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