calldata->arg.seqid);
2553         nfs4_put_state_owner(sp);
2554         nfs_sb_deactive(sb);
2555         kfree(calldata);
2556 }
2557
2558 static void nfs4_close_done(struct rpc_task *task, void *data)
2559 {
2560         struct nfs4_closedata *calldata = data;
2561         struct nfs4_state *state = calldata->state;
2562         struct nfs_server *server = NFS_SERVER(calldata->inode);
2563
2564         dprintk("%s: begin!\n", __func__);
2565         if (!nfs4_sequence_done(task, &calldata->res.seq_res))
2566                 return;
2567         trace_nfs4_close(state, &calldata->arg, &calldata->res, task->tk_status);
2568         /* hmm. we are done with the inode, and in the process of freeing
2569          * the state_owner. we keep this around to process errors
2570          */
2571         switch (task->tk_status) {
2572                 case 0:
2573                         if (calldata->roc)
2574                                 pnfs_roc_set_barrier(state->inode,
2575                                                      calldata->roc_barrier);
2576                         nfs_clear_open_stateid(state, &calldata->res.stateid, 0);
2577                         renew_lease(server, calldata->timestamp);
2578                         goto out_release;
2579                 case -NFS4ERR_ADMIN_REVOKED:
2580                 case -NFS4ERR_STALE_STATEID:
2581                 case -NFS4ERR_OLD_STATEID:
2582                 case -NFS4ERR_BAD_STATEID:
2583                 case -NFS4ERR_EXPIRED:
2584                         if (calldata->arg.fmode == 0)
2585                                 break;
2586                 default:
2587                         if (nfs4_async_handle_error(task, server, state) == -EAGAIN) {
2588                                 rpc_restart_call_prepare(task);
2589                                 goto out_release;
2590                         }
2591         }
2592         nfs_clear_open_stateid(state, NULL, calldata->arg.fmode);
2593 out_release:
2594         nfs_release_seqid(calldata->arg.seqid);
2595         nfs_refresh_inode(calldata->inode, calldata->res.fattr);
2596         dprintk("%s: done, ret = %d!\n", __func__, task->tk_status);
2597 }
2598
2599 static void nfs4_close_prepare(struct rpc_task *task, void *data)
2600 {
2601         struct nfs4_closedata *calldata = data;
2602         struct nfs4_state *state = calldata->state;
2603         struct inode *inode = calldata->inode;
2604         bool is_rdonly, is_wronly, is_rdwr;
2605         int call_close = 0;
2606
2607         dprintk("%s: begin!\n", __func__);
2608         if (nfs_wait_on_sequence(calldata->arg.seqid, task) != 0)
2609                 goto out_wait;
2610
2611         task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_DOWNGRADE];
2612         spin_lock(&state->owner->so_lock);
2613         is_rdwr = test_bit(NFS_O_RDWR_STATE, &state->flags);
2614         is_rdonly = test_bit(NFS_O_RDONLY_STATE, &state->flags);
2615         is_wronly = test_bit(NFS_O_WRONLY_STATE, &state->flags);
2616         /* Calculate the current open share mode */
2617         calldata->arg.fmode = 0;
2618         if (is_rdonly || is_rdwr)
2619                 calldata->arg.fmode |= FMODE_READ;
2620         if (is_wronly || is_rdwr)
2621                 calldata->arg.fmode |= FMODE_WRITE;
2622         /* Calculate the change in open mode */
2623         if (state->n_rdwr == 0) {
2624                 if (state->n_rdonly == 0) {
2625                         call_close |= is_rdonly || is_rdwr;
2626                         calldata->arg.fmode &= ~FMODE_READ;
2627                 }
2628                 if (state->n_wronly == 0) {
2629                         call_close |= is_wronly || is_rdwr;
2630                         calldata->arg.fmode &= ~FMODE_WRITE;
2631                 }
2632         }
2633         if (!nfs4_valid_open_stateid(state))
2634                 call_close = 0;
2635         spin_unlock(&state->owner->so_lock);
2636
2637         if (!call_close) {
2638                 /* Note: exit _without_ calling nfs4_close_done */
2639                 goto out_no_action;
2640         }
2641
2642         if (calldata->arg.fmode == 0) {
2643                 task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE];
2644                 if (calldata->roc &&
2645                     pnfs_roc_drain(inode, &calldata->roc_barrier, task)) {
2646                         nfs_release_seqid(calldata->arg.seqid);
2647                         goto out_wait;
2648                     }
2649         }
2650
2651         nfs_fattr_init(calldata->res.fattr);
2652         calldata->timestamp = jiffies;
2653         if (nfs4_setup_sequence(NFS_SERVER(inode),
2654                                 &calldata->arg.seq_args,
2655                                 &calldata->res.seq_res,
2656                                 task) != 0)
2657                 nfs_release_seqid(calldata->arg.seqid);
2658         dprintk("%s: done!\n", __func__);
2659         return;
2660 out_no_action:
2661         task->tk_action = NULL;
2662 out_wait:
2663         nfs4_sequence_done(task, &calldata->res.seq_res);
2664 }
2665
2666 static const struct rpc_call_ops nfs4_close_ops = {
2667         .rpc_call_prepare = nfs4_close_prepare,
2668         .rpc_call_done = nfs4_close_done,
2669         .rpc_release = nfs4_free_closedata,
2670 };
2671
2672 static bool nfs4_state_has_opener(struct nfs4_state *state)
2673 {
2674         /* first check existing openers */
2675         if (test_bit(NFS_O_RDONLY_STATE, &state->flags) != 0 &&
2676             state->n_rdonly != 0)
2677                 return true;
2678
2679         if (test_bit(NFS_O_WRONLY_STATE, &state->flags) != 0 &&
2680             state->n_wronly != 0)
2681                 return true;
2682
2683         if (test_bit(NFS_O_RDWR_STATE, &state->flags) != 0 &&
2684             state->n_rdwr != 0)
2685                 return true;
2686
2687         return false;
2688 }
2689
2690 static bool nfs4_roc(struct inode *inode)
2691 {
2692         struct nfs_inode *nfsi = NFS_I(inode);
2693         struct nfs_open_context *ctx;
2694         struct nfs4_state *state;
2695
2696         spin_lock(&inode->i_lock);
2697         list_for_each_entry(ctx, &nfsi->open_files, list) {
2698                 state = ctx->state;
2699                 if (state == NULL)
2700                         continue;
2701                 if (nfs4_state_has_opener(state)) {
2702                         spin_unlock(&inode->i_lock);
2703                         return false;
2704                 }
2705         }
2706         spin_unlock(&inode->i_lock);
2707
2708         if (nfs4_check_delegation(inode, FMODE_READ))
2709                 return false;
2710
2711         return pnfs_roc(inode);
2712 }
2713
2714 /*
2715  * It is possible for data to be read/written from a mem-mapped file
2716  * after the sys_close call (which hits the vfs layer as a flush).
2717  * This means that we can't safely call nfsv4 close on a file until
2718  * the inode is cleared. This in turn means that we are not good
2719  * NFSv4 citizens - we do not indicate to the server to update the file's
2720  * share state even when we are done with one of the three share
2721  * stateid's in the inode.
2722  *
2723  * NOTE: Caller must be holding the sp->so_owner semaphore!
2724  */
2725 int nfs4_do_close(struct nfs4_state *state, gfp_t gfp_mask, int wait)
2726 {
2727         struct nfs_server *server = NFS_SERVER(state->inode);
2728         struct nfs4_closedata *calldata;
2729         struct nfs4_state_owner *sp = state->owner;
2730         struct rpc_task *task;
2731         struct rpc_message msg = {
2732                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE],
2733                 .rpc_cred = state->owner->so_cred,
2734         };
2735         struct rpc_task_setup task_setup_data = {
2736                 .rpc_client = server->client,
2737                 .rpc_message = &msg,
2738                 .callback_ops = &nfs4_close_ops,
2739                 .workqueue = nfsiod_workqueue,
2740                 .flags = RPC_TASK_ASYNC,
2741         };
2742         int status = -ENOMEM;
2743
2744         nfs4_state_protect(server->nfs_client, NFS_SP4_MACH_CRED_CLEANUP,
2745                 &task_setup_data.rpc_client, &msg);
2746
2747         calldata = kzalloc(sizeof(*calldata), gfp_mask);
2748         if (calldata == NULL)
2749                 goto out;
2750         nfs4_init_sequence(&calldata->arg.seq_args, &calldata->res.seq_res, 1);
2751         calldata->inode = state->inode;
2752         calldata->state = state;
2753         calldata->arg.fh = NFS_FH(state->inode);
2754         calldata->arg.stateid = &state->open_stateid;
2755         /* Serialization for the sequence id */
2756         calldata->arg.seqid = nfs_alloc_seqid(&state->owner->so_seqid, gfp_mask);
2757         if (calldata->arg.seqid == NULL)
2758                 goto out_free_calldata;
2759         calldata->arg.fmode = 0;
2760         calldata->arg.bitmask = server->cache_consistency_bitmask;
2761         calldata->res.fattr = &calldata->fattr;
2762         calldata->res.seqid = calldata->arg.seqid;
2763         calldata->res.server = server;
2764         calldata->roc = nfs4_roc(state->inode);
2765         nfs_sb_active(calldata->inode->i_sb);
2766
2767         msg.rpc_argp = &calldata->arg;
2768         msg.rpc_resp = &calldata->res;
2769         task_setup_data.callback_data = calldata;
2770         task = rpc_run_task(&task_setup_data);
2771         if (IS_ERR(task))
2772                 return PTR_ERR(task);
2773         status = 0;
2774         if (wait)
2775                 status = rpc_wait_for_completion_task(task);
2776         rpc_put_task(task);
2777         return status;
2778 out_free_calldata:
2779         kfree(calldata);
2780 out:
2781         nfs4_put_open_state(state);
2782         nfs4_put_state_owner(sp);
2783         return status;
2784 }
2785
2786 static struct inode *
2787 nfs4_atomic_open(struct inode *dir, struct nfs_open_context *ctx,
2788                 int open_flags, struct iattr *attr, int *opened)
2789 {
2790         struct nfs4_state *state;
2791         struct nfs4_label l = {0, 0, 0, NULL}, *label = NULL;
2792
2793         label = nfs4_label_init_security(dir, ctx->dentry, attr, &l);
2794
2795         /* Protect against concurrent sillydeletes */
2796         state = nfs4_do_open(dir, ctx, open_flags, attr, label, opened);
2797
2798         nfs4_label_release_security(label);
2799
2800         if (IS_ERR(state))
2801                 return ERR_CAST(state);
2802         return state->inode;
2803 }
2804
2805 static void nfs4_close_context(struct nfs_open_context *ctx, int is_sync)
2806 {
2807         if (ctx->state == NULL)
2808                 return;
2809         if (is_sync)
2810                 nfs4_close_sync(ctx->state, ctx->mode);
2811         else
2812                 nfs4_close_state(ctx->state, ctx->mode);
2813 }
2814
2815 #define FATTR4_WORD1_NFS40_MASK (2*FATTR4_WORD1_MOUNTED_ON_FILEID - 1UL)
2816 #define FATTR4_WORD2_NFS41_MASK (2*FATTR4_WORD2_SUPPATTR_EXCLCREAT - 1UL)
2817 #define FATTR4_WORD2_NFS42_MASK (2*FATTR4_WORD2_SECURITY_LABEL - 1UL)
2818
2819 static int _nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
2820 {
2821         struct nfs4_server_caps_arg args = {
2822                 .fhandle = fhandle,
2823         };
2824         struct nfs4_server_caps_res res = {};
2825         struct rpc_message msg = {
2826                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_SERVER_CAPS],
2827                 .rpc_argp = &args,
2828                 .rpc_resp = &res,
2829         };
2830         int status;
2831
2832         status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
2833         if (status == 0) {
2834                 /* Sanity check the server answers */
2835                 switch (server->nfs_client->cl_minorversion) {
2836                 case 0:
2837                         res.attr_bitmask[1] &= FATTR4_WORD1_NFS40_MASK;
2838                         res.attr_bitmask[2] = 0;
2839                         break;
2840                 case 1:
2841                         res.attr_bitmask[2] &= FATTR4_WORD2_NFS41_MASK;
2842                         break;
2843                 case 2:
2844                         res.attr_bitmask[2] &= FATTR4_WORD2_NFS42_MASK;
2845                 }
2846                 memcpy(server->attr_bitmask, res.attr_bitmask, sizeof(server->attr_bitmask));
2847                 server->caps &= ~(NFS_CAP_ACLS|NFS_CAP_HARDLINKS|
2848                                 NFS_CAP_SYMLINKS|NFS_CAP_FILEID|
2849                                 NFS_CAP_MODE|NFS_CAP_NLINK|NFS_CAP_OWNER|
2850                                 NFS_CAP_OWNER_GROUP|NFS_CAP_ATIME|
2851                                 NFS_CAP_CTIME|NFS_CAP_MTIME|
2852                                 NFS_CAP_SECURITY_LABEL);
2853                 if (res.attr_bitmask[0] & FATTR4_WORD0_ACL &&
2854                                 res.acl_bitmask & ACL4_SUPPORT_ALLOW_ACL)
2855                         server->caps |= NFS_CAP_ACLS;
2856                 if (res.has_links != 0)
2857                         server->caps |= NFS_CAP_HARDLINKS;
2858                 if (res.has_symlinks != 0)
2859                         server->caps |= NFS_CAP_SYMLINKS;
2860                 if (res.attr_bitmask[0] & FATTR4_WORD0_FILEID)
2861                         server->caps |= NFS_CAP_FILEID;
2862                 if (res.attr_bitmask[1] & FATTR4_WORD1_MODE)
2863                         server->caps |= NFS_CAP_MODE;
2864                 if (res.attr_bitmask[1] & FATTR4_WORD1_NUMLINKS)
2865                         server->caps |= NFS_CAP_NLINK;
2866                 if (res.attr_bitmask[1] & FATTR4_WORD1_OWNER)
2867                         server->caps |= NFS_CAP_OWNER;
2868                 if (res.attr_bitmask[1] & FATTR4_WORD1_OWNER_GROUP)
2869                         server->caps |= NFS_CAP_OWNER_GROUP;
2870                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_ACCESS)
2871                         server->caps |= NFS_CAP_ATIME;
2872                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_METADATA)
2873                         server->caps |= NFS_CAP_CTIME;
2874                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_MODIFY)
2875                         server->caps |= NFS_CAP_MTIME;
2876 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
2877                 if (res.attr_bitmask[2] & FATTR4_WORD2_SECURITY_LABEL)
2878                         server->caps |= NFS_CAP_SECURITY_LABEL;
2879 #endif
2880                 memcpy(server->attr_bitmask_nl, res.attr_bitmask,
2881                                 sizeof(server->attr_bitmask));
2882                 server->attr_bitmask_nl[2] &= ~FATTR4_WORD2_SECURITY_LABEL;
2883
2884                 memcpy(server->cache_consistency_bitmask, res.attr_bitmask, sizeof(server->cache_consistency_bitmask));
2885                 server->cache_consistency_bitmask[0] &= FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;
2886                 server->cache_consistency_bitmask[1] &= FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;
2887                 server->cache_consistency_bitmask[2] = 0;
2888                 server->acl_bitmask = res.acl_bitmask;
2889                 server->fh_expire_type = res.fh_expire_type;
2890         }
2891
2892         return status;
2893 }
2894
2895 int nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
2896 {
2897         struct nfs4_exception exception = { };
2898         int err;
2899         do {
2900                 err = nfs4_handle_exception(server,
2901                                 _nfs4_server_capabilities(server, fhandle),
2902                                 &exception);
2903         } while (exception.retry);
2904         return err;
2905 }
2906
2907 static int _nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
2908                 struct nfs_fsinfo *info)
2909 {
2910         u32 bitmask[3];
2911         struct nfs4_lookup_root_arg args = {
2912                 .bitmask = bitmask,
2913         };
2914         struct nfs4_lookup_res res = {
2915                 .server = server,
2916                 .fattr = info->fattr,
2917                 .fh = fhandle,
2918         };
2919         struct rpc_message msg = {
2920                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOOKUP_ROOT],
2921                 .rpc_argp = &args,
2922                 .rpc_resp = &res,
2923         };
2924
2925         bitmask[0] = nfs4_fattr_bitmap[0];
2926         bitmask[1] = nfs4_fattr_bitmap[1];
2927         /*
2928          * Process the label in the upcoming getfattr
2929          */
2930         bitmask[2] = nfs4_fattr_bitmap[2] & ~FATTR4_WORD2_SECURITY_LABEL;
2931
2932         nfs_fattr_init(info->fattr);
2933         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
2934 }
2935
2936 static int nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
2937                 struct nfs_fsinfo *info)
2938 {
2939         struct nfs4_exception exception = { };
2940         int err;
2941         do {
2942                 err = _nfs4_lookup_root(server, fhandle, info);
2943                 trace_nfs4_lookup_root(server, fhandle, info->fattr, err);
2944                 switch (err) {
2945                 case 0:
2946                 case -NFS4ERR_WRONGSEC:
2947                         goto out;
2948                 default:
2949                         err = nfs4_handle_exception(server, err, &exception);
2950                 }
2951         } while (exception.retry);
2952 out:
2953         return err;
2954 }
2955
2956 static int nfs4_lookup_root_sec(struct nfs_server *server, struct nfs_fh *fhandle,
2957                                 struct nfs_fsinfo *info, rpc_authflavor_t flavor)
2958 {
2959         struct rpc_auth_create_args auth_args = {
2960                 .pseudoflavor = flavor,
2961         };
2962         struct rpc_auth *auth;
2963         int ret;
2964
2965         auth = rpcauth_create(&auth_args, server->client);
2966         if (IS_ERR(auth)) {
2967                 ret = -EACCES;
2968                 goto out;
2969         }
2970         ret = nfs4_lookup_root(server, fhandle, info);
2971 out:
2972         return ret;
2973 }
2974
2975 /*
2976  * Retry pseudoroot lookup with various security flavors.  We do this when:
2977  *
2978  *   NFSv4.0: the PUTROOTFH operation returns NFS4ERR_WRONGSEC
2979  *   NFSv4.1: the server does not support the SECINFO_NO_NAME operation
2980  *
2981  * Returns zero on success, or a negative NFS4ERR value, or a
2982  * negative errno value.
2983  */
2984 static int nfs4_find_root_sec(struct nfs_server *server, struct nfs_fh *fhandle,
2985                               struct nfs_fsinfo *info)
2986 {
2987         /* Per 3530bis 15.33.5 */
2988         static const rpc_authflavor_t flav_array[] = {
2989                 RPC_AUTH_GSS_KRB5P,
2990                 RPC_AUTH_GSS_KRB5I,
2991                 RPC_AUTH_GSS_KRB5,
2992                 RPC_AUTH_UNIX,                  /* courtesy */
2993                 RPC_AUTH_NULL,
2994         };
2995         int status = -EPERM;
2996         size_t i;
2997
2998         if (server->auth_info.flavor_len > 0) {
2999                 /* try each flavor specified by user */
3000                 for (i = 0; i < server->auth_info.flavor_len; i++) {
3001                         status = nfs4_lookup_root_sec(server, fhandle, info,
3002                                                 server->auth_info.flavors[i]);
3003                         if (status == -NFS4ERR_WRONGSEC || status == -EACCES)
3004                                 continue;
3005                         break;
3006                 }
3007         } else {
3008                 /* no flavors specified by user, try default list */
3009                 for (i = 0; i < ARRAY_SIZE(flav_array); i++) {
3010                         status = nfs4_lookup_root_sec(server, fhandle, info,
3011                                                       flav_array[i]);
3012                         if (status == -NFS4ERR_WRONGSEC || status == -EACCES)
3013                                 continue;
3014                         break;
3015                 }
3016         }
3017
3018         /*
3019          * -EACCESS could mean that the user doesn't have correct permissions
3020          * to access the mount.  It could also mean that we tried to mount
3021          * with a gss auth flavor, but rpc.gssd isn't running.  Either way,
3022          * existing mount programs don't handle -EACCES very well so it should
3023          * be mapped to -EPERM instead.
3024          */
3025         if (status == -EACCES)
3026                 status = -EPERM;
3027         return status;
3028 }
3029
3030 static int nfs4_do_find_root_sec(struct nfs_server *server,
3031                 struct nfs_fh *fhandle, struct nfs_fsinfo *info)
3032 {
3033         int mv = server->nfs_client->cl_minorversion;
3034         return nfs_v4_minor_ops[mv]->find_root_sec(server, fhandle, info);
3035 }
3036
3037 /**
3038  * nfs4_proc_get_rootfh - get file handle for server's pseudoroot
3039  * @server: initialized nfs_server handle
3040  * @fhandle: we fill in the pseudo-fs root file handle
3041  * @info: we fill in an FSINFO struct
3042  * @auth_probe: probe the auth flavours
3043  *
3044  * Returns zero on success, or a negative errno.
3045  */
3046 int nfs4_proc_get_rootfh(struct nfs_server *server, struct nfs_fh *fhandle,
3047                          struct nfs_fsinfo *info,
3048                          bool auth_probe)
3049 {
3050         int status;
3051
3052         switch (auth_probe) {
3053         case false:
3054                 status = nfs4_lookup_root(server, fhandle, info);
3055                 if (status != -NFS4ERR_WRONGSEC)
3056                         break;
3057         default:
3058                 status = nfs4_do_find_root_sec(server, fhandle, info);
3059         }
3060
3061         if (status == 0)
3062                 status = nfs4_server_capabilities(server, fhandle);
3063         if (status == 0)
3064                 status = nfs4_do_fsinfo(server, fhandle, info);
3065
3066         return nfs4_map_errors(status);
3067 }
3068
3069 static int nfs4_proc_get_root(struct nfs_server *server, struct nfs_fh *mntfh,
3070                               struct nfs_fsinfo *info)
3071 {
3072         int error;
3073         struct nfs_fattr *fattr = info->fattr;
3074         struct nfs4_label *label = NULL;
3075
3076         error = nfs4_server_capabilities(server, mntfh);
3077         if (error < 0) {
3078                 dprintk("nfs4_get_root: getcaps error = %d\n", -error);
3079                 return error;
3080         }
3081
3082         label = nfs4_label_alloc(server, GFP_KERNEL);
3083         if (IS_ERR(label))
3084                 return PTR_ERR(label);
3085
3086         error = nfs4_proc_getattr(server, mntfh, fattr, label);
3087         if (error < 0) {
3088                 dprintk("nfs4_get_root: getattr error = %d\n", -error);
3089                 goto err_free_label;
3090         }
3091
3092         if (fattr->valid & NFS_ATTR_FATTR_FSID &&
3093             !nfs_fsid_equal(&server->fsid, &fattr->fsid))
3094                 memcpy(&server->fsid, &fattr->fsid, sizeof(server->fsid));
3095
3096 err_free_label:
3097         nfs4_label_free(label);
3098
3099         return error;
3100 }
3101
3102 /*
3103  * Get locations and (maybe) other attributes of a referral.
3104  * Note that we'll actually follow the referral later when
3105  * we detect fsid mismatch in inode revalidation
3106  */
3107 static int nfs4_get_referral(struct rpc_clnt *client, struct inode *dir,
3108                              const struct qstr *name, struct nfs_fattr *fattr,
3109                              struct nfs_fh *fhandle)
3110 {
3111         int status = -ENOMEM;
3112         struct page *page = NULL;
3113         struct nfs4_fs_locations *locations = NULL;
3114
3115         page = alloc_page(GFP_KERNEL);
3116         if (page == NULL)
3117                 goto out;
3118         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
3119         if (locations == NULL)
3120                 goto out;
3121
3122         status = nfs4_proc_fs_locations(client, dir, name, locations, page);
3123         if (status != 0)
3124                 goto out;
3125
3126         /*
3127          * If the fsid didn't change, this is a migration event, not a
3128          * referral.  Cause us to drop into the exception handler, which
3129          * will kick off migration recovery.
3130          */
3131         if (nfs_fsid_equal(&NFS_SERVER(dir)->fsid, &locations->fattr.fsid)) {
3132                 dprintk("%s: server did not return a different fsid for"
3133                         " a referral at %s\n", __func__, name->name);
3134                 status = -NFS4ERR_MOVED;
3135                 goto out;
3136         }
3137         /* Fixup attributes for the nfs_lookup() call to nfs_fhget() */
3138         nfs_fixup_referral_attributes(&locations->fattr);
3139
3140         /* replace the lookup nfs_fattr with the locations nfs_fattr */
3141         memcpy(fattr, &locations->fattr, sizeof(struct nfs_fattr));
3142         memset(fhandle, 0, sizeof(struct nfs_fh));
3143 out:
3144         if (page)
3145                 __free_page(page);
3146         kfree(locations);
3147         return status;
3148 }
3149
3150 static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle,
3151                                 struct nfs_fattr *fattr, struct nfs4_label *label)
3152 {
3153         struct nfs4_getattr_arg args = {
3154                 .fh = fhandle,
3155                 .bitmask = server->attr_bitmask,
3156         };
3157         struct nfs4_getattr_res res = {
3158                 .fattr = fattr,
3159                 .label = label,
3160                 .server = server,
3161         };
3162         struct rpc_message msg = {
3163                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETATTR],
3164                 .rpc_argp = &args,
3165                 .rpc_resp = &res,
3166         };
3167
3168         args.bitmask = nfs4_bitmask(server, label);
3169
3170         nfs_fattr_init(fattr);
3171         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3172 }
3173
3174 static int nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle,
3175                                 struct nfs_fattr *fattr, struct nfs4_label *label)
3176 {
3177         struct nfs4_exception exception = { };
3178         int err;
3179         do {
3180                 err = _nfs4_proc_getattr(server, fhandle, fattr, label);
3181                 trace_nfs4_getattr(server, fhandle, fattr, err);
3182                 err = nfs4_handle_exception(server, err,
3183                                 &exception);
3184         } while (exception.retry);
3185         return err;
3186 }
3187
3188 /*
3189  * The file is not closed if it is opened due to the a request to change
3190  * the size of the file. The open call will not be needed once the
3191  * VFS layer lookup-intents are implemented.
3192  *
3193  * Close is called when the inode is destroyed.
3194  * If we haven't opened the file for O_WRONLY, we
3195  * need to in the size_change case to obtain a stateid.
3196  *
3197  * Got race?
3198  * Because OPEN is always done by name in nfsv4, it is
3199  * possible that we opened a different file by the same
3200  * name.  We can recognize this race condition, but we
3201  * can't do anything about it besides returning an error.