fs/nfs/inode.c

   1 /*
   2  *  linux/fs/nfs/inode.c
   3  *
   4  *  Copyright (C) 1992  Rick Sladkey
   5  *
   6  *  nfs inode and superblock handling functions
   7  *
   8  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
   9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
  10  *
  11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
  12  *  J.S.Peatfield@damtp.cam.ac.uk
  13  *
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/init.h>
  18 #include <linux/sched.h>
  19 #include <linux/time.h>
  20 #include <linux/kernel.h>
  21 #include <linux/mm.h>
  22 #include <linux/string.h>
  23 #include <linux/stat.h>
  24 #include <linux/errno.h>
  25 #include <linux/unistd.h>
  26 #include <linux/sunrpc/clnt.h>
  27 #include <linux/sunrpc/stats.h>
  28 #include <linux/sunrpc/metrics.h>
  29 #include <linux/nfs_fs.h>
  30 #include <linux/nfs_mount.h>
  31 #include <linux/nfs4_mount.h>
  32 #include <linux/lockd/bind.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/mount.h>
  35 #include <linux/nfs_idmap.h>
  36 #include <linux/vfs.h>
  37 #include <linux/inet.h>
  38 #include <linux/nfs_xdr.h>
  39 #include <linux/slab.h>
  40
  41 #include <asm/system.h>
  42 #include <asm/uaccess.h>
  43
  44 #include "nfs4_fs.h"
  45 #include "callback.h"
  46 #include "delegation.h"
  47 #include "iostat.h"
  48 #include "internal.h"
  49 #include "fscache.h"
  50 #include "dns_resolve.h"
  51
  52 #define NFSDBG_FACILITY         NFSDBG_VFS
  53
  54 #define NFS_64_BIT_INODE_NUMBERS_ENABLED        1
  55
  56 /* Default is to see 64-bit inode numbers */
  57 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
  58
  59 static void nfs_invalidate_inode(struct inode *);
  60 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
  61
  62 static struct kmem_cache * nfs_inode_cachep;
  63
  64 static inline unsigned long
  65 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
  66 {
  67         return nfs_fileid_to_ino_t(fattr->fileid);
  68 }
  69
  70 /**
  71  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
  72  * @word: long word containing the bit lock
  73  */
  74 int nfs_wait_bit_killable(void *word)
  75 {
  76         if (fatal_signal_pending(current))
  77                 return -ERESTARTSYS;
  78         schedule();
  79         return 0;
  80 }
  81
  82 /**
  83  * nfs_compat_user_ino64 - returns the user-visible inode number
  84  * @fileid: 64-bit fileid
  85  *
  86  * This function returns a 32-bit inode number if the boot parameter
  87  * nfs.enable_ino64 is zero.
  88  */
  89 u64 nfs_compat_user_ino64(u64 fileid)
  90 {
  91         int ino;
  92
  93         if (enable_ino64)
  94                 return fileid;
  95         ino = fileid;
  96         if (sizeof(ino) < sizeof(fileid))
  97                 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
  98         return ino;
  99 }
 100
 101 void nfs_clear_inode(struct inode *inode)
 102 {
 103         /*
 104          * The following should never happen...
 105          */
 106         BUG_ON(nfs_have_writebacks(inode));
 107         BUG_ON(!list_empty(&NFS_I(inode)->open_files));
 108         nfs_zap_acl_cache(inode);
 109         nfs_access_zap_cache(inode);
 110         nfs_fscache_release_inode_cookie(inode);
 111 }
 112
 113 /**
 114  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
 115  */
 116 int nfs_sync_mapping(struct address_space *mapping)
 117 {
 118         int ret = 0;
 119
 120         if (mapping->nrpages != 0) {
 121                 unmap_mapping_range(mapping, 0, 0, 0);
 122                 ret = nfs_wb_all(mapping->host);
 123         }
 124         return ret;
 125 }
 126
 127 /*
 128  * Invalidate the local caches
 129  */
 130 static void nfs_zap_caches_locked(struct inode *inode)
 131 {
 132         struct nfs_inode *nfsi = NFS_I(inode);
 133         int mode = inode->i_mode;
 134
 135         nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
 136
 137         nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 138         nfsi->attrtimeo_timestamp = jiffies;
 139
 140         memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
 141         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
 142                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
 143         else
 144                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
 145 }
 146
 147 void nfs_zap_caches(struct inode *inode)
 148 {
 149         spin_lock(&inode->i_lock);
 150         nfs_zap_caches_locked(inode);
 151         spin_unlock(&inode->i_lock);
 152 }
 153
 154 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
 155 {
 156         if (mapping->nrpages != 0) {
 157                 spin_lock(&inode->i_lock);
 158                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
 159                 spin_unlock(&inode->i_lock);
 160         }
 161 }
 162
 163 void nfs_zap_acl_cache(struct inode *inode)
 164 {
 165         void (*clear_acl_cache)(struct inode *);
 166
 167         clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
 168         if (clear_acl_cache != NULL)
 169                 clear_acl_cache(inode);
 170         spin_lock(&inode->i_lock);
 171         NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
 172         spin_unlock(&inode->i_lock);
 173 }
 174
 175 void nfs_invalidate_atime(struct inode *inode)
 176 {
 177         spin_lock(&inode->i_lock);
 178         NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
 179         spin_unlock(&inode->i_lock);
 180 }
 181
 182 /*
 183  * Invalidate, but do not unhash, the inode.
 184  * NB: must be called with inode->i_lock held!
 185  */
 186 static void nfs_invalidate_inode(struct inode *inode)
 187 {
 188         set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 189         nfs_zap_caches_locked(inode);
 190 }
 191
 192 struct nfs_find_desc {
 193         struct nfs_fh           *fh;
 194         struct nfs_fattr        *fattr;
 195 };
 196
 197 /*
 198  * In NFSv3 we can have 64bit inode numbers. In order to support
 199  * this, and re-exported directories (also seen in NFSv2)
 200  * we are forced to allow 2 different inodes to have the same
 201  * i_ino.
 202  */
 203 static int
 204 nfs_find_actor(struct inode *inode, void *opaque)
 205 {
 206         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
 207         struct nfs_fh           *fh = desc->fh;
 208         struct nfs_fattr        *fattr = desc->fattr;
 209
 210         if (NFS_FILEID(inode) != fattr->fileid)
 211                 return 0;
 212         if (nfs_compare_fh(NFS_FH(inode), fh))
 213                 return 0;
 214         if (is_bad_inode(inode) || NFS_STALE(inode))
 215                 return 0;
 216         return 1;
 217 }
 218
 219 static int
 220 nfs_init_locked(struct inode *inode, void *opaque)
 221 {
 222         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
 223         struct nfs_fattr        *fattr = desc->fattr;
 224
 225         set_nfs_fileid(inode, fattr->fileid);
 226         nfs_copy_fh(NFS_FH(inode), desc->fh);
 227         return 0;
 228 }
 229
 230 /* Don't use READDIRPLUS on directories that we believe are too large */
 231 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
 232
 233 /*
 234  * This is our front-end to iget that looks up inodes by file handle
 235  * instead of inode number.
 236  */
 237 struct inode *
 238 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
 239 {
 240         struct nfs_find_desc desc = {
 241                 .fh     = fh,
 242                 .fattr  = fattr
 243         };
 244         struct inode *inode = ERR_PTR(-ENOENT);
 245         unsigned long hash;
 246
 247         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
 248                 goto out_no_inode;
 249         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
 250                 goto out_no_inode;
 251
 252         hash = nfs_fattr_to_ino_t(fattr);
 253
 254         inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
 255         if (inode == NULL) {
 256                 inode = ERR_PTR(-ENOMEM);
 257                 goto out_no_inode;
 258         }
 259
 260         if (inode->i_state & I_NEW) {
 261                 struct nfs_inode *nfsi = NFS_I(inode);
 262                 unsigned long now = jiffies;
 263
 264                 /* We set i_ino for the few things that still rely on it,
 265                  * such as stat(2) */
 266                 inode->i_ino = hash;
 267
 268                 /* We can't support update_atime(), since the server will reset it */
 269                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 270                 inode->i_mode = fattr->mode;
 271                 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
 272                                 && nfs_server_capable(inode, NFS_CAP_MODE))
 273                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 274                                 | NFS_INO_INVALID_ACCESS
 275                                 | NFS_INO_INVALID_ACL;
 276                 /* Why so? Because we want revalidate for devices/FIFOs, and
 277                  * that's precisely what we have in nfs_file_inode_operations.
 278                  */
 279                 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
 280                 if (S_ISREG(inode->i_mode)) {
 281                         inode->i_fop = &nfs_file_operations;
 282                         inode->i_data.a_ops = &nfs_file_aops;
 283                         inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
 284                 } else if (S_ISDIR(inode->i_mode)) {
 285                         inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
 286                         inode->i_fop = &nfs_dir_operations;
 287                         if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
 288                             && fattr->size <= NFS_LIMIT_READDIRPLUS)
 289                                 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 290                         /* Deal with crossing mountpoints */
 291                         if ((fattr->valid & NFS_ATTR_FATTR_FSID)
 292                                         && !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
 293                                 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
 294                                         inode->i_op = &nfs_referral_inode_operations;
 295                                 else
 296                                         inode->i_op = &nfs_mountpoint_inode_operations;
 297                                 inode->i_fop = NULL;
 298                                 set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
 299                         }
 300                 } else if (S_ISLNK(inode->i_mode))
 301                         inode->i_op = &nfs_symlink_inode_operations;
 302                 else
 303                         init_special_inode(inode, inode->i_mode, fattr->rdev);
 304
 305                 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
 306                 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
 307                 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
 308                 nfsi->change_attr = 0;
 309                 inode->i_size = 0;
 310                 inode->i_nlink = 0;
 311                 inode->i_uid = -2;
 312                 inode->i_gid = -2;
 313                 inode->i_blocks = 0;
 314                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 315
 316                 nfsi->read_cache_jiffies = fattr->time_start;
 317                 nfsi->attr_gencount = fattr->gencount;
 318                 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
 319                         inode->i_atime = fattr->atime;
 320                 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
 321                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
 322                 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
 323                         inode->i_mtime = fattr->mtime;
 324                 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
 325                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 326                                 | NFS_INO_INVALID_DATA;
 327                 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 328                         inode->i_ctime = fattr->ctime;
 329                 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
 330                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 331                                 | NFS_INO_INVALID_ACCESS
 332                                 | NFS_INO_INVALID_ACL;
 333                 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 334                         nfsi->change_attr = fattr->change_attr;
 335                 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
 336                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 337                                 | NFS_INO_INVALID_DATA;
 338                 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
 339                         inode->i_size = nfs_size_to_loff_t(fattr->size);
 340                 else
 341                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 342                                 | NFS_INO_INVALID_DATA
 343                                 | NFS_INO_REVAL_PAGECACHE;
 344                 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
 345                         inode->i_nlink = fattr->nlink;
 346                 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
 347                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
 348                 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
 349                         inode->i_uid = fattr->uid;
 350                 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
 351                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 352                                 | NFS_INO_INVALID_ACCESS
 353                                 | NFS_INO_INVALID_ACL;
 354                 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
 355                         inode->i_gid = fattr->gid;
 356                 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
 357                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 358                                 | NFS_INO_INVALID_ACCESS
 359                                 | NFS_INO_INVALID_ACL;
 360                 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
 361                         inode->i_blocks = fattr->du.nfs2.blocks;
 362                 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
 363                         /*
 364                          * report the blocks in 512byte units
 365                          */
 366                         inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
 367                 }
 368                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 369                 nfsi->attrtimeo_timestamp = now;
 370                 nfsi->access_cache = RB_ROOT;
 371
 372                 nfs_fscache_init_inode_cookie(inode);
 373
 374                 unlock_new_inode(inode);
 375         } else
 376                 nfs_refresh_inode(inode, fattr);
 377         dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
 378                 inode->i_sb->s_id,
 379                 (long long)NFS_FILEID(inode),
 380                 atomic_read(&inode->i_count));
 381
 382 out:
 383         return inode;
 384
 385 out_no_inode:
 386         dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
 387         goto out;
 388 }
 389
 390 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
 391
 392 int
 393 nfs_setattr(struct dentry *dentry, struct iattr *attr)
 394 {
 395         struct inode *inode = dentry->d_inode;
 396         struct nfs_fattr *fattr;
 397         int error = -ENOMEM;
 398
 399         nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
 400
 401         /* skip mode change if it's just for clearing setuid/setgid */
 402         if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
 403                 attr->ia_valid &= ~ATTR_MODE;
 404
 405         if (attr->ia_valid & ATTR_SIZE) {
 406                 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
 407                         attr->ia_valid &= ~ATTR_SIZE;
 408         }
 409
 410         /* Optimization: if the end result is no change, don't RPC */
 411         attr->ia_valid &= NFS_VALID_ATTRS;
 412         if ((attr->ia_valid & ~ATTR_FILE) == 0)
 413                 return 0;
 414
 415         /* Write all dirty data */
 416         if (S_ISREG(inode->i_mode))
 417                 nfs_wb_all(inode);
 418
 419         fattr = nfs_alloc_fattr();
 420         if (fattr == NULL)
 421                 goto out;
 422         /*
 423          * Return any delegations if we're going to change ACLs
 424          */
 425         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
 426                 nfs_inode_return_delegation(inode);
 427         error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
 428         if (error == 0)
 429                 nfs_refresh_inode(inode, fattr);
 430         nfs_free_fattr(fattr);
 431 out:
 432         return error;
 433 }
 434
 435 /**
 436  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
 437  * @inode: inode of the file used
 438  * @offset: file offset to start truncating
 439  *
 440  * This is a copy of the common vmtruncate, but with the locking
 441  * corrected to take into account the fact that NFS requires
 442  * inode->i_size to be updated under the inode->i_lock.
 443  */
 444 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
 445 {
 446         loff_t oldsize;
 447         int err;
 448
 449         err = inode_newsize_ok(inode, offset);
 450         if (err)
 451                 goto out;
 452
 453         spin_lock(&inode->i_lock);
 454         oldsize = inode->i_size;
 455         i_size_write(inode, offset);
 456         spin_unlock(&inode->i_lock);
 457
 458         truncate_pagecache(inode, oldsize, offset);
 459 out:
 460         return err;
 461 }
 462
 463 /**
 464  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
 465  * @inode: pointer to struct inode
 466  * @attr: pointer to struct iattr
 467  *
 468  * Note: we do this in the *proc.c in order to ensure that
 469  *       it works for things like exclusive creates too.
 470  */
 471 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
 472 {
 473         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
 474                 spin_lock(&inode->i_lock);
 475                 if ((attr->ia_valid & ATTR_MODE) != 0) {
 476                         int mode = attr->ia_mode & S_IALLUGO;
 477                         mode |= inode->i_mode & ~S_IALLUGO;
 478                         inode->i_mode = mode;
 479                 }
 480                 if ((attr->ia_valid & ATTR_UID) != 0)
 481                         inode->i_uid = attr->ia_uid;
 482                 if ((attr->ia_valid & ATTR_GID) != 0)
 483                         inode->i_gid = attr->ia_gid;
 484                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
 485                 spin_unlock(&inode->i_lock);
 486         }
 487         if ((attr->ia_valid & ATTR_SIZE) != 0) {
 488                 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
 489                 nfs_vmtruncate(inode, attr->ia_size);
 490         }
 491 }
 492
 493 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 494 {
 495         struct inode *inode = dentry->d_inode;
 496         int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
 497         int err;
 498
 499         /* Flush out writes to the server in order to update c/mtime.  */
 500         if (S_ISREG(inode->i_mode)) {
 501                 err = filemap_write_and_wait(inode->i_mapping);
 502                 if (err)
 503                         goto out;
 504         }
 505
 506         /*
 507          * We may force a getattr if the user cares about atime.
 508          *
 509          * Note that we only have to check the vfsmount flags here:
 510          *  - NFS always sets S_NOATIME by so checking it would give a
 511          *    bogus result
 512          *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
 513          *    no point in checking those.
 514          */
 515         if ((mnt->mnt_flags & MNT_NOATIME) ||
 516             ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 517                 need_atime = 0;
 518
 519         if (need_atime)
 520                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 521         else
 522                 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
 523         if (!err) {
 524                 generic_fillattr(inode, stat);
 525                 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
 526         }
 527 out:
 528         return err;
 529 }
 530
 531 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
 532 {
 533         atomic_set(&l_ctx->count, 1);
 534         l_ctx->lockowner = current->files;
 535         l_ctx->pid = current->tgid;
 536         INIT_LIST_HEAD(&l_ctx->list);
 537 }
 538
 539 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
 540 {
 541         struct nfs_lock_context *pos;
 542
 543         list_for_each_entry(pos, &ctx->lock_context.list, list) {
 544                 if (pos->lockowner != current->files)
 545                         continue;
 546                 if (pos->pid != current->tgid)
 547                         continue;
 548                 atomic_inc(&pos->count);
 549                 return pos;
 550         }
 551         return NULL;
 552 }
 553
 554 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
 555 {
 556         struct nfs_lock_context *res, *new = NULL;
 557         struct inode *inode = ctx->path.dentry->d_inode;
 558
 559         spin_lock(&inode->i_lock);
 560         res = __nfs_find_lock_context(ctx);
 561         if (res == NULL) {
 562                 spin_unlock(&inode->i_lock);
 563                 new = kmalloc(sizeof(*new), GFP_KERNEL);
 564                 if (new == NULL)
 565                         return NULL;
 566                 nfs_init_lock_context(new);
 567                 spin_lock(&inode->i_lock);
 568                 res = __nfs_find_lock_context(ctx);
 569                 if (res == NULL) {
 570                         list_add_tail(&new->list, &ctx->lock_context.list);
 571                         new->open_context = ctx;
 572                         res = new;
 573                         new = NULL;
 574                 }
 575         }
 576         spin_unlock(&inode->i_lock);
 577         kfree(new);
 578         return res;
 579 }
 580
 581 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
 582 {
 583         struct nfs_open_context *ctx = l_ctx->open_context;
 584         struct inode *inode = ctx->path.dentry->d_inode;
 585
 586         if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
 587                 return;
 588         list_del(&l_ctx->list);
 589         spin_unlock(&inode->i_lock);
 590         kfree(l_ctx);
 591 }
 592
 593 /**
 594  * nfs_close_context - Common close_context() routine NFSv2/v3
 595  * @ctx: pointer to context
 596  * @is_sync: is this a synchronous close
 597  *
 598  * always ensure that the attributes are up to date if we're mounted
 599  * with close-to-open semantics
 600  */
 601 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
 602 {
 603         struct inode *inode;
 604         struct nfs_server *server;
 605
 606         if (!(ctx->mode & FMODE_WRITE))
 607                 return;
 608         if (!is_sync)
 609                 return;
 610         inode = ctx->path.dentry->d_inode;
 611         if (!list_empty(&NFS_I(inode)->open_files))
 612                 return;
 613         server = NFS_SERVER(inode);
 614         if (server->flags & NFS_MOUNT_NOCTO)
 615                 return;
 616         nfs_revalidate_inode(server, inode);
 617 }
 618
 619 static struct nfs_open_context *alloc_nfs_open_context(struct path *path, struct rpc_cred *cred)
 620 {
 621         struct nfs_open_context *ctx;
 622
 623         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 624         if (ctx != NULL) {
 625                 ctx->path = *path;
 626                 path_get(&ctx->path);
 627                 ctx->cred = get_rpccred(cred);
 628                 ctx->state = NULL;
 629                 ctx->flags = 0;
 630                 ctx->error = 0;
 631                 ctx->dir_cookie = 0;
 632                 nfs_init_lock_context(&ctx->lock_context);
 633                 ctx->lock_context.open_context = ctx;
 634         }
 635         return ctx;
 636 }
 637
 638 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
 639 {
 640         if (ctx != NULL)
 641                 atomic_inc(&ctx->lock_context.count);
 642         return ctx;
 643 }
 644
 645 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
 646 {
 647         struct inode *inode = ctx->path.dentry->d_inode;
 648
 649         if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
 650                 return;
 651         list_del(&ctx->list);
 652         spin_unlock(&inode->i_lock);
 653         NFS_PROTO(inode)->close_context(ctx, is_sync);
 654         if (ctx->cred != NULL)
 655                 put_rpccred(ctx->cred);
 656         path_put(&ctx->path);
 657         kfree(ctx);
 658 }
 659
 660 void put_nfs_open_context(struct nfs_open_context *ctx)
 661 {
 662         __put_nfs_open_context(ctx, 0);
 663 }
 664
 665 /*
 666  * Ensure that mmap has a recent RPC credential for use when writing out
 667  * shared pages
 668  */
 669 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
 670 {
 671         struct inode *inode = filp->f_path.dentry->d_inode;
 672         struct nfs_inode *nfsi = NFS_I(inode);
 673
 674         filp->private_data = get_nfs_open_context(ctx);
 675         spin_lock(&inode->i_lock);
 676         list_add(&ctx->list, &nfsi->open_files);
 677         spin_unlock(&inode->i_lock);
 678 }
 679
 680 /*
 681  * Given an inode, search for an open context with the desired characteristics
 682  */
 683 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
 684 {
 685         struct nfs_inode *nfsi = NFS_I(inode);
 686         struct nfs_open_context *pos, *ctx = NULL;
 687
 688         spin_lock(&inode->i_lock);
 689         list_for_each_entry(pos, &nfsi->open_files, list) {
 690                 if (cred != NULL && pos->cred != cred)
 691                         continue;
 692                 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
 693                         continue;
 694                 ctx = get_nfs_open_context(pos);
 695                 break;
 696         }
 697         spin_unlock(&inode->i_lock);
 698         return ctx;
 699 }
 700
 701 static void nfs_file_clear_open_context(struct file *filp)
 702 {
 703         struct inode *inode = filp->f_path.dentry->d_inode;
 704         struct nfs_open_context *ctx = nfs_file_open_context(filp);
 705
 706         if (ctx) {
 707                 filp->private_data = NULL;
 708                 spin_lock(&inode->i_lock);
 709                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
 710                 spin_unlock(&inode->i_lock);
 711                 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
 712         }
 713 }
 714
 715 /*
 716  * These allocate and release file read/write context information.
 717  */
 718 int nfs_open(struct inode *inode, struct file *filp)
 719 {
 720         struct nfs_open_context *ctx;
 721         struct rpc_cred *cred;
 722
 723         cred = rpc_lookup_cred();
 724         if (IS_ERR(cred))
 725                 return PTR_ERR(cred);
 726         ctx = alloc_nfs_open_context(&filp->f_path, cred);
 727         put_rpccred(cred);
 728         if (ctx == NULL)
 729                 return -ENOMEM;
 730         ctx->mode = filp->f_mode;
 731         nfs_file_set_open_context(filp, ctx);
 732         put_nfs_open_context(ctx);
 733         nfs_fscache_set_inode_cookie(inode, filp);
 734         return 0;
 735 }
 736
 737 int nfs_release(struct inode *inode, struct file *filp)
 738 {
 739         nfs_file_clear_open_context(filp);
 740         return 0;
 741 }
 742
 743 /*
 744  * This function is called whenever some part of NFS notices that
 745  * the cached attributes have to be refreshed.
 746  */
 747 int
 748 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 749 {
 750         int              status = -ESTALE;
 751         struct nfs_fattr *fattr = NULL;
 752         struct nfs_inode *nfsi = NFS_I(inode);
 753
 754         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
 755                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
 756
 757         if (is_bad_inode(inode))
 758                 goto out;
 759         if (NFS_STALE(inode))
 760                 goto out;
 761
 762         status = -ENOMEM;
 763         fattr = nfs_alloc_fattr();
 764         if (fattr == NULL)
 765                 goto out;
 766
 767         nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
 768         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
 769         if (status != 0) {
 770                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
 771                          inode->i_sb->s_id,
 772                          (long long)NFS_FILEID(inode), status);
 773                 if (status == -ESTALE) {
 774                         nfs_zap_caches(inode);
 775                         if (!S_ISDIR(inode->i_mode))
 776                                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 777                 }
 778                 goto out;
 779         }
 780
 781         status = nfs_refresh_inode(inode, fattr);
 782         if (status) {
 783                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
 784                          inode->i_sb->s_id,
 785                          (long long)NFS_FILEID(inode), status);
 786                 goto out;
 787         }
 788
 789         if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
 790                 nfs_zap_acl_cache(inode);
 791
 792         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
 793                 inode->i_sb->s_id,
 794                 (long long)NFS_FILEID(inode));
 795
 796  out:
 797         nfs_free_fattr(fattr);
 798         return status;
 799 }
 800
 801 int nfs_attribute_timeout(struct inode *inode)
 802 {
 803         struct nfs_inode *nfsi = NFS_I(inode);
 804
 805         return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
 806 }
 807
 808 static int nfs_attribute_cache_expired(struct inode *inode)
 809 {
 810         if (nfs_have_delegated_attributes(inode))
 811                 return 0;
 812         return nfs_attribute_timeout(inode);
 813 }
 814
 815 /**
 816  * nfs_revalidate_inode - Revalidate the inode attributes
 817  * @server - pointer to nfs_server struct
 818  * @inode - pointer to inode struct
 819  *
 820  * Updates inode attribute information by retrieving the data from the server.
 821  */
 822 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 823 {
 824         if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
 825                         && !nfs_attribute_cache_expired(inode))
 826                 return NFS_STALE(inode) ? -ESTALE : 0;
 827         return __nfs_revalidate_inode(server, inode);
 828 }
 829
 830 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
 831 {
 832         struct nfs_inode *nfsi = NFS_I(inode);
 833
 834         if (mapping->nrpages != 0) {
 835                 int ret = invalidate_inode_pages2(mapping);
 836                 if (ret < 0)
 837                         return ret;
 838         }
 839         spin_lock(&inode->i_lock);
 840         nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
 841         if (S_ISDIR(inode->i_mode))
 842                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 843         spin_unlock(&inode->i_lock);
 844         nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
 845         nfs_fscache_reset_inode_cookie(inode);
 846         dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
 847                         inode->i_sb->s_id, (long long)NFS_FILEID(inode));
 848         return 0;
 849 }
 850
 851 /**
 852  * nfs_revalidate_mapping - Revalidate the pagecache
 853  * @inode - pointer to host inode
 854  * @mapping - pointer to mapping
 855  */
 856 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
 857 {
 858         struct nfs_inode *nfsi = NFS_I(inode);
 859         int ret = 0;
 860
 861         if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
 862                         || nfs_attribute_cache_expired(inode)
 863                         || NFS_STALE(inode)) {
 864                 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 865                 if (ret < 0)
 866                         goto out;
 867         }
 868         if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
 869                 ret = nfs_invalidate_mapping(inode, mapping);
 870 out:
 871         return ret;
 872 }
 873
 874 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 875 {
 876         struct nfs_inode *nfsi = NFS_I(inode);
 877
 878         if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
 879                         && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 880                         && nfsi->change_attr == fattr->pre_change_attr) {
 881                 nfsi->change_attr = fattr->change_attr;
 882                 if (S_ISDIR(inode->i_mode))
 883                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 884         }
 885         /* If we have atomic WCC data, we may update some attributes */
 886         if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
 887                         && (fattr->valid & NFS_ATTR_FATTR_CTIME)
 888                         && timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
 889                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
 890
 891         if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
 892                         && (fattr->valid & NFS_ATTR_FATTR_MTIME)
 893                         && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
 894                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
 895                         if (S_ISDIR(inode->i_mode))
 896                                 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 897         }
 898         if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
 899                         && (fattr->valid & NFS_ATTR_FATTR_SIZE)
 900                         && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
 901                         && nfsi->npages == 0)
 902                         i_size_write(inode, nfs_size_to_loff_t(fattr->size));
 903 }
 904
 905 /**
 906  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
 907  * @inode - pointer to inode
 908  * @fattr - updated attributes
 909  *
 910  * Verifies the attribute cache. If we have just changed the attributes,
 911  * so that fattr carries weak cache consistency data, then it may
 912  * also update the ctime/mtime/change_attribute.
 913  */
 914 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
 915 {
 916         struct nfs_inode *nfsi = NFS_I(inode);
 917         loff_t cur_size, new_isize;
 918         unsigned long invalid = 0;
 919
 920
 921         /* Has the inode gone and changed behind our back? */
 922         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
 923                 return -EIO;
 924         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
 925                 return -EIO;
 926
 927         if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
 928                         nfsi->change_attr != fattr->change_attr)
 929                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 930
 931         /* Verify a few of the more important attributes */
 932         if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
 933                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 934
 935         if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
 936                 cur_size = i_size_read(inode);
 937                 new_isize = nfs_size_to_loff_t(fattr->size);
 938                 if (cur_size != new_isize && nfsi->npages == 0)
 939                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 940         }
 941
 942         /* Have any file permissions changed? */
 943         if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
 944                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 945         if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
 946                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 947         if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
 948                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 949
 950         /* Has the link count changed? */
 951         if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
 952                 invalid |= NFS_INO_INVALID_ATTR;
 953
 954         if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
 955                 invalid |= NFS_INO_INVALID_ATIME;
 956
 957         if (invalid != 0)
 958                 nfsi->cache_validity |= invalid;
 959
 960         nfsi->read_cache_jiffies = fattr->time_start;
 961         return 0;
 962 }
 963
 964 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
 965 {
 966         if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
 967                 return 0;
 968         return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
 969 }
 970
 971 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
 972 {
 973         if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
 974                 return 0;
 975         return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
 976 }
 977
 978 static atomic_long_t nfs_attr_generation_counter;
 979
 980 static unsigned long nfs_read_attr_generation_counter(void)
 981 {
 982         return atomic_long_read(&nfs_attr_generation_counter);
 983 }
 984
 985 unsigned long nfs_inc_attr_generation_counter(void)
 986 {
 987         return atomic_long_inc_return(&nfs_attr_generation_counter);
 988 }
 989
 990 void nfs_fattr_init(struct nfs_fattr *fattr)
 991 {
 992         fattr->valid = 0;
 993         fattr->time_start = jiffies;
 994         fattr->gencount = nfs_inc_attr_generation_counter();
 995 }
 996
 997 struct nfs_fattr *nfs_alloc_fattr(void)
 998 {
 999         struct nfs_fattr *fattr;
1000
1001         fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1002         if (fattr != NULL)
1003                 nfs_fattr_init(fattr);
1004         return fattr;
1005 }
1006
1007 struct nfs_fh *nfs_alloc_fhandle(void)
1008 {
1009         struct nfs_fh *fh;
1010
1011         fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1012         if (fh != NULL)
1013                 fh->size = 0;
1014         return fh;
1015 }
1016
1017 /**
1018  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1019  * @inode - pointer to inode
1020  * @fattr - attributes
1021  *
1022  * Attempt to divine whether or not an RPC call reply carrying stale
1023  * attributes got scheduled after another call carrying updated ones.
1024  *
1025  * To do so, the function first assumes that a more recent ctime means
1026  * that the attributes in fattr are newer, however it also attempt to
1027  * catch the case where ctime either didn't change, or went backwards
1028  * (if someone reset the clock on the server) by looking at whether
1029  * or not this RPC call was started after the inode was last updated.
1030  * Note also the check for wraparound of 'attr_gencount'
1031  *
1032  * The function returns 'true' if it thinks the attributes in 'fattr' are
1033  * more recent than the ones cached in the inode.
1034  *
1035  */
1036 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1037 {
1038         const struct nfs_inode *nfsi = NFS_I(inode);
1039
1040         return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1041                 nfs_ctime_need_update(inode, fattr) ||
1042                 nfs_size_need_update(inode, fattr) ||
1043                 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1044 }
1045
1046 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1047 {
1048         if (nfs_inode_attrs_need_update(inode, fattr))
1049                 return nfs_update_inode(inode, fattr);
1050         return nfs_check_inode_attributes(inode, fattr);
1051 }
1052
1053 /**
1054  * nfs_refresh_inode - try to update the inode attribute cache
1055  * @inode - pointer to inode
1056  * @fattr - updated attributes
1057  *
1058  * Check that an RPC call that returned attributes has not overlapped with
1059  * other recent updates of the inode metadata, then decide whether it is
1060  * safe to do a full update of the inode attributes, or whether just to
1061  * call nfs_check_inode_attributes.
1062  */
1063 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1064 {
1065         int status;
1066
1067         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1068                 return 0;
1069         spin_lock(&inode->i_lock);
1070         status = nfs_refresh_inode_locked(inode, fattr);
1071         spin_unlock(&inode->i_lock);
1072
1073         return status;
1074 }
1075
1076 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1077 {
1078         struct nfs_inode *nfsi = NFS_I(inode);
1079
1080         nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1081         if (S_ISDIR(inode->i_mode))
1082                 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1083         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1084                 return 0;
1085         return nfs_refresh_inode_locked(inode, fattr);
1086 }
1087
1088 /**
1089  * nfs_post_op_update_inode - try to update the inode attribute cache
1090  * @inode - pointer to inode
1091  * @fattr - updated attributes
1092  *
1093  * After an operation that has changed the inode metadata, mark the
1094  * attribute cache as being invalid, then try to update it.
1095  *
1096  * NB: if the server didn't return any post op attributes, this
1097  * function will force the retrieval of attributes before the next
1098  * NFS request.  Thus it should be used only for operations that
1099  * are expected to change one or more attributes, to avoid
1100  * unnecessary NFS requests and trips through nfs_update_inode().
1101  */
1102 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1103 {
1104         int status;
1105
1106         spin_lock(&inode->i_lock);
1107         status = nfs_post_op_update_inode_locked(inode, fattr);
1108         spin_unlock(&inode->i_lock);
1109         return status;
1110 }
1111
1112 /**
1113  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1114  * @inode - pointer to inode
1115  * @fattr - updated attributes
1116  *
1117  * After an operation that has changed the inode metadata, mark the
1118  * attribute cache as being invalid, then try to update it. Fake up
1119  * weak cache consistency data, if none exist.
1120  *
1121  * This function is mainly designed to be used by the ->write_done() functions.
1122  */
1123 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1124 {
1125         int status;
1126
1127         spin_lock(&inode->i_lock);
1128         /* Don't do a WCC update if these attributes are already stale */
1129         if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1130                         !nfs_inode_attrs_need_update(inode, fattr)) {
1131                 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1132                                 | NFS_ATTR_FATTR_PRESIZE
1133                                 | NFS_ATTR_FATTR_PREMTIME
1134                                 | NFS_ATTR_FATTR_PRECTIME);
1135                 goto out_noforce;
1136         }
1137         if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1138                         (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1139                 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1140                 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1141         }
1142         if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1143                         (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1144                 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1145                 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1146         }
1147         if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1148                         (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1149                 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1150                 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1151         }
1152         if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1153                         (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1154                 fattr->pre_size = i_size_read(inode);
1155                 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1156         }
1157 out_noforce:
1158         status = nfs_post_op_update_inode_locked(inode, fattr);
1159         spin_unlock(&inode->i_lock);
1160         return status;
1161 }
1162
1163 /*
1164  * Many nfs protocol calls return the new file attributes after
1165  * an operation.  Here we update the inode to reflect the state
1166  * of the server's inode.
1167  *
1168  * This is a bit tricky because we have to make sure all dirty pages
1169  * have been sent off to the server before calling invalidate_inode_pages.
1170  * To make sure no other process adds more write requests while we try
1171  * our best to flush them, we make them sleep during the attribute refresh.
1172  *
1173  * A very similar scenario holds for the dir cache.
1174  */
1175 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1176 {
1177         struct nfs_server *server;
1178         struct nfs_inode *nfsi = NFS_I(inode);
1179         loff_t cur_isize, new_isize;
1180         unsigned long invalid = 0;
1181         unsigned long now = jiffies;
1182         unsigned long save_cache_validity;
1183
1184         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1185                         __func__, inode->i_sb->s_id, inode->i_ino,
1186                         atomic_read(&inode->i_count), fattr->valid);
1187
1188         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1189                 goto out_fileid;
1190
1191         /*
1192          * Make sure the inode's type hasn't changed.
1193          */
1194         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1195                 goto out_changed;
1196
1197         server = NFS_SERVER(inode);
1198         /* Update the fsid? */
1199         if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1200                         !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1201                         !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1202                 server->fsid = fattr->fsid;
1203
1204         /*
1205          * Update the read time so we don't revalidate too often.
1206          */
1207         nfsi->read_cache_jiffies = fattr->time_start;
1208
1209         save_cache_validity = nfsi->cache_validity;
1210         nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1211                         | NFS_INO_INVALID_ATIME
1212                         | NFS_INO_REVAL_FORCED
1213                         | NFS_INO_REVAL_PAGECACHE);
1214
1215         /* Do atomic weak cache consistency updates */
1216         nfs_wcc_update_inode(inode, fattr);
1217
1218         /* More cache consistency checks */
1219         if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1220                 if (nfsi->change_attr != fattr->change_attr) {
1221                         dprintk("NFS: change_attr change on server for file %s/%ld\n",
1222                                         inode->i_sb->s_id, inode->i_ino);
1223                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1224                         if (S_ISDIR(inode->i_mode))
1225                                 nfs_force_lookup_revalidate(inode);
1226                         nfsi->change_attr = fattr->change_attr;
1227                 }
1228         } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1229                 invalid |= save_cache_validity;
1230
1231         if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1232                 /* NFSv2/v3: Check if the mtime agrees */
1233                 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1234                         dprintk("NFS: mtime change on server for file %s/%ld\n",
1235                                         inode->i_sb->s_id, inode->i_ino);
1236                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1237                         if (S_ISDIR(inode->i_mode))
1238                                 nfs_force_lookup_revalidate(inode);
1239                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1240                 }
1241         } else if (server->caps & NFS_CAP_MTIME)
1242                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1243                                 | NFS_INO_INVALID_DATA
1244                                 | NFS_INO_REVAL_PAGECACHE
1245                                 | NFS_INO_REVAL_FORCED);
1246
1247         if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1248                 /* If ctime has changed we should definitely clear access+acl caches */
1249                 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1250                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1251                         /* and probably clear data for a directory too as utimes can cause
1252                          * havoc with our cache.
1253                          */
1254                         if (S_ISDIR(inode->i_mode)) {
1255                                 invalid |= NFS_INO_INVALID_DATA;
1256                                 nfs_force_lookup_revalidate(inode);
1257                         }
1258                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1259                 }
1260         } else if (server->caps & NFS_CAP_CTIME)
1261                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1262                                 | NFS_INO_INVALID_ACCESS
1263                                 | NFS_INO_INVALID_ACL
1264                                 | NFS_INO_REVAL_FORCED);
1265
1266         /* Check if our cached file size is stale */
1267         if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1268                 new_isize = nfs_size_to_loff_t(fattr->size);
1269                 cur_isize = i_size_read(inode);
1270                 if (new_isize != cur_isize) {
1271                         /* Do we perhaps have any outstanding writes, or has
1272                          * the file grown beyond our last write? */
1273                         if (nfsi->npages == 0 || new_isize > cur_isize) {
1274                                 i_size_write(inode, new_isize);
1275                                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1276                         }
1277                         dprintk("NFS: isize change on server for file %s/%ld\n",
1278                                         inode->i_sb->s_id, inode->i_ino);
1279                 }
1280         } else
1281                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1282                                 | NFS_INO_REVAL_PAGECACHE
1283                                 | NFS_INO_REVAL_FORCED);
1284
1285
1286         if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1287                 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1288         else if (server->caps & NFS_CAP_ATIME)
1289                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1290                                 | NFS_INO_REVAL_FORCED);
1291
1292         if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1293                 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1294                         umode_t newmode = inode->i_mode & S_IFMT;
1295                         newmode |= fattr->mode & S_IALLUGO;
1296                         inode->i_mode = newmode;
1297                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1298                 }
1299         } else if (server->caps & NFS_CAP_MODE)
1300                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1301                                 | NFS_INO_INVALID_ACCESS
1302                                 | NFS_INO_INVALID_ACL
1303                                 | NFS_INO_REVAL_FORCED);
1304
1305         if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1306                 if (inode->i_uid != fattr->uid) {
1307                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1308                         inode->i_uid = fattr->uid;
1309                 }
1310         } else if (server->caps & NFS_CAP_OWNER)
1311                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1312                                 | NFS_INO_INVALID_ACCESS
1313                                 | NFS_INO_INVALID_ACL
1314                                 | NFS_INO_REVAL_FORCED);
1315
1316         if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1317                 if (inode->i_gid != fattr->gid) {
1318                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1319                         inode->i_gid = fattr->gid;
1320                 }
1321         } else if (server->caps & NFS_CAP_OWNER_GROUP)
1322                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1323                                 | NFS_INO_INVALID_ACCESS
1324                                 | NFS_INO_INVALID_ACL
1325                                 | NFS_INO_REVAL_FORCED);
1326
1327         if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1328                 if (inode->i_nlink != fattr->nlink) {
1329                         invalid |= NFS_INO_INVALID_ATTR;
1330                         if (S_ISDIR(inode->i_mode))
1331                                 invalid |= NFS_INO_INVALID_DATA;
1332                         inode->i_nlink = fattr->nlink;
1333                 }
1334         } else if (server->caps & NFS_CAP_NLINK)
1335                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1336                                 | NFS_INO_REVAL_FORCED);
1337
1338         if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1339                 /*
1340                  * report the blocks in 512byte units
1341                  */
1342                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1343         }
1344         if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1345                 inode->i_blocks = fattr->du.nfs2.blocks;
1346
1347         /* Update attrtimeo value if we're out of the unstable period */
1348         if (invalid & NFS_INO_INVALID_ATTR) {
1349                 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1350                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1351                 nfsi->attrtimeo_timestamp = now;
1352                 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1353         } else {
1354                 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1355                         if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1356                                 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1357                         nfsi->attrtimeo_timestamp = now;
1358                 }
1359         }
1360         invalid &= ~NFS_INO_INVALID_ATTR;
1361         /* Don't invalidate the data if we were to blame */
1362         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1363                                 || S_ISLNK(inode->i_mode)))
1364                 invalid &= ~NFS_INO_INVALID_DATA;
1365         if (!nfs_have_delegation(inode, FMODE_READ) ||
1366                         (save_cache_validity & NFS_INO_REVAL_FORCED))
1367                 nfsi->cache_validity |= invalid;
1368
1369         return 0;
1370  out_changed:
1371         /*
1372          * Big trouble! The inode has become a different object.
1373          */
1374         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1375                         __func__, inode->i_ino, inode->i_mode, fattr->mode);
1376  out_err:
1377         /*
1378          * No need to worry about unhashing the dentry, as the
1379          * lookup validation will know that the inode is bad.
1380          * (But we fall through to invalidate the caches.)
1381          */
1382         nfs_invalidate_inode(inode);
1383         return -ESTALE;
1384
1385  out_fileid:
1386         printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1387                 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1388                 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1389                 (long long)nfsi->fileid, (long long)fattr->fileid);
1390         goto out_err;
1391 }
1392
1393
1394 #ifdef CONFIG_NFS_V4
1395
1396 /*
1397  * Clean out any remaining NFSv4 state that might be left over due
1398  * to open() calls that passed nfs_atomic_lookup, but failed to call
1399  * nfs_open().
1400  */
1401 void nfs4_clear_inode(struct inode *inode)
1402 {
1403         /* If we are holding a delegation, return it! */
1404         nfs_inode_return_delegation_noreclaim(inode);
1405         /* First call standard NFS clear_inode() code */
1406         nfs_clear_inode(inode);
1407 }
1408 #endif
1409
1410 struct inode *nfs_alloc_inode(struct super_block *sb)
1411 {
1412         struct nfs_inode *nfsi;
1413         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1414         if (!nfsi)
1415                 return NULL;
1416         nfsi->flags = 0UL;
1417         nfsi->cache_validity = 0UL;
1418 #ifdef CONFIG_NFS_V3_ACL
1419         nfsi->acl_access = ERR_PTR(-EAGAIN);
1420         nfsi->acl_default = ERR_PTR(-EAGAIN);
1421 #endif
1422 #ifdef CONFIG_NFS_V4
1423         nfsi->nfs4_acl = NULL;
1424 #endif /* CONFIG_NFS_V4 */
1425         return &nfsi->vfs_inode;
1426 }
1427
1428 void nfs_destroy_inode(struct inode *inode)
1429 {
1430         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1431 }
1432
1433 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1434 {
1435 #ifdef CONFIG_NFS_V4
1436         INIT_LIST_HEAD(&nfsi->open_states);
1437         nfsi->delegation = NULL;
1438         nfsi->delegation_state = 0;
1439         init_rwsem(&nfsi->rwsem);
1440 #endif
1441 }
1442
1443 static void init_once(void *foo)
1444 {
1445         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1446
1447         inode_init_once(&nfsi->vfs_inode);
1448         INIT_LIST_HEAD(&nfsi->open_files);
1449         INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1450         INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1451         INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1452         nfsi->npages = 0;
1453         nfsi->ncommit = 0;
1454         atomic_set(&nfsi->silly_count, 1);
1455         INIT_HLIST_HEAD(&nfsi->silly_list);
1456         init_waitqueue_head(&nfsi->waitqueue);
1457         nfs4_init_once(nfsi);
1458 }
1459
1460 static int __init nfs_init_inodecache(void)
1461 {
1462         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1463                                              sizeof(struct nfs_inode),
1464                                              0, (SLAB_RECLAIM_ACCOUNT|
1465                                                 SLAB_MEM_SPREAD),
1466                                              init_once);
1467         if (nfs_inode_cachep == NULL)
1468                 return -ENOMEM;
1469
1470         return 0;
1471 }
1472
1473 static void nfs_destroy_inodecache(void)
1474 {
1475         kmem_cache_destroy(nfs_inode_cachep);
1476 }
1477
1478 struct workqueue_struct *nfsiod_workqueue;
1479
1480 /*
1481  * start up the nfsiod workqueue
1482  */
1483 static int nfsiod_start(void)
1484 {
1485         struct workqueue_struct *wq;
1486         dprintk("RPC:       creating workqueue nfsiod\n");
1487         wq = create_singlethread_workqueue("nfsiod");
1488         if (wq == NULL)
1489                 return -ENOMEM;
1490         nfsiod_workqueue = wq;
1491         return 0;
1492 }
1493
1494 /*
1495  * Destroy the nfsiod workqueue
1496  */
1497 static void nfsiod_stop(void)
1498 {
1499         struct workqueue_struct *wq;
1500
1501         wq = nfsiod_workqueue;
1502         if (wq == NULL)
1503                 return;
1504         nfsiod_workqueue = NULL;
1505         destroy_workqueue(wq);
1506 }
1507
1508 /*
1509  * Initialize NFS
1510  */
1511 static int __init init_nfs_fs(void)
1512 {
1513         int err;
1514
1515         err = nfs_dns_resolver_init();
1516         if (err < 0)
1517                 goto out8;
1518
1519         err = nfs_fscache_register();
1520         if (err < 0)
1521                 goto out7;
1522
1523         err = nfsiod_start();
1524         if (err)
1525                 goto out6;
1526
1527         err = nfs_fs_proc_init();
1528         if (err)
1529                 goto out5;
1530
1531         err = nfs_init_nfspagecache();
1532         if (err)
1533                 goto out4;
1534
1535         err = nfs_init_inodecache();
1536         if (err)
1537                 goto out3;
1538
1539         err = nfs_init_readpagecache();
1540         if (err)
1541                 goto out2;
1542
1543         err = nfs_init_writepagecache();
1544         if (err)
1545                 goto out1;
1546
1547         err = nfs_init_directcache();
1548         if (err)
1549                 goto out0;
1550
1551 #ifdef CONFIG_PROC_FS
1552         rpc_proc_register(&nfs_rpcstat);
1553 #endif
1554         if ((err = register_nfs_fs()) != 0)
1555                 goto out;
1556         return 0;
1557 out:
1558 #ifdef CONFIG_PROC_FS
1559         rpc_proc_unregister("nfs");
1560 #endif
1561         nfs_destroy_directcache();
1562 out0:
1563         nfs_destroy_writepagecache();
1564 out1:
1565         nfs_destroy_readpagecache();
1566 out2:
1567         nfs_destroy_inodecache();
1568 out3:
1569         nfs_destroy_nfspagecache();
1570 out4:
1571         nfs_fs_proc_exit();
1572 out5:
1573         nfsiod_stop();
1574 out6:
1575         nfs_fscache_unregister();
1576 out7:
1577         nfs_dns_resolver_destroy();
1578 out8:
1579         return err;
1580 }
1581
1582 static void __exit exit_nfs_fs(void)
1583 {
1584         nfs_destroy_directcache();
1585         nfs_destroy_writepagecache();
1586         nfs_destroy_readpagecache();
1587         nfs_destroy_inodecache();
1588         nfs_destroy_nfspagecache();
1589         nfs_fscache_unregister();
1590         nfs_dns_resolver_destroy();
1591 #ifdef CONFIG_PROC_FS
1592         rpc_proc_unregister("nfs");
1593 #endif
1594         unregister_nfs_fs();
1595         nfs_fs_proc_exit();
1596         nfsiod_stop();
1597 }
1598
1599 /* Not quite true; I just maintain it */
1600 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1601 MODULE_LICENSE("GPL");
1602 module_param(enable_ino64, bool, 0644);
1603
1604 module_init(init_nfs_fs)
1605 module_exit(exit_nfs_fs)