fs/nfs/file.c

   1 /*
   2  *  linux/fs/nfs/file.c
   3  *
   4  *  Copyright (C) 1992  Rick Sladkey
   5  *
   6  *  Changes Copyright (C) 1994 by Florian La Roche
   7  *   - Do not copy data too often around in the kernel.
   8  *   - In nfs_file_read the return value of kmalloc wasn't checked.
   9  *   - Put in a better version of read look-ahead buffering. Original idea
  10  *     and implementation by Wai S Kok elekokws@ee.nus.sg.
  11  *
  12  *  Expire cache on write to a file by Wai S Kok (Oct 1994).
  13  *
  14  *  Total rewrite of read side for new NFS buffer cache.. Linus.
  15  *
  16  *  nfs regular file handling functions
  17  */
  18
  19 #include <linux/time.h>
  20 #include <linux/kernel.h>
  21 #include <linux/errno.h>
  22 #include <linux/fcntl.h>
  23 #include <linux/stat.h>
  24 #include <linux/nfs_fs.h>
  25 #include <linux/nfs_mount.h>
  26 #include <linux/mm.h>
  27 #include <linux/slab.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/aio.h>
  31
  32 #include <asm/uaccess.h>
  33 #include <asm/system.h>
  34
  35 #include "delegation.h"
  36 #include "internal.h"
  37 #include "iostat.h"
  38
  39 #define NFSDBG_FACILITY         NFSDBG_FILE
  40
  41 static int nfs_file_open(struct inode *, struct file *);
  42 static int nfs_file_release(struct inode *, struct file *);
  43 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
  44 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  45 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
  46                                         struct pipe_inode_info *pipe,
  47                                         size_t count, unsigned int flags);
  48 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
  49                                 unsigned long nr_segs, loff_t pos);
  50 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
  51                                 unsigned long nr_segs, loff_t pos);
  52 static int  nfs_file_flush(struct file *, fl_owner_t id);
  53 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  54 static int nfs_check_flags(int flags);
  55 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  56 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  57 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
  58
  59 static struct vm_operations_struct nfs_file_vm_ops;
  60
  61 const struct file_operations nfs_file_operations = {
  62         .llseek         = nfs_file_llseek,
  63         .read           = do_sync_read,
  64         .write          = do_sync_write,
  65         .aio_read       = nfs_file_read,
  66         .aio_write      = nfs_file_write,
  67         .mmap           = nfs_file_mmap,
  68         .open           = nfs_file_open,
  69         .flush          = nfs_file_flush,
  70         .release        = nfs_file_release,
  71         .fsync          = nfs_fsync,
  72         .lock           = nfs_lock,
  73         .flock          = nfs_flock,
  74         .splice_read    = nfs_file_splice_read,
  75         .check_flags    = nfs_check_flags,
  76         .setlease       = nfs_setlease,
  77 };
  78
  79 const struct inode_operations nfs_file_inode_operations = {
  80         .permission     = nfs_permission,
  81         .getattr        = nfs_getattr,
  82         .setattr        = nfs_setattr,
  83 };
  84
  85 #ifdef CONFIG_NFS_V3
  86 const struct inode_operations nfs3_file_inode_operations = {
  87         .permission     = nfs_permission,
  88         .getattr        = nfs_getattr,
  89         .setattr        = nfs_setattr,
  90         .listxattr      = nfs3_listxattr,
  91         .getxattr       = nfs3_getxattr,
  92         .setxattr       = nfs3_setxattr,
  93         .removexattr    = nfs3_removexattr,
  94 };
  95 #endif  /* CONFIG_NFS_v3 */
  96
  97 /* Hack for future NFS swap support */
  98 #ifndef IS_SWAPFILE
  99 # define IS_SWAPFILE(inode)     (0)
 100 #endif
 101
 102 static int nfs_check_flags(int flags)
 103 {
 104         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 105                 return -EINVAL;
 106
 107         return 0;
 108 }
 109
 110 /*
 111  * Open file
 112  */
 113 static int
 114 nfs_file_open(struct inode *inode, struct file *filp)
 115 {
 116         int res;
 117
 118         res = nfs_check_flags(filp->f_flags);
 119         if (res)
 120                 return res;
 121
 122         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 123         lock_kernel();
 124         res = NFS_PROTO(inode)->file_open(inode, filp);
 125         unlock_kernel();
 126         return res;
 127 }
 128
 129 static int
 130 nfs_file_release(struct inode *inode, struct file *filp)
 131 {
 132         /* Ensure that dirty pages are flushed out with the right creds */
 133         if (filp->f_mode & FMODE_WRITE)
 134                 filemap_fdatawrite(filp->f_mapping);
 135         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 136         return NFS_PROTO(inode)->file_release(inode, filp);
 137 }
 138
 139 /**
 140  * nfs_revalidate_size - Revalidate the file size
 141  * @inode - pointer to inode struct
 142  * @file - pointer to struct file
 143  *
 144  * Revalidates the file length. This is basically a wrapper around
 145  * nfs_revalidate_inode() that takes into account the fact that we may
 146  * have cached writes (in which case we don't care about the server's
 147  * idea of what the file length is), or O_DIRECT (in which case we
 148  * shouldn't trust the cache).
 149  */
 150 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 151 {
 152         struct nfs_server *server = NFS_SERVER(inode);
 153         struct nfs_inode *nfsi = NFS_I(inode);
 154
 155         if (server->flags & NFS_MOUNT_NOAC)
 156                 goto force_reval;
 157         if (filp->f_flags & O_DIRECT)
 158                 goto force_reval;
 159         if (nfsi->npages != 0)
 160                 return 0;
 161         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 162                 return 0;
 163 force_reval:
 164         return __nfs_revalidate_inode(server, inode);
 165 }
 166
 167 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 168 {
 169         /* origin == SEEK_END => we must revalidate the cached file length */
 170         if (origin == SEEK_END) {
 171                 struct inode *inode = filp->f_mapping->host;
 172                 int retval = nfs_revalidate_file_size(inode, filp);
 173                 if (retval < 0)
 174                         return (loff_t)retval;
 175         }
 176         return remote_llseek(filp, offset, origin);
 177 }
 178
 179 /*
 180  * Flush all dirty pages, and check for write errors.
 181  *
 182  */
 183 static int
 184 nfs_file_flush(struct file *file, fl_owner_t id)
 185 {
 186         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 187         struct inode    *inode = file->f_path.dentry->d_inode;
 188         int             status;
 189
 190         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 191
 192         if ((file->f_mode & FMODE_WRITE) == 0)
 193                 return 0;
 194         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 195         lock_kernel();
 196         /* Ensure that data+attribute caches are up to date after close() */
 197         status = nfs_wb_all(inode);
 198         if (!status) {
 199                 status = ctx->error;
 200                 ctx->error = 0;
 201                 if (!status)
 202                         nfs_revalidate_inode(NFS_SERVER(inode), inode);
 203         }
 204         unlock_kernel();
 205         return status;
 206 }
 207
 208 static ssize_t
 209 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
 210                 unsigned long nr_segs, loff_t pos)
 211 {
 212         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 213         struct inode * inode = dentry->d_inode;
 214         ssize_t result;
 215         size_t count = iov_length(iov, nr_segs);
 216
 217 #ifdef CONFIG_NFS_DIRECTIO
 218         if (iocb->ki_filp->f_flags & O_DIRECT)
 219                 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
 220 #endif
 221
 222         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 223                 dentry->d_parent->d_name.name, dentry->d_name.name,
 224                 (unsigned long) count, (unsigned long) pos);
 225
 226         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 227         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
 228         if (!result)
 229                 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
 230         return result;
 231 }
 232
 233 static ssize_t
 234 nfs_file_splice_read(struct file *filp, loff_t *ppos,
 235                      struct pipe_inode_info *pipe, size_t count,
 236                      unsigned int flags)
 237 {
 238         struct dentry *dentry = filp->f_path.dentry;
 239         struct inode *inode = dentry->d_inode;
 240         ssize_t res;
 241
 242         dfprintk(VFS, "nfs: splice_read(%s/%s, %lu@%Lu)\n",
 243                 dentry->d_parent->d_name.name, dentry->d_name.name,
 244                 (unsigned long) count, (unsigned long long) *ppos);
 245
 246         res = nfs_revalidate_mapping(inode, filp->f_mapping);
 247         if (!res)
 248                 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
 249         return res;
 250 }
 251
 252 static int
 253 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 254 {
 255         struct dentry *dentry = file->f_path.dentry;
 256         struct inode *inode = dentry->d_inode;
 257         int     status;
 258
 259         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 260                 dentry->d_parent->d_name.name, dentry->d_name.name);
 261
 262         status = nfs_revalidate_mapping(inode, file->f_mapping);
 263         if (!status) {
 264                 vma->vm_ops = &nfs_file_vm_ops;
 265                 vma->vm_flags |= VM_CAN_NONLINEAR;
 266                 file_accessed(file);
 267         }
 268         return status;
 269 }
 270
 271 /*
 272  * Flush any dirty pages for this process, and check for write errors.
 273  * The return status from this call provides a reliable indication of
 274  * whether any write errors occurred for this process.
 275  */
 276 static int
 277 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 278 {
 279         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 280         struct inode *inode = dentry->d_inode;
 281         int status;
 282
 283         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 284
 285         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 286         lock_kernel();
 287         status = nfs_wb_all(inode);
 288         if (!status) {
 289                 status = ctx->error;
 290                 ctx->error = 0;
 291         }
 292         unlock_kernel();
 293         return status;
 294 }
 295
 296 /*
 297  * This does the "real" work of the write. The generic routine has
 298  * allocated the page, locked it, done all the page alignment stuff
 299  * calculations etc. Now we should just copy the data from user
 300  * space and write it back to the real medium..
 301  *
 302  * If the writer ends up delaying the write, the writer needs to
 303  * increment the page use counts until he is done with the page.
 304  */
 305 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 306 {
 307         return nfs_flush_incompatible(file, page);
 308 }
 309
 310 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 311 {
 312         long status;
 313
 314         lock_kernel();
 315         status = nfs_updatepage(file, page, offset, to-offset);
 316         unlock_kernel();
 317         return status;
 318 }
 319
 320 static void nfs_invalidate_page(struct page *page, unsigned long offset)
 321 {
 322         if (offset != 0)
 323                 return;
 324         /* Cancel any unstarted writes on this page */
 325         nfs_wb_page_cancel(page->mapping->host, page);
 326 }
 327
 328 static int nfs_release_page(struct page *page, gfp_t gfp)
 329 {
 330         /* If PagePrivate() is set, then the page is not freeable */
 331         return 0;
 332 }
 333
 334 static int nfs_launder_page(struct page *page)
 335 {
 336         return nfs_wb_page(page->mapping->host, page);
 337 }
 338
 339 const struct address_space_operations nfs_file_aops = {
 340         .readpage = nfs_readpage,
 341         .readpages = nfs_readpages,
 342         .set_page_dirty = __set_page_dirty_nobuffers,
 343         .writepage = nfs_writepage,
 344         .writepages = nfs_writepages,
 345         .prepare_write = nfs_prepare_write,
 346         .commit_write = nfs_commit_write,
 347         .invalidatepage = nfs_invalidate_page,
 348         .releasepage = nfs_release_page,
 349 #ifdef CONFIG_NFS_DIRECTIO
 350         .direct_IO = nfs_direct_IO,
 351 #endif
 352         .launder_page = nfs_launder_page,
 353 };
 354
 355 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 356 {
 357         struct file *filp = vma->vm_file;
 358         unsigned pagelen;
 359         int ret = -EINVAL;
 360
 361         lock_page(page);
 362         if (page->mapping != vma->vm_file->f_path.dentry->d_inode->i_mapping)
 363                 goto out_unlock;
 364         pagelen = nfs_page_length(page);
 365         if (pagelen == 0)
 366                 goto out_unlock;
 367         ret = nfs_prepare_write(filp, page, 0, pagelen);
 368         if (!ret)
 369                 ret = nfs_commit_write(filp, page, 0, pagelen);
 370 out_unlock:
 371         unlock_page(page);
 372         return ret;
 373 }
 374
 375 static struct vm_operations_struct nfs_file_vm_ops = {
 376         .fault = filemap_fault,
 377         .page_mkwrite = nfs_vm_page_mkwrite,
 378 };
 379
 380 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
 381                                 unsigned long nr_segs, loff_t pos)
 382 {
 383         struct dentry * dentry = iocb->ki_filp->f_path.dentry;
 384         struct inode * inode = dentry->d_inode;
 385         ssize_t result;
 386         size_t count = iov_length(iov, nr_segs);
 387
 388 #ifdef CONFIG_NFS_DIRECTIO
 389         if (iocb->ki_filp->f_flags & O_DIRECT)
 390                 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
 391 #endif
 392
 393         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%Ld)\n",
 394                 dentry->d_parent->d_name.name, dentry->d_name.name,
 395                 inode->i_ino, (unsigned long) count, (long long) pos);
 396
 397         result = -EBUSY;
 398         if (IS_SWAPFILE(inode))
 399                 goto out_swapfile;
 400         /*
 401          * O_APPEND implies that we must revalidate the file length.
 402          */
 403         if (iocb->ki_filp->f_flags & O_APPEND) {
 404                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 405                 if (result)
 406                         goto out;
 407         }
 408
 409         result = count;
 410         if (!count)
 411                 goto out;
 412
 413         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
 414         result = generic_file_aio_write(iocb, iov, nr_segs, pos);
 415         /* Return error values for O_SYNC and IS_SYNC() */
 416         if (result >= 0 && (IS_SYNC(inode) || (iocb->ki_filp->f_flags & O_SYNC))) {
 417                 int err = nfs_fsync(iocb->ki_filp, dentry, 1);
 418                 if (err < 0)
 419                         result = err;
 420         }
 421 out:
 422         return result;
 423
 424 out_swapfile:
 425         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 426         goto out;
 427 }
 428
 429 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 430 {
 431         struct inode *inode = filp->f_mapping->host;
 432         int status = 0;
 433
 434         lock_kernel();
 435         /* Try local locking first */
 436         posix_test_lock(filp, fl);
 437         if (fl->fl_type != F_UNLCK) {
 438                 /* found a conflict */
 439                 goto out;
 440         }
 441
 442         if (nfs_have_delegation(inode, FMODE_READ))
 443                 goto out_noconflict;
 444
 445         if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
 446                 goto out_noconflict;
 447
 448         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 449 out:
 450         unlock_kernel();
 451         return status;
 452 out_noconflict:
 453         fl->fl_type = F_UNLCK;
 454         goto out;
 455 }
 456
 457 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 458 {
 459         int res = 0;
 460         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 461                 case FL_POSIX:
 462                         res = posix_lock_file_wait(file, fl);
 463                         break;
 464                 case FL_FLOCK:
 465                         res = flock_lock_file_wait(file, fl);
 466                         break;
 467                 default:
 468                         BUG();
 469         }
 470         if (res < 0)
 471                 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
 472                         " - error %d!\n",
 473                                 __FUNCTION__, res);
 474         return res;
 475 }
 476
 477 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 478 {
 479         struct inode *inode = filp->f_mapping->host;
 480         int status;
 481
 482         /*
 483          * Flush all pending writes before doing anything
 484          * with locks..
 485          */
 486         nfs_sync_mapping(filp->f_mapping);
 487
 488         /* NOTE: special case
 489          *      If we're signalled while cleaning up locks on process exit, we
 490          *      still need to complete the unlock.
 491          */
 492         lock_kernel();
 493         /* Use local locking if mounted with "-onolock" */
 494         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 495                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 496         else
 497                 status = do_vfs_lock(filp, fl);
 498         unlock_kernel();
 499         return status;
 500 }
 501
 502 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 503 {
 504         struct inode *inode = filp->f_mapping->host;
 505         int status;
 506
 507         /*
 508          * Flush all pending writes before doing anything
 509          * with locks..
 510          */
 511         status = nfs_sync_mapping(filp->f_mapping);
 512         if (status != 0)
 513                 goto out;
 514
 515         lock_kernel();
 516         /* Use local locking if mounted with "-onolock" */
 517         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 518                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 519                 /* If we were signalled we still need to ensure that
 520                  * we clean up any state on the server. We therefore
 521                  * record the lock call as having succeeded in order to
 522                  * ensure that locks_remove_posix() cleans it out when
 523                  * the process exits.
 524                  */
 525                 if (status == -EINTR || status == -ERESTARTSYS)
 526                         do_vfs_lock(filp, fl);
 527         } else
 528                 status = do_vfs_lock(filp, fl);
 529         unlock_kernel();
 530         if (status < 0)
 531                 goto out;
 532         /*
 533          * Make sure we clear the cache whenever we try to get the lock.
 534          * This makes locking act as a cache coherency point.
 535          */
 536         nfs_sync_mapping(filp->f_mapping);
 537         nfs_zap_caches(inode);
 538 out:
 539         return status;
 540 }
 541
 542 /*
 543  * Lock a (portion of) a file
 544  */
 545 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 546 {
 547         struct inode * inode = filp->f_mapping->host;
 548
 549         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 550                         inode->i_sb->s_id, inode->i_ino,
 551                         fl->fl_type, fl->fl_flags,
 552                         (long long)fl->fl_start, (long long)fl->fl_end);
 553         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
 554
 555         /* No mandatory locks over NFS */
 556         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
 557             fl->fl_type != F_UNLCK)
 558                 return -ENOLCK;
 559
 560         if (IS_GETLK(cmd))
 561                 return do_getlk(filp, cmd, fl);
 562         if (fl->fl_type == F_UNLCK)
 563                 return do_unlk(filp, cmd, fl);
 564         return do_setlk(filp, cmd, fl);
 565 }
 566
 567 /*
 568  * Lock a (portion of) a file
 569  */
 570 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 571 {
 572         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 573                         filp->f_path.dentry->d_inode->i_sb->s_id,
 574                         filp->f_path.dentry->d_inode->i_ino,
 575                         fl->fl_type, fl->fl_flags);
 576
 577         /*
 578          * No BSD flocks over NFS allowed.
 579          * Note: we could try to fake a POSIX lock request here by
 580          * using ((u32) filp | 0x80000000) or some such as the pid.
 581          * Not sure whether that would be unique, though, or whether
 582          * that would break in other places.
 583          */
 584         if (!(fl->fl_flags & FL_FLOCK))
 585                 return -ENOLCK;
 586
 587         /* We're simulating flock() locks using posix locks on the server */
 588         fl->fl_owner = (fl_owner_t)filp;
 589         fl->fl_start = 0;
 590         fl->fl_end = OFFSET_MAX;
 591
 592         if (fl->fl_type == F_UNLCK)
 593                 return do_unlk(filp, cmd, fl);
 594         return do_setlk(filp, cmd, fl);
 595 }
 596
 597 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
 598 {
 599         /*
 600          * There is no protocol support for leases, so we have no way
 601          * to implement them correctly in the face of opens by other
 602          * clients.
 603          */
 604         return -EINVAL;
 605 }