linux/fs/nfs/file.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/nfs/file.c
   4 *
   5 *  Copyright (C) 1992  Rick Sladkey
   6 *
   7 *  Changes Copyright (C) 1994 by Florian La Roche
   8 *   - Do not copy data too often around in the kernel.
   9 *   - In nfs_file_read the return value of kmalloc wasn't checked.
  10 *   - Put in a better version of read look-ahead buffering. Original idea
  11 *     and implementation by Wai S Kok elekokws@ee.nus.sg.
  12 *
  13 *  Expire cache on write to a file by Wai S Kok (Oct 1994).
  14 *
  15 *  Total rewrite of read side for new NFS buffer cache.. Linus.
  16 *
  17 *  nfs regular file handling functions
  18 */
  19
  20#include <linux/module.h>
  21#include <linux/time.h>
  22#include <linux/kernel.h>
  23#include <linux/errno.h>
  24#include <linux/fcntl.h>
  25#include <linux/stat.h>
  26#include <linux/nfs_fs.h>
  27#include <linux/nfs_mount.h>
  28#include <linux/mm.h>
  29#include <linux/pagemap.h>
  30#include <linux/gfp.h>
  31#include <linux/swap.h>
  32
  33#include <linux/uaccess.h>
  34
  35#include "delegation.h"
  36#include "internal.h"
  37#include "iostat.h"
  38#include "fscache.h"
  39#include "pnfs.h"
  40
  41#include "nfstrace.h"
  42
  43#define NFSDBG_FACILITY         NFSDBG_FILE
  44
  45static const struct vm_operations_struct nfs_file_vm_ops;
  46
  47/* Hack for future NFS swap support */
  48#ifndef IS_SWAPFILE
  49# define IS_SWAPFILE(inode)     (0)
  50#endif
  51
  52int nfs_check_flags(int flags)
  53{
  54        if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
  55                return -EINVAL;
  56
  57        return 0;
  58}
  59EXPORT_SYMBOL_GPL(nfs_check_flags);
  60
  61/*
  62 * Open file
  63 */
  64static int
  65nfs_file_open(struct inode *inode, struct file *filp)
  66{
  67        int res;
  68
  69        dprintk("NFS: open file(%pD2)\n", filp);
  70
  71        nfs_inc_stats(inode, NFSIOS_VFSOPEN);
  72        res = nfs_check_flags(filp->f_flags);
  73        if (res)
  74                return res;
  75
  76        res = nfs_open(inode, filp);
  77        return res;
  78}
  79
  80int
  81nfs_file_release(struct inode *inode, struct file *filp)
  82{
  83        dprintk("NFS: release(%pD2)\n", filp);
  84
  85        nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
  86        nfs_file_clear_open_context(filp);
  87        return 0;
  88}
  89EXPORT_SYMBOL_GPL(nfs_file_release);
  90
  91/**
  92 * nfs_revalidate_size - Revalidate the file size
  93 * @inode: pointer to inode struct
  94 * @filp: pointer to struct file
  95 *
  96 * Revalidates the file length. This is basically a wrapper around
  97 * nfs_revalidate_inode() that takes into account the fact that we may
  98 * have cached writes (in which case we don't care about the server's
  99 * idea of what the file length is), or O_DIRECT (in which case we
 100 * shouldn't trust the cache).
 101 */
 102static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 103{
 104        struct nfs_server *server = NFS_SERVER(inode);
 105
 106        if (filp->f_flags & O_DIRECT)
 107                goto force_reval;
 108        if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
 109                goto force_reval;
 110        return 0;
 111force_reval:
 112        return __nfs_revalidate_inode(server, inode);
 113}
 114
 115loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
 116{
 117        dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
 118                        filp, offset, whence);
 119
 120        /*
 121         * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
 122         * the cached file length
 123         */
 124        if (whence != SEEK_SET && whence != SEEK_CUR) {
 125                struct inode *inode = filp->f_mapping->host;
 126
 127                int retval = nfs_revalidate_file_size(inode, filp);
 128                if (retval < 0)
 129                        return (loff_t)retval;
 130        }
 131
 132        return generic_file_llseek(filp, offset, whence);
 133}
 134EXPORT_SYMBOL_GPL(nfs_file_llseek);
 135
 136/*
 137 * Flush all dirty pages, and check for write errors.
 138 */
 139static int
 140nfs_file_flush(struct file *file, fl_owner_t id)
 141{
 142        struct inode    *inode = file_inode(file);
 143
 144        dprintk("NFS: flush(%pD2)\n", file);
 145
 146        nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 147        if ((file->f_mode & FMODE_WRITE) == 0)
 148                return 0;
 149
 150        /* Flush writes to the server and return any errors */
 151        return nfs_wb_all(inode);
 152}
 153
 154ssize_t
 155nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
 156{
 157        struct inode *inode = file_inode(iocb->ki_filp);
 158        ssize_t result;
 159
 160        if (iocb->ki_flags & IOCB_DIRECT)
 161                return nfs_file_direct_read(iocb, to);
 162
 163        dprintk("NFS: read(%pD2, %zu@%lu)\n",
 164                iocb->ki_filp,
 165                iov_iter_count(to), (unsigned long) iocb->ki_pos);
 166
 167        nfs_start_io_read(inode);
 168        result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 169        if (!result) {
 170                result = generic_file_read_iter(iocb, to);
 171                if (result > 0)
 172                        nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
 173        }
 174        nfs_end_io_read(inode);
 175        return result;
 176}
 177EXPORT_SYMBOL_GPL(nfs_file_read);
 178
 179int
 180nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 181{
 182        struct inode *inode = file_inode(file);
 183        int     status;
 184
 185        dprintk("NFS: mmap(%pD2)\n", file);
 186
 187        /* Note: generic_file_mmap() returns ENOSYS on nommu systems
 188         *       so we call that before revalidating the mapping
 189         */
 190        status = generic_file_mmap(file, vma);
 191        if (!status) {
 192                vma->vm_ops = &nfs_file_vm_ops;
 193                status = nfs_revalidate_mapping(inode, file->f_mapping);
 194        }
 195        return status;
 196}
 197EXPORT_SYMBOL_GPL(nfs_file_mmap);
 198
 199/*
 200 * Flush any dirty pages for this process, and check for write errors.
 201 * The return status from this call provides a reliable indication of
 202 * whether any write errors occurred for this process.
 203 */
 204static int
 205nfs_file_fsync_commit(struct file *file, int datasync)
 206{
 207        struct inode *inode = file_inode(file);
 208        int ret;
 209
 210        dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
 211
 212        nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 213        ret = nfs_commit_inode(inode, FLUSH_SYNC);
 214        if (ret < 0)
 215                return ret;
 216        return file_check_and_advance_wb_err(file);
 217}
 218
 219int
 220nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 221{
 222        struct nfs_open_context *ctx = nfs_file_open_context(file);
 223        struct inode *inode = file_inode(file);
 224        int ret;
 225
 226        trace_nfs_fsync_enter(inode);
 227
 228        for (;;) {
 229                ret = file_write_and_wait_range(file, start, end);
 230                if (ret != 0)
 231                        break;
 232                ret = nfs_file_fsync_commit(file, datasync);
 233                if (ret != 0)
 234                        break;
 235                ret = pnfs_sync_inode(inode, !!datasync);
 236                if (ret != 0)
 237                        break;
 238                if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
 239                        break;
 240                /*
 241                 * If nfs_file_fsync_commit detected a server reboot, then
 242                 * resend all dirty pages that might have been covered by
 243                 * the NFS_CONTEXT_RESEND_WRITES flag
 244                 */
 245                start = 0;
 246                end = LLONG_MAX;
 247        }
 248
 249        trace_nfs_fsync_exit(inode, ret);
 250        return ret;
 251}
 252EXPORT_SYMBOL_GPL(nfs_file_fsync);
 253
 254/*
 255 * Decide whether a read/modify/write cycle may be more efficient
 256 * then a modify/write/read cycle when writing to a page in the
 257 * page cache.
 258 *
 259 * Some pNFS layout drivers can only read/write at a certain block
 260 * granularity like all block devices and therefore we must perform
 261 * read/modify/write whenever a page hasn't read yet and the data
 262 * to be written there is not aligned to a block boundary and/or
 263 * smaller than the block size.
 264 *
 265 * The modify/write/read cycle may occur if a page is read before
 266 * being completely filled by the writer.  In this situation, the
 267 * page must be completely written to stable storage on the server
 268 * before it can be refilled by reading in the page from the server.
 269 * This can lead to expensive, small, FILE_SYNC mode writes being
 270 * done.
 271 *
 272 * It may be more efficient to read the page first if the file is
 273 * open for reading in addition to writing, the page is not marked
 274 * as Uptodate, it is not dirty or waiting to be committed,
 275 * indicating that it was previously allocated and then modified,
 276 * that there were valid bytes of data in that range of the file,
 277 * and that the new data won't completely replace the old data in
 278 * that range of the file.
 279 */
 280static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
 281{
 282        unsigned int pglen = nfs_page_length(page);
 283        unsigned int offset = pos & (PAGE_SIZE - 1);
 284        unsigned int end = offset + len;
 285
 286        return !pglen || (end >= pglen && !offset);
 287}
 288
 289static bool nfs_want_read_modify_write(struct file *file, struct page *page,
 290                        loff_t pos, unsigned int len)
 291{
 292        /*
 293         * Up-to-date pages, those with ongoing or full-page write
 294         * don't need read/modify/write
 295         */
 296        if (PageUptodate(page) || PagePrivate(page) ||
 297            nfs_full_page_write(page, pos, len))
 298                return false;
 299
 300        if (pnfs_ld_read_whole_page(file->f_mapping->host))
 301                return true;
 302        /* Open for reading too? */
 303        if (file->f_mode & FMODE_READ)
 304                return true;
 305        return false;
 306}
 307
 308/*
 309 * This does the "real" work of the write. We must allocate and lock the
 310 * page to be sent back to the generic routine, which then copies the
 311 * data from user space.
 312 *
 313 * If the writer ends up delaying the write, the writer needs to
 314 * increment the page use counts until he is done with the page.
 315 */
 316static int nfs_write_begin(struct file *file, struct address_space *mapping,
 317                        loff_t pos, unsigned len, unsigned flags,
 318                        struct page **pagep, void **fsdata)
 319{
 320        int ret;
 321        pgoff_t index = pos >> PAGE_SHIFT;
 322        struct page *page;
 323        int once_thru = 0;
 324
 325        dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
 326                file, mapping->host->i_ino, len, (long long) pos);
 327
 328start:
 329        page = grab_cache_page_write_begin(mapping, index, flags);
 330        if (!page)
 331                return -ENOMEM;
 332        *pagep = page;
 333
 334        ret = nfs_flush_incompatible(file, page);
 335        if (ret) {
 336                unlock_page(page);
 337                put_page(page);
 338        } else if (!once_thru &&
 339                   nfs_want_read_modify_write(file, page, pos, len)) {
 340                once_thru = 1;
 341                ret = nfs_readpage(file, page);
 342                put_page(page);
 343                if (!ret)
 344                        goto start;
 345        }
 346        return ret;
 347}
 348
 349static int nfs_write_end(struct file *file, struct address_space *mapping,
 350                        loff_t pos, unsigned len, unsigned copied,
 351                        struct page *page, void *fsdata)
 352{
 353        unsigned offset = pos & (PAGE_SIZE - 1);
 354        struct nfs_open_context *ctx = nfs_file_open_context(file);
 355        int status;
 356
 357        dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
 358                file, mapping->host->i_ino, len, (long long) pos);
 359
 360        /*
 361         * Zero any uninitialised parts of the page, and then mark the page
 362         * as up to date if it turns out that we're extending the file.
 363         */
 364        if (!PageUptodate(page)) {
 365                unsigned pglen = nfs_page_length(page);
 366                unsigned end = offset + copied;
 367
 368                if (pglen == 0) {
 369                        zero_user_segments(page, 0, offset,
 370                                        end, PAGE_SIZE);
 371                        SetPageUptodate(page);
 372                } else if (end >= pglen) {
 373                        zero_user_segment(page, end, PAGE_SIZE);
 374                        if (offset == 0)
 375                                SetPageUptodate(page);
 376                } else
 377                        zero_user_segment(page, pglen, PAGE_SIZE);
 378        }
 379
 380        status = nfs_updatepage(file, page, offset, copied);
 381
 382        unlock_page(page);
 383        put_page(page);
 384
 385        if (status < 0)
 386                return status;
 387        NFS_I(mapping->host)->write_io += copied;
 388
 389        if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
 390                status = nfs_wb_all(mapping->host);
 391                if (status < 0)
 392                        return status;
 393        }
 394
 395        return copied;
 396}
 397
 398/*
 399 * Partially or wholly invalidate a page
 400 * - Release the private state associated with a page if undergoing complete
 401 *   page invalidation
 402 * - Called if either PG_private or PG_fscache is set on the page
 403 * - Caller holds page lock
 404 */
 405static void nfs_invalidate_page(struct page *page, unsigned int offset,
 406                                unsigned int length)
 407{
 408        dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
 409                 page, offset, length);
 410
 411        if (offset != 0 || length < PAGE_SIZE)
 412                return;
 413        /* Cancel any unstarted writes on this page */
 414        nfs_wb_page_cancel(page_file_mapping(page)->host, page);
 415
 416        nfs_fscache_invalidate_page(page, page->mapping->host);
 417}
 418
 419/*
 420 * Attempt to release the private state associated with a page
 421 * - Called if either PG_private or PG_fscache is set on the page
 422 * - Caller holds page lock
 423 * - Return true (may release page) or false (may not)
 424 */
 425static int nfs_release_page(struct page *page, gfp_t gfp)
 426{
 427        dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
 428
 429        /* If PagePrivate() is set, then the page is not freeable */
 430        if (PagePrivate(page))
 431                return 0;
 432        return nfs_fscache_release_page(page, gfp);
 433}
 434
 435static void nfs_check_dirty_writeback(struct page *page,
 436                                bool *dirty, bool *writeback)
 437{
 438        struct nfs_inode *nfsi;
 439        struct address_space *mapping = page_file_mapping(page);
 440
 441        if (!mapping || PageSwapCache(page))
 442                return;
 443
 444        /*
 445         * Check if an unstable page is currently being committed and
 446         * if so, have the VM treat it as if the page is under writeback
 447         * so it will not block due to pages that will shortly be freeable.
 448         */
 449        nfsi = NFS_I(mapping->host);
 450        if (atomic_read(&nfsi->commit_info.rpcs_out)) {
 451                *writeback = true;
 452                return;
 453        }
 454
 455        /*
 456         * If PagePrivate() is set, then the page is not freeable and as the
 457         * inode is not being committed, it's not going to be cleaned in the
 458         * near future so treat it as dirty
 459         */
 460        if (PagePrivate(page))
 461                *dirty = true;
 462}
 463
 464/*
 465 * Attempt to clear the private state associated with a page when an error
 466 * occurs that requires the cached contents of an inode to be written back or
 467 * destroyed
 468 * - Called if either PG_private or fscache is set on the page
 469 * - Caller holds page lock
 470 * - Return 0 if successful, -error otherwise
 471 */
 472static int nfs_launder_page(struct page *page)
 473{
 474        struct inode *inode = page_file_mapping(page)->host;
 475        struct nfs_inode *nfsi = NFS_I(inode);
 476
 477        dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
 478                inode->i_ino, (long long)page_offset(page));
 479
 480        nfs_fscache_wait_on_page_write(nfsi, page);
 481        return nfs_wb_page(inode, page);
 482}
 483
 484static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
 485                                                sector_t *span)
 486{
 487        unsigned long blocks;
 488        long long isize;
 489        struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
 490        struct inode *inode = file->f_mapping->host;
 491
 492        spin_lock(&inode->i_lock);
 493        blocks = inode->i_blocks;
 494        isize = inode->i_size;
 495        spin_unlock(&inode->i_lock);
 496        if (blocks*512 < isize) {
 497                pr_warn("swap activate: swapfile has holes\n");
 498                return -EINVAL;
 499        }
 500
 501        *span = sis->pages;
 502
 503        return rpc_clnt_swap_activate(clnt);
 504}
 505
 506static void nfs_swap_deactivate(struct file *file)
 507{
 508        struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
 509
 510        rpc_clnt_swap_deactivate(clnt);
 511}
 512
 513const struct address_space_operations nfs_file_aops = {
 514        .readpage = nfs_readpage,
 515        .readpages = nfs_readpages,
 516        .set_page_dirty = __set_page_dirty_nobuffers,
 517        .writepage = nfs_writepage,
 518        .writepages = nfs_writepages,
 519        .write_begin = nfs_write_begin,
 520        .write_end = nfs_write_end,
 521        .invalidatepage = nfs_invalidate_page,
 522        .releasepage = nfs_release_page,
 523        .direct_IO = nfs_direct_IO,
 524#ifdef CONFIG_MIGRATION
 525        .migratepage = nfs_migrate_page,
 526#endif
 527        .launder_page = nfs_launder_page,
 528        .is_dirty_writeback = nfs_check_dirty_writeback,
 529        .error_remove_page = generic_error_remove_page,
 530        .swap_activate = nfs_swap_activate,
 531        .swap_deactivate = nfs_swap_deactivate,
 532};
 533
 534/*
 535 * Notification that a PTE pointing to an NFS page is about to be made
 536 * writable, implying that someone is about to modify the page through a
 537 * shared-writable mapping
 538 */
 539static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
 540{
 541        struct page *page = vmf->page;
 542        struct file *filp = vmf->vma->vm_file;
 543        struct inode *inode = file_inode(filp);
 544        unsigned pagelen;
 545        vm_fault_t ret = VM_FAULT_NOPAGE;
 546        struct address_space *mapping;
 547
 548        dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
 549                filp, filp->f_mapping->host->i_ino,
 550                (long long)page_offset(page));
 551
 552        sb_start_pagefault(inode->i_sb);
 553
 554        /* make sure the cache has finished storing the page */
 555        nfs_fscache_wait_on_page_write(NFS_I(inode), page);
 556
 557        wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
 558                        nfs_wait_bit_killable, TASK_KILLABLE);
 559
 560        lock_page(page);
 561        mapping = page_file_mapping(page);
 562        if (mapping != inode->i_mapping)
 563                goto out_unlock;
 564
 565        wait_on_page_writeback(page);
 566
 567        pagelen = nfs_page_length(page);
 568        if (pagelen == 0)
 569                goto out_unlock;
 570
 571        ret = VM_FAULT_LOCKED;
 572        if (nfs_flush_incompatible(filp, page) == 0 &&
 573            nfs_updatepage(filp, page, 0, pagelen) == 0)
 574                goto out;
 575
 576        ret = VM_FAULT_SIGBUS;
 577out_unlock:
 578        unlock_page(page);
 579out:
 580        sb_end_pagefault(inode->i_sb);
 581        return ret;
 582}
 583
 584static const struct vm_operations_struct nfs_file_vm_ops = {
 585        .fault = filemap_fault,
 586        .map_pages = filemap_map_pages,
 587        .page_mkwrite = nfs_vm_page_mkwrite,
 588};
 589
 590static int nfs_need_check_write(struct file *filp, struct inode *inode)
 591{
 592        struct nfs_open_context *ctx;
 593
 594        ctx = nfs_file_open_context(filp);
 595        if (nfs_ctx_key_to_expire(ctx, inode))
 596                return 1;
 597        return 0;
 598}
 599
 600ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
 601{
 602        struct file *file = iocb->ki_filp;
 603        struct inode *inode = file_inode(file);
 604        unsigned long written = 0;
 605        ssize_t result;
 606
 607        result = nfs_key_timeout_notify(file, inode);
 608        if (result)
 609                return result;
 610
 611        if (iocb->ki_flags & IOCB_DIRECT)
 612                return nfs_file_direct_write(iocb, from);
 613
 614        dprintk("NFS: write(%pD2, %zu@%Ld)\n",
 615                file, iov_iter_count(from), (long long) iocb->ki_pos);
 616
 617        if (IS_SWAPFILE(inode))
 618                goto out_swapfile;
 619        /*
 620         * O_APPEND implies that we must revalidate the file length.
 621         */
 622        if (iocb->ki_flags & IOCB_APPEND) {
 623                result = nfs_revalidate_file_size(inode, file);
 624                if (result)
 625                        goto out;
 626        }
 627        if (iocb->ki_pos > i_size_read(inode))
 628                nfs_revalidate_mapping(inode, file->f_mapping);
 629
 630        nfs_start_io_write(inode);
 631        result = generic_write_checks(iocb, from);
 632        if (result > 0) {
 633                current->backing_dev_info = inode_to_bdi(inode);
 634                result = generic_perform_write(file, from, iocb->ki_pos);
 635                current->backing_dev_info = NULL;
 636        }
 637        nfs_end_io_write(inode);
 638        if (result <= 0)
 639                goto out;
 640
 641        written = result;
 642        iocb->ki_pos += written;
 643        result = generic_write_sync(iocb, written);
 644        if (result < 0)
 645                goto out;
 646
 647        /* Return error values */
 648        if (nfs_need_check_write(file, inode)) {
 649                int err = nfs_wb_all(inode);
 650                if (err < 0)
 651                        result = err;
 652        }
 653        nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
 654out:
 655        return result;
 656
 657out_swapfile:
 658        printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 659        return -ETXTBSY;
 660}
 661EXPORT_SYMBOL_GPL(nfs_file_write);
 662
 663static int
 664do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
 665{
 666        struct inode *inode = filp->f_mapping->host;
 667        int status = 0;
 668        unsigned int saved_type = fl->fl_type;
 669
 670        /* Try local locking first */
 671        posix_test_lock(filp, fl);
 672        if (fl->fl_type != F_UNLCK) {
 673                /* found a conflict */
 674                goto out;
 675        }
 676        fl->fl_type = saved_type;
 677
 678        if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
 679                goto out_noconflict;
 680
 681        if (is_local)
 682                goto out_noconflict;
 683
 684        status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 685out:
 686        return status;
 687out_noconflict:
 688        fl->fl_type = F_UNLCK;
 689        goto out;
 690}
 691
 692static int
 693do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
 694{
 695        struct inode *inode = filp->f_mapping->host;
 696        struct nfs_lock_context *l_ctx;
 697        int status;
 698
 699        /*
 700         * Flush all pending writes before doing anything
 701         * with locks..
 702         */
 703        nfs_wb_all(inode);
 704
 705        l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
 706        if (!IS_ERR(l_ctx)) {
 707                status = nfs_iocounter_wait(l_ctx);
 708                nfs_put_lock_context(l_ctx);
 709                /*  NOTE: special case
 710                 *      If we're signalled while cleaning up locks on process exit, we
 711                 *      still need to complete the unlock.
 712                 */
 713                if (status < 0 && !(fl->fl_flags & FL_CLOSE))
 714                        return status;
 715        }
 716
 717        /*
 718         * Use local locking if mounted with "-onolock" or with appropriate
 719         * "-olocal_lock="
 720         */
 721        if (!is_local)
 722                status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 723        else
 724                status = locks_lock_file_wait(filp, fl);
 725        return status;
 726}
 727
 728static int
 729do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
 730{
 731        struct inode *inode = filp->f_mapping->host;
 732        int status;
 733
 734        /*
 735         * Flush all pending writes before doing anything
 736         * with locks..
 737         */
 738        status = nfs_sync_mapping(filp->f_mapping);
 739        if (status != 0)
 740                goto out;
 741
 742        /*
 743         * Use local locking if mounted with "-onolock" or with appropriate
 744         * "-olocal_lock="
 745         */
 746        if (!is_local)
 747                status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 748        else
 749                status = locks_lock_file_wait(filp, fl);
 750        if (status < 0)
 751                goto out;
 752
 753        /*
 754         * Invalidate cache to prevent missing any changes.  If
 755         * the file is mapped, clear the page cache as well so
 756         * those mappings will be loaded.
 757         *
 758         * This makes locking act as a cache coherency point.
 759         */
 760        nfs_sync_mapping(filp->f_mapping);
 761        if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
 762                nfs_zap_caches(inode);
 763                if (mapping_mapped(filp->f_mapping))
 764                        nfs_revalidate_mapping(inode, filp->f_mapping);
 765        }
 766out:
 767        return status;
 768}
 769
 770/*
 771 * Lock a (portion of) a file
 772 */
 773int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 774{
 775        struct inode *inode = filp->f_mapping->host;
 776        int ret = -ENOLCK;
 777        int is_local = 0;
 778
 779        dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
 780                        filp, fl->fl_type, fl->fl_flags,
 781                        (long long)fl->fl_start, (long long)fl->fl_end);
 782
 783        nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 784
 785        /* No mandatory locks over NFS */
 786        if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
 787                goto out_err;
 788
 789        if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
 790                is_local = 1;
 791
 792        if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
 793                ret = NFS_PROTO(inode)->lock_check_bounds(fl);
 794                if (ret < 0)
 795                        goto out_err;
 796        }
 797
 798        if (IS_GETLK(cmd))
 799                ret = do_getlk(filp, cmd, fl, is_local);
 800        else if (fl->fl_type == F_UNLCK)
 801                ret = do_unlk(filp, cmd, fl, is_local);
 802        else
 803                ret = do_setlk(filp, cmd, fl, is_local);
 804out_err:
 805        return ret;
 806}
 807EXPORT_SYMBOL_GPL(nfs_lock);
 808
 809/*
 810 * Lock a (portion of) a file
 811 */
 812int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 813{
 814        struct inode *inode = filp->f_mapping->host;
 815        int is_local = 0;
 816
 817        dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
 818                        filp, fl->fl_type, fl->fl_flags);
 819
 820        if (!(fl->fl_flags & FL_FLOCK))
 821                return -ENOLCK;
 822
 823        /*
 824         * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
 825         * any standard. In principle we might be able to support LOCK_MAND
 826         * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
 827         * NFS code is not set up for it.
 828         */
 829        if (fl->fl_type & LOCK_MAND)
 830                return -EINVAL;
 831
 832        if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
 833                is_local = 1;
 834
 835        /* We're simulating flock() locks using posix locks on the server */
 836        if (fl->fl_type == F_UNLCK)
 837                return do_unlk(filp, cmd, fl, is_local);
 838        return do_setlk(filp, cmd, fl, is_local);
 839}
 840EXPORT_SYMBOL_GPL(nfs_flock);
 841
 842const struct file_operations nfs_file_operations = {
 843        .llseek         = nfs_file_llseek,
 844        .read_iter      = nfs_file_read,
 845        .write_iter     = nfs_file_write,
 846        .mmap           = nfs_file_mmap,
 847        .open           = nfs_file_open,
 848        .flush          = nfs_file_flush,
 849        .release        = nfs_file_release,
 850        .fsync          = nfs_file_fsync,
 851        .lock           = nfs_lock,
 852        .flock          = nfs_flock,
 853        .splice_read    = generic_file_splice_read,
 854        .splice_write   = iter_file_splice_write,
 855        .check_flags    = nfs_check_flags,
 856        .setlease       = simple_nosetlease,
 857};
 858EXPORT_SYMBOL_GPL(nfs_file_operations);
 859