linux/fs/nfs/inode.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/nfs/inode.c
   4 *
   5 *  Copyright (C) 1992  Rick Sladkey
   6 *
   7 *  nfs inode and superblock handling functions
   8 *
   9 *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
  10 *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
  11 *
  12 *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
  13 *  J.S.Peatfield@damtp.cam.ac.uk
  14 *
  15 */
  16
  17#include <linux/module.h>
  18#include <linux/init.h>
  19#include <linux/sched/signal.h>
  20#include <linux/time.h>
  21#include <linux/kernel.h>
  22#include <linux/mm.h>
  23#include <linux/string.h>
  24#include <linux/stat.h>
  25#include <linux/errno.h>
  26#include <linux/unistd.h>
  27#include <linux/sunrpc/clnt.h>
  28#include <linux/sunrpc/stats.h>
  29#include <linux/sunrpc/metrics.h>
  30#include <linux/nfs_fs.h>
  31#include <linux/nfs_mount.h>
  32#include <linux/nfs4_mount.h>
  33#include <linux/lockd/bind.h>
  34#include <linux/seq_file.h>
  35#include <linux/mount.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#include <linux/freezer.h>
  42#include <linux/uaccess.h>
  43#include <linux/iversion.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 "pnfs.h"
  52#include "nfs.h"
  53#include "netns.h"
  54#include "sysfs.h"
  55
  56#include "nfstrace.h"
  57
  58#define NFSDBG_FACILITY         NFSDBG_VFS
  59
  60#define NFS_64_BIT_INODE_NUMBERS_ENABLED        1
  61
  62/* Default is to see 64-bit inode numbers */
  63static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
  64
  65static int nfs_update_inode(struct inode *, struct nfs_fattr *);
  66
  67static struct kmem_cache * nfs_inode_cachep;
  68
  69static inline unsigned long
  70nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
  71{
  72        return nfs_fileid_to_ino_t(fattr->fileid);
  73}
  74
  75static int nfs_wait_killable(int mode)
  76{
  77        freezable_schedule_unsafe();
  78        if (signal_pending_state(mode, current))
  79                return -ERESTARTSYS;
  80        return 0;
  81}
  82
  83int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
  84{
  85        return nfs_wait_killable(mode);
  86}
  87EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
  88
  89/**
  90 * nfs_compat_user_ino64 - returns the user-visible inode number
  91 * @fileid: 64-bit fileid
  92 *
  93 * This function returns a 32-bit inode number if the boot parameter
  94 * nfs.enable_ino64 is zero.
  95 */
  96u64 nfs_compat_user_ino64(u64 fileid)
  97{
  98#ifdef CONFIG_COMPAT
  99        compat_ulong_t ino;
 100#else   
 101        unsigned long ino;
 102#endif
 103
 104        if (enable_ino64)
 105                return fileid;
 106        ino = fileid;
 107        if (sizeof(ino) < sizeof(fileid))
 108                ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
 109        return ino;
 110}
 111
 112int nfs_drop_inode(struct inode *inode)
 113{
 114        return NFS_STALE(inode) || generic_drop_inode(inode);
 115}
 116EXPORT_SYMBOL_GPL(nfs_drop_inode);
 117
 118void nfs_clear_inode(struct inode *inode)
 119{
 120        /*
 121         * The following should never happen...
 122         */
 123        WARN_ON_ONCE(nfs_have_writebacks(inode));
 124        WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
 125        nfs_zap_acl_cache(inode);
 126        nfs_access_zap_cache(inode);
 127        nfs_fscache_clear_inode(inode);
 128}
 129EXPORT_SYMBOL_GPL(nfs_clear_inode);
 130
 131void nfs_evict_inode(struct inode *inode)
 132{
 133        truncate_inode_pages_final(&inode->i_data);
 134        clear_inode(inode);
 135        nfs_clear_inode(inode);
 136}
 137
 138int nfs_sync_inode(struct inode *inode)
 139{
 140        inode_dio_wait(inode);
 141        return nfs_wb_all(inode);
 142}
 143EXPORT_SYMBOL_GPL(nfs_sync_inode);
 144
 145/**
 146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
 147 * @mapping: pointer to struct address_space
 148 */
 149int nfs_sync_mapping(struct address_space *mapping)
 150{
 151        int ret = 0;
 152
 153        if (mapping->nrpages != 0) {
 154                unmap_mapping_range(mapping, 0, 0, 0);
 155                ret = nfs_wb_all(mapping->host);
 156        }
 157        return ret;
 158}
 159
 160static int nfs_attribute_timeout(struct inode *inode)
 161{
 162        struct nfs_inode *nfsi = NFS_I(inode);
 163
 164        return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
 165}
 166
 167static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
 168{
 169        unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
 170
 171        /* Special case for the pagecache or access cache */
 172        if (flags == NFS_INO_REVAL_PAGECACHE &&
 173            !(cache_validity & NFS_INO_REVAL_FORCED))
 174                return false;
 175        return (cache_validity & flags) != 0;
 176}
 177
 178static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
 179{
 180        unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
 181
 182        if ((cache_validity & flags) != 0)
 183                return true;
 184        if (nfs_attribute_timeout(inode))
 185                return true;
 186        return false;
 187}
 188
 189bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
 190{
 191        if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
 192                return nfs_check_cache_invalid_delegated(inode, flags);
 193
 194        return nfs_check_cache_invalid_not_delegated(inode, flags);
 195}
 196EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
 197
 198static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
 199{
 200        struct nfs_inode *nfsi = NFS_I(inode);
 201        bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
 202
 203        if (have_delegation) {
 204                if (!(flags & NFS_INO_REVAL_FORCED))
 205                        flags &= ~NFS_INO_INVALID_OTHER;
 206                flags &= ~(NFS_INO_INVALID_CHANGE
 207                                | NFS_INO_INVALID_SIZE
 208                                | NFS_INO_REVAL_PAGECACHE
 209                                | NFS_INO_INVALID_XATTR);
 210        }
 211
 212        if (inode->i_mapping->nrpages == 0)
 213                flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
 214        nfsi->cache_validity |= flags;
 215        if (flags & NFS_INO_INVALID_DATA)
 216                nfs_fscache_invalidate(inode);
 217}
 218
 219/*
 220 * Invalidate the local caches
 221 */
 222static void nfs_zap_caches_locked(struct inode *inode)
 223{
 224        struct nfs_inode *nfsi = NFS_I(inode);
 225        int mode = inode->i_mode;
 226
 227        nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
 228
 229        nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 230        nfsi->attrtimeo_timestamp = jiffies;
 231
 232        memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
 233        if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 234                nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
 235                                        | NFS_INO_INVALID_DATA
 236                                        | NFS_INO_INVALID_ACCESS
 237                                        | NFS_INO_INVALID_ACL
 238                                        | NFS_INO_INVALID_XATTR
 239                                        | NFS_INO_REVAL_PAGECACHE);
 240        } else
 241                nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
 242                                        | NFS_INO_INVALID_ACCESS
 243                                        | NFS_INO_INVALID_ACL
 244                                        | NFS_INO_INVALID_XATTR
 245                                        | NFS_INO_REVAL_PAGECACHE);
 246        nfs_zap_label_cache_locked(nfsi);
 247}
 248
 249void nfs_zap_caches(struct inode *inode)
 250{
 251        spin_lock(&inode->i_lock);
 252        nfs_zap_caches_locked(inode);
 253        spin_unlock(&inode->i_lock);
 254}
 255
 256void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
 257{
 258        if (mapping->nrpages != 0) {
 259                spin_lock(&inode->i_lock);
 260                nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
 261                spin_unlock(&inode->i_lock);
 262        }
 263}
 264
 265void nfs_zap_acl_cache(struct inode *inode)
 266{
 267        void (*clear_acl_cache)(struct inode *);
 268
 269        clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
 270        if (clear_acl_cache != NULL)
 271                clear_acl_cache(inode);
 272        spin_lock(&inode->i_lock);
 273        NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
 274        spin_unlock(&inode->i_lock);
 275}
 276EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
 277
 278void nfs_invalidate_atime(struct inode *inode)
 279{
 280        spin_lock(&inode->i_lock);
 281        nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
 282        spin_unlock(&inode->i_lock);
 283}
 284EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
 285
 286/*
 287 * Invalidate, but do not unhash, the inode.
 288 * NB: must be called with inode->i_lock held!
 289 */
 290static void nfs_set_inode_stale_locked(struct inode *inode)
 291{
 292        set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
 293        nfs_zap_caches_locked(inode);
 294        trace_nfs_set_inode_stale(inode);
 295}
 296
 297void nfs_set_inode_stale(struct inode *inode)
 298{
 299        spin_lock(&inode->i_lock);
 300        nfs_set_inode_stale_locked(inode);
 301        spin_unlock(&inode->i_lock);
 302}
 303
 304struct nfs_find_desc {
 305        struct nfs_fh           *fh;
 306        struct nfs_fattr        *fattr;
 307};
 308
 309/*
 310 * In NFSv3 we can have 64bit inode numbers. In order to support
 311 * this, and re-exported directories (also seen in NFSv2)
 312 * we are forced to allow 2 different inodes to have the same
 313 * i_ino.
 314 */
 315static int
 316nfs_find_actor(struct inode *inode, void *opaque)
 317{
 318        struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
 319        struct nfs_fh           *fh = desc->fh;
 320        struct nfs_fattr        *fattr = desc->fattr;
 321
 322        if (NFS_FILEID(inode) != fattr->fileid)
 323                return 0;
 324        if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
 325                return 0;
 326        if (nfs_compare_fh(NFS_FH(inode), fh))
 327                return 0;
 328        if (is_bad_inode(inode) || NFS_STALE(inode))
 329                return 0;
 330        return 1;
 331}
 332
 333static int
 334nfs_init_locked(struct inode *inode, void *opaque)
 335{
 336        struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
 337        struct nfs_fattr        *fattr = desc->fattr;
 338
 339        set_nfs_fileid(inode, fattr->fileid);
 340        inode->i_mode = fattr->mode;
 341        nfs_copy_fh(NFS_FH(inode), desc->fh);
 342        return 0;
 343}
 344
 345#ifdef CONFIG_NFS_V4_SECURITY_LABEL
 346static void nfs_clear_label_invalid(struct inode *inode)
 347{
 348        spin_lock(&inode->i_lock);
 349        NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
 350        spin_unlock(&inode->i_lock);
 351}
 352
 353void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
 354                                        struct nfs4_label *label)
 355{
 356        int error;
 357
 358        if (label == NULL)
 359                return;
 360
 361        if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
 362                error = security_inode_notifysecctx(inode, label->label,
 363                                label->len);
 364                if (error)
 365                        printk(KERN_ERR "%s() %s %d "
 366                                        "security_inode_notifysecctx() %d\n",
 367                                        __func__,
 368                                        (char *)label->label,
 369                                        label->len, error);
 370                nfs_clear_label_invalid(inode);
 371        }
 372}
 373
 374struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
 375{
 376        struct nfs4_label *label = NULL;
 377        int minor_version = server->nfs_client->cl_minorversion;
 378
 379        if (minor_version < 2)
 380                return label;
 381
 382        if (!(server->caps & NFS_CAP_SECURITY_LABEL))
 383                return label;
 384
 385        label = kzalloc(sizeof(struct nfs4_label), flags);
 386        if (label == NULL)
 387                return ERR_PTR(-ENOMEM);
 388
 389        label->label = kzalloc(NFS4_MAXLABELLEN, flags);
 390        if (label->label == NULL) {
 391                kfree(label);
 392                return ERR_PTR(-ENOMEM);
 393        }
 394        label->len = NFS4_MAXLABELLEN;
 395
 396        return label;
 397}
 398EXPORT_SYMBOL_GPL(nfs4_label_alloc);
 399#else
 400void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
 401                                        struct nfs4_label *label)
 402{
 403}
 404#endif
 405EXPORT_SYMBOL_GPL(nfs_setsecurity);
 406
 407/* Search for inode identified by fh, fileid and i_mode in inode cache. */
 408struct inode *
 409nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
 410{
 411        struct nfs_find_desc desc = {
 412                .fh     = fh,
 413                .fattr  = fattr,
 414        };
 415        struct inode *inode;
 416        unsigned long hash;
 417
 418        if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
 419            !(fattr->valid & NFS_ATTR_FATTR_TYPE))
 420                return NULL;
 421
 422        hash = nfs_fattr_to_ino_t(fattr);
 423        inode = ilookup5(sb, hash, nfs_find_actor, &desc);
 424
 425        dprintk("%s: returning %p\n", __func__, inode);
 426        return inode;
 427}
 428
 429/*
 430 * This is our front-end to iget that looks up inodes by file handle
 431 * instead of inode number.
 432 */
 433struct inode *
 434nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
 435{
 436        struct nfs_find_desc desc = {
 437                .fh     = fh,
 438                .fattr  = fattr
 439        };
 440        struct inode *inode = ERR_PTR(-ENOENT);
 441        unsigned long hash;
 442
 443        nfs_attr_check_mountpoint(sb, fattr);
 444
 445        if (nfs_attr_use_mounted_on_fileid(fattr))
 446                fattr->fileid = fattr->mounted_on_fileid;
 447        else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
 448                goto out_no_inode;
 449        if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
 450                goto out_no_inode;
 451
 452        hash = nfs_fattr_to_ino_t(fattr);
 453
 454        inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
 455        if (inode == NULL) {
 456                inode = ERR_PTR(-ENOMEM);
 457                goto out_no_inode;
 458        }
 459
 460        if (inode->i_state & I_NEW) {
 461                struct nfs_inode *nfsi = NFS_I(inode);
 462                unsigned long now = jiffies;
 463
 464                /* We set i_ino for the few things that still rely on it,
 465                 * such as stat(2) */
 466                inode->i_ino = hash;
 467
 468                /* We can't support update_atime(), since the server will reset it */
 469                inode->i_flags |= S_NOATIME|S_NOCMTIME;
 470                inode->i_mode = fattr->mode;
 471                nfsi->cache_validity = 0;
 472                if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
 473                                && nfs_server_capable(inode, NFS_CAP_MODE))
 474                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
 475                /* Why so? Because we want revalidate for devices/FIFOs, and
 476                 * that's precisely what we have in nfs_file_inode_operations.
 477                 */
 478                inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
 479                if (S_ISREG(inode->i_mode)) {
 480                        inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
 481                        inode->i_data.a_ops = &nfs_file_aops;
 482                } else if (S_ISDIR(inode->i_mode)) {
 483                        inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
 484                        inode->i_fop = &nfs_dir_operations;
 485                        inode->i_data.a_ops = &nfs_dir_aops;
 486                        /* Deal with crossing mountpoints */
 487                        if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
 488                                        fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
 489                                if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
 490                                        inode->i_op = &nfs_referral_inode_operations;
 491                                else
 492                                        inode->i_op = &nfs_mountpoint_inode_operations;
 493                                inode->i_fop = NULL;
 494                                inode->i_flags |= S_AUTOMOUNT;
 495                        }
 496                } else if (S_ISLNK(inode->i_mode)) {
 497                        inode->i_op = &nfs_symlink_inode_operations;
 498                        inode_nohighmem(inode);
 499                } else
 500                        init_special_inode(inode, inode->i_mode, fattr->rdev);
 501
 502                memset(&inode->i_atime, 0, sizeof(inode->i_atime));
 503                memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
 504                memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
 505                inode_set_iversion_raw(inode, 0);
 506                inode->i_size = 0;
 507                clear_nlink(inode);
 508                inode->i_uid = make_kuid(&init_user_ns, -2);
 509                inode->i_gid = make_kgid(&init_user_ns, -2);
 510                inode->i_blocks = 0;
 511                memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
 512                nfsi->write_io = 0;
 513                nfsi->read_io = 0;
 514
 515                nfsi->read_cache_jiffies = fattr->time_start;
 516                nfsi->attr_gencount = fattr->gencount;
 517                if (fattr->valid & NFS_ATTR_FATTR_ATIME)
 518                        inode->i_atime = fattr->atime;
 519                else if (nfs_server_capable(inode, NFS_CAP_ATIME))
 520                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
 521                if (fattr->valid & NFS_ATTR_FATTR_MTIME)
 522                        inode->i_mtime = fattr->mtime;
 523                else if (nfs_server_capable(inode, NFS_CAP_MTIME))
 524                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
 525                if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 526                        inode->i_ctime = fattr->ctime;
 527                else if (nfs_server_capable(inode, NFS_CAP_CTIME))
 528                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
 529                if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
 530                        inode_set_iversion_raw(inode, fattr->change_attr);
 531                else
 532                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
 533                if (fattr->valid & NFS_ATTR_FATTR_SIZE)
 534                        inode->i_size = nfs_size_to_loff_t(fattr->size);
 535                else
 536                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
 537                if (fattr->valid & NFS_ATTR_FATTR_NLINK)
 538                        set_nlink(inode, fattr->nlink);
 539                else if (nfs_server_capable(inode, NFS_CAP_NLINK))
 540                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
 541                if (fattr->valid & NFS_ATTR_FATTR_OWNER)
 542                        inode->i_uid = fattr->uid;
 543                else if (nfs_server_capable(inode, NFS_CAP_OWNER))
 544                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
 545                if (fattr->valid & NFS_ATTR_FATTR_GROUP)
 546                        inode->i_gid = fattr->gid;
 547                else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
 548                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
 549                if (nfs_server_capable(inode, NFS_CAP_XATTR))
 550                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
 551                if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
 552                        inode->i_blocks = fattr->du.nfs2.blocks;
 553                if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
 554                        /*
 555                         * report the blocks in 512byte units
 556                         */
 557                        inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
 558                }
 559
 560                if (nfsi->cache_validity != 0)
 561                        nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
 562
 563                nfs_setsecurity(inode, fattr, label);
 564
 565                nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 566                nfsi->attrtimeo_timestamp = now;
 567                nfsi->access_cache = RB_ROOT;
 568
 569                nfs_fscache_init_inode(inode);
 570
 571                unlock_new_inode(inode);
 572        } else {
 573                int err = nfs_refresh_inode(inode, fattr);
 574                if (err < 0) {
 575                        iput(inode);
 576                        inode = ERR_PTR(err);
 577                        goto out_no_inode;
 578                }
 579        }
 580        dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
 581                inode->i_sb->s_id,
 582                (unsigned long long)NFS_FILEID(inode),
 583                nfs_display_fhandle_hash(fh),
 584                atomic_read(&inode->i_count));
 585
 586out:
 587        return inode;
 588
 589out_no_inode:
 590        dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
 591        goto out;
 592}
 593EXPORT_SYMBOL_GPL(nfs_fhget);
 594
 595#define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
 596
 597int
 598nfs_setattr(struct dentry *dentry, struct iattr *attr)
 599{
 600        struct inode *inode = d_inode(dentry);
 601        struct nfs_fattr *fattr;
 602        int error = 0;
 603
 604        nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
 605
 606        /* skip mode change if it's just for clearing setuid/setgid */
 607        if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
 608                attr->ia_valid &= ~ATTR_MODE;
 609
 610        if (attr->ia_valid & ATTR_SIZE) {
 611                BUG_ON(!S_ISREG(inode->i_mode));
 612
 613                error = inode_newsize_ok(inode, attr->ia_size);
 614                if (error)
 615                        return error;
 616
 617                if (attr->ia_size == i_size_read(inode))
 618                        attr->ia_valid &= ~ATTR_SIZE;
 619        }
 620
 621        /* Optimization: if the end result is no change, don't RPC */
 622        attr->ia_valid &= NFS_VALID_ATTRS;
 623        if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
 624                return 0;
 625
 626        trace_nfs_setattr_enter(inode);
 627
 628        /* Write all dirty data */
 629        if (S_ISREG(inode->i_mode))
 630                nfs_sync_inode(inode);
 631
 632        fattr = nfs_alloc_fattr();
 633        if (fattr == NULL) {
 634                error = -ENOMEM;
 635                goto out;
 636        }
 637
 638        error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
 639        if (error == 0)
 640                error = nfs_refresh_inode(inode, fattr);
 641        nfs_free_fattr(fattr);
 642out:
 643        trace_nfs_setattr_exit(inode, error);
 644        return error;
 645}
 646EXPORT_SYMBOL_GPL(nfs_setattr);
 647
 648/**
 649 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
 650 * @inode: inode of the file used
 651 * @offset: file offset to start truncating
 652 *
 653 * This is a copy of the common vmtruncate, but with the locking
 654 * corrected to take into account the fact that NFS requires
 655 * inode->i_size to be updated under the inode->i_lock.
 656 * Note: must be called with inode->i_lock held!
 657 */
 658static int nfs_vmtruncate(struct inode * inode, loff_t offset)
 659{
 660        int err;
 661
 662        err = inode_newsize_ok(inode, offset);
 663        if (err)
 664                goto out;
 665
 666        i_size_write(inode, offset);
 667        /* Optimisation */
 668        if (offset == 0)
 669                NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
 670                                NFS_INO_DATA_INVAL_DEFER);
 671        NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
 672
 673        spin_unlock(&inode->i_lock);
 674        truncate_pagecache(inode, offset);
 675        spin_lock(&inode->i_lock);
 676out:
 677        return err;
 678}
 679
 680/**
 681 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
 682 * @inode: pointer to struct inode
 683 * @attr: pointer to struct iattr
 684 * @fattr: pointer to struct nfs_fattr
 685 *
 686 * Note: we do this in the *proc.c in order to ensure that
 687 *       it works for things like exclusive creates too.
 688 */
 689void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
 690                struct nfs_fattr *fattr)
 691{
 692        /* Barrier: bump the attribute generation count. */
 693        nfs_fattr_set_barrier(fattr);
 694
 695        spin_lock(&inode->i_lock);
 696        NFS_I(inode)->attr_gencount = fattr->gencount;
 697        if ((attr->ia_valid & ATTR_SIZE) != 0) {
 698                nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
 699                nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
 700                nfs_vmtruncate(inode, attr->ia_size);
 701        }
 702        if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
 703                NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
 704                if ((attr->ia_valid & ATTR_MODE) != 0) {
 705                        int mode = attr->ia_mode & S_IALLUGO;
 706                        mode |= inode->i_mode & ~S_IALLUGO;
 707                        inode->i_mode = mode;
 708                }
 709                if ((attr->ia_valid & ATTR_UID) != 0)
 710                        inode->i_uid = attr->ia_uid;
 711                if ((attr->ia_valid & ATTR_GID) != 0)
 712                        inode->i_gid = attr->ia_gid;
 713                if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 714                        inode->i_ctime = fattr->ctime;
 715                else
 716                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
 717                                        | NFS_INO_INVALID_CTIME);
 718                nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
 719                                | NFS_INO_INVALID_ACL);
 720        }
 721        if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
 722                NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
 723                                | NFS_INO_INVALID_CTIME);
 724                if (fattr->valid & NFS_ATTR_FATTR_ATIME)
 725                        inode->i_atime = fattr->atime;
 726                else if (attr->ia_valid & ATTR_ATIME_SET)
 727                        inode->i_atime = attr->ia_atime;
 728                else
 729                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
 730
 731                if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 732                        inode->i_ctime = fattr->ctime;
 733                else
 734                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
 735                                        | NFS_INO_INVALID_CTIME);
 736        }
 737        if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
 738                NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
 739                                | NFS_INO_INVALID_CTIME);
 740                if (fattr->valid & NFS_ATTR_FATTR_MTIME)
 741                        inode->i_mtime = fattr->mtime;
 742                else if (attr->ia_valid & ATTR_MTIME_SET)
 743                        inode->i_mtime = attr->ia_mtime;
 744                else
 745                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
 746
 747                if (fattr->valid & NFS_ATTR_FATTR_CTIME)
 748                        inode->i_ctime = fattr->ctime;
 749                else
 750                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
 751                                        | NFS_INO_INVALID_CTIME);
 752        }
 753        if (fattr->valid)
 754                nfs_update_inode(inode, fattr);
 755        spin_unlock(&inode->i_lock);
 756}
 757EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
 758
 759static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
 760{
 761        struct dentry *parent;
 762
 763        if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
 764                return;
 765        parent = dget_parent(dentry);
 766        nfs_force_use_readdirplus(d_inode(parent));
 767        dput(parent);
 768}
 769
 770static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
 771{
 772        struct dentry *parent;
 773
 774        if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
 775                return;
 776        parent = dget_parent(dentry);
 777        nfs_advise_use_readdirplus(d_inode(parent));
 778        dput(parent);
 779}
 780
 781static bool nfs_need_revalidate_inode(struct inode *inode)
 782{
 783        if (NFS_I(inode)->cache_validity &
 784                        (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
 785                return true;
 786        if (nfs_attribute_cache_expired(inode))
 787                return true;
 788        return false;
 789}
 790
 791int nfs_getattr(const struct path *path, struct kstat *stat,
 792                u32 request_mask, unsigned int query_flags)
 793{
 794        struct inode *inode = d_inode(path->dentry);
 795        struct nfs_server *server = NFS_SERVER(inode);
 796        unsigned long cache_validity;
 797        int err = 0;
 798        bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
 799        bool do_update = false;
 800
 801        trace_nfs_getattr_enter(inode);
 802
 803        if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
 804                nfs_readdirplus_parent_cache_hit(path->dentry);
 805                goto out_no_update;
 806        }
 807
 808        /* Flush out writes to the server in order to update c/mtime.  */
 809        if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
 810                        S_ISREG(inode->i_mode)) {
 811                err = filemap_write_and_wait(inode->i_mapping);
 812                if (err)
 813                        goto out;
 814        }
 815
 816        /*
 817         * We may force a getattr if the user cares about atime.
 818         *
 819         * Note that we only have to check the vfsmount flags here:
 820         *  - NFS always sets S_NOATIME by so checking it would give a
 821         *    bogus result
 822         *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
 823         *    no point in checking those.
 824         */
 825        if ((path->mnt->mnt_flags & MNT_NOATIME) ||
 826            ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
 827                request_mask &= ~STATX_ATIME;
 828
 829        /* Is the user requesting attributes that might need revalidation? */
 830        if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
 831                                        STATX_MTIME|STATX_UID|STATX_GID|
 832                                        STATX_SIZE|STATX_BLOCKS)))
 833                goto out_no_revalidate;
 834
 835        /* Check whether the cached attributes are stale */
 836        do_update |= force_sync || nfs_attribute_cache_expired(inode);
 837        cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
 838        do_update |= cache_validity &
 839                (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
 840        if (request_mask & STATX_ATIME)
 841                do_update |= cache_validity & NFS_INO_INVALID_ATIME;
 842        if (request_mask & (STATX_CTIME|STATX_MTIME))
 843                do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
 844        if (request_mask & STATX_BLOCKS)
 845                do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
 846        if (do_update) {
 847                /* Update the attribute cache */
 848                if (!(server->flags & NFS_MOUNT_NOAC))
 849                        nfs_readdirplus_parent_cache_miss(path->dentry);
 850                else
 851                        nfs_readdirplus_parent_cache_hit(path->dentry);
 852                err = __nfs_revalidate_inode(server, inode);
 853                if (err)
 854                        goto out;
 855        } else
 856                nfs_readdirplus_parent_cache_hit(path->dentry);
 857out_no_revalidate:
 858        /* Only return attributes that were revalidated. */
 859        stat->result_mask &= request_mask;
 860out_no_update:
 861        generic_fillattr(inode, stat);
 862        stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
 863        if (S_ISDIR(inode->i_mode))
 864                stat->blksize = NFS_SERVER(inode)->dtsize;
 865out:
 866        trace_nfs_getattr_exit(inode, err);
 867        return err;
 868}
 869EXPORT_SYMBOL_GPL(nfs_getattr);
 870
 871static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
 872{
 873        refcount_set(&l_ctx->count, 1);
 874        l_ctx->lockowner = current->files;
 875        INIT_LIST_HEAD(&l_ctx->list);
 876        atomic_set(&l_ctx->io_count, 0);
 877}
 878
 879static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
 880{
 881        struct nfs_lock_context *pos;
 882
 883        list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
 884                if (pos->lockowner != current->files)
 885                        continue;
 886                if (refcount_inc_not_zero(&pos->count))
 887                        return pos;
 888        }
 889        return NULL;
 890}
 891
 892struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
 893{
 894        struct nfs_lock_context *res, *new = NULL;
 895        struct inode *inode = d_inode(ctx->dentry);
 896
 897        rcu_read_lock();
 898        res = __nfs_find_lock_context(ctx);
 899        rcu_read_unlock();
 900        if (res == NULL) {
 901                new = kmalloc(sizeof(*new), GFP_KERNEL);
 902                if (new == NULL)
 903                        return ERR_PTR(-ENOMEM);
 904                nfs_init_lock_context(new);
 905                spin_lock(&inode->i_lock);
 906                res = __nfs_find_lock_context(ctx);
 907                if (res == NULL) {
 908                        new->open_context = get_nfs_open_context(ctx);
 909                        if (new->open_context) {
 910                                list_add_tail_rcu(&new->list,
 911                                                &ctx->lock_context.list);
 912                                res = new;
 913                                new = NULL;
 914                        } else
 915                                res = ERR_PTR(-EBADF);
 916                }
 917                spin_unlock(&inode->i_lock);
 918                kfree(new);
 919        }
 920        return res;
 921}
 922EXPORT_SYMBOL_GPL(nfs_get_lock_context);
 923
 924void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
 925{
 926        struct nfs_open_context *ctx = l_ctx->open_context;
 927        struct inode *inode = d_inode(ctx->dentry);
 928
 929        if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
 930                return;
 931        list_del_rcu(&l_ctx->list);
 932        spin_unlock(&inode->i_lock);
 933        put_nfs_open_context(ctx);
 934        kfree_rcu(l_ctx, rcu_head);
 935}
 936EXPORT_SYMBOL_GPL(nfs_put_lock_context);
 937
 938/**
 939 * nfs_close_context - Common close_context() routine NFSv2/v3
 940 * @ctx: pointer to context
 941 * @is_sync: is this a synchronous close
 942 *
 943 * Ensure that the attributes are up to date if we're mounted
 944 * with close-to-open semantics and we have cached data that will
 945 * need to be revalidated on open.
 946 */
 947void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
 948{
 949        struct nfs_inode *nfsi;
 950        struct inode *inode;
 951        struct nfs_server *server;
 952
 953        if (!(ctx->mode & FMODE_WRITE))
 954                return;
 955        if (!is_sync)
 956                return;
 957        inode = d_inode(ctx->dentry);
 958        if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
 959                return;
 960        nfsi = NFS_I(inode);
 961        if (inode->i_mapping->nrpages == 0)
 962                return;
 963        if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
 964                return;
 965        if (!list_empty(&nfsi->open_files))
 966                return;
 967        server = NFS_SERVER(inode);
 968        if (server->flags & NFS_MOUNT_NOCTO)
 969                return;
 970        nfs_revalidate_inode(server, inode);
 971}
 972EXPORT_SYMBOL_GPL(nfs_close_context);
 973
 974struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
 975                                                fmode_t f_mode,
 976                                                struct file *filp)
 977{
 978        struct nfs_open_context *ctx;
 979
 980        ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 981        if (!ctx)
 982                return ERR_PTR(-ENOMEM);
 983        nfs_sb_active(dentry->d_sb);
 984        ctx->dentry = dget(dentry);
 985        if (filp)
 986                ctx->cred = get_cred(filp->f_cred);
 987        else
 988                ctx->cred = get_current_cred();
 989        ctx->ll_cred = NULL;
 990        ctx->state = NULL;
 991        ctx->mode = f_mode;
 992        ctx->flags = 0;
 993        ctx->error = 0;
 994        ctx->flock_owner = (fl_owner_t)filp;
 995        nfs_init_lock_context(&ctx->lock_context);
 996        ctx->lock_context.open_context = ctx;
 997        INIT_LIST_HEAD(&ctx->list);
 998        ctx->mdsthreshold = NULL;
 999        return ctx;
1000}
1001EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1002
1003struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1004{
1005        if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1006                return ctx;
1007        return NULL;
1008}
1009EXPORT_SYMBOL_GPL(get_nfs_open_context);
1010
1011static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1012{
1013        struct inode *inode = d_inode(ctx->dentry);
1014        struct super_block *sb = ctx->dentry->d_sb;
1015
1016        if (!refcount_dec_and_test(&ctx->lock_context.count))
1017                return;
1018        if (!list_empty(&ctx->list)) {
1019                spin_lock(&inode->i_lock);
1020                list_del_rcu(&ctx->list);
1021                spin_unlock(&inode->i_lock);
1022        }
1023        if (inode != NULL)
1024                NFS_PROTO(inode)->close_context(ctx, is_sync);
1025        put_cred(ctx->cred);
1026        dput(ctx->dentry);
1027        nfs_sb_deactive(sb);
1028        put_rpccred(ctx->ll_cred);
1029        kfree(ctx->mdsthreshold);
1030        kfree_rcu(ctx, rcu_head);
1031}
1032
1033void put_nfs_open_context(struct nfs_open_context *ctx)
1034{
1035        __put_nfs_open_context(ctx, 0);
1036}
1037EXPORT_SYMBOL_GPL(put_nfs_open_context);
1038
1039static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1040{
1041        __put_nfs_open_context(ctx, 1);
1042}
1043
1044/*
1045 * Ensure that mmap has a recent RPC credential for use when writing out
1046 * shared pages
1047 */
1048void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1049{
1050        struct inode *inode = d_inode(ctx->dentry);
1051        struct nfs_inode *nfsi = NFS_I(inode);
1052
1053        spin_lock(&inode->i_lock);
1054        if (list_empty(&nfsi->open_files) &&
1055            (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1056                nfsi->cache_validity |= NFS_INO_INVALID_DATA |
1057                        NFS_INO_REVAL_FORCED;
1058        list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1059        spin_unlock(&inode->i_lock);
1060}
1061EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1062
1063void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1064{
1065        filp->private_data = get_nfs_open_context(ctx);
1066        if (list_empty(&ctx->list))
1067                nfs_inode_attach_open_context(ctx);
1068}
1069EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1070
1071/*
1072 * Given an inode, search for an open context with the desired characteristics
1073 */
1074struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1075{
1076        struct nfs_inode *nfsi = NFS_I(inode);
1077        struct nfs_open_context *pos, *ctx = NULL;
1078
1079        rcu_read_lock();
1080        list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1081                if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1082                        continue;
1083                if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1084                        continue;
1085                ctx = get_nfs_open_context(pos);
1086                if (ctx)
1087                        break;
1088        }
1089        rcu_read_unlock();
1090        return ctx;
1091}
1092
1093void nfs_file_clear_open_context(struct file *filp)
1094{
1095        struct nfs_open_context *ctx = nfs_file_open_context(filp);
1096
1097        if (ctx) {
1098                struct inode *inode = d_inode(ctx->dentry);
1099
1100                /*
1101                 * We fatal error on write before. Try to writeback
1102                 * every page again.
1103                 */
1104                if (ctx->error < 0)
1105                        invalidate_inode_pages2(inode->i_mapping);
1106                filp->private_data = NULL;
1107                put_nfs_open_context_sync(ctx);
1108        }
1109}
1110
1111/*
1112 * These allocate and release file read/write context information.
1113 */
1114int nfs_open(struct inode *inode, struct file *filp)
1115{
1116        struct nfs_open_context *ctx;
1117
1118        ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1119        if (IS_ERR(ctx))
1120                return PTR_ERR(ctx);
1121        nfs_file_set_open_context(filp, ctx);
1122        put_nfs_open_context(ctx);
1123        nfs_fscache_open_file(inode, filp);
1124        return 0;
1125}
1126EXPORT_SYMBOL_GPL(nfs_open);
1127
1128/*
1129 * This function is called whenever some part of NFS notices that
1130 * the cached attributes have to be refreshed.
1131 */
1132int
1133__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1134{
1135        int              status = -ESTALE;
1136        struct nfs4_label *label = NULL;
1137        struct nfs_fattr *fattr = NULL;
1138        struct nfs_inode *nfsi = NFS_I(inode);
1139
1140        dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1141                inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1142
1143        trace_nfs_revalidate_inode_enter(inode);
1144
1145        if (is_bad_inode(inode))
1146                goto out;
1147        if (NFS_STALE(inode))
1148                goto out;
1149
1150        /* pNFS: Attributes aren't updated until we layoutcommit */
1151        if (S_ISREG(inode->i_mode)) {
1152                status = pnfs_sync_inode(inode, false);
1153                if (status)
1154                        goto out;
1155        }
1156
1157        status = -ENOMEM;
1158        fattr = nfs_alloc_fattr();
1159        if (fattr == NULL)
1160                goto out;
1161
1162        nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1163
1164        label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1165        if (IS_ERR(label)) {
1166                status = PTR_ERR(label);
1167                goto out;
1168        }
1169
1170        status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1171                        label, inode);
1172        if (status != 0) {
1173                dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1174                         inode->i_sb->s_id,
1175                         (unsigned long long)NFS_FILEID(inode), status);
1176                switch (status) {
1177                case -ETIMEDOUT:
1178                        /* A soft timeout occurred. Use cached information? */
1179                        if (server->flags & NFS_MOUNT_SOFTREVAL)
1180                                status = 0;
1181                        break;
1182                case -ESTALE:
1183                        if (!S_ISDIR(inode->i_mode))
1184                                nfs_set_inode_stale(inode);
1185                        else
1186                                nfs_zap_caches(inode);
1187                }
1188                goto err_out;
1189        }
1190
1191        status = nfs_refresh_inode(inode, fattr);
1192        if (status) {
1193                dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1194                         inode->i_sb->s_id,
1195                         (unsigned long long)NFS_FILEID(inode), status);
1196                goto err_out;
1197        }
1198
1199        if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1200                nfs_zap_acl_cache(inode);
1201
1202        nfs_setsecurity(inode, fattr, label);
1203
1204        dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1205                inode->i_sb->s_id,
1206                (unsigned long long)NFS_FILEID(inode));
1207
1208err_out:
1209        nfs4_label_free(label);
1210out:
1211        nfs_free_fattr(fattr);
1212        trace_nfs_revalidate_inode_exit(inode, status);
1213        return status;
1214}
1215
1216int nfs_attribute_cache_expired(struct inode *inode)
1217{
1218        if (nfs_have_delegated_attributes(inode))
1219                return 0;
1220        return nfs_attribute_timeout(inode);
1221}
1222
1223/**
1224 * nfs_revalidate_inode - Revalidate the inode attributes
1225 * @server: pointer to nfs_server struct
1226 * @inode: pointer to inode struct
1227 *
1228 * Updates inode attribute information by retrieving the data from the server.
1229 */
1230int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1231{
1232        if (!nfs_need_revalidate_inode(inode))
1233                return NFS_STALE(inode) ? -ESTALE : 0;
1234        return __nfs_revalidate_inode(server, inode);
1235}
1236EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1237
1238static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1239{
1240        struct nfs_inode *nfsi = NFS_I(inode);
1241        int ret;
1242
1243        if (mapping->nrpages != 0) {
1244                if (S_ISREG(inode->i_mode)) {
1245                        ret = nfs_sync_mapping(mapping);
1246                        if (ret < 0)
1247                                return ret;
1248                }
1249                ret = invalidate_inode_pages2(mapping);
1250                if (ret < 0)
1251                        return ret;
1252        }
1253        if (S_ISDIR(inode->i_mode)) {
1254                spin_lock(&inode->i_lock);
1255                memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1256                spin_unlock(&inode->i_lock);
1257        }
1258        nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1259        nfs_fscache_wait_on_invalidate(inode);
1260
1261        dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1262                        inode->i_sb->s_id,
1263                        (unsigned long long)NFS_FILEID(inode));
1264        return 0;
1265}
1266
1267bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1268{
1269        return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1270                NFS_STALE(inode);
1271}
1272
1273int nfs_revalidate_mapping_rcu(struct inode *inode)
1274{
1275        struct nfs_inode *nfsi = NFS_I(inode);
1276        unsigned long *bitlock = &nfsi->flags;
1277        int ret = 0;
1278
1279        if (IS_SWAPFILE(inode))
1280                goto out;
1281        if (nfs_mapping_need_revalidate_inode(inode)) {
1282                ret = -ECHILD;
1283                goto out;
1284        }
1285        spin_lock(&inode->i_lock);
1286        if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1287            (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1288                ret = -ECHILD;
1289        spin_unlock(&inode->i_lock);
1290out:
1291        return ret;
1292}
1293
1294/**
1295 * nfs_revalidate_mapping - Revalidate the pagecache
1296 * @inode: pointer to host inode
1297 * @mapping: pointer to mapping
1298 */
1299int nfs_revalidate_mapping(struct inode *inode,
1300                struct address_space *mapping)
1301{
1302        struct nfs_inode *nfsi = NFS_I(inode);
1303        unsigned long *bitlock = &nfsi->flags;
1304        int ret = 0;
1305
1306        /* swapfiles are not supposed to be shared. */
1307        if (IS_SWAPFILE(inode))
1308                goto out;
1309
1310        if (nfs_mapping_need_revalidate_inode(inode)) {
1311                ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1312                if (ret < 0)
1313                        goto out;
1314        }
1315
1316        /*
1317         * We must clear NFS_INO_INVALID_DATA first to ensure that
1318         * invalidations that come in while we're shooting down the mappings
1319         * are respected. But, that leaves a race window where one revalidator
1320         * can clear the flag, and then another checks it before the mapping
1321         * gets invalidated. Fix that by serializing access to this part of
1322         * the function.
1323         *
1324         * At the same time, we need to allow other tasks to see whether we
1325         * might be in the middle of invalidating the pages, so we only set
1326         * the bit lock here if it looks like we're going to be doing that.
1327         */
1328        for (;;) {
1329                ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1330                                         nfs_wait_bit_killable, TASK_KILLABLE);
1331                if (ret)
1332                        goto out;
1333                spin_lock(&inode->i_lock);
1334                if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1335                        spin_unlock(&inode->i_lock);
1336                        continue;
1337                }
1338                if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1339                        break;
1340                spin_unlock(&inode->i_lock);
1341                goto out;
1342        }
1343
1344        set_bit(NFS_INO_INVALIDATING, bitlock);
1345        smp_wmb();
1346        nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA|
1347                        NFS_INO_DATA_INVAL_DEFER);
1348        spin_unlock(&inode->i_lock);
1349        trace_nfs_invalidate_mapping_enter(inode);
1350        ret = nfs_invalidate_mapping(inode, mapping);
1351        trace_nfs_invalidate_mapping_exit(inode, ret);
1352
1353        clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1354        smp_mb__after_atomic();
1355        wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1356out:
1357        return ret;
1358}
1359
1360static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1361{
1362        struct inode *inode = &nfsi->vfs_inode;
1363
1364        if (!S_ISREG(inode->i_mode))
1365                return false;
1366        if (list_empty(&nfsi->open_files))
1367                return false;
1368        return inode_is_open_for_write(inode);
1369}
1370
1371static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1372{
1373        return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1374}
1375
1376static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1377{
1378        struct timespec64 ts;
1379
1380        if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1381                        && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1382                        && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1383                inode_set_iversion_raw(inode, fattr->change_attr);
1384                if (S_ISDIR(inode->i_mode))
1385                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1386                else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1387                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1388        }
1389        /* If we have atomic WCC data, we may update some attributes */
1390        ts = inode->i_ctime;
1391        if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1392                        && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1393                        && timespec64_equal(&ts, &fattr->pre_ctime)) {
1394                inode->i_ctime = fattr->ctime;
1395        }
1396
1397        ts = inode->i_mtime;
1398        if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1399                        && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1400                        && timespec64_equal(&ts, &fattr->pre_mtime)) {
1401                inode->i_mtime = fattr->mtime;
1402                if (S_ISDIR(inode->i_mode))
1403                        nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1404        }
1405        if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1406                        && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1407                        && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1408                        && !nfs_have_writebacks(inode)) {
1409                i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1410        }
1411}
1412
1413/**
1414 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1415 * @inode: pointer to inode
1416 * @fattr: updated attributes
1417 *
1418 * Verifies the attribute cache. If we have just changed the attributes,
1419 * so that fattr carries weak cache consistency data, then it may
1420 * also update the ctime/mtime/change_attribute.
1421 */
1422static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1423{
1424        struct nfs_inode *nfsi = NFS_I(inode);
1425        loff_t cur_size, new_isize;
1426        unsigned long invalid = 0;
1427        struct timespec64 ts;
1428
1429        if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1430                return 0;
1431
1432        if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1433                /* Only a mounted-on-fileid? Just exit */
1434                if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1435                        return 0;
1436        /* Has the inode gone and changed behind our back? */
1437        } else if (nfsi->fileid != fattr->fileid) {
1438                /* Is this perhaps the mounted-on fileid? */
1439                if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1440                    nfsi->fileid == fattr->mounted_on_fileid)
1441                        return 0;
1442                return -ESTALE;
1443        }
1444        if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1445                return -ESTALE;
1446
1447
1448        if (!nfs_file_has_buffered_writers(nfsi)) {
1449                /* Verify a few of the more important attributes */
1450                if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1451                        invalid |= NFS_INO_INVALID_CHANGE
1452                                | NFS_INO_REVAL_PAGECACHE;
1453
1454                ts = inode->i_mtime;
1455                if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1456                        invalid |= NFS_INO_INVALID_MTIME;
1457
1458                ts = inode->i_ctime;
1459                if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1460                        invalid |= NFS_INO_INVALID_CTIME;
1461
1462                if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1463                        cur_size = i_size_read(inode);
1464                        new_isize = nfs_size_to_loff_t(fattr->size);
1465                        if (cur_size != new_isize)
1466                                invalid |= NFS_INO_INVALID_SIZE
1467                                        | NFS_INO_REVAL_PAGECACHE;
1468                }
1469        }
1470
1471        /* Have any file permissions changed? */
1472        if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1473                invalid |= NFS_INO_INVALID_ACCESS
1474                        | NFS_INO_INVALID_ACL
1475                        | NFS_INO_INVALID_OTHER;
1476        if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1477                invalid |= NFS_INO_INVALID_ACCESS
1478                        | NFS_INO_INVALID_ACL
1479                        | NFS_INO_INVALID_OTHER;
1480        if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1481                invalid |= NFS_INO_INVALID_ACCESS
1482                        | NFS_INO_INVALID_ACL
1483                        | NFS_INO_INVALID_OTHER;
1484
1485        /* Has the link count changed? */
1486        if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1487                invalid |= NFS_INO_INVALID_OTHER;
1488
1489        ts = inode->i_atime;
1490        if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1491                invalid |= NFS_INO_INVALID_ATIME;
1492
1493        if (invalid != 0)
1494                nfs_set_cache_invalid(inode, invalid);
1495
1496        nfsi->read_cache_jiffies = fattr->time_start;
1497        return 0;
1498}
1499
1500static atomic_long_t nfs_attr_generation_counter;
1501
1502static unsigned long nfs_read_attr_generation_counter(void)
1503{
1504        return atomic_long_read(&nfs_attr_generation_counter);
1505}
1506
1507unsigned long nfs_inc_attr_generation_counter(void)
1508{
1509        return atomic_long_inc_return(&nfs_attr_generation_counter);
1510}
1511EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1512
1513void nfs_fattr_init(struct nfs_fattr *fattr)
1514{
1515        fattr->valid = 0;
1516        fattr->time_start = jiffies;
1517        fattr->gencount = nfs_inc_attr_generation_counter();
1518        fattr->owner_name = NULL;
1519        fattr->group_name = NULL;
1520}
1521EXPORT_SYMBOL_GPL(nfs_fattr_init);
1522
1523/**
1524 * nfs_fattr_set_barrier
1525 * @fattr: attributes
1526 *
1527 * Used to set a barrier after an attribute was updated. This
1528 * barrier ensures that older attributes from RPC calls that may
1529 * have raced with our update cannot clobber these new values.
1530 * Note that you are still responsible for ensuring that other
1531 * operations which change the attribute on the server do not
1532 * collide.
1533 */
1534void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1535{
1536        fattr->gencount = nfs_inc_attr_generation_counter();
1537}
1538
1539struct nfs_fattr *nfs_alloc_fattr(void)
1540{
1541        struct nfs_fattr *fattr;
1542
1543        fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1544        if (fattr != NULL)
1545                nfs_fattr_init(fattr);
1546        return fattr;
1547}
1548EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1549
1550struct nfs_fh *nfs_alloc_fhandle(void)
1551{
1552        struct nfs_fh *fh;
1553
1554        fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1555        if (fh != NULL)
1556                fh->size = 0;
1557        return fh;
1558}
1559EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1560
1561#ifdef NFS_DEBUG
1562/*
1563 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1564 *                             in the same way that wireshark does
1565 *
1566 * @fh: file handle
1567 *
1568 * For debugging only.
1569 */
1570u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1571{
1572        /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1573         * not on the result */
1574        return nfs_fhandle_hash(fh);
1575}
1576EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1577
1578/*
1579 * _nfs_display_fhandle - display an NFS file handle on the console
1580 *
1581 * @fh: file handle to display
1582 * @caption: display caption
1583 *
1584 * For debugging only.
1585 */
1586void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1587{
1588        unsigned short i;
1589
1590        if (fh == NULL || fh->size == 0) {
1591                printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1592                return;
1593        }
1594
1595        printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1596               caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1597        for (i = 0; i < fh->size; i += 16) {
1598                __be32 *pos = (__be32 *)&fh->data[i];
1599
1600                switch ((fh->size - i - 1) >> 2) {
1601                case 0:
1602                        printk(KERN_DEFAULT " %08x\n",
1603                                be32_to_cpup(pos));
1604                        break;
1605                case 1:
1606                        printk(KERN_DEFAULT " %08x %08x\n",
1607                                be32_to_cpup(pos), be32_to_cpup(pos + 1));
1608                        break;
1609                case 2:
1610                        printk(KERN_DEFAULT " %08x %08x %08x\n",
1611                                be32_to_cpup(pos), be32_to_cpup(pos + 1),
1612                                be32_to_cpup(pos + 2));
1613                        break;
1614                default:
1615                        printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1616                                be32_to_cpup(pos), be32_to_cpup(pos + 1),
1617                                be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1618                }
1619        }
1620}
1621EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1622#endif
1623
1624/**
1625 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1626 * @inode: pointer to inode
1627 * @fattr: attributes
1628 *
1629 * Attempt to divine whether or not an RPC call reply carrying stale
1630 * attributes got scheduled after another call carrying updated ones.
1631 *
1632 * To do so, the function first assumes that a more recent ctime means
1633 * that the attributes in fattr are newer, however it also attempt to
1634 * catch the case where ctime either didn't change, or went backwards
1635 * (if someone reset the clock on the server) by looking at whether
1636 * or not this RPC call was started after the inode was last updated.
1637 * Note also the check for wraparound of 'attr_gencount'
1638 *
1639 * The function returns 'true' if it thinks the attributes in 'fattr' are
1640 * more recent than the ones cached in the inode.
1641 *
1642 */
1643static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1644{
1645        const struct nfs_inode *nfsi = NFS_I(inode);
1646
1647        return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1648                ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1649}
1650
1651static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1652{
1653        int ret;
1654
1655        trace_nfs_refresh_inode_enter(inode);
1656
1657        if (nfs_inode_attrs_need_update(inode, fattr))
1658                ret = nfs_update_inode(inode, fattr);
1659        else
1660                ret = nfs_check_inode_attributes(inode, fattr);
1661
1662        trace_nfs_refresh_inode_exit(inode, ret);
1663        return ret;
1664}
1665
1666/**
1667 * nfs_refresh_inode - try to update the inode attribute cache
1668 * @inode: pointer to inode
1669 * @fattr: updated attributes
1670 *
1671 * Check that an RPC call that returned attributes has not overlapped with
1672 * other recent updates of the inode metadata, then decide whether it is
1673 * safe to do a full update of the inode attributes, or whether just to
1674 * call nfs_check_inode_attributes.
1675 */
1676int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1677{
1678        int status;
1679
1680        if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1681                return 0;
1682        spin_lock(&inode->i_lock);
1683        status = nfs_refresh_inode_locked(inode, fattr);
1684        spin_unlock(&inode->i_lock);
1685
1686        return status;
1687}
1688EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1689
1690static int nfs_post_op_update_inode_locked(struct inode *inode,
1691                struct nfs_fattr *fattr, unsigned int invalid)
1692{
1693        if (S_ISDIR(inode->i_mode))
1694                invalid |= NFS_INO_INVALID_DATA;
1695        nfs_set_cache_invalid(inode, invalid);
1696        if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1697                return 0;
1698        return nfs_refresh_inode_locked(inode, fattr);
1699}
1700
1701/**
1702 * nfs_post_op_update_inode - try to update the inode attribute cache
1703 * @inode: pointer to inode
1704 * @fattr: updated attributes
1705 *
1706 * After an operation that has changed the inode metadata, mark the
1707 * attribute cache as being invalid, then try to update it.
1708 *
1709 * NB: if the server didn't return any post op attributes, this
1710 * function will force the retrieval of attributes before the next
1711 * NFS request.  Thus it should be used only for operations that
1712 * are expected to change one or more attributes, to avoid
1713 * unnecessary NFS requests and trips through nfs_update_inode().
1714 */
1715int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1716{
1717        int status;
1718
1719        spin_lock(&inode->i_lock);
1720        nfs_fattr_set_barrier(fattr);
1721        status = nfs_post_op_update_inode_locked(inode, fattr,
1722                        NFS_INO_INVALID_CHANGE
1723                        | NFS_INO_INVALID_CTIME
1724                        | NFS_INO_REVAL_FORCED);
1725        spin_unlock(&inode->i_lock);
1726
1727        return status;
1728}
1729EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1730
1731/**
1732 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1733 * @inode: pointer to inode
1734 * @fattr: updated attributes
1735 *
1736 * After an operation that has changed the inode metadata, mark the
1737 * attribute cache as being invalid, then try to update it. Fake up
1738 * weak cache consistency data, if none exist.
1739 *
1740 * This function is mainly designed to be used by the ->write_done() functions.
1741 */
1742int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1743{
1744        int status;
1745
1746        /* Don't do a WCC update if these attributes are already stale */
1747        if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1748                        !nfs_inode_attrs_need_update(inode, fattr)) {
1749                fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1750                                | NFS_ATTR_FATTR_PRESIZE
1751                                | NFS_ATTR_FATTR_PREMTIME
1752                                | NFS_ATTR_FATTR_PRECTIME);
1753                goto out_noforce;
1754        }
1755        if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1756                        (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1757                fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1758                fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1759        }
1760        if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1761                        (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1762                fattr->pre_ctime = inode->i_ctime;
1763                fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1764        }
1765        if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1766                        (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1767                fattr->pre_mtime = inode->i_mtime;
1768                fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1769        }
1770        if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1771                        (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1772                fattr->pre_size = i_size_read(inode);
1773                fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1774        }
1775out_noforce:
1776        status = nfs_post_op_update_inode_locked(inode, fattr,
1777                        NFS_INO_INVALID_CHANGE
1778                        | NFS_INO_INVALID_CTIME
1779                        | NFS_INO_INVALID_MTIME
1780                        | NFS_INO_INVALID_BLOCKS);
1781        return status;
1782}
1783
1784/**
1785 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1786 * @inode: pointer to inode
1787 * @fattr: updated attributes
1788 *
1789 * After an operation that has changed the inode metadata, mark the
1790 * attribute cache as being invalid, then try to update it. Fake up
1791 * weak cache consistency data, if none exist.
1792 *
1793 * This function is mainly designed to be used by the ->write_done() functions.
1794 */
1795int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1796{
1797        int status;
1798
1799        spin_lock(&inode->i_lock);
1800        nfs_fattr_set_barrier(fattr);
1801        status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1802        spin_unlock(&inode->i_lock);
1803        return status;
1804}
1805EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1806
1807
1808/*
1809 * Many nfs protocol calls return the new file attributes after
1810 * an operation.  Here we update the inode to reflect the state
1811 * of the server's inode.
1812 *
1813 * This is a bit tricky because we have to make sure all dirty pages
1814 * have been sent off to the server before calling invalidate_inode_pages.
1815 * To make sure no other process adds more write requests while we try
1816 * our best to flush them, we make them sleep during the attribute refresh.
1817 *
1818 * A very similar scenario holds for the dir cache.
1819 */
1820static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1821{
1822        struct nfs_server *server;
1823        struct nfs_inode *nfsi = NFS_I(inode);
1824        loff_t cur_isize, new_isize;
1825        unsigned long invalid = 0;
1826        unsigned long now = jiffies;
1827        unsigned long save_cache_validity;
1828        bool have_writers = nfs_file_has_buffered_writers(nfsi);
1829        bool cache_revalidated = true;
1830        bool attr_changed = false;
1831        bool have_delegation;
1832
1833        dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1834                        __func__, inode->i_sb->s_id, inode->i_ino,
1835                        nfs_display_fhandle_hash(NFS_FH(inode)),
1836                        atomic_read(&inode->i_count), fattr->valid);
1837
1838        if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1839                /* Only a mounted-on-fileid? Just exit */
1840                if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1841                        return 0;
1842        /* Has the inode gone and changed behind our back? */
1843        } else if (nfsi->fileid != fattr->fileid) {
1844                /* Is this perhaps the mounted-on fileid? */
1845                if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1846                    nfsi->fileid == fattr->mounted_on_fileid)
1847                        return 0;
1848                printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1849                        "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1850                        NFS_SERVER(inode)->nfs_client->cl_hostname,
1851                        inode->i_sb->s_id, (long long)nfsi->fileid,
1852                        (long long)fattr->fileid);
1853                goto out_err;
1854        }
1855
1856        /*
1857         * Make sure the inode's type hasn't changed.
1858         */
1859        if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1860                /*
1861                * Big trouble! The inode has become a different object.
1862                */
1863                printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1864                                __func__, inode->i_ino, inode->i_mode, fattr->mode);
1865                goto out_err;
1866        }
1867
1868        server = NFS_SERVER(inode);
1869        /* Update the fsid? */
1870        if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1871                        !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1872                        !IS_AUTOMOUNT(inode))
1873                server->fsid = fattr->fsid;
1874
1875        /* Save the delegation state before clearing cache_validity */
1876        have_delegation = nfs_have_delegated_attributes(inode);
1877
1878        /*
1879         * Update the read time so we don't revalidate too often.
1880         */
1881        nfsi->read_cache_jiffies = fattr->time_start;
1882
1883        save_cache_validity = nfsi->cache_validity;
1884        nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1885                        | NFS_INO_INVALID_ATIME
1886                        | NFS_INO_REVAL_FORCED
1887                        | NFS_INO_REVAL_PAGECACHE
1888                        | NFS_INO_INVALID_BLOCKS);
1889
1890        /* Do atomic weak cache consistency updates */
1891        nfs_wcc_update_inode(inode, fattr);
1892
1893        if (pnfs_layoutcommit_outstanding(inode)) {
1894                nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1895                cache_revalidated = false;
1896        }
1897
1898        /* More cache consistency checks */
1899        if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1900                if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1901                        /* Could it be a race with writeback? */
1902                        if (!(have_writers || have_delegation)) {
1903                                invalid |= NFS_INO_INVALID_DATA
1904                                        | NFS_INO_INVALID_ACCESS
1905                                        | NFS_INO_INVALID_ACL
1906                                        | NFS_INO_INVALID_XATTR;
1907                                /* Force revalidate of all attributes */
1908                                save_cache_validity |= NFS_INO_INVALID_CTIME
1909                                        | NFS_INO_INVALID_MTIME
1910                                        | NFS_INO_INVALID_SIZE
1911                                        | NFS_INO_INVALID_OTHER;
1912                                if (S_ISDIR(inode->i_mode))
1913                                        nfs_force_lookup_revalidate(inode);
1914                                dprintk("NFS: change_attr change on server for file %s/%ld\n",
1915                                                inode->i_sb->s_id,
1916                                                inode->i_ino);
1917                        } else if (!have_delegation)
1918                                nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1919                        inode_set_iversion_raw(inode, fattr->change_attr);
1920                        attr_changed = true;
1921                }
1922        } else {
1923                nfsi->cache_validity |= save_cache_validity &
1924                                (NFS_INO_INVALID_CHANGE
1925                                | NFS_INO_REVAL_PAGECACHE
1926                                | NFS_INO_REVAL_FORCED);
1927                cache_revalidated = false;
1928        }
1929
1930        if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1931                inode->i_mtime = fattr->mtime;
1932        } else if (server->caps & NFS_CAP_MTIME) {
1933                nfsi->cache_validity |= save_cache_validity &
1934                                (NFS_INO_INVALID_MTIME
1935                                | NFS_INO_REVAL_FORCED);
1936                cache_revalidated = false;
1937        }
1938
1939        if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1940                inode->i_ctime = fattr->ctime;
1941        } else if (server->caps & NFS_CAP_CTIME) {
1942                nfsi->cache_validity |= save_cache_validity &
1943                                (NFS_INO_INVALID_CTIME
1944                                | NFS_INO_REVAL_FORCED);
1945                cache_revalidated = false;
1946        }
1947
1948        /* Check if our cached file size is stale */
1949        if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1950                new_isize = nfs_size_to_loff_t(fattr->size);
1951                cur_isize = i_size_read(inode);
1952                if (new_isize != cur_isize && !have_delegation) {
1953                        /* Do we perhaps have any outstanding writes, or has
1954                         * the file grown beyond our last write? */
1955                        if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1956                                i_size_write(inode, new_isize);
1957                                if (!have_writers)
1958                                        invalid |= NFS_INO_INVALID_DATA;
1959                                attr_changed = true;
1960                        }
1961                        dprintk("NFS: isize change on server for file %s/%ld "
1962                                        "(%Ld to %Ld)\n",
1963                                        inode->i_sb->s_id,
1964                                        inode->i_ino,
1965                                        (long long)cur_isize,
1966                                        (long long)new_isize);
1967                }
1968        } else {
1969                nfsi->cache_validity |= save_cache_validity &
1970                                (NFS_INO_INVALID_SIZE
1971                                | NFS_INO_REVAL_PAGECACHE
1972                                | NFS_INO_REVAL_FORCED);
1973                cache_revalidated = false;
1974        }
1975
1976
1977        if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1978                inode->i_atime = fattr->atime;
1979        else if (server->caps & NFS_CAP_ATIME) {
1980                nfsi->cache_validity |= save_cache_validity &
1981                                (NFS_INO_INVALID_ATIME
1982                                | NFS_INO_REVAL_FORCED);
1983                cache_revalidated = false;
1984        }
1985
1986        if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1987                if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1988                        umode_t newmode = inode->i_mode & S_IFMT;
1989                        newmode |= fattr->mode & S_IALLUGO;
1990                        inode->i_mode = newmode;
1991                        invalid |= NFS_INO_INVALID_ACCESS
1992                                | NFS_INO_INVALID_ACL;
1993                        attr_changed = true;
1994                }
1995        } else if (server->caps & NFS_CAP_MODE) {
1996                nfsi->cache_validity |= save_cache_validity &
1997                                (NFS_INO_INVALID_OTHER
1998                                | NFS_INO_REVAL_FORCED);
1999                cache_revalidated = false;
2000        }
2001
2002        if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2003                if (!uid_eq(inode->i_uid, fattr->uid)) {
2004                        invalid |= NFS_INO_INVALID_ACCESS
2005                                | NFS_INO_INVALID_ACL;
2006                        inode->i_uid = fattr->uid;
2007                        attr_changed = true;
2008                }
2009        } else if (server->caps & NFS_CAP_OWNER) {
2010                nfsi->cache_validity |= save_cache_validity &
2011                                (NFS_INO_INVALID_OTHER
2012                                | NFS_INO_REVAL_FORCED);
2013                cache_revalidated = false;
2014        }
2015
2016        if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2017                if (!gid_eq(inode->i_gid, fattr->gid)) {
2018                        invalid |= NFS_INO_INVALID_ACCESS
2019                                | NFS_INO_INVALID_ACL;
2020                        inode->i_gid = fattr->gid;
2021                        attr_changed = true;
2022                }
2023        } else if (server->caps & NFS_CAP_OWNER_GROUP) {
2024                nfsi->cache_validity |= save_cache_validity &
2025                                (NFS_INO_INVALID_OTHER
2026                                | NFS_INO_REVAL_FORCED);
2027                cache_revalidated = false;
2028        }
2029
2030        if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2031                if (inode->i_nlink != fattr->nlink) {
2032                        if (S_ISDIR(inode->i_mode))
2033                                invalid |= NFS_INO_INVALID_DATA;
2034                        set_nlink(inode, fattr->nlink);
2035                        attr_changed = true;
2036                }
2037        } else if (server->caps & NFS_CAP_NLINK) {
2038                nfsi->cache_validity |= save_cache_validity &
2039                                (NFS_INO_INVALID_OTHER
2040                                | NFS_INO_REVAL_FORCED);
2041                cache_revalidated = false;
2042        }
2043
2044        if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2045                /*
2046                 * report the blocks in 512byte units
2047                 */
2048                inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2049        } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2050                inode->i_blocks = fattr->du.nfs2.blocks;
2051        else {
2052                nfsi->cache_validity |= save_cache_validity &
2053                                (NFS_INO_INVALID_BLOCKS
2054                                | NFS_INO_REVAL_FORCED);
2055                cache_revalidated = false;
2056        }
2057
2058        /* Update attrtimeo value if we're out of the unstable period */
2059        if (attr_changed) {
2060                invalid &= ~NFS_INO_INVALID_ATTR;
2061                nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2062                nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2063                nfsi->attrtimeo_timestamp = now;
2064                /* Set barrier to be more recent than all outstanding updates */
2065                nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2066        } else {
2067                if (cache_revalidated) {
2068                        if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2069                                nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2070                                nfsi->attrtimeo <<= 1;
2071                                if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2072                                        nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2073                        }
2074                        nfsi->attrtimeo_timestamp = now;
2075                }
2076                /* Set the barrier to be more recent than this fattr */
2077                if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2078                        nfsi->attr_gencount = fattr->gencount;
2079        }
2080
2081        /* Don't invalidate the data if we were to blame */
2082        if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2083                                || S_ISLNK(inode->i_mode)))
2084                invalid &= ~NFS_INO_INVALID_DATA;
2085        nfs_set_cache_invalid(inode, invalid);
2086
2087        return 0;
2088 out_err:
2089        /*
2090         * No need to worry about unhashing the dentry, as the
2091         * lookup validation will know that the inode is bad.
2092         * (But we fall through to invalidate the caches.)
2093         */
2094        nfs_set_inode_stale_locked(inode);
2095        return -ESTALE;
2096}
2097
2098struct inode *nfs_alloc_inode(struct super_block *sb)
2099{
2100        struct nfs_inode *nfsi;
2101        nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2102        if (!nfsi)
2103                return NULL;
2104        nfsi->flags = 0UL;
2105        nfsi->cache_validity = 0UL;
2106#if IS_ENABLED(CONFIG_NFS_V4)
2107        nfsi->nfs4_acl = NULL;
2108#endif /* CONFIG_NFS_V4 */
2109#ifdef CONFIG_NFS_V4_2
2110        nfsi->xattr_cache = NULL;
2111#endif
2112        return &nfsi->vfs_inode;
2113}
2114EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2115
2116void nfs_free_inode(struct inode *inode)
2117{
2118        kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2119}
2120EXPORT_SYMBOL_GPL(nfs_free_inode);
2121
2122static inline void nfs4_init_once(struct nfs_inode *nfsi)
2123{
2124#if IS_ENABLED(CONFIG_NFS_V4)
2125        INIT_LIST_HEAD(&nfsi->open_states);
2126        nfsi->delegation = NULL;
2127        init_rwsem(&nfsi->rwsem);
2128        nfsi->layout = NULL;
2129#endif
2130}
2131
2132static void init_once(void *foo)
2133{
2134        struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2135
2136        inode_init_once(&nfsi->vfs_inode);
2137        INIT_LIST_HEAD(&nfsi->open_files);
2138        INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2139        INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2140        INIT_LIST_HEAD(&nfsi->commit_info.list);
2141        atomic_long_set(&nfsi->nrequests, 0);
2142        atomic_long_set(&nfsi->commit_info.ncommit, 0);
2143        atomic_set(&nfsi->commit_info.rpcs_out, 0);
2144        init_rwsem(&nfsi->rmdir_sem);
2145        mutex_init(&nfsi->commit_mutex);
2146        nfs4_init_once(nfsi);
2147        nfsi->cache_change_attribute = 0;
2148}
2149
2150static int __init nfs_init_inodecache(void)
2151{
2152        nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2153                                             sizeof(struct nfs_inode),
2154                                             0, (SLAB_RECLAIM_ACCOUNT|
2155                                                SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2156                                             init_once);
2157        if (nfs_inode_cachep == NULL)
2158                return -ENOMEM;
2159
2160        return 0;
2161}
2162
2163static void nfs_destroy_inodecache(void)
2164{
2165        /*
2166         * Make sure all delayed rcu free inodes are flushed before we
2167         * destroy cache.
2168         */
2169        rcu_barrier();
2170        kmem_cache_destroy(nfs_inode_cachep);
2171}
2172
2173struct workqueue_struct *nfsiod_workqueue;
2174EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2175
2176/*
2177 * start up the nfsiod workqueue
2178 */
2179static int nfsiod_start(void)
2180{
2181        struct workqueue_struct *wq;
2182        dprintk("RPC:       creating workqueue nfsiod\n");
2183        wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2184        if (wq == NULL)
2185                return -ENOMEM;
2186        nfsiod_workqueue = wq;
2187        return 0;
2188}
2189
2190/*
2191 * Destroy the nfsiod workqueue
2192 */
2193static void nfsiod_stop(void)
2194{
2195        struct workqueue_struct *wq;
2196
2197        wq = nfsiod_workqueue;
2198        if (wq == NULL)
2199                return;
2200        nfsiod_workqueue = NULL;
2201        destroy_workqueue(wq);
2202}
2203
2204unsigned int nfs_net_id;
2205EXPORT_SYMBOL_GPL(nfs_net_id);
2206
2207static int nfs_net_init(struct net *net)
2208{
2209        nfs_clients_init(net);
2210        return nfs_fs_proc_net_init(net);
2211}
2212
2213static void nfs_net_exit(struct net *net)
2214{
2215        nfs_fs_proc_net_exit(net);
2216        nfs_clients_exit(net);
2217}
2218
2219static struct pernet_operations nfs_net_ops = {
2220        .init = nfs_net_init,
2221        .exit = nfs_net_exit,
2222        .id   = &nfs_net_id,
2223        .size = sizeof(struct nfs_net),
2224};
2225
2226/*
2227 * Initialize NFS
2228 */
2229static int __init init_nfs_fs(void)
2230{
2231        int err;
2232
2233        err = nfs_sysfs_init();
2234        if (err < 0)
2235                goto out10;
2236
2237        err = register_pernet_subsys(&nfs_net_ops);
2238        if (err < 0)
2239                goto out9;
2240
2241        err = nfs_fscache_register();
2242        if (err < 0)
2243                goto out8;
2244
2245        err = nfsiod_start();
2246        if (err)
2247                goto out7;
2248
2249        err = nfs_fs_proc_init();
2250        if (err)
2251                goto out6;
2252
2253        err = nfs_init_nfspagecache();
2254        if (err)
2255                goto out5;
2256
2257        err = nfs_init_inodecache();
2258        if (err)
2259                goto out4;
2260
2261        err = nfs_init_readpagecache();
2262        if (err)
2263                goto out3;
2264
2265        err = nfs_init_writepagecache();
2266        if (err)
2267                goto out2;
2268
2269        err = nfs_init_directcache();
2270        if (err)
2271                goto out1;
2272
2273        rpc_proc_register(&init_net, &nfs_rpcstat);
2274
2275        err = register_nfs_fs();
2276        if (err)
2277                goto out0;
2278
2279        return 0;
2280out0:
2281        rpc_proc_unregister(&init_net, "nfs");
2282        nfs_destroy_directcache();
2283out1:
2284        nfs_destroy_writepagecache();
2285out2:
2286        nfs_destroy_readpagecache();
2287out3:
2288        nfs_destroy_inodecache();
2289out4:
2290        nfs_destroy_nfspagecache();
2291out5:
2292        nfs_fs_proc_exit();
2293out6:
2294        nfsiod_stop();
2295out7:
2296        nfs_fscache_unregister();
2297out8:
2298        unregister_pernet_subsys(&nfs_net_ops);
2299out9:
2300        nfs_sysfs_exit();
2301out10:
2302        return err;
2303}
2304
2305static void __exit exit_nfs_fs(void)
2306{
2307        nfs_destroy_directcache();
2308        nfs_destroy_writepagecache();
2309        nfs_destroy_readpagecache();
2310        nfs_destroy_inodecache();
2311        nfs_destroy_nfspagecache();
2312        nfs_fscache_unregister();
2313        unregister_pernet_subsys(&nfs_net_ops);
2314        rpc_proc_unregister(&init_net, "nfs");
2315        unregister_nfs_fs();
2316        nfs_fs_proc_exit();
2317        nfsiod_stop();
2318        nfs_sysfs_exit();
2319}
2320
2321/* Not quite true; I just maintain it */
2322MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2323MODULE_LICENSE("GPL");
2324module_param(enable_ino64, bool, 0644);
2325
2326module_init(init_nfs_fs)
2327module_exit(exit_nfs_fs)
2328