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_SB(sb)->nfs_client->rpc_ops->file_ops;
 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                 inode->i_version = 0;
 322                 inode->i_size = 0;
 323                 clear_nlink(inode);
 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                         inode->i_version = 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                         set_nlink(inode, 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->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->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->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 dentry *dentry, 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                 nfs_sb_active(dentry->d_sb);
 639                 ctx->dentry = dget(dentry);
 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->dentry->d_inode;
 662         struct super_block *sb = ctx->dentry->d_sb;
 663
 664         if (!list_empty(&ctx->list)) {
 665                 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
 666                         return;
 667                 list_del(&ctx->list);
 668                 spin_unlock(&inode->i_lock);
 669         } else if (!atomic_dec_and_test(&ctx->lock_context.count))
 670                 return;
 671         if (inode != NULL)
 672                 NFS_PROTO(inode)->close_context(ctx, is_sync);
 673         if (ctx->cred != NULL)
 674                 put_rpccred(ctx->cred);
 675         dput(ctx->dentry);
 676         nfs_sb_deactive(sb);
 677         kfree(ctx);
 678 }
 679
 680 void put_nfs_open_context(struct nfs_open_context *ctx)
 681 {
 682         __put_nfs_open_context(ctx, 0);
 683 }
 684
 685 /*
 686  * Ensure that mmap has a recent RPC credential for use when writing out
 687  * shared pages
 688  */
 689 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
 690 {
 691         struct inode *inode = filp->f_path.dentry->d_inode;
 692         struct nfs_inode *nfsi = NFS_I(inode);
 693
 694         filp->private_data = get_nfs_open_context(ctx);
 695         spin_lock(&inode->i_lock);
 696         list_add(&ctx->list, &nfsi->open_files);
 697         spin_unlock(&inode->i_lock);
 698 }
 699
 700 /*
 701  * Given an inode, search for an open context with the desired characteristics
 702  */
 703 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
 704 {
 705         struct nfs_inode *nfsi = NFS_I(inode);
 706         struct nfs_open_context *pos, *ctx = NULL;
 707
 708         spin_lock(&inode->i_lock);
 709         list_for_each_entry(pos, &nfsi->open_files, list) {
 710                 if (cred != NULL && pos->cred != cred)
 711                         continue;
 712                 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
 713                         continue;
 714                 ctx = get_nfs_open_context(pos);
 715                 break;
 716         }
 717         spin_unlock(&inode->i_lock);
 718         return ctx;
 719 }
 720
 721 static void nfs_file_clear_open_context(struct file *filp)
 722 {
 723         struct inode *inode = filp->f_path.dentry->d_inode;
 724         struct nfs_open_context *ctx = nfs_file_open_context(filp);
 725
 726         if (ctx) {
 727                 filp->private_data = NULL;
 728                 spin_lock(&inode->i_lock);
 729                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
 730                 spin_unlock(&inode->i_lock);
 731                 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
 732         }
 733 }
 734
 735 /*
 736  * These allocate and release file read/write context information.
 737  */
 738 int nfs_open(struct inode *inode, struct file *filp)
 739 {
 740         struct nfs_open_context *ctx;
 741         struct rpc_cred *cred;
 742
 743         cred = rpc_lookup_cred();
 744         if (IS_ERR(cred))
 745                 return PTR_ERR(cred);
 746         ctx = alloc_nfs_open_context(filp->f_path.dentry, cred, filp->f_mode);
 747         put_rpccred(cred);
 748         if (ctx == NULL)
 749                 return -ENOMEM;
 750         nfs_file_set_open_context(filp, ctx);
 751         put_nfs_open_context(ctx);
 752         nfs_fscache_set_inode_cookie(inode, filp);
 753         return 0;
 754 }
 755
 756 int nfs_release(struct inode *inode, struct file *filp)
 757 {
 758         nfs_file_clear_open_context(filp);
 759         return 0;
 760 }
 761
 762 /*
 763  * This function is called whenever some part of NFS notices that
 764  * the cached attributes have to be refreshed.
 765  */
 766 int
 767 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 768 {
 769         int              status = -ESTALE;
 770         struct nfs_fattr *fattr = NULL;
 771         struct nfs_inode *nfsi = NFS_I(inode);
 772
 773         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
 774                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
 775
 776         if (is_bad_inode(inode))
 777                 goto out;
 778         if (NFS_STALE(inode))
 779                 goto out;
 780
 781         status = -ENOMEM;
 782         fattr = nfs_alloc_fattr();
 783         if (fattr == NULL)
 784                 goto out;
 785
 786         nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
 787         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
 788         if (status != 0) {
 789                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
 790                          inode->i_sb->s_id,
 791                          (long long)NFS_FILEID(inode), status);
 792                 if (status == -ESTALE) {
 793                         nfs_zap_caches(inode);
 794                         if (!S_ISDIR(inode->i_mode))
 795                                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 796                 }
 797                 goto out;
 798         }
 799
 800         status = nfs_refresh_inode(inode, fattr);
 801         if (status) {
 802                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
 803                          inode->i_sb->s_id,
 804                          (long long)NFS_FILEID(inode), status);
 805                 goto out;
 806         }
 807
 808         if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
 809                 nfs_zap_acl_cache(inode);
 810
 811         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
 812                 inode->i_sb->s_id,
 813                 (long long)NFS_FILEID(inode));
 814
 815  out:
 816         nfs_free_fattr(fattr);
 817         return status;
 818 }
 819
 820 int nfs_attribute_timeout(struct inode *inode)
 821 {
 822         struct nfs_inode *nfsi = NFS_I(inode);
 823
 824         return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
 825 }
 826
 827 static int nfs_attribute_cache_expired(struct inode *inode)
 828 {
 829         if (nfs_have_delegated_attributes(inode))
 830                 return 0;
 831         return nfs_attribute_timeout(inode);
 832 }
 833
 834 /**
 835  * nfs_revalidate_inode - Revalidate the inode attributes
 836  * @server - pointer to nfs_server struct
 837  * @inode - pointer to inode struct
 838  *
 839  * Updates inode attribute information by retrieving the data from the server.
 840  */
 841 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 842 {
 843         if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
 844                         && !nfs_attribute_cache_expired(inode))
 845                 return NFS_STALE(inode) ? -ESTALE : 0;
 846         return __nfs_revalidate_inode(server, inode);
 847 }
 848
 849 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
 850 {
 851         struct nfs_inode *nfsi = NFS_I(inode);
 852
 853         if (mapping->nrpages != 0) {
 854                 int ret = invalidate_inode_pages2(mapping);
 855                 if (ret < 0)
 856                         return ret;
 857         }
 858         spin_lock(&inode->i_lock);
 859         nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
 860         if (S_ISDIR(inode->i_mode))
 861                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 862         spin_unlock(&inode->i_lock);
 863         nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
 864         nfs_fscache_reset_inode_cookie(inode);
 865         dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
 866                         inode->i_sb->s_id, (long long)NFS_FILEID(inode));
 867         return 0;
 868 }
 869
 870 /**
 871  * nfs_revalidate_mapping - Revalidate the pagecache
 872  * @inode - pointer to host inode
 873  * @mapping - pointer to mapping
 874  */
 875 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
 876 {
 877         struct nfs_inode *nfsi = NFS_I(inode);
 878         int ret = 0;
 879
 880         if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
 881                         || nfs_attribute_cache_expired(inode)
 882                         || NFS_STALE(inode)) {
 883                 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 884                 if (ret < 0)
 885                         goto out;
 886         }
 887         if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
 888                 ret = nfs_invalidate_mapping(inode, mapping);
 889 out:
 890         return ret;
 891 }
 892
 893 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 894 {
 895         struct nfs_inode *nfsi = NFS_I(inode);
 896         unsigned long ret = 0;
 897
 898         if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
 899                         && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 900                         && inode->i_version == fattr->pre_change_attr) {
 901                 inode->i_version = fattr->change_attr;
 902                 if (S_ISDIR(inode->i_mode))
 903                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 904                 ret |= NFS_INO_INVALID_ATTR;
 905         }
 906         /* If we have atomic WCC data, we may update some attributes */
 907         if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
 908                         && (fattr->valid & NFS_ATTR_FATTR_CTIME)
 909                         && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
 910                 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
 911                 ret |= NFS_INO_INVALID_ATTR;
 912         }
 913
 914         if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
 915                         && (fattr->valid & NFS_ATTR_FATTR_MTIME)
 916                         && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
 917                 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
 918                 if (S_ISDIR(inode->i_mode))
 919                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 920                 ret |= NFS_INO_INVALID_ATTR;
 921         }
 922         if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
 923                         && (fattr->valid & NFS_ATTR_FATTR_SIZE)
 924                         && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
 925                         && nfsi->npages == 0) {
 926                 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
 927                 ret |= NFS_INO_INVALID_ATTR;
 928         }
 929         return ret;
 930 }
 931
 932 /**
 933  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
 934  * @inode - pointer to inode
 935  * @fattr - updated attributes
 936  *
 937  * Verifies the attribute cache. If we have just changed the attributes,
 938  * so that fattr carries weak cache consistency data, then it may
 939  * also update the ctime/mtime/change_attribute.
 940  */
 941 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
 942 {
 943         struct nfs_inode *nfsi = NFS_I(inode);
 944         loff_t cur_size, new_isize;
 945         unsigned long invalid = 0;
 946
 947
 948         /* Has the inode gone and changed behind our back? */
 949         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
 950                 return -EIO;
 951         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
 952                 return -EIO;
 953
 954         if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
 955                         inode->i_version != fattr->change_attr)
 956                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 957
 958         /* Verify a few of the more important attributes */
 959         if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
 960                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 961
 962         if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
 963                 cur_size = i_size_read(inode);
 964                 new_isize = nfs_size_to_loff_t(fattr->size);
 965                 if (cur_size != new_isize && nfsi->npages == 0)
 966                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 967         }
 968
 969         /* Have any file permissions changed? */
 970         if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
 971                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 972         if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
 973                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 974         if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
 975                 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 976
 977         /* Has the link count changed? */
 978         if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
 979                 invalid |= NFS_INO_INVALID_ATTR;
 980
 981         if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
 982                 invalid |= NFS_INO_INVALID_ATIME;
 983
 984         if (invalid != 0)
 985                 nfsi->cache_validity |= invalid;
 986
 987         nfsi->read_cache_jiffies = fattr->time_start;
 988         return 0;
 989 }
 990
 991 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
 992 {
 993         if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
 994                 return 0;
 995         return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
 996 }
 997
 998 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
 999 {
1000         if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1001                 return 0;
1002         return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1003 }
1004
1005 static atomic_long_t nfs_attr_generation_counter;
1006
1007 static unsigned long nfs_read_attr_generation_counter(void)
1008 {
1009         return atomic_long_read(&nfs_attr_generation_counter);
1010 }
1011
1012 unsigned long nfs_inc_attr_generation_counter(void)
1013 {
1014         return atomic_long_inc_return(&nfs_attr_generation_counter);
1015 }
1016
1017 void nfs_fattr_init(struct nfs_fattr *fattr)
1018 {
1019         fattr->valid = 0;
1020         fattr->time_start = jiffies;
1021         fattr->gencount = nfs_inc_attr_generation_counter();
1022 }
1023
1024 struct nfs_fattr *nfs_alloc_fattr(void)
1025 {
1026         struct nfs_fattr *fattr;
1027
1028         fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1029         if (fattr != NULL)
1030                 nfs_fattr_init(fattr);
1031         return fattr;
1032 }
1033
1034 struct nfs_fh *nfs_alloc_fhandle(void)
1035 {
1036         struct nfs_fh *fh;
1037
1038         fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1039         if (fh != NULL)
1040                 fh->size = 0;
1041         return fh;
1042 }
1043
1044 /**
1045  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1046  * @inode - pointer to inode
1047  * @fattr - attributes
1048  *
1049  * Attempt to divine whether or not an RPC call reply carrying stale
1050  * attributes got scheduled after another call carrying updated ones.
1051  *
1052  * To do so, the function first assumes that a more recent ctime means
1053  * that the attributes in fattr are newer, however it also attempt to
1054  * catch the case where ctime either didn't change, or went backwards
1055  * (if someone reset the clock on the server) by looking at whether
1056  * or not this RPC call was started after the inode was last updated.
1057  * Note also the check for wraparound of 'attr_gencount'
1058  *
1059  * The function returns 'true' if it thinks the attributes in 'fattr' are
1060  * more recent than the ones cached in the inode.
1061  *
1062  */
1063 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1064 {
1065         const struct nfs_inode *nfsi = NFS_I(inode);
1066
1067         return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1068                 nfs_ctime_need_update(inode, fattr) ||
1069                 nfs_size_need_update(inode, fattr) ||
1070                 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1071 }
1072
1073 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1074 {
1075         if (nfs_inode_attrs_need_update(inode, fattr))
1076                 return nfs_update_inode(inode, fattr);
1077         return nfs_check_inode_attributes(inode, fattr);
1078 }
1079
1080 /**
1081  * nfs_refresh_inode - try to update the inode attribute cache
1082  * @inode - pointer to inode
1083  * @fattr - updated attributes
1084  *
1085  * Check that an RPC call that returned attributes has not overlapped with
1086  * other recent updates of the inode metadata, then decide whether it is
1087  * safe to do a full update of the inode attributes, or whether just to
1088  * call nfs_check_inode_attributes.
1089  */
1090 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1091 {
1092         int status;
1093
1094         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1095                 return 0;
1096         spin_lock(&inode->i_lock);
1097         status = nfs_refresh_inode_locked(inode, fattr);
1098         spin_unlock(&inode->i_lock);
1099
1100         return status;
1101 }
1102
1103 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1104 {
1105         struct nfs_inode *nfsi = NFS_I(inode);
1106
1107         nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1108         if (S_ISDIR(inode->i_mode))
1109                 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1110         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1111                 return 0;
1112         return nfs_refresh_inode_locked(inode, fattr);
1113 }
1114
1115 /**
1116  * nfs_post_op_update_inode - try to update the inode attribute cache
1117  * @inode - pointer to inode
1118  * @fattr - updated attributes
1119  *
1120  * After an operation that has changed the inode metadata, mark the
1121  * attribute cache as being invalid, then try to update it.
1122  *
1123  * NB: if the server didn't return any post op attributes, this
1124  * function will force the retrieval of attributes before the next
1125  * NFS request.  Thus it should be used only for operations that
1126  * are expected to change one or more attributes, to avoid
1127  * unnecessary NFS requests and trips through nfs_update_inode().
1128  */
1129 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1130 {
1131         int status;
1132
1133         spin_lock(&inode->i_lock);
1134         status = nfs_post_op_update_inode_locked(inode, fattr);
1135         spin_unlock(&inode->i_lock);
1136         return status;
1137 }
1138
1139 /**
1140  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1141  * @inode - pointer to inode
1142  * @fattr - updated attributes
1143  *
1144  * After an operation that has changed the inode metadata, mark the
1145  * attribute cache as being invalid, then try to update it. Fake up
1146  * weak cache consistency data, if none exist.
1147  *
1148  * This function is mainly designed to be used by the ->write_done() functions.
1149  */
1150 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1151 {
1152         int status;
1153
1154         spin_lock(&inode->i_lock);
1155         /* Don't do a WCC update if these attributes are already stale */
1156         if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1157                         !nfs_inode_attrs_need_update(inode, fattr)) {
1158                 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1159                                 | NFS_ATTR_FATTR_PRESIZE
1160                                 | NFS_ATTR_FATTR_PREMTIME
1161                                 | NFS_ATTR_FATTR_PRECTIME);
1162                 goto out_noforce;
1163         }
1164         if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1165                         (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1166                 fattr->pre_change_attr = inode->i_version;
1167                 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1168         }
1169         if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1170                         (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1171                 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1172                 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1173         }
1174         if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1175                         (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1176                 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1177                 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1178         }
1179         if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1180                         (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1181                 fattr->pre_size = i_size_read(inode);
1182                 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1183         }
1184 out_noforce:
1185         status = nfs_post_op_update_inode_locked(inode, fattr);
1186         spin_unlock(&inode->i_lock);
1187         return status;
1188 }
1189
1190 /*
1191  * Many nfs protocol calls return the new file attributes after
1192  * an operation.  Here we update the inode to reflect the state
1193  * of the server's inode.
1194  *
1195  * This is a bit tricky because we have to make sure all dirty pages
1196  * have been sent off to the server before calling invalidate_inode_pages.
1197  * To make sure no other process adds more write requests while we try
1198  * our best to flush them, we make them sleep during the attribute refresh.
1199  *
1200  * A very similar scenario holds for the dir cache.
1201  */
1202 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1203 {
1204         struct nfs_server *server;
1205         struct nfs_inode *nfsi = NFS_I(inode);
1206         loff_t cur_isize, new_isize;
1207         unsigned long invalid = 0;
1208         unsigned long now = jiffies;
1209         unsigned long save_cache_validity;
1210
1211         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1212                         __func__, inode->i_sb->s_id, inode->i_ino,
1213                         atomic_read(&inode->i_count), fattr->valid);
1214
1215         if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1216                 goto out_fileid;
1217
1218         /*
1219          * Make sure the inode's type hasn't changed.
1220          */
1221         if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1222                 goto out_changed;
1223
1224         server = NFS_SERVER(inode);
1225         /* Update the fsid? */
1226         if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1227                         !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1228                         !IS_AUTOMOUNT(inode))
1229                 server->fsid = fattr->fsid;
1230
1231         /*
1232          * Update the read time so we don't revalidate too often.
1233          */
1234         nfsi->read_cache_jiffies = fattr->time_start;
1235
1236         save_cache_validity = nfsi->cache_validity;
1237         nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1238                         | NFS_INO_INVALID_ATIME
1239                         | NFS_INO_REVAL_FORCED
1240                         | NFS_INO_REVAL_PAGECACHE);
1241
1242         /* Do atomic weak cache consistency updates */
1243         invalid |= nfs_wcc_update_inode(inode, fattr);
1244
1245         /* More cache consistency checks */
1246         if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1247                 if (inode->i_version != fattr->change_attr) {
1248                         dprintk("NFS: change_attr change on server for file %s/%ld\n",
1249                                         inode->i_sb->s_id, inode->i_ino);
1250                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1251                         if (S_ISDIR(inode->i_mode))
1252                                 nfs_force_lookup_revalidate(inode);
1253                         inode->i_version = fattr->change_attr;
1254                 }
1255         } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1256                 invalid |= save_cache_validity;
1257
1258         if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1259                 /* NFSv2/v3: Check if the mtime agrees */
1260                 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1261                         dprintk("NFS: mtime change on server for file %s/%ld\n",
1262                                         inode->i_sb->s_id, inode->i_ino);
1263                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1264                         if (S_ISDIR(inode->i_mode))
1265                                 nfs_force_lookup_revalidate(inode);
1266                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1267                 }
1268         } else if (server->caps & NFS_CAP_MTIME)
1269                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1270                                 | NFS_INO_INVALID_DATA
1271                                 | NFS_INO_REVAL_PAGECACHE
1272                                 | NFS_INO_REVAL_FORCED);
1273
1274         if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1275                 /* If ctime has changed we should definitely clear access+acl caches */
1276                 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1277                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1278                         /* and probably clear data for a directory too as utimes can cause
1279                          * havoc with our cache.
1280                          */
1281                         if (S_ISDIR(inode->i_mode)) {
1282                                 invalid |= NFS_INO_INVALID_DATA;
1283                                 nfs_force_lookup_revalidate(inode);
1284                         }
1285                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1286                 }
1287         } else if (server->caps & NFS_CAP_CTIME)
1288                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1289                                 | NFS_INO_INVALID_ACCESS
1290                                 | NFS_INO_INVALID_ACL
1291                                 | NFS_INO_REVAL_FORCED);
1292
1293         /* Check if our cached file size is stale */
1294         if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1295                 new_isize = nfs_size_to_loff_t(fattr->size);
1296                 cur_isize = i_size_read(inode);
1297                 if (new_isize != cur_isize) {
1298                         /* Do we perhaps have any outstanding writes, or has
1299                          * the file grown beyond our last write? */
1300                         if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1301                              new_isize > cur_isize) {
1302                                 i_size_write(inode, new_isize);
1303                                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1304                         }
1305                         dprintk("NFS: isize change on server for file %s/%ld "
1306                                         "(%Ld to %Ld)\n",
1307                                         inode->i_sb->s_id,
1308                                         inode->i_ino,
1309                                         (long long)cur_isize,
1310                                         (long long)new_isize);
1311                 }
1312         } else
1313                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1314                                 | NFS_INO_REVAL_PAGECACHE
1315                                 | NFS_INO_REVAL_FORCED);
1316
1317
1318         if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1319                 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1320         else if (server->caps & NFS_CAP_ATIME)
1321                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1322                                 | NFS_INO_REVAL_FORCED);
1323
1324         if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1325                 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1326                         umode_t newmode = inode->i_mode & S_IFMT;
1327                         newmode |= fattr->mode & S_IALLUGO;
1328                         inode->i_mode = newmode;
1329                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1330                 }
1331         } else if (server->caps & NFS_CAP_MODE)
1332                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1333                                 | NFS_INO_INVALID_ACCESS
1334                                 | NFS_INO_INVALID_ACL
1335                                 | NFS_INO_REVAL_FORCED);
1336
1337         if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1338                 if (inode->i_uid != fattr->uid) {
1339                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1340                         inode->i_uid = fattr->uid;
1341                 }
1342         } else if (server->caps & NFS_CAP_OWNER)
1343                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1344                                 | NFS_INO_INVALID_ACCESS
1345                                 | NFS_INO_INVALID_ACL
1346                                 | NFS_INO_REVAL_FORCED);
1347
1348         if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1349                 if (inode->i_gid != fattr->gid) {
1350                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1351                         inode->i_gid = fattr->gid;
1352                 }
1353         } else if (server->caps & NFS_CAP_OWNER_GROUP)
1354                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1355                                 | NFS_INO_INVALID_ACCESS
1356                                 | NFS_INO_INVALID_ACL
1357                                 | NFS_INO_REVAL_FORCED);
1358
1359         if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1360                 if (inode->i_nlink != fattr->nlink) {
1361                         invalid |= NFS_INO_INVALID_ATTR;
1362                         if (S_ISDIR(inode->i_mode))
1363                                 invalid |= NFS_INO_INVALID_DATA;
1364                         set_nlink(inode, fattr->nlink);
1365                 }
1366         } else if (server->caps & NFS_CAP_NLINK)
1367                 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1368                                 | NFS_INO_REVAL_FORCED);
1369
1370         if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1371                 /*
1372                  * report the blocks in 512byte units
1373                  */
1374                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1375         }
1376         if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1377                 inode->i_blocks = fattr->du.nfs2.blocks;
1378
1379         /* Update attrtimeo value if we're out of the unstable period */
1380         if (invalid & NFS_INO_INVALID_ATTR) {
1381                 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1382                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1383                 nfsi->attrtimeo_timestamp = now;
1384                 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1385         } else {
1386                 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1387                         if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1388                                 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1389                         nfsi->attrtimeo_timestamp = now;
1390                 }
1391         }
1392         invalid &= ~NFS_INO_INVALID_ATTR;
1393         /* Don't invalidate the data if we were to blame */
1394         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1395                                 || S_ISLNK(inode->i_mode)))
1396                 invalid &= ~NFS_INO_INVALID_DATA;
1397         if (!nfs_have_delegation(inode, FMODE_READ) ||
1398                         (save_cache_validity & NFS_INO_REVAL_FORCED))
1399                 nfsi->cache_validity |= invalid;
1400
1401         return 0;
1402  out_changed:
1403         /*
1404          * Big trouble! The inode has become a different object.
1405          */
1406         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1407                         __func__, inode->i_ino, inode->i_mode, fattr->mode);
1408  out_err:
1409         /*
1410          * No need to worry about unhashing the dentry, as the
1411          * lookup validation will know that the inode is bad.
1412          * (But we fall through to invalidate the caches.)
1413          */
1414         nfs_invalidate_inode(inode);
1415         return -ESTALE;
1416
1417  out_fileid:
1418         printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1419                 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1420                 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1421                 (long long)nfsi->fileid, (long long)fattr->fileid);
1422         goto out_err;
1423 }
1424
1425
1426 #ifdef CONFIG_NFS_V4
1427
1428 /*
1429  * Clean out any remaining NFSv4 state that might be left over due
1430  * to open() calls that passed nfs_atomic_lookup, but failed to call
1431  * nfs_open().
1432  */
1433 void nfs4_evict_inode(struct inode *inode)
1434 {
1435         truncate_inode_pages(&inode->i_data, 0);
1436         end_writeback(inode);
1437         pnfs_return_layout(inode);
1438         pnfs_destroy_layout(NFS_I(inode));
1439         /* If we are holding a delegation, return it! */
1440         nfs_inode_return_delegation_noreclaim(inode);
1441         /* First call standard NFS clear_inode() code */
1442         nfs_clear_inode(inode);
1443 }
1444 #endif
1445
1446 struct inode *nfs_alloc_inode(struct super_block *sb)
1447 {
1448         struct nfs_inode *nfsi;
1449         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1450         if (!nfsi)
1451                 return NULL;
1452         nfsi->flags = 0UL;
1453         nfsi->cache_validity = 0UL;
1454 #ifdef CONFIG_NFS_V3_ACL
1455         nfsi->acl_access = ERR_PTR(-EAGAIN);
1456         nfsi->acl_default = ERR_PTR(-EAGAIN);
1457 #endif
1458 #ifdef CONFIG_NFS_V4
1459         nfsi->nfs4_acl = NULL;
1460 #endif /* CONFIG_NFS_V4 */
1461         return &nfsi->vfs_inode;
1462 }
1463
1464 static void nfs_i_callback(struct rcu_head *head)
1465 {
1466         struct inode *inode = container_of(head, struct inode, i_rcu);
1467         INIT_LIST_HEAD(&inode->i_dentry);
1468         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1469 }
1470
1471 void nfs_destroy_inode(struct inode *inode)
1472 {
1473         call_rcu(&inode->i_rcu, nfs_i_callback);
1474 }
1475
1476 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1477 {
1478 #ifdef CONFIG_NFS_V4
1479         INIT_LIST_HEAD(&nfsi->open_states);
1480         nfsi->delegation = NULL;
1481         nfsi->delegation_state = 0;
1482         init_rwsem(&nfsi->rwsem);
1483         nfsi->layout = NULL;
1484         atomic_set(&nfsi->commits_outstanding, 0);
1485 #endif
1486 }
1487
1488 static void init_once(void *foo)
1489 {
1490         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1491
1492         inode_init_once(&nfsi->vfs_inode);
1493         INIT_LIST_HEAD(&nfsi->open_files);
1494         INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1495         INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1496         INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1497         nfsi->npages = 0;
1498         nfsi->ncommit = 0;
1499         atomic_set(&nfsi->silly_count, 1);
1500         INIT_HLIST_HEAD(&nfsi->silly_list);
1501         init_waitqueue_head(&nfsi->waitqueue);
1502         nfs4_init_once(nfsi);
1503 }
1504
1505 static int __init nfs_init_inodecache(void)
1506 {
1507         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1508                                              sizeof(struct nfs_inode),
1509                                              0, (SLAB_RECLAIM_ACCOUNT|
1510                                                 SLAB_MEM_SPREAD),
1511                                              init_once);
1512         if (nfs_inode_cachep == NULL)
1513                 return -ENOMEM;
1514
1515         return 0;
1516 }
1517
1518 static void nfs_destroy_inodecache(void)
1519 {
1520         kmem_cache_destroy(nfs_inode_cachep);
1521 }
1522
1523 struct workqueue_struct *nfsiod_workqueue;
1524
1525 /*
1526  * start up the nfsiod workqueue
1527  */
1528 static int nfsiod_start(void)
1529 {
1530         struct workqueue_struct *wq;
1531         dprintk("RPC:       creating workqueue nfsiod\n");
1532         wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1533         if (wq == NULL)
1534                 return -ENOMEM;
1535         nfsiod_workqueue = wq;
1536         return 0;
1537 }
1538
1539 /*
1540  * Destroy the nfsiod workqueue
1541  */
1542 static void nfsiod_stop(void)
1543 {
1544         struct workqueue_struct *wq;
1545
1546         wq = nfsiod_workqueue;
1547         if (wq == NULL)
1548                 return;
1549         nfsiod_workqueue = NULL;
1550         destroy_workqueue(wq);
1551 }
1552
1553 /*
1554  * Initialize NFS
1555  */
1556 static int __init init_nfs_fs(void)
1557 {
1558         int err;
1559
1560         err = nfs_idmap_init();
1561         if (err < 0)
1562                 goto out9;
1563
1564         err = nfs_dns_resolver_init();
1565         if (err < 0)
1566                 goto out8;
1567
1568         err = nfs_fscache_register();
1569         if (err < 0)
1570                 goto out7;
1571
1572         err = nfsiod_start();
1573         if (err)
1574                 goto out6;
1575
1576         err = nfs_fs_proc_init();
1577         if (err)
1578                 goto out5;
1579
1580         err = nfs_init_nfspagecache();
1581         if (err)
1582                 goto out4;
1583
1584         err = nfs_init_inodecache();
1585         if (err)
1586                 goto out3;
1587
1588         err = nfs_init_readpagecache();
1589         if (err)
1590                 goto out2;
1591
1592         err = nfs_init_writepagecache();
1593         if (err)
1594                 goto out1;
1595
1596         err = nfs_init_directcache();
1597         if (err)
1598                 goto out0;
1599
1600 #ifdef CONFIG_PROC_FS
1601         rpc_proc_register(&nfs_rpcstat);
1602 #endif
1603         if ((err = register_nfs_fs()) != 0)
1604                 goto out;
1605         return 0;
1606 out:
1607 #ifdef CONFIG_PROC_FS
1608         rpc_proc_unregister("nfs");
1609 #endif
1610         nfs_destroy_directcache();
1611 out0:
1612         nfs_destroy_writepagecache();
1613 out1:
1614         nfs_destroy_readpagecache();
1615 out2:
1616         nfs_destroy_inodecache();
1617 out3:
1618         nfs_destroy_nfspagecache();
1619 out4:
1620         nfs_fs_proc_exit();
1621 out5:
1622         nfsiod_stop();
1623 out6:
1624         nfs_fscache_unregister();
1625 out7:
1626         nfs_dns_resolver_destroy();
1627 out8:
1628         nfs_idmap_quit();
1629 out9:
1630         return err;
1631 }
1632
1633 static void __exit exit_nfs_fs(void)
1634 {
1635         nfs_destroy_directcache();
1636         nfs_destroy_writepagecache();
1637         nfs_destroy_readpagecache();
1638         nfs_destroy_inodecache();
1639         nfs_destroy_nfspagecache();
1640         nfs_fscache_unregister();
1641         nfs_dns_resolver_destroy();
1642         nfs_idmap_quit();
1643 #ifdef CONFIG_PROC_FS
1644         rpc_proc_unregister("nfs");
1645 #endif
1646         nfs_cleanup_cb_ident_idr();
1647         unregister_nfs_fs();
1648         nfs_fs_proc_exit();
1649         nfsiod_stop();
1650 }
1651
1652 /* Not quite true; I just maintain it */
1653 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1654 MODULE_LICENSE("GPL");
1655 module_param(enable_ino64, bool, 0644);
1656
1657 module_init(init_nfs_fs)
1658 module_exit(exit_nfs_fs)