linux/fs/nfs/nfs4file.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/nfs/file.c
   4 *
   5 *  Copyright (C) 1992  Rick Sladkey
   6 */
   7#include <linux/fs.h>
   8#include <linux/file.h>
   9#include <linux/falloc.h>
  10#include <linux/mount.h>
  11#include <linux/nfs_fs.h>
  12#include <linux/nfs_ssc.h>
  13#include "delegation.h"
  14#include "internal.h"
  15#include "iostat.h"
  16#include "fscache.h"
  17#include "pnfs.h"
  18
  19#include "nfstrace.h"
  20
  21#ifdef CONFIG_NFS_V4_2
  22#include "nfs42.h"
  23#endif
  24
  25#define NFSDBG_FACILITY         NFSDBG_FILE
  26
  27static int
  28nfs4_file_open(struct inode *inode, struct file *filp)
  29{
  30        struct nfs_open_context *ctx;
  31        struct dentry *dentry = file_dentry(filp);
  32        struct dentry *parent = NULL;
  33        struct inode *dir;
  34        unsigned openflags = filp->f_flags;
  35        struct iattr attr;
  36        int err;
  37
  38        /*
  39         * If no cached dentry exists or if it's negative, NFSv4 handled the
  40         * opens in ->lookup() or ->create().
  41         *
  42         * We only get this far for a cached positive dentry.  We skipped
  43         * revalidation, so handle it here by dropping the dentry and returning
  44         * -EOPENSTALE.  The VFS will retry the lookup/create/open.
  45         */
  46
  47        dprintk("NFS: open file(%pd2)\n", dentry);
  48
  49        err = nfs_check_flags(openflags);
  50        if (err)
  51                return err;
  52
  53        if ((openflags & O_ACCMODE) == 3)
  54                return nfs_open(inode, filp);
  55
  56        /* We can't create new files here */
  57        openflags &= ~(O_CREAT|O_EXCL);
  58
  59        parent = dget_parent(dentry);
  60        dir = d_inode(parent);
  61
  62        ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
  63        err = PTR_ERR(ctx);
  64        if (IS_ERR(ctx))
  65                goto out;
  66
  67        attr.ia_valid = ATTR_OPEN;
  68        if (openflags & O_TRUNC) {
  69                attr.ia_valid |= ATTR_SIZE;
  70                attr.ia_size = 0;
  71                filemap_write_and_wait(inode->i_mapping);
  72        }
  73
  74        inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
  75        if (IS_ERR(inode)) {
  76                err = PTR_ERR(inode);
  77                switch (err) {
  78                default:
  79                        goto out_put_ctx;
  80                case -ENOENT:
  81                case -ESTALE:
  82                case -EISDIR:
  83                case -ENOTDIR:
  84                case -ELOOP:
  85                        goto out_drop;
  86                }
  87        }
  88        if (inode != d_inode(dentry))
  89                goto out_drop;
  90
  91        nfs_file_set_open_context(filp, ctx);
  92        nfs_fscache_open_file(inode, filp);
  93        err = 0;
  94
  95out_put_ctx:
  96        put_nfs_open_context(ctx);
  97out:
  98        dput(parent);
  99        return err;
 100
 101out_drop:
 102        d_drop(dentry);
 103        err = -EOPENSTALE;
 104        goto out_put_ctx;
 105}
 106
 107/*
 108 * Flush all dirty pages, and check for write errors.
 109 */
 110static int
 111nfs4_file_flush(struct file *file, fl_owner_t id)
 112{
 113        struct inode    *inode = file_inode(file);
 114        errseq_t since;
 115
 116        dprintk("NFS: flush(%pD2)\n", file);
 117
 118        nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 119        if ((file->f_mode & FMODE_WRITE) == 0)
 120                return 0;
 121
 122        /*
 123         * If we're holding a write delegation, then check if we're required
 124         * to flush the i/o on close. If not, then just start the i/o now.
 125         */
 126        if (!nfs4_delegation_flush_on_close(inode))
 127                return filemap_fdatawrite(file->f_mapping);
 128
 129        /* Flush writes to the server and return any errors */
 130        since = filemap_sample_wb_err(file->f_mapping);
 131        nfs_wb_all(inode);
 132        return filemap_check_wb_err(file->f_mapping, since);
 133}
 134
 135#ifdef CONFIG_NFS_V4_2
 136static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
 137                                      struct file *file_out, loff_t pos_out,
 138                                      size_t count, unsigned int flags)
 139{
 140        struct nfs42_copy_notify_res *cn_resp = NULL;
 141        struct nl4_server *nss = NULL;
 142        nfs4_stateid *cnrs = NULL;
 143        ssize_t ret;
 144        bool sync = false;
 145
 146        /* Only offload copy if superblock is the same */
 147        if (file_in->f_op != &nfs4_file_operations)
 148                return -EXDEV;
 149        if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY) ||
 150            !nfs_server_capable(file_inode(file_in), NFS_CAP_COPY))
 151                return -EOPNOTSUPP;
 152        if (file_inode(file_in) == file_inode(file_out))
 153                return -EOPNOTSUPP;
 154        /* if the copy size if smaller than 2 RPC payloads, make it
 155         * synchronous
 156         */
 157        if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
 158                sync = true;
 159retry:
 160        if (!nfs42_files_from_same_server(file_in, file_out)) {
 161                /* for inter copy, if copy size if smaller than 12 RPC
 162                 * payloads, fallback to traditional copy. There are
 163                 * 14 RPCs during an NFSv4.x mount between source/dest
 164                 * servers.
 165                 */
 166                if (sync ||
 167                        count <= 14 * NFS_SERVER(file_inode(file_in))->rsize)
 168                        return -EOPNOTSUPP;
 169                cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res),
 170                                GFP_NOFS);
 171                if (unlikely(cn_resp == NULL))
 172                        return -ENOMEM;
 173
 174                ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
 175                if (ret) {
 176                        ret = -EOPNOTSUPP;
 177                        goto out;
 178                }
 179                nss = &cn_resp->cnr_src;
 180                cnrs = &cn_resp->cnr_stateid;
 181        }
 182        ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
 183                                nss, cnrs, sync);
 184out:
 185        if (!nfs42_files_from_same_server(file_in, file_out))
 186                kfree(cn_resp);
 187        if (ret == -EAGAIN)
 188                goto retry;
 189        return ret;
 190}
 191
 192static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
 193                                    struct file *file_out, loff_t pos_out,
 194                                    size_t count, unsigned int flags)
 195{
 196        ssize_t ret;
 197
 198        ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
 199                                     flags);
 200        if (ret == -EOPNOTSUPP || ret == -EXDEV)
 201                ret = generic_copy_file_range(file_in, pos_in, file_out,
 202                                              pos_out, count, flags);
 203        return ret;
 204}
 205
 206static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
 207{
 208        loff_t ret;
 209
 210        switch (whence) {
 211        case SEEK_HOLE:
 212        case SEEK_DATA:
 213                ret = nfs42_proc_llseek(filep, offset, whence);
 214                if (ret != -ENOTSUPP)
 215                        return ret;
 216                fallthrough;
 217        default:
 218                return nfs_file_llseek(filep, offset, whence);
 219        }
 220}
 221
 222static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
 223{
 224        struct inode *inode = file_inode(filep);
 225        long ret;
 226
 227        if (!S_ISREG(inode->i_mode))
 228                return -EOPNOTSUPP;
 229
 230        if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
 231                return -EOPNOTSUPP;
 232
 233        ret = inode_newsize_ok(inode, offset + len);
 234        if (ret < 0)
 235                return ret;
 236
 237        if (mode & FALLOC_FL_PUNCH_HOLE)
 238                return nfs42_proc_deallocate(filep, offset, len);
 239        return nfs42_proc_allocate(filep, offset, len);
 240}
 241
 242static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
 243                struct file *dst_file, loff_t dst_off, loff_t count,
 244                unsigned int remap_flags)
 245{
 246        struct inode *dst_inode = file_inode(dst_file);
 247        struct nfs_server *server = NFS_SERVER(dst_inode);
 248        struct inode *src_inode = file_inode(src_file);
 249        unsigned int bs = server->clone_blksize;
 250        bool same_inode = false;
 251        int ret;
 252
 253        /* NFS does not support deduplication. */
 254        if (remap_flags & REMAP_FILE_DEDUP)
 255                return -EOPNOTSUPP;
 256
 257        if (remap_flags & ~REMAP_FILE_ADVISORY)
 258                return -EINVAL;
 259
 260        if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode))
 261                return -ETXTBSY;
 262
 263        /* check alignment w.r.t. clone_blksize */
 264        ret = -EINVAL;
 265        if (bs) {
 266                if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
 267                        goto out;
 268                if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
 269                        goto out;
 270        }
 271
 272        if (src_inode == dst_inode)
 273                same_inode = true;
 274
 275        /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
 276        if (same_inode) {
 277                inode_lock(src_inode);
 278        } else if (dst_inode < src_inode) {
 279                inode_lock_nested(dst_inode, I_MUTEX_PARENT);
 280                inode_lock_nested(src_inode, I_MUTEX_CHILD);
 281        } else {
 282                inode_lock_nested(src_inode, I_MUTEX_PARENT);
 283                inode_lock_nested(dst_inode, I_MUTEX_CHILD);
 284        }
 285
 286        /* flush all pending writes on both src and dst so that server
 287         * has the latest data */
 288        ret = nfs_sync_inode(src_inode);
 289        if (ret)
 290                goto out_unlock;
 291        ret = nfs_sync_inode(dst_inode);
 292        if (ret)
 293                goto out_unlock;
 294
 295        ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
 296
 297        /* truncate inode page cache of the dst range so that future reads can fetch
 298         * new data from server */
 299        if (!ret)
 300                truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
 301
 302out_unlock:
 303        if (same_inode) {
 304                inode_unlock(src_inode);
 305        } else if (dst_inode < src_inode) {
 306                inode_unlock(src_inode);
 307                inode_unlock(dst_inode);
 308        } else {
 309                inode_unlock(dst_inode);
 310                inode_unlock(src_inode);
 311        }
 312out:
 313        return ret < 0 ? ret : count;
 314}
 315
 316static int read_name_gen = 1;
 317#define SSC_READ_NAME_BODY "ssc_read_%d"
 318
 319static struct file *__nfs42_ssc_open(struct vfsmount *ss_mnt,
 320                struct nfs_fh *src_fh, nfs4_stateid *stateid)
 321{
 322        struct nfs_fattr fattr;
 323        struct file *filep, *res;
 324        struct nfs_server *server;
 325        struct inode *r_ino = NULL;
 326        struct nfs_open_context *ctx;
 327        struct nfs4_state_owner *sp;
 328        char *read_name = NULL;
 329        int len, status = 0;
 330
 331        server = NFS_SERVER(ss_mnt->mnt_root->d_inode);
 332
 333        nfs_fattr_init(&fattr);
 334
 335        status = nfs4_proc_getattr(server, src_fh, &fattr, NULL, NULL);
 336        if (status < 0) {
 337                res = ERR_PTR(status);
 338                goto out;
 339        }
 340
 341        res = ERR_PTR(-ENOMEM);
 342        len = strlen(SSC_READ_NAME_BODY) + 16;
 343        read_name = kzalloc(len, GFP_NOFS);
 344        if (read_name == NULL)
 345                goto out;
 346        snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
 347
 348        r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, &fattr,
 349                        NULL);
 350        if (IS_ERR(r_ino)) {
 351                res = ERR_CAST(r_ino);
 352                goto out_free_name;
 353        }
 354
 355        filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ,
 356                                     r_ino->i_fop);
 357        if (IS_ERR(filep)) {
 358                res = ERR_CAST(filep);
 359                goto out_free_name;
 360        }
 361        filep->f_mode |= FMODE_READ;
 362
 363        ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
 364                                        filep);
 365        if (IS_ERR(ctx)) {
 366                res = ERR_CAST(ctx);
 367                goto out_filep;
 368        }
 369
 370        res = ERR_PTR(-EINVAL);
 371        sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
 372        if (sp == NULL)
 373                goto out_ctx;
 374
 375        ctx->state = nfs4_get_open_state(r_ino, sp);
 376        if (ctx->state == NULL)
 377                goto out_stateowner;
 378
 379        set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags);
 380        memcpy(&ctx->state->open_stateid.other, &stateid->other,
 381               NFS4_STATEID_OTHER_SIZE);
 382        update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
 383        set_bit(NFS_OPEN_STATE, &ctx->state->flags);
 384
 385        nfs_file_set_open_context(filep, ctx);
 386        put_nfs_open_context(ctx);
 387
 388        file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
 389        res = filep;
 390out_free_name:
 391        kfree(read_name);
 392out:
 393        return res;
 394out_stateowner:
 395        nfs4_put_state_owner(sp);
 396out_ctx:
 397        put_nfs_open_context(ctx);
 398out_filep:
 399        fput(filep);
 400        goto out_free_name;
 401}
 402
 403static void __nfs42_ssc_close(struct file *filep)
 404{
 405        struct nfs_open_context *ctx = nfs_file_open_context(filep);
 406
 407        ctx->state->flags = 0;
 408}
 409
 410static const struct nfs4_ssc_client_ops nfs4_ssc_clnt_ops_tbl = {
 411        .sco_open = __nfs42_ssc_open,
 412        .sco_close = __nfs42_ssc_close,
 413};
 414
 415/**
 416 * nfs42_ssc_register_ops - Wrapper to register NFS_V4 ops in nfs_common
 417 *
 418 * Return values:
 419 *   None
 420 */
 421void nfs42_ssc_register_ops(void)
 422{
 423        nfs42_ssc_register(&nfs4_ssc_clnt_ops_tbl);
 424}
 425
 426/**
 427 * nfs42_ssc_unregister_ops - wrapper to un-register NFS_V4 ops in nfs_common
 428 *
 429 * Return values:
 430 *   None.
 431 */
 432void nfs42_ssc_unregister_ops(void)
 433{
 434        nfs42_ssc_unregister(&nfs4_ssc_clnt_ops_tbl);
 435}
 436#endif /* CONFIG_NFS_V4_2 */
 437
 438const struct file_operations nfs4_file_operations = {
 439        .read_iter      = nfs_file_read,
 440        .write_iter     = nfs_file_write,
 441        .mmap           = nfs_file_mmap,
 442        .open           = nfs4_file_open,
 443        .flush          = nfs4_file_flush,
 444        .release        = nfs_file_release,
 445        .fsync          = nfs_file_fsync,
 446        .lock           = nfs_lock,
 447        .flock          = nfs_flock,
 448        .splice_read    = generic_file_splice_read,
 449        .splice_write   = iter_file_splice_write,
 450        .check_flags    = nfs_check_flags,
 451        .setlease       = simple_nosetlease,
 452#ifdef CONFIG_NFS_V4_2
 453        .copy_file_range = nfs4_copy_file_range,
 454        .llseek         = nfs4_file_llseek,
 455        .fallocate      = nfs42_fallocate,
 456        .remap_file_range = nfs42_remap_file_range,
 457#else
 458        .llseek         = nfs_file_llseek,
 459#endif
 460};
 461