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 #include <linux/compat.h>
  41
  42 #include <asm/system.h>
  43 #include <asm/uaccess.h>
  44
  45 #include "nfs4_fs.h"
  46 #include "callback.h"
  47 #include "delegation.h"
  48 #include "iostat.h"
  49 #include "internal.h"
  50 #include "fscache.h"
  51 #include "dns_resolve.h"
  52 #include "pnfs.h"
  53
  54 #define NFSDBG_FACILITY         NFSDBG_VFS
  55
  56 #define NFS_64_BIT_INODE_NUMBERS_ENABLED        1
  57
  58 /* Default is to see 64-bit inode numbers */
  59 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
  60
  61 static void nfs_invalidate_inode(struct inode *);
  62 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
  63
  64 static struct kmem_cache * nfs_inode_cachep;
  65
  66 static inline unsigned long
  67 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
  68 {
  69         return nfs_fileid_to_ino_t(fattr->fileid);
  70 }
  71
  72 /**
  73  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
  74  * @word: long word containing the bit lock
  75  */
  76 int nfs_wait_bit_killable(void *word)
  77 {
  78         if (fatal_signal_pending(current))
  79                 return -ERESTARTSYS;
  80         schedule();
  81         return 0;
  82 }
  83
  84 /**
  85  * nfs_compat_user_ino64 - returns the user-visible inode number
  86  * @fileid: 64-bit fileid
  87  *
  88  * This function returns a 32-bit inode number if the boot parameter
  89  * nfs.enable_ino64 is zero.
  90  */
  91 u64 nfs_compat_user_ino64(u64 fileid)
  92 {
  93 #ifdef CONFIG_COMPAT
  94         compat_ulong_t ino;
  95 #else
  96         unsigned long ino;
  97 #endif
  98
  99         if (enable_ino64)
 100                 return fileid;
 101         ino = fileid;
 102         if (sizeof(ino) < sizeof(fileid))
 103                 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
 104         return ino;
 105 }
 106
 107 static void nfs_clear_inode(struct inode *inode)
 108 {
 109         /*
 110          * The following should never happen...
 111          */
 112         BUG_ON(nfs_have_writebacks(inode));
 113         BUG_ON(!list_empty(&NFS_I(inode)->open_files));
 114         nfs_zap_acl_cache(inode);
 115         nfs_access_zap_cache(inode);
 116         nfs_fscache_release_inode_cookie(inode);
 117 }
 118
 119 void nfs_evict_inode(struct inode *inode)
 120 {
 121         truncate_inode_pages(&inode->i_data, 0);
 122         end_writeback(inode);
 123         nfs_clear_inode(inode);
 124 }
 125
 126 /**
 127  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
 128  */
 129 int nfs_sync_mapping(struct address_space *mapping)
 130 {
 131         int ret = 0;
 132
 133         if (mapping->nrpages != 0) {
 134                 unmap_mapping_range(mapping, 0, 0, 0);
 135                 ret = nfs_wb_all(mapping->host);
 136         }
 137         return ret;
 138 }
 139
 140 /*
 141  * Invalidate the local caches
 142  */
 143 static void nfs_zap_caches_locked(struct inode *inode)
 144 {
 145         struct nfs_inode *nfsi = NFS_I(inode);
 146         int mode = inode->i_mode;
 147
 148         nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
 149
 150         nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 151         nfsi->attrtimeo_timestamp = jiffies;
 152
 153         memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
 154         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
 155                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
 156         else
 157                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
 158 }
 159
 160 void nfs_zap_caches(struct inode *inode)
 161 {
 162         spin_lock(&inode->i_lock);
 163         nfs_zap_caches_locked(inode);
 164         spin_unlock(&inode->i_lock);
 165 }
 166
 167 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
 168 {
 169         if (mapping->nrpages != 0) {
 170                 spin_lock(&inode->i_lock);
 171                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
 172                 spin_unlock(&inode->i_lock);
 173         }
 174 }
 175
 176 void nfs_zap_acl_cache(struct inode *inode)
 177 {
 178         void (*clear_acl_cache)(struct inode *);
 179
 180         clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
 181         if (clear_acl_cache != NULL)
 182                 clear_acl_cache(inode);
 183         spin_lock(&inode->i_lock);
 184         NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
 185         spin_unlock(&inode->i_lock);
 186 }
 187
 188 void nfs_invalidate_atime(struct inode *inode)
 189 {
 190         spin_lock(&inode->i_lock);
 191         NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
 192         spin_unlock(&inode->i_lock);
 193 }
 194
 195 /*
 196  * Invalidate, but do not unhash, the inode.
 197  * NB: must be called with inode->i_lock held!
 198  */
 199 static void nfs_invalidate_inode(struct inode *inode)
 200 {
 201         set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 202         nfs_zap_caches_locked(inode);
 203 }
 204
 205 struct nfs_find_desc {
 206         struct nfs_fh           *fh;
 207         struct nfs_fattr        *fattr;
 208 };
 209
 210 /*
 211  * In NFSv3 we can have 64bit inode numbers. In order to support
 212  * this, and re-exported directories (also seen in NFSv2)
 213  * we are forced to allow 2 different inodes to have the same
 214  * i_ino.
 215  */
 216 static int
 217 nfs_find_actor(struct inode *inode, void *opaque)
 218 {
 219         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
 220         struct nfs_fh           *fh = desc->fh;
 221         struct nfs_fattr        *fattr = desc->fattr;
 222
 223         if (NFS_FILEID(inode) != fattr->fileid)
 224                 return 0;
 225         if (nfs_compare_fh(NFS_FH(inode), fh))
 226                 return 0;
 227         if (is_bad_inode(inode) || NFS_STALE(inode))
 228                 return 0;
 229         return 1;
 230 }
 231
 232 static int
 233 nfs_init_locked(struct inode *inode, void *opaque)
 234 {
 235         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
 236         struct nfs_fattr        *fattr = desc->fattr;
 237
 238         set_nfs_fileid(inode, fattr->fileid);
 239         nfs_copy_fh(NFS_FH(inode), desc->fh);
 240         return 0;
 241 }
 242
 243 /*
 244  * This is our front-end to iget that looks up inodes by file handle
 245  * instead of inode number.
 246  */
 247 struct inode *
 248 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
 249 {
 250         struct nfs_find_desc desc = {
 251                 .fh     = fh,
 252                 .fattr  = fattr
 253         };
 254         struct inode *inode = ERR_PTR(-ENOENT);
 255         unsigned long hash;
 256
 257         nfs_attr_check_mountpoint(sb, fattr);
 258
 259         if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
 260             !nfs_attr_use_mounted_on_fileid(fattr))
 261                 goto out_no_inode;
 262         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
 263                 goto out_no_inode;
 264
 265         hash = nfs_fattr_to_ino_t(fattr);
 266
 267         inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
 268         if (inode == NULL) {
 269                 inode = ERR_PTR(-ENOMEM);
 270                 goto out_no_inode;
 271         }
 272
 273         if (inode->i_state & I_NEW) {
 274                 struct nfs_inode *nfsi = NFS_I(inode);
 275                 unsigned long now = jiffies;
 276
 277                 /* We set i_ino for the few things that still rely on it,
 278                  * such as stat(2) */
 279                 inode->i_ino = hash;
 280
 281                 /* We can't support update_atime(), since the server will reset it */
 282                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 283                 inode->i_mode = fattr->mode;
 284                 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
 285                                 && nfs_server_capable(inode, NFS_CAP_MODE))
 286                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 287                                 | NFS_INO_INVALID_ACCESS
 288                                 | NFS_INO_INVALID_ACL;
 289                 /* Why so? Because we want revalidate for devices/FIFOs, and
 290                  * that's precisely what we have in nfs_file_inode_operations.
 291                  */
 292                 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
 293                 if (S_ISREG(inode->i_mode)) {
 294                         inode->i_fop = &nfs_file_operations;
 295                         inode->i_data.a_ops = &nfs_file_aops;
 296                         inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
 297                 } else if (S_ISDIR(inode->i_mode)) {
 298                         inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
 299                         inode->i_fop = &nfs_dir_operations;
 300                         inode->i_data.a_ops = &nfs_dir_aops;
 301                         if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS))
 302                                 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 303                         /* Deal with crossing mountpoints */
 304                         if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
 305                                         fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
 306                                 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
 307                                         inode->i_op = &nfs_referral_inode_operations;
 308                                 else
 309                                         inode->i_op = &nfs_mountpoint_inode_operations;
 310                                 inode->i_fop = NULL;
 311                                 inode->i_flags |= S_AUTOMOUNT;
 312                         }
 313                 } else if (S_ISLNK(inode->i_mode))
 314                         inode->i_op = &nfs_symlink_inode_operations;
 315                 else
 316                         init_special_inode(inode, inode->i_mode, fattr->rdev);
 317
 318                 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
 319                 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
 320                 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
 321                 nfsi->change_attr = 0;
 322                 inode->i_size = 0;
 323                 inode->i_nlink = 0;
 324                 inode->i_uid = -2;
 325                 inode->i_gid = -2;
 326                 inode->i_blocks = 0;
 327                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 328
 329                 nfsi->read_cache_jiffies = fattr->time_start;
 330                 nfsi->attr_gencount = fattr->gencount;
 331                 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
 332                         inode->i_atime = fattr->atime;
 333                 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
 334                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
 335                 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
 336                         inode->i_mtime = fattr->mtime;
 337                 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
 338                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 339                                 | NFS_INO_INVALID_DATA;
 340                 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 341                         inode->i_ctime = fattr->ctime;
 342                 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
 343                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 344                                 | NFS_INO_INVALID_ACCESS
 345                                 | NFS_INO_INVALID_ACL;
 346                 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 347                         nfsi->change_attr = fattr->change_attr;
 348                 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
 349                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 350                                 | NFS_INO_INVALID_DATA;
 351                 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
 352                         inode->i_size = nfs_size_to_loff_t(fattr->size);
 353                 else
 354                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 355                                 | NFS_INO_INVALID_DATA
 356                                 | NFS_INO_REVAL_PAGECACHE;
 357                 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
 358                         inode->i_nlink = fattr->nlink;
 359                 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
 360                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
 361                 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
 362                         inode->i_uid = fattr->uid;
 363                 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
 364                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 365                                 | NFS_INO_INVALID_ACCESS
 366                                 | NFS_INO_INVALID_ACL;
 367                 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
 368                         inode->i_gid = fattr->gid;
 369                 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
 370                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR
 371                                 | NFS_INO_INVALID_ACCESS
 372                                 | NFS_INO_INVALID_ACL;
 373                 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
 374                         inode->i_blocks = fattr->du.nfs2.blocks;
 375                 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
 376                         /*
 377                          * report the blocks in 512byte units
 378                          */
 379                         inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
 380                 }
 381                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 382                 nfsi->attrtimeo_timestamp = now;
 383                 nfsi->access_cache = RB_ROOT;
 384
 385                 nfs_fscache_init_inode_cookie(inode);
 386
 387                 unlock_new_inode(inode);
 388         } else
 389                 nfs_refresh_inode(inode, fattr);
 390         dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
 391                 inode->i_sb->s_id,
 392                 (long long)NFS_FILEID(inode),
 393                 atomic_read(&inode->i_count));
 394
 395 out:
 396         return inode;
 397
 398 out_no_inode:
 399         dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
 400         goto out;
 401 }
 402
 403 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
 404
 405 int
 406 nfs_setattr(struct dentry *dentry, struct iattr *attr)
 407 {
 408         struct inode *inode = dentry->d_inode;
 409         struct nfs_fattr *fattr;
 410         int error = -ENOMEM;
 411
 412         nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
 413
 414         /* skip mode change if it's just for clearing setuid/setgid */
 415         if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
 416                 attr->ia_valid &= ~ATTR_MODE;
 417
 418         if (attr->ia_valid & ATTR_SIZE) {
 419                 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
 420                         attr->ia_valid &= ~ATTR_SIZE;
 421         }
 422
 423         /* Optimization: if the end result is no change, don't RPC */
 424         attr->ia_valid &= NFS_VALID_ATTRS;
 425         if ((attr->ia_valid & ~ATTR_FILE) == 0)
 426                 return 0;
 427
 428         /* Write all dirty data */
 429         if (S_ISREG(inode->i_mode))
 430                 nfs_wb_all(inode);
 431
 432         fattr = nfs_alloc_fattr();
 433         if (fattr == NULL)
 434                 goto out;
 435         /*
 436          * Return any delegations if we're going to change ACLs
 437          */
 438         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
 439                 nfs_inode_return_delegation(inode);
 440         error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
 441         if (error == 0)
 442                 nfs_refresh_inode(inode, fattr);
 443         nfs_free_fattr(fattr);
 444 out:
 445         return error;
 446 }
 447
 448 /**
 449  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
 450  * @inode: inode of the file used
 451  * @offset: file offset to start truncating
 452  *
 453  * This is a copy of the common vmtruncate, but with the locking
 454  * corrected to take into account the fact that NFS requires
 455  * inode->i_size to be updated under the inode->i_lock.
 456  */
 457 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
 458 {
 459         loff_t oldsize;
 460         int err;
 461
 462         err = inode_newsize_ok(inode, offset);
 463         if (err)
 464                 goto out;
 465
 466         spin_lock(&inode->i_lock);
 467         oldsize = inode->i_size;
 468         i_size_write(inode, offset);
 469         spin_unlock(&inode->i_lock);
 470
 471         truncate_pagecache(inode, oldsize, offset);
 472 out:
 473         return err;
 474 }
 475
 476 /**
 477  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
 478  * @inode: pointer to struct inode
 479  * @attr: pointer to struct iattr
 480  *
 481  * Note: we do this in the *proc.c in order to ensure that
 482  *       it works for things like exclusive creates too.
 483  */
 484 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
 485 {
 486         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
 487                 spin_lock(&inode->i_lock);
 488                 if ((attr->ia_valid & ATTR_MODE) != 0) {
 489                         int mode = attr->ia_mode & S_IALLUGO;
 490                         mode |= inode->i_mode & ~S_IALLUGO;
 491                         inode->i_mode = mode;
 492                 }
 493                 if ((attr->ia_valid & ATTR_UID) != 0)
 494                         inode->i_uid = attr->ia_uid;
 495                 if ((attr->ia_valid & ATTR_GID) != 0)
 496                         inode->i_gid = attr->ia_gid;
 497                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
 498                 spin_unlock(&inode->i_lock);
 499         }
 500         if ((attr->ia_valid & ATTR_SIZE) != 0) {
 501                 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
 502                 nfs_vmtruncate(inode, attr->ia_size);
 503         }
 504 }
 505
 506 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 507 {
 508         struct inode *inode = dentry->d_inode;
 509         int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
 510         int err;
 511
 512         /* Flush out writes to the server in order to update c/mtime.  */
 513         if (S_ISREG(inode->i_mode)) {
 514                 err = filemap_write_and_wait(inode->i_mapping);
 515                 if (err)
 516                         goto out;
 517         }
 518
 519         /*
 520          * We may force a getattr if the user cares about atime.
 521          *
 522          * Note that we only have to check the vfsmount flags here:
 523          *  - NFS always sets S_NOATIME by so checking it would give a
 524          *    bogus result
 525          *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
 526          *    no point in checking those.
 527          */
 528         if ((mnt->mnt_flags & MNT_NOATIME) ||
 529             ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 530                 need_atime = 0;
 531
 532         if (need_atime)
 533                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 534         else
 535                 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
 536         if (!err) {
 537                 generic_fillattr(inode, stat);
 538                 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
 539         }
 540 out:
 541         return err;
 542 }
 543
 544 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
 545 {
 546         atomic_set(&l_ctx->count, 1);
 547         l_ctx->lockowner = current->files;
 548         l_ctx->pid = current->tgid;
 549         INIT_LIST_HEAD(&l_ctx->list);
 550 }
 551
 552 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
 553 {
 554         struct nfs_lock_context *pos;
 555
 556         list_for_each_entry(pos, &ctx->lock_context.list, list) {
 557                 if (pos->lockowner != current->files)
 558                         continue;
 559                 if (pos->pid != current->tgid)
 560                         continue;
 561                 atomic_inc(&pos->count);
 562                 return pos;
 563         }
 564         return NULL;
 565 }
 566
 567 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
 568 {
 569         struct nfs_lock_context *res, *new = NULL;
 570         struct inode *inode = ctx->path.dentry->d_inode;
 571
 572         spin_lock(&inode->i_lock);
 573         res = __nfs_find_lock_context(ctx);
 574         if (res == NULL) {
 575                 spin_unlock(&inode->i_lock);
 576                 new = kmalloc(sizeof(*new), GFP_KERNEL);
 577                 if (new == NULL)
 578                         return NULL;
 579                 nfs_init_lock_context(new);
 580                 spin_lock(&inode->i_lock);
 581                 res = __nfs_find_lock_context(ctx);
 582                 if (res == NULL) {
 583                         list_add_tail(&new->list, &ctx->lock_context.list);
 584                         new->open_context = ctx;
 585                         res = new;
 586                         new = NULL;
 587                 }
 588         }
 589         spin_unlock(&inode->i_lock);
 590         kfree(new);
 591         return res;
 592 }
 593
 594 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
 595 {
 596         struct nfs_open_context *ctx = l_ctx->open_context;
 597         struct inode *inode = ctx->path.dentry->d_inode;
 598
 599         if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
 600                 return;
 601         list_del(&l_ctx->list);
 602         spin_unlock(&inode->i_lock);
 603         kfree(l_ctx);
 604 }
 605
 606 /**
 607  * nfs_close_context - Common close_context() routine NFSv2/v3
 608  * @ctx: pointer to context
 609  * @is_sync: is this a synchronous close
 610  *
 611  * always ensure that the attributes are up to date if we're mounted
 612  * with close-to-open semantics
 613  */
 614 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
 615 {
 616         struct inode *inode;
 617         struct nfs_server *server;
 618
 619         if (!(ctx->mode & FMODE_WRITE))
 620                 return;
 621         if (!is_sync)
 622                 return;
 623         inode = ctx->path.dentry->d_inode;
 624         if (!list_empty(&NFS_I(inode)->open_files))
 625                 return;
 626         server = NFS_SERVER(inode);
 627         if (server->flags & NFS_MOUNT_NOCTO)
 628                 return;
 629         nfs_revalidate_inode(server, inode);
 630 }
 631
 632 struct nfs_open_context *alloc_nfs_open_context(struct path *path, struct rpc_cred *cred, fmode_t f_mode)
 633 {
 634         struct nfs_open_context *ctx;
 635
 636         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 637         if (ctx != NULL) {
 638                 ctx->path = *path;
 639                 path_get(&ctx->path);
 640                 ctx->cred = get_rpccred(cred);
 641                 ctx->state = NULL;
 642                 ctx->mode = f_mode;
 643                 ctx->flags = 0;
 644                 ctx->error = 0;
 645                 nfs_init_lock_context(&ctx->lock_context);
 646                 ctx->lock_context.open_context = ctx;
 647                 INIT_LIST_HEAD(&ctx->list);
 648         }
 649         return ctx;
 650 }
 651
 652 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
 653 {
 654         if (ctx != NULL)
 655                 atomic_inc(&ctx->lock_context.count);
 656         return ctx;
 657 }
 658
 659 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
 660 {
 661         struct inode *inode = ctx->path.dentry->d_inode;
 662
 663         if (!list_empty(&ctx->list)) {
 664                 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
 665                         return;
 666                 list_del(&ctx->list);
 667                 spin_unlock(&inode->i_lock);
 668         } else if (!atomic_dec_and_test(&ctx->lock_context.count))
 669                 return;
 670         if (inode != NULL)
 671                 NFS_PROTO(inode)->close_context(ctx, is_sync);
 672         if (ctx->cred != NULL)
 673                 put_rpccred(ctx->cred);
 674         path_put(&ctx->path);
 675         kfree(ctx);
 676 }
 677
 678 void put_nfs_open_context(struct nfs_open_context *ctx)
 679 {
 680         __put_nfs_open_context(ctx, 0);
 681 }
 682
 683 /*
 684  * Ensure that mmap has a recent RPC credential for use when writing out
 685  * shared pages
 686  */
 687 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
 688 {
 689         struct inode *inode = filp->f_path.dentry->d_inode;
 690         struct nfs_inode *nfsi = NFS_I(inode);
 691
 692         filp->private_data = get_nfs_open_context(ctx);
 693         spin_lock(&inode->i_lock);
 694         list_add(&ctx->list, &nfsi->open_files);
 695         spin_unlock(&inode->i_lock);
 696 }
 697
 698 /*
 699  * Given an inode, search for an open context with the desired characteristics
 700  */
 701 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
 702 {
 703         struct nfs_inode *nfsi = NFS_I(inode);
 704         struct nfs_open_context *pos, *ctx = NULL;
 705
 706         spin_lock(&inode->i_lock);
 707         list_for_each_entry(pos, &nfsi->open_files, list) {
 708                 if (cred != NULL && pos->cred != cred)
 709                         continue;
 710                 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
 711                         continue;
 712                 ctx = get_nfs_open_context(pos);
 713                 break;
 714         }
 715         spin_unlock(&inode->i_lock);
 716         return ctx;
 717 }
 718
 719 static void nfs_file_clear_open_context(struct file *filp)
 720 {
 721         struct inode *inode = filp->f_path.dentry->d_inode;
 722         struct nfs_open_context *ctx = nfs_file_open_context(filp);
 723
 724         if (ctx) {
 725                 filp->private_data = NULL;
 726                 spin_lock(&inode->i_lock);
 727                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
 728                 spin_unlock(&inode->i_lock);
 729                 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
 730         }
 731 }
 732
 733 /*
 734  * These allocate and release file read/write context information.
 735  */
 736 int nfs_open(struct inode *inode, struct file *filp)
 737 {
 738         struct nfs_open_context *ctx;
 739         struct rpc_cred *cred;
 740
 741         cred = rpc_lookup_cred();
 742         if (IS_ERR(cred))
 743                 return PTR_ERR(cred);
 744         ctx = alloc_nfs_open_context(&filp->f_path, cred, filp->f_mode);
 745         put_rpccred(cred);
 746         if (ctx == NULL)
 747                 return -ENOMEM;
 748         nfs_file_set_open_context(filp, ctx);
 749         put_nfs_open_context(ctx);
 750         nfs_fscache_set_inode_cookie(inode, filp);
 751         return 0;
 752 }
 753
 754 int nfs_release(struct inode *inode, struct file *filp)
 755 {
 756         nfs_file_clear_open_context(filp);
 757         return 0;
 758 }
 759
 760 /*
 761  * This function is called whenever some part of NFS notices that
 762  * the cached attributes have to be refreshed.
 763  */
 764 int
 765 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 766 {
 767         int              status = -ESTALE;
 768         struct nfs_fattr *fattr = NULL;
 769         struct nfs_inode *nfsi = NFS_I(inode);
 770
 771         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
 772                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
 773
 774         if (is_bad_inode(inode))
 775                 goto out;
 776         if (NFS_STALE(inode))
 777                 goto out;
 778
 779         status = -ENOMEM;
 780         fattr = nfs_alloc_fattr();
 781         if (fattr == NULL)
 782                 goto out;
 783
 784         nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
 785         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
 786         if (status != 0) {
 787                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
 788                          inode->i_sb->s_id,
 789                          (long long)NFS_FILEID(inode), status);
 790                 if (status == -ESTALE) {
 791                         nfs_zap_caches(inode);
 792                         if (!S_ISDIR(inode->i_mode))
 793                                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 794                 }
 795                 goto out;
 796         }
 797
 798         status = nfs_refresh_inode(inode, fattr);
 799         if (status) {
 800                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
 801                          inode->i_sb->s_id,
 802                          (long long)NFS_FILEID(inode), status);
 803                 goto out;
 804         }
 805
 806         if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
 807                 nfs_zap_acl_cache(inode);
 808
 809         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
 810                 inode->i_sb->s_id,
 811                 (long long)NFS_FILEID(inode));
 812
 813  out:
 814         nfs_free_fattr(fattr);
 815         return status;
 816 }
 817
 818 int nfs_attribute_timeout(struct inode *inode)
 819 {
 820         struct nfs_inode *nfsi = NFS_I(inode);
 821
 822         return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
 823 }
 824
 825 static int nfs_attribute_cache_expired(struct inode *inode)
 826 {
 827         if (nfs_have_delegated_attributes(inode))
 828                 return 0;
 829         return nfs_attribute_timeout(inode);
 830 }
 831
 832 /**
 833  * nfs_revalidate_inode - Revalidate the inode attributes
 834  * @server - pointer to nfs_server struct
 835  * @inode - pointer to inode struct
 836  *
 837  * Updates inode attribute information by retrieving the data from the server.
 838  */
 839 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 840 {
 841         if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
 842                         && !nfs_attribute_cache_expired(inode))
 843                 return NFS_STALE(inode) ? -ESTALE : 0;
 844         return __nfs_revalidate_inode(server, inode);
 845 }
 846
 847 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
 848 {
 849         struct nfs_inode *nfsi = NFS_I(inode);
 850
 851         if (mapping->nrpages != 0) {
 852                 int ret = invalidate_inode_pages2(mapping);
 853                 if (ret < 0)
 854                         return ret;
 855         }
 856         spin_lock(&inode->i_lock);
 857         nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
 858         if (S_ISDIR(inode->i_mode))
 859                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 860         spin_unlock(&inode->i_lock);
 861         nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
 862         nfs_fscache_reset_inode_cookie(inode);
 863         dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
 864                         inode->i_sb->s_id, (long long)NFS_FILEID(inode));
 865         return 0;
 866 }
 867
 868 /**
 869  * nfs_revalidate_mapping - Revalidate the pagecache
 870  * @inode - pointer to host inode
 871  * @mapping - pointer to mapping
 872  */
 873 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
 874 {
 875         struct nfs_inode *nfsi = NFS_I(inode);
 876         int ret = 0;
 877
 878         if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
 879                         || nfs_attribute_cache_expired(inode)
 880                         || NFS_STALE(inode)) {
 881                 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 882                 if (ret < 0)
 883                         goto out;
 884         }
 885         if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
 886                 ret = nfs_invalidate_mapping(inode, mapping);
 887 out:
 888         return ret;
 889 }
 890
 891 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 892 {
 893         struct nfs_inode *nfsi = NFS_I(inode);
 894         unsigned long ret = 0;
 895
 896         if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
 897                         && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 898                         && nfsi->change_attr == fattr->pre_change_attr) {
 899                 nfsi->change_attr = fattr->change_attr;
 900                 if (S_ISDIR(inode->i_mode))
 901                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 902                 ret |= NFS_INO_INVALID_ATTR;
 903         }
 904         /* If we have atomic WCC data, we may update some attributes */
 905         if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
 906                         && (fattr->valid & NFS_ATTR_FATTR_CTIME)
 907                         && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
 908                 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
 909                 ret |= NFS_INO_INVALID_ATTR;
 910         }
 911
 912         if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
 913                         && (fattr->valid & NFS_ATTR_FATTR_MTIME)
 914                         && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
 915                 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
 916                 if (S_ISDIR(inode->i_mode))
 917                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 918                 ret |= NFS_INO_INVALID_ATTR;
 919         }
 920         if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
 921                         && (fattr->valid & NFS_ATTR_FATTR_SIZE)
 922                         && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
 923                         && nfsi->npages == 0) {
 924                 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
 925                 ret |= NFS_INO_INVALID_ATTR;
 926         }
 927         return ret;
 928 }
 929
 930 /**
 931  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
 932  * @inode - pointer to inode
 933  * @fattr - updated attributes
 934  *
 935  * Verifies the attribute cache. If we have just changed the attributes,
 936  * so that fattr carries weak cache consistency data, then it may
 937  * also update the ctime/mtime/change_attribute.
 938  */
 939 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
 940 {
 941         struct nfs_inode *nfsi = NFS_I(inode);
 942         loff_t cur_size, new_isize;
 943         unsigned long invalid = 0;
 944
 945
 946         /* Has the inode gone and changed behind our back? */
 947         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
 948                 return -EIO;
 949         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
 950                 return -EIO;
 951
 952         if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
 953                         nfsi->change_attr != fattr->change_attr)
 954                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 955
 956         /* Verify a few of the more important attributes */
 957         if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
 958                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 959
 960         if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
 961                 cur_size = i_size_read(inode);
 962                 new_isize = nfs_size_to_loff_t(fattr->size);
 963                 if (cur_size != new_isize && nfsi->npages == 0)
 964                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 965         }
 966
 967         /* Have any file permissions changed? */
 968         if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
 969                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 970         if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
 971                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 972         if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
 973                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 974
 975         /* Has the link count changed? */
 976         if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
 977                 invalid |= NFS_INO_INVALID_ATTR;
 978
 979         if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
 980                 invalid |= NFS_INO_INVALID_ATIME;
 981
 982         if (invalid != 0)
 983                 nfsi->cache_validity |= invalid;
 984
 985         nfsi->read_cache_jiffies = fattr->time_start;
 986         return 0;
 987 }
 988
 989 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
 990 {
 991         if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
 992                 return 0;
 993         return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
 994 }
 995
 996 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
 997 {
 998         if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
 999                 return 0;
1000         return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1001 }
1002
1003 static atomic_long_t nfs_attr_generation_counter;
1004
1005 static unsigned long nfs_read_attr_generation_counter(void)
1006 {
1007         return atomic_long_read(&nfs_attr_generation_counter);
1008 }
1009
1010 unsigned long nfs_inc_attr_generation_counter(void)
1011 {
1012         return atomic_long_inc_return(&nfs_attr_generation_counter);
1013 }
1014
1015 void nfs_fattr_init(struct nfs_fattr *fattr)
1016 {
1017         fattr->valid = 0;
1018         fattr->time_start = jiffies;
1019         fattr->gencount = nfs_inc_attr_generation_counter();
1020 }
1021
1022 struct nfs_fattr *nfs_alloc_fattr(void)
1023 {
1024         struct nfs_fattr *fattr;
1025
1026         fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1027         if (fattr != NULL)
1028                 nfs_fattr_init(fattr);
1029         return fattr;
1030 }
1031
1032 struct nfs_fh *nfs_alloc_fhandle(void)
1033 {
1034         struct nfs_fh *fh;
1035
1036         fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1037         if (fh != NULL)
1038                 fh->size = 0;
1039         return fh;
1040 }
1041
1042 /**
1043  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1044  * @inode - pointer to inode
1045  * @fattr - attributes
1046  *
1047  * Attempt to divine whether or not an RPC call reply carrying stale
1048  * attributes got scheduled after another call carrying updated ones.
1049  *
1050  * To do so, the function first assumes that a more recent ctime means
1051  * that the attributes in fattr are newer, however it also attempt to
1052  * catch the case where ctime either didn't change, or went backwards
1053  * (if someone reset the clock on the server) by looking at whether
1054  * or not this RPC call was started after the inode was last updated.
1055  * Note also the check for wraparound of 'attr_gencount'
1056  *
1057  * The function returns 'true' if it thinks the attributes in 'fattr' are
1058  * more recent than the ones cached in the inode.
1059  *
1060  */
1061 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1062 {
1063         const struct nfs_inode *nfsi = NFS_I(inode);
1064
1065         return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1066                 nfs_ctime_need_update(inode, fattr) ||
1067                 nfs_size_need_update(inode, fattr) ||
1068                 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1069 }
1070
1071 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1072 {
1073         if (nfs_inode_attrs_need_update(inode, fattr))
1074                 return nfs_update_inode(inode, fattr);
1075         return nfs_check_inode_attributes(inode, fattr);
1076 }
1077
1078 /**
1079  * nfs_refresh_inode - try to update the inode attribute cache
1080  * @inode - pointer to inode
1081  * @fattr - updated attributes
1082  *
1083  * Check that an RPC call that returned attributes has not overlapped with
1084  * other recent updates of the inode metadata, then decide whether it is
1085  * safe to do a full update of the inode attributes, or whether just to
1086  * call nfs_check_inode_attributes.
1087  */
1088 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1089 {
1090         int status;
1091
1092         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1093                 return 0;
1094         spin_lock(&inode->i_lock);
1095         status = nfs_refresh_inode_locked(inode, fattr);
1096         spin_unlock(&inode->i_lock);
1097
1098         return status;
1099 }
1100
1101 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1102 {
1103         struct nfs_inode *nfsi = NFS_I(inode);
1104
1105         nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1106         if (S_ISDIR(inode->i_mode))
1107                 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1108         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1109                 return 0;
1110         return nfs_refresh_inode_locked(inode, fattr);
1111 }
1112
1113 /**
1114  * nfs_post_op_update_inode - try to update the inode attribute cache
1115  * @inode - pointer to inode
1116  * @fattr - updated attributes
1117  *
1118  * After an operation that has changed the inode metadata, mark the
1119  * attribute cache as being invalid, then try to update it.
1120  *
1121  * NB: if the server didn't return any post op attributes, this
1122  * function will force the retrieval of attributes before the next
1123  * NFS request.  Thus it should be used only for operations that
1124  * are expected to change one or more attributes, to avoid
1125  * unnecessary NFS requests and trips through nfs_update_inode().
1126  */
1127 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1128 {
1129         int status;
1130
1131         spin_lock(&inode->i_lock);
1132         status = nfs_post_op_update_inode_locked(inode, fattr);
1133         spin_unlock(&inode->i_lock);
1134         return status;
1135 }
1136
1137 /**
1138  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1139  * @inode - pointer to inode
1140  * @fattr - updated attributes
1141  *
1142  * After an operation that has changed the inode metadata, mark the
1143  * attribute cache as being invalid, then try to update it. Fake up
1144  * weak cache consistency data, if none exist.
1145  *
1146  * This function is mainly designed to be used by the ->write_done() functions.
1147  */
1148 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1149 {
1150         int status;
1151
1152         spin_lock(&inode->i_lock);
1153         /* Don't do a WCC update if these attributes are already stale */
1154         if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1155                         !nfs_inode_attrs_need_update(inode, fattr)) {
1156                 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1157                                 | NFS_ATTR_FATTR_PRESIZE
1158                                 | NFS_ATTR_FATTR_PREMTIME
1159                                 | NFS_ATTR_FATTR_PRECTIME);
1160                 goto out_noforce;
1161         }
1162         if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1163                         (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1164                 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1165                 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1166         }
1167         if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1168                         (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1169                 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1170                 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1171         }
1172         if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1173                         (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1174                 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1175                 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1176         }
1177         if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1178                         (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1179                 fattr->pre_size = i_size_read(inode);
1180                 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1181         }
1182 out_noforce:
1183         status = nfs_post_op_update_inode_locked(inode, fattr);
1184         spin_unlock(&inode->i_lock);
1185         return status;
1186 }
1187
1188 /*
1189  * Many nfs protocol calls return the new file attributes after
1190  * an operation.  Here we update the inode to reflect the state
1191  * of the server's inode.
1192  *
1193  * This is a bit tricky because we have to make sure all dirty pages
1194  * have been sent off to the server before calling invalidate_inode_pages.
1195  * To make sure no other process adds more write requests while we try
1196  * our best to flush them, we make them sleep during the attribute refresh.
1197  *
1198  * A very similar scenario holds for the dir cache.
1199  */
1200 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1201 {
1202         struct nfs_server *server;
1203         struct nfs_inode *nfsi = NFS_I(inode);
1204         loff_t cur_isize, new_isize;
1205         unsigned long invalid = 0;
1206         unsigned long now = jiffies;
1207         unsigned long save_cache_validity;
1208
1209         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1210                         __func__, inode->i_sb->s_id, inode->i_ino,
1211                         atomic_read(&inode->i_count), fattr->valid);
1212
1213         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1214                 goto out_fileid;
1215
1216         /*
1217          * Make sure the inode's type hasn't changed.
1218          */
1219         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1220                 goto out_changed;
1221
1222         server = NFS_SERVER(inode);
1223         /* Update the fsid? */
1224         if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1225                         !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1226                         !IS_AUTOMOUNT(inode))
1227                 server->fsid = fattr->fsid;
1228
1229         /*
1230          * Update the read time so we don't revalidate too often.
1231          */
1232         nfsi->read_cache_jiffies = fattr->time_start;
1233
1234         save_cache_validity = nfsi->cache_validity;
1235         nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1236                         | NFS_INO_INVALID_ATIME
1237                         | NFS_INO_REVAL_FORCED
1238                         | NFS_INO_REVAL_PAGECACHE);
1239
1240         /* Do atomic weak cache consistency updates */
1241         invalid |= nfs_wcc_update_inode(inode, fattr);
1242
1243         /* More cache consistency checks */
1244         if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1245                 if (nfsi->change_attr != fattr->change_attr) {
1246                         dprintk("NFS: change_attr change on server for file %s/%ld\n",
1247                                         inode->i_sb->s_id, inode->i_ino);
1248                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1249                         if (S_ISDIR(inode->i_mode))
1250                                 nfs_force_lookup_revalidate(inode);
1251                         nfsi->change_attr = fattr->change_attr;
1252                 }
1253         } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1254                 invalid |= save_cache_validity;
1255
1256         if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1257                 /* NFSv2/v3: Check if the mtime agrees */
1258                 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1259                         dprintk("NFS: mtime change on server for file %s/%ld\n",
1260                                         inode->i_sb->s_id, inode->i_ino);
1261                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1262                         if (S_ISDIR(inode->i_mode))
1263                                 nfs_force_lookup_revalidate(inode);
1264                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1265                 }
1266         } else if (server->caps & NFS_CAP_MTIME)
1267                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1268                                 | NFS_INO_INVALID_DATA
1269                                 | NFS_INO_REVAL_PAGECACHE
1270                                 | NFS_INO_REVAL_FORCED);
1271
1272         if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1273                 /* If ctime has changed we should definitely clear access+acl caches */
1274                 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1275                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1276                         /* and probably clear data for a directory too as utimes can cause
1277                          * havoc with our cache.
1278                          */
1279                         if (S_ISDIR(inode->i_mode)) {
1280                                 invalid |= NFS_INO_INVALID_DATA;
1281                                 nfs_force_lookup_revalidate(inode);
1282                         }
1283                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1284                 }
1285         } else if (server->caps & NFS_CAP_CTIME)
1286                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1287                                 | NFS_INO_INVALID_ACCESS
1288                                 | NFS_INO_INVALID_ACL
1289                                 | NFS_INO_REVAL_FORCED);
1290
1291         /* Check if our cached file size is stale */
1292         if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1293                 new_isize = nfs_size_to_loff_t(fattr->size);
1294                 cur_isize = i_size_read(inode);
1295                 if (new_isize != cur_isize) {
1296                         /* Do we perhaps have any outstanding writes, or has
1297                          * the file grown beyond our last write? */
1298                         if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1299                              new_isize > cur_isize) {
1300                                 i_size_write(inode, new_isize);
1301                                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1302                         }
1303                         dprintk("NFS: isize change on server for file %s/%ld\n",
1304                                         inode->i_sb->s_id, inode->i_ino);
1305                 }
1306         } else
1307                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1308                                 | NFS_INO_REVAL_PAGECACHE
1309                                 | NFS_INO_REVAL_FORCED);
1310
1311
1312         if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1313                 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1314         else if (server->caps & NFS_CAP_ATIME)
1315                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1316                                 | NFS_INO_REVAL_FORCED);
1317
1318         if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1319                 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1320                         umode_t newmode = inode->i_mode & S_IFMT;
1321                         newmode |= fattr->mode & S_IALLUGO;
1322                         inode->i_mode = newmode;
1323                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1324                 }
1325         } else if (server->caps & NFS_CAP_MODE)
1326                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1327                                 | NFS_INO_INVALID_ACCESS
1328                                 | NFS_INO_INVALID_ACL
1329                                 | NFS_INO_REVAL_FORCED);
1330
1331         if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1332                 if (inode->i_uid != fattr->uid) {
1333                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1334                         inode->i_uid = fattr->uid;
1335                 }
1336         } else if (server->caps & NFS_CAP_OWNER)
1337                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1338                                 | NFS_INO_INVALID_ACCESS
1339                                 | NFS_INO_INVALID_ACL
1340                                 | NFS_INO_REVAL_FORCED);
1341
1342         if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1343                 if (inode->i_gid != fattr->gid) {
1344                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1345                         inode->i_gid = fattr->gid;
1346                 }
1347         } else if (server->caps & NFS_CAP_OWNER_GROUP)
1348                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1349                                 | NFS_INO_INVALID_ACCESS
1350                                 | NFS_INO_INVALID_ACL
1351                                 | NFS_INO_REVAL_FORCED);
1352
1353         if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1354                 if (inode->i_nlink != fattr->nlink) {
1355                         invalid |= NFS_INO_INVALID_ATTR;
1356                         if (S_ISDIR(inode->i_mode))
1357                                 invalid |= NFS_INO_INVALID_DATA;
1358                         inode->i_nlink = fattr->nlink;
1359                 }
1360         } else if (server->caps & NFS_CAP_NLINK)
1361                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1362                                 | NFS_INO_REVAL_FORCED);
1363
1364         if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1365                 /*
1366                  * report the blocks in 512byte units
1367                  */
1368                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1369         }
1370         if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1371                 inode->i_blocks = fattr->du.nfs2.blocks;
1372
1373         /* Update attrtimeo value if we're out of the unstable period */
1374         if (invalid & NFS_INO_INVALID_ATTR) {
1375                 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1376                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1377                 nfsi->attrtimeo_timestamp = now;
1378                 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1379         } else {
1380                 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1381                         if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1382                                 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1383                         nfsi->attrtimeo_timestamp = now;
1384                 }
1385         }
1386         invalid &= ~NFS_INO_INVALID_ATTR;
1387         /* Don't invalidate the data if we were to blame */
1388         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1389                                 || S_ISLNK(inode->i_mode)))
1390                 invalid &= ~NFS_INO_INVALID_DATA;
1391         if (!nfs_have_delegation(inode, FMODE_READ) ||
1392                         (save_cache_validity & NFS_INO_REVAL_FORCED))
1393                 nfsi->cache_validity |= invalid;
1394
1395         return 0;
1396  out_changed:
1397         /*
1398          * Big trouble! The inode has become a different object.
1399          */
1400         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1401                         __func__, inode->i_ino, inode->i_mode, fattr->mode);
1402  out_err:
1403         /*
1404          * No need to worry about unhashing the dentry, as the
1405          * lookup validation will know that the inode is bad.
1406          * (But we fall through to invalidate the caches.)
1407          */
1408         nfs_invalidate_inode(inode);
1409         return -ESTALE;
1410
1411  out_fileid:
1412         printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1413                 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1414                 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1415                 (long long)nfsi->fileid, (long long)fattr->fileid);
1416         goto out_err;
1417 }
1418
1419
1420 #ifdef CONFIG_NFS_V4
1421
1422 /*
1423  * Clean out any remaining NFSv4 state that might be left over due
1424  * to open() calls that passed nfs_atomic_lookup, but failed to call
1425  * nfs_open().
1426  */
1427 void nfs4_evict_inode(struct inode *inode)
1428 {
1429         pnfs_destroy_layout(NFS_I(inode));
1430         truncate_inode_pages(&inode->i_data, 0);
1431         end_writeback(inode);
1432         /* If we are holding a delegation, return it! */
1433         nfs_inode_return_delegation_noreclaim(inode);
1434         /* First call standard NFS clear_inode() code */
1435         nfs_clear_inode(inode);
1436 }
1437 #endif
1438
1439 struct inode *nfs_alloc_inode(struct super_block *sb)
1440 {
1441         struct nfs_inode *nfsi;
1442         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1443         if (!nfsi)
1444                 return NULL;
1445         nfsi->flags = 0UL;
1446         nfsi->cache_validity = 0UL;
1447 #ifdef CONFIG_NFS_V3_ACL
1448         nfsi->acl_access = ERR_PTR(-EAGAIN);
1449         nfsi->acl_default = ERR_PTR(-EAGAIN);
1450 #endif
1451 #ifdef CONFIG_NFS_V4
1452         nfsi->nfs4_acl = NULL;
1453 #endif /* CONFIG_NFS_V4 */
1454         return &nfsi->vfs_inode;
1455 }
1456
1457 static void nfs_i_callback(struct rcu_head *head)
1458 {
1459         struct inode *inode = container_of(head, struct inode, i_rcu);
1460         INIT_LIST_HEAD(&inode->i_dentry);
1461         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1462 }
1463
1464 void nfs_destroy_inode(struct inode *inode)
1465 {
1466         call_rcu(&inode->i_rcu, nfs_i_callback);
1467 }
1468
1469 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1470 {
1471 #ifdef CONFIG_NFS_V4
1472         INIT_LIST_HEAD(&nfsi->open_states);
1473         nfsi->delegation = NULL;
1474         nfsi->delegation_state = 0;
1475         init_rwsem(&nfsi->rwsem);
1476         nfsi->layout = NULL;
1477         atomic_set(&nfsi->commits_outstanding, 0);
1478 #endif
1479 }
1480
1481 static void init_once(void *foo)
1482 {
1483         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1484
1485         inode_init_once(&nfsi->vfs_inode);
1486         INIT_LIST_HEAD(&nfsi->open_files);
1487         INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1488         INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1489         INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1490         nfsi->npages = 0;
1491         nfsi->ncommit = 0;
1492         atomic_set(&nfsi->silly_count, 1);
1493         INIT_HLIST_HEAD(&nfsi->silly_list);
1494         init_waitqueue_head(&nfsi->waitqueue);
1495         nfs4_init_once(nfsi);
1496 }
1497
1498 static int __init nfs_init_inodecache(void)
1499 {
1500         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1501                                              sizeof(struct nfs_inode),
1502                                              0, (SLAB_RECLAIM_ACCOUNT|
1503                                                 SLAB_MEM_SPREAD),
1504                                              init_once);
1505         if (nfs_inode_cachep == NULL)
1506                 return -ENOMEM;
1507
1508         return 0;
1509 }
1510
1511 static void nfs_destroy_inodecache(void)
1512 {
1513         kmem_cache_destroy(nfs_inode_cachep);
1514 }
1515
1516 struct workqueue_struct *nfsiod_workqueue;
1517
1518 /*
1519  * start up the nfsiod workqueue
1520  */
1521 static int nfsiod_start(void)
1522 {
1523         struct workqueue_struct *wq;
1524         dprintk("RPC:       creating workqueue nfsiod\n");
1525         wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1526         if (wq == NULL)
1527                 return -ENOMEM;
1528         nfsiod_workqueue = wq;
1529         return 0;
1530 }
1531
1532 /*
1533  * Destroy the nfsiod workqueue
1534  */
1535 static void nfsiod_stop(void)
1536 {
1537         struct workqueue_struct *wq;
1538
1539         wq = nfsiod_workqueue;
1540         if (wq == NULL)
1541                 return;
1542         nfsiod_workqueue = NULL;
1543         destroy_workqueue(wq);
1544 }
1545
1546 /*
1547  * Initialize NFS
1548  */
1549 static int __init init_nfs_fs(void)
1550 {
1551         int err;
1552
1553         err = nfs_idmap_init();
1554         if (err < 0)
1555                 goto out9;
1556
1557         err = nfs_dns_resolver_init();
1558         if (err < 0)
1559                 goto out8;
1560
1561         err = nfs_fscache_register();
1562         if (err < 0)
1563                 goto out7;
1564
1565         err = nfsiod_start();
1566         if (err)
1567                 goto out6;
1568
1569         err = nfs_fs_proc_init();
1570         if (err)
1571                 goto out5;
1572
1573         err = nfs_init_nfspagecache();
1574         if (err)
1575                 goto out4;
1576
1577         err = nfs_init_inodecache();
1578         if (err)
1579                 goto out3;
1580
1581         err = nfs_init_readpagecache();
1582         if (err)
1583                 goto out2;
1584
1585         err = nfs_init_writepagecache();
1586         if (err)
1587                 goto out1;
1588
1589         err = nfs_init_directcache();
1590         if (err)
1591                 goto out0;
1592
1593 #ifdef CONFIG_PROC_FS
1594         rpc_proc_register(&nfs_rpcstat);
1595 #endif
1596         if ((err = register_nfs_fs()) != 0)
1597                 goto out;
1598         return 0;
1599 out:
1600 #ifdef CONFIG_PROC_FS
1601         rpc_proc_unregister("nfs");
1602 #endif
1603         nfs_destroy_directcache();
1604 out0:
1605         nfs_destroy_writepagecache();
1606 out1:
1607         nfs_destroy_readpagecache();
1608 out2:
1609         nfs_destroy_inodecache();
1610 out3:
1611         nfs_destroy_nfspagecache();
1612 out4:
1613         nfs_fs_proc_exit();
1614 out5:
1615         nfsiod_stop();
1616 out6:
1617         nfs_fscache_unregister();
1618 out7:
1619         nfs_dns_resolver_destroy();
1620 out8:
1621         nfs_idmap_quit();
1622 out9:
1623         return err;
1624 }
1625
1626 static void __exit exit_nfs_fs(void)
1627 {
1628         nfs_destroy_directcache();
1629         nfs_destroy_writepagecache();
1630         nfs_destroy_readpagecache();
1631         nfs_destroy_inodecache();
1632         nfs_destroy_nfspagecache();
1633         nfs_fscache_unregister();
1634         nfs_dns_resolver_destroy();
1635         nfs_idmap_quit();
1636 #ifdef CONFIG_PROC_FS
1637         rpc_proc_unregister("nfs");
1638 #endif
1639         nfs_cleanup_cb_ident_idr();
1640         unregister_nfs_fs();
1641         nfs_fs_proc_exit();
1642         nfsiod_stop();
1643 }
1644
1645 /* Not quite true; I just maintain it */
1646 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1647 MODULE_LICENSE("GPL");
1648 module_param(enable_ino64, bool, 0644);
1649
1650 module_init(init_nfs_fs)
1651 module_exit(exit_nfs_fs)