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
  31 #include <asm/uaccess.h>
  32 #include <asm/system.h>
  33
  34 #include "delegation.h"
  35
  36 #define NFSDBG_FACILITY         NFSDBG_FILE
  37
  38 static int nfs_file_open(struct inode *, struct file *);
  39 static int nfs_file_release(struct inode *, struct file *);
  40 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
  41 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  42 static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *);
  43 static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t);
  44 static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t);
  45 static int  nfs_file_flush(struct file *);
  46 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  47 static int nfs_check_flags(int flags);
  48 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
  49 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
  50
  51 struct file_operations nfs_file_operations = {
  52         .llseek         = nfs_file_llseek,
  53         .read           = do_sync_read,
  54         .write          = do_sync_write,
  55         .aio_read               = nfs_file_read,
  56         .aio_write              = nfs_file_write,
  57         .mmap           = nfs_file_mmap,
  58         .open           = nfs_file_open,
  59         .flush          = nfs_file_flush,
  60         .release        = nfs_file_release,
  61         .fsync          = nfs_fsync,
  62         .lock           = nfs_lock,
  63         .flock          = nfs_flock,
  64         .sendfile       = nfs_file_sendfile,
  65         .check_flags    = nfs_check_flags,
  66 };
  67
  68 struct inode_operations nfs_file_inode_operations = {
  69         .permission     = nfs_permission,
  70         .getattr        = nfs_getattr,
  71         .setattr        = nfs_setattr,
  72 };
  73
  74 #ifdef CONFIG_NFS_V3
  75 struct inode_operations nfs3_file_inode_operations = {
  76         .permission     = nfs_permission,
  77         .getattr        = nfs_getattr,
  78         .setattr        = nfs_setattr,
  79         .listxattr      = nfs3_listxattr,
  80         .getxattr       = nfs3_getxattr,
  81         .setxattr       = nfs3_setxattr,
  82         .removexattr    = nfs3_removexattr,
  83 };
  84 #endif  /* CONFIG_NFS_v3 */
  85
  86 /* Hack for future NFS swap support */
  87 #ifndef IS_SWAPFILE
  88 # define IS_SWAPFILE(inode)     (0)
  89 #endif
  90
  91 static int nfs_check_flags(int flags)
  92 {
  93         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
  94                 return -EINVAL;
  95
  96         return 0;
  97 }
  98
  99 /*
 100  * Open file
 101  */
 102 static int
 103 nfs_file_open(struct inode *inode, struct file *filp)
 104 {
 105         struct nfs_server *server = NFS_SERVER(inode);
 106         int (*open)(struct inode *, struct file *);
 107         int res;
 108
 109         res = nfs_check_flags(filp->f_flags);
 110         if (res)
 111                 return res;
 112
 113         lock_kernel();
 114         /* Do NFSv4 open() call */
 115         if ((open = server->rpc_ops->file_open) != NULL)
 116                 res = open(inode, filp);
 117         unlock_kernel();
 118         return res;
 119 }
 120
 121 static int
 122 nfs_file_release(struct inode *inode, struct file *filp)
 123 {
 124         /* Ensure that dirty pages are flushed out with the right creds */
 125         if (filp->f_mode & FMODE_WRITE)
 126                 filemap_fdatawrite(filp->f_mapping);
 127         return NFS_PROTO(inode)->file_release(inode, filp);
 128 }
 129
 130 /**
 131  * nfs_revalidate_file - Revalidate the page cache & related metadata
 132  * @inode - pointer to inode struct
 133  * @file - pointer to file
 134  */
 135 static int nfs_revalidate_file(struct inode *inode, struct file *filp)
 136 {
 137         int retval = 0;
 138
 139         if ((NFS_FLAGS(inode) & NFS_INO_REVAL_PAGECACHE) || nfs_attribute_timeout(inode))
 140                 retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 141         nfs_revalidate_mapping(inode, filp->f_mapping);
 142         return 0;
 143 }
 144
 145 /**
 146  * nfs_revalidate_size - Revalidate the file size
 147  * @inode - pointer to inode struct
 148  * @file - pointer to struct file
 149  *
 150  * Revalidates the file length. This is basically a wrapper around
 151  * nfs_revalidate_inode() that takes into account the fact that we may
 152  * have cached writes (in which case we don't care about the server's
 153  * idea of what the file length is), or O_DIRECT (in which case we
 154  * shouldn't trust the cache).
 155  */
 156 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 157 {
 158         struct nfs_server *server = NFS_SERVER(inode);
 159         struct nfs_inode *nfsi = NFS_I(inode);
 160
 161         if (server->flags & NFS_MOUNT_NOAC)
 162                 goto force_reval;
 163         if (filp->f_flags & O_DIRECT)
 164                 goto force_reval;
 165         if (nfsi->npages != 0)
 166                 return 0;
 167         if (!(NFS_FLAGS(inode) & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 168                 return 0;
 169 force_reval:
 170         return __nfs_revalidate_inode(server, inode);
 171 }
 172
 173 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 174 {
 175         /* origin == SEEK_END => we must revalidate the cached file length */
 176         if (origin == 2) {
 177                 struct inode *inode = filp->f_mapping->host;
 178                 int retval = nfs_revalidate_file_size(inode, filp);
 179                 if (retval < 0)
 180                         return (loff_t)retval;
 181         }
 182         return remote_llseek(filp, offset, origin);
 183 }
 184
 185 /*
 186  * Flush all dirty pages, and check for write errors.
 187  *
 188  */
 189 static int
 190 nfs_file_flush(struct file *file)
 191 {
 192         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 193         struct inode    *inode = file->f_dentry->d_inode;
 194         int             status;
 195
 196         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 197
 198         if ((file->f_mode & FMODE_WRITE) == 0)
 199                 return 0;
 200         lock_kernel();
 201         /* Ensure that data+attribute caches are up to date after close() */
 202         status = nfs_wb_all(inode);
 203         if (!status) {
 204                 status = ctx->error;
 205                 ctx->error = 0;
 206                 if (!status && !nfs_have_delegation(inode, FMODE_READ))
 207                         __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 208         }
 209         unlock_kernel();
 210         return status;
 211 }
 212
 213 static ssize_t
 214 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
 215 {
 216         struct dentry * dentry = iocb->ki_filp->f_dentry;
 217         struct inode * inode = dentry->d_inode;
 218         ssize_t result;
 219
 220 #ifdef CONFIG_NFS_DIRECTIO
 221         if (iocb->ki_filp->f_flags & O_DIRECT)
 222                 return nfs_file_direct_read(iocb, buf, count, pos);
 223 #endif
 224
 225         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 226                 dentry->d_parent->d_name.name, dentry->d_name.name,
 227                 (unsigned long) count, (unsigned long) pos);
 228
 229         result = nfs_revalidate_file(inode, iocb->ki_filp);
 230         if (!result)
 231                 result = generic_file_aio_read(iocb, buf, count, pos);
 232         return result;
 233 }
 234
 235 static ssize_t
 236 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 237                 read_actor_t actor, void *target)
 238 {
 239         struct dentry *dentry = filp->f_dentry;
 240         struct inode *inode = dentry->d_inode;
 241         ssize_t res;
 242
 243         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 244                 dentry->d_parent->d_name.name, dentry->d_name.name,
 245                 (unsigned long) count, (unsigned long long) *ppos);
 246
 247         res = nfs_revalidate_file(inode, filp);
 248         if (!res)
 249                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 250         return res;
 251 }
 252
 253 static int
 254 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 255 {
 256         struct dentry *dentry = file->f_dentry;
 257         struct inode *inode = dentry->d_inode;
 258         int     status;
 259
 260         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 261                 dentry->d_parent->d_name.name, dentry->d_name.name);
 262
 263         status = nfs_revalidate_file(inode, file);
 264         if (!status)
 265                 status = generic_file_mmap(file, vma);
 266         return status;
 267 }
 268
 269 /*
 270  * Flush any dirty pages for this process, and check for write errors.
 271  * The return status from this call provides a reliable indication of
 272  * whether any write errors occurred for this process.
 273  */
 274 static int
 275 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 276 {
 277         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 278         struct inode *inode = dentry->d_inode;
 279         int status;
 280
 281         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 282
 283         lock_kernel();
 284         status = nfs_wb_all(inode);
 285         if (!status) {
 286                 status = ctx->error;
 287                 ctx->error = 0;
 288         }
 289         unlock_kernel();
 290         return status;
 291 }
 292
 293 /*
 294  * This does the "real" work of the write. The generic routine has
 295  * allocated the page, locked it, done all the page alignment stuff
 296  * calculations etc. Now we should just copy the data from user
 297  * space and write it back to the real medium..
 298  *
 299  * If the writer ends up delaying the write, the writer needs to
 300  * increment the page use counts until he is done with the page.
 301  */
 302 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 303 {
 304         return nfs_flush_incompatible(file, page);
 305 }
 306
 307 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 308 {
 309         long status;
 310
 311         lock_kernel();
 312         status = nfs_updatepage(file, page, offset, to-offset);
 313         unlock_kernel();
 314         return status;
 315 }
 316
 317 struct address_space_operations nfs_file_aops = {
 318         .readpage = nfs_readpage,
 319         .readpages = nfs_readpages,
 320         .set_page_dirty = __set_page_dirty_nobuffers,
 321         .writepage = nfs_writepage,
 322         .writepages = nfs_writepages,
 323         .prepare_write = nfs_prepare_write,
 324         .commit_write = nfs_commit_write,
 325 #ifdef CONFIG_NFS_DIRECTIO
 326         .direct_IO = nfs_direct_IO,
 327 #endif
 328 };
 329
 330 /*
 331  * Write to a file (through the page cache).
 332  */
 333 static ssize_t
 334 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 335 {
 336         struct dentry * dentry = iocb->ki_filp->f_dentry;
 337         struct inode * inode = dentry->d_inode;
 338         ssize_t result;
 339
 340 #ifdef CONFIG_NFS_DIRECTIO
 341         if (iocb->ki_filp->f_flags & O_DIRECT)
 342                 return nfs_file_direct_write(iocb, buf, count, pos);
 343 #endif
 344
 345         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
 346                 dentry->d_parent->d_name.name, dentry->d_name.name,
 347                 inode->i_ino, (unsigned long) count, (unsigned long) pos);
 348
 349         result = -EBUSY;
 350         if (IS_SWAPFILE(inode))
 351                 goto out_swapfile;
 352         /*
 353          * O_APPEND implies that we must revalidate the file length.
 354          */
 355         if (iocb->ki_filp->f_flags & O_APPEND) {
 356                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 357                 if (result)
 358                         goto out;
 359         }
 360         nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 361
 362         result = count;
 363         if (!count)
 364                 goto out;
 365
 366         result = generic_file_aio_write(iocb, buf, count, pos);
 367 out:
 368         return result;
 369
 370 out_swapfile:
 371         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 372         goto out;
 373 }
 374
 375 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 376 {
 377         struct inode *inode = filp->f_mapping->host;
 378         int status = 0;
 379
 380         lock_kernel();
 381         /* Use local locking if mounted with "-onolock" */
 382         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 383                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 384         else {
 385                 struct file_lock *cfl = posix_test_lock(filp, fl);
 386
 387                 fl->fl_type = F_UNLCK;
 388                 if (cfl != NULL)
 389                         memcpy(fl, cfl, sizeof(*fl));
 390         }
 391         unlock_kernel();
 392         return status;
 393 }
 394
 395 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 396 {
 397         int res = 0;
 398         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 399                 case FL_POSIX:
 400                         res = posix_lock_file_wait(file, fl);
 401                         break;
 402                 case FL_FLOCK:
 403                         res = flock_lock_file_wait(file, fl);
 404                         break;
 405                 default:
 406                         BUG();
 407         }
 408         if (res < 0)
 409                 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
 410                                 __FUNCTION__);
 411         return res;
 412 }
 413
 414 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 415 {
 416         struct inode *inode = filp->f_mapping->host;
 417         sigset_t oldset;
 418         int status;
 419
 420         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 421         /*
 422          * Flush all pending writes before doing anything
 423          * with locks..
 424          */
 425         filemap_fdatawrite(filp->f_mapping);
 426         down(&inode->i_sem);
 427         nfs_wb_all(inode);
 428         up(&inode->i_sem);
 429         filemap_fdatawait(filp->f_mapping);
 430
 431         /* NOTE: special case
 432          *      If we're signalled while cleaning up locks on process exit, we
 433          *      still need to complete the unlock.
 434          */
 435         lock_kernel();
 436         /* Use local locking if mounted with "-onolock" */
 437         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 438                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 439         else
 440                 status = do_vfs_lock(filp, fl);
 441         unlock_kernel();
 442         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 443         return status;
 444 }
 445
 446 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 447 {
 448         struct inode *inode = filp->f_mapping->host;
 449         sigset_t oldset;
 450         int status;
 451
 452         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 453         /*
 454          * Flush all pending writes before doing anything
 455          * with locks..
 456          */
 457         status = filemap_fdatawrite(filp->f_mapping);
 458         if (status == 0) {
 459                 down(&inode->i_sem);
 460                 status = nfs_wb_all(inode);
 461                 up(&inode->i_sem);
 462                 if (status == 0)
 463                         status = filemap_fdatawait(filp->f_mapping);
 464         }
 465         if (status < 0)
 466                 goto out;
 467
 468         lock_kernel();
 469         /* Use local locking if mounted with "-onolock" */
 470         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 471                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 472                 /* If we were signalled we still need to ensure that
 473                  * we clean up any state on the server. We therefore
 474                  * record the lock call as having succeeded in order to
 475                  * ensure that locks_remove_posix() cleans it out when
 476                  * the process exits.
 477                  */
 478                 if (status == -EINTR || status == -ERESTARTSYS)
 479                         do_vfs_lock(filp, fl);
 480         } else
 481                 status = do_vfs_lock(filp, fl);
 482         unlock_kernel();
 483         if (status < 0)
 484                 goto out;
 485         /*
 486          * Make sure we clear the cache whenever we try to get the lock.
 487          * This makes locking act as a cache coherency point.
 488          */
 489         filemap_fdatawrite(filp->f_mapping);
 490         down(&inode->i_sem);
 491         nfs_wb_all(inode);      /* we may have slept */
 492         up(&inode->i_sem);
 493         filemap_fdatawait(filp->f_mapping);
 494         nfs_zap_caches(inode);
 495 out:
 496         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 497         return status;
 498 }
 499
 500 /*
 501  * Lock a (portion of) a file
 502  */
 503 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 504 {
 505         struct inode * inode = filp->f_mapping->host;
 506
 507         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 508                         inode->i_sb->s_id, inode->i_ino,
 509                         fl->fl_type, fl->fl_flags,
 510                         (long long)fl->fl_start, (long long)fl->fl_end);
 511
 512         if (!inode)
 513                 return -EINVAL;
 514
 515         /* No mandatory locks over NFS */
 516         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
 517                 return -ENOLCK;
 518
 519         if (IS_GETLK(cmd))
 520                 return do_getlk(filp, cmd, fl);
 521         if (fl->fl_type == F_UNLCK)
 522                 return do_unlk(filp, cmd, fl);
 523         return do_setlk(filp, cmd, fl);
 524 }
 525
 526 /*
 527  * Lock a (portion of) a file
 528  */
 529 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 530 {
 531         struct inode * inode = filp->f_mapping->host;
 532
 533         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 534                         inode->i_sb->s_id, inode->i_ino,
 535                         fl->fl_type, fl->fl_flags);
 536
 537         if (!inode)
 538                 return -EINVAL;
 539
 540         /*
 541          * No BSD flocks over NFS allowed.
 542          * Note: we could try to fake a POSIX lock request here by
 543          * using ((u32) filp | 0x80000000) or some such as the pid.
 544          * Not sure whether that would be unique, though, or whether
 545          * that would break in other places.
 546          */
 547         if (!(fl->fl_flags & FL_FLOCK))
 548                 return -ENOLCK;
 549
 550         /* We're simulating flock() locks using posix locks on the server */
 551         fl->fl_owner = (fl_owner_t)filp;
 552         fl->fl_start = 0;
 553         fl->fl_end = OFFSET_MAX;
 554
 555         if (fl->fl_type == F_UNLCK)
 556                 return do_unlk(filp, cmd, fl);
 557         return do_setlk(filp, cmd, fl);
 558 }