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         struct nfs_inode *nfsi = NFS_I(inode);
 138         int retval = 0;
 139
 140         if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) || nfs_attribute_timeout(inode))
 141                 retval = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 142         nfs_revalidate_mapping(inode, filp->f_mapping);
 143         return 0;
 144 }
 145
 146 /**
 147  * nfs_revalidate_size - Revalidate the file size
 148  * @inode - pointer to inode struct
 149  * @file - pointer to struct file
 150  *
 151  * Revalidates the file length. This is basically a wrapper around
 152  * nfs_revalidate_inode() that takes into account the fact that we may
 153  * have cached writes (in which case we don't care about the server's
 154  * idea of what the file length is), or O_DIRECT (in which case we
 155  * shouldn't trust the cache).
 156  */
 157 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
 158 {
 159         struct nfs_server *server = NFS_SERVER(inode);
 160         struct nfs_inode *nfsi = NFS_I(inode);
 161
 162         if (server->flags & NFS_MOUNT_NOAC)
 163                 goto force_reval;
 164         if (filp->f_flags & O_DIRECT)
 165                 goto force_reval;
 166         if (nfsi->npages != 0)
 167                 return 0;
 168         if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
 169                 return 0;
 170 force_reval:
 171         return __nfs_revalidate_inode(server, inode);
 172 }
 173
 174 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
 175 {
 176         /* origin == SEEK_END => we must revalidate the cached file length */
 177         if (origin == 2) {
 178                 struct inode *inode = filp->f_mapping->host;
 179                 int retval = nfs_revalidate_file_size(inode, filp);
 180                 if (retval < 0)
 181                         return (loff_t)retval;
 182         }
 183         return remote_llseek(filp, offset, origin);
 184 }
 185
 186 /*
 187  * Flush all dirty pages, and check for write errors.
 188  *
 189  */
 190 static int
 191 nfs_file_flush(struct file *file)
 192 {
 193         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 194         struct inode    *inode = file->f_dentry->d_inode;
 195         int             status;
 196
 197         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 198
 199         if ((file->f_mode & FMODE_WRITE) == 0)
 200                 return 0;
 201         lock_kernel();
 202         /* Ensure that data+attribute caches are up to date after close() */
 203         status = nfs_wb_all(inode);
 204         if (!status) {
 205                 status = ctx->error;
 206                 ctx->error = 0;
 207                 if (!status && !nfs_have_delegation(inode, FMODE_READ))
 208                         __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 209         }
 210         unlock_kernel();
 211         return status;
 212 }
 213
 214 static ssize_t
 215 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
 216 {
 217         struct dentry * dentry = iocb->ki_filp->f_dentry;
 218         struct inode * inode = dentry->d_inode;
 219         ssize_t result;
 220
 221 #ifdef CONFIG_NFS_DIRECTIO
 222         if (iocb->ki_filp->f_flags & O_DIRECT)
 223                 return nfs_file_direct_read(iocb, buf, count, pos);
 224 #endif
 225
 226         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 227                 dentry->d_parent->d_name.name, dentry->d_name.name,
 228                 (unsigned long) count, (unsigned long) pos);
 229
 230         result = nfs_revalidate_file(inode, iocb->ki_filp);
 231         if (!result)
 232                 result = generic_file_aio_read(iocb, buf, count, pos);
 233         return result;
 234 }
 235
 236 static ssize_t
 237 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 238                 read_actor_t actor, void *target)
 239 {
 240         struct dentry *dentry = filp->f_dentry;
 241         struct inode *inode = dentry->d_inode;
 242         ssize_t res;
 243
 244         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 245                 dentry->d_parent->d_name.name, dentry->d_name.name,
 246                 (unsigned long) count, (unsigned long long) *ppos);
 247
 248         res = nfs_revalidate_file(inode, filp);
 249         if (!res)
 250                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 251         return res;
 252 }
 253
 254 static int
 255 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 256 {
 257         struct dentry *dentry = file->f_dentry;
 258         struct inode *inode = dentry->d_inode;
 259         int     status;
 260
 261         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 262                 dentry->d_parent->d_name.name, dentry->d_name.name);
 263
 264         status = nfs_revalidate_file(inode, file);
 265         if (!status)
 266                 status = generic_file_mmap(file, vma);
 267         return status;
 268 }
 269
 270 /*
 271  * Flush any dirty pages for this process, and check for write errors.
 272  * The return status from this call provides a reliable indication of
 273  * whether any write errors occurred for this process.
 274  */
 275 static int
 276 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 277 {
 278         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 279         struct inode *inode = dentry->d_inode;
 280         int status;
 281
 282         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 283
 284         lock_kernel();
 285         status = nfs_wb_all(inode);
 286         if (!status) {
 287                 status = ctx->error;
 288                 ctx->error = 0;
 289         }
 290         unlock_kernel();
 291         return status;
 292 }
 293
 294 /*
 295  * This does the "real" work of the write. The generic routine has
 296  * allocated the page, locked it, done all the page alignment stuff
 297  * calculations etc. Now we should just copy the data from user
 298  * space and write it back to the real medium..
 299  *
 300  * If the writer ends up delaying the write, the writer needs to
 301  * increment the page use counts until he is done with the page.
 302  */
 303 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 304 {
 305         return nfs_flush_incompatible(file, page);
 306 }
 307
 308 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 309 {
 310         long status;
 311
 312         lock_kernel();
 313         status = nfs_updatepage(file, page, offset, to-offset);
 314         unlock_kernel();
 315         return status;
 316 }
 317
 318 struct address_space_operations nfs_file_aops = {
 319         .readpage = nfs_readpage,
 320         .readpages = nfs_readpages,
 321         .set_page_dirty = __set_page_dirty_nobuffers,
 322         .writepage = nfs_writepage,
 323         .writepages = nfs_writepages,
 324         .prepare_write = nfs_prepare_write,
 325         .commit_write = nfs_commit_write,
 326 #ifdef CONFIG_NFS_DIRECTIO
 327         .direct_IO = nfs_direct_IO,
 328 #endif
 329 };
 330
 331 /*
 332  * Write to a file (through the page cache).
 333  */
 334 static ssize_t
 335 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 336 {
 337         struct dentry * dentry = iocb->ki_filp->f_dentry;
 338         struct inode * inode = dentry->d_inode;
 339         ssize_t result;
 340
 341 #ifdef CONFIG_NFS_DIRECTIO
 342         if (iocb->ki_filp->f_flags & O_DIRECT)
 343                 return nfs_file_direct_write(iocb, buf, count, pos);
 344 #endif
 345
 346         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
 347                 dentry->d_parent->d_name.name, dentry->d_name.name,
 348                 inode->i_ino, (unsigned long) count, (unsigned long) pos);
 349
 350         result = -EBUSY;
 351         if (IS_SWAPFILE(inode))
 352                 goto out_swapfile;
 353         /*
 354          * O_APPEND implies that we must revalidate the file length.
 355          */
 356         if (iocb->ki_filp->f_flags & O_APPEND) {
 357                 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
 358                 if (result)
 359                         goto out;
 360         }
 361         nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
 362
 363         result = count;
 364         if (!count)
 365                 goto out;
 366
 367         result = generic_file_aio_write(iocb, buf, count, pos);
 368 out:
 369         return result;
 370
 371 out_swapfile:
 372         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 373         goto out;
 374 }
 375
 376 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 377 {
 378         struct inode *inode = filp->f_mapping->host;
 379         int status = 0;
 380
 381         lock_kernel();
 382         /* Use local locking if mounted with "-onolock" */
 383         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 384                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 385         else {
 386                 struct file_lock *cfl = posix_test_lock(filp, fl);
 387
 388                 fl->fl_type = F_UNLCK;
 389                 if (cfl != NULL)
 390                         memcpy(fl, cfl, sizeof(*fl));
 391         }
 392         unlock_kernel();
 393         return status;
 394 }
 395
 396 static int do_vfs_lock(struct file *file, struct file_lock *fl)
 397 {
 398         int res = 0;
 399         switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
 400                 case FL_POSIX:
 401                         res = posix_lock_file_wait(file, fl);
 402                         break;
 403                 case FL_FLOCK:
 404                         res = flock_lock_file_wait(file, fl);
 405                         break;
 406                 default:
 407                         BUG();
 408         }
 409         if (res < 0)
 410                 printk(KERN_WARNING "%s: VFS is out of sync with lock manager!\n",
 411                                 __FUNCTION__);
 412         return res;
 413 }
 414
 415 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 416 {
 417         struct inode *inode = filp->f_mapping->host;
 418         sigset_t oldset;
 419         int status;
 420
 421         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 422         /*
 423          * Flush all pending writes before doing anything
 424          * with locks..
 425          */
 426         filemap_fdatawrite(filp->f_mapping);
 427         down(&inode->i_sem);
 428         nfs_wb_all(inode);
 429         up(&inode->i_sem);
 430         filemap_fdatawait(filp->f_mapping);
 431
 432         /* NOTE: special case
 433          *      If we're signalled while cleaning up locks on process exit, we
 434          *      still need to complete the unlock.
 435          */
 436         lock_kernel();
 437         /* Use local locking if mounted with "-onolock" */
 438         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
 439                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 440         else
 441                 status = do_vfs_lock(filp, fl);
 442         unlock_kernel();
 443         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 444         return status;
 445 }
 446
 447 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 448 {
 449         struct inode *inode = filp->f_mapping->host;
 450         sigset_t oldset;
 451         int status;
 452
 453         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 454         /*
 455          * Flush all pending writes before doing anything
 456          * with locks..
 457          */
 458         status = filemap_fdatawrite(filp->f_mapping);
 459         if (status == 0) {
 460                 down(&inode->i_sem);
 461                 status = nfs_wb_all(inode);
 462                 up(&inode->i_sem);
 463                 if (status == 0)
 464                         status = filemap_fdatawait(filp->f_mapping);
 465         }
 466         if (status < 0)
 467                 goto out;
 468
 469         lock_kernel();
 470         /* Use local locking if mounted with "-onolock" */
 471         if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) {
 472                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 473                 /* If we were signalled we still need to ensure that
 474                  * we clean up any state on the server. We therefore
 475                  * record the lock call as having succeeded in order to
 476                  * ensure that locks_remove_posix() cleans it out when
 477                  * the process exits.
 478                  */
 479                 if (status == -EINTR || status == -ERESTARTSYS)
 480                         do_vfs_lock(filp, fl);
 481         } else
 482                 status = do_vfs_lock(filp, fl);
 483         unlock_kernel();
 484         if (status < 0)
 485                 goto out;
 486         /*
 487          * Make sure we clear the cache whenever we try to get the lock.
 488          * This makes locking act as a cache coherency point.
 489          */
 490         filemap_fdatawrite(filp->f_mapping);
 491         down(&inode->i_sem);
 492         nfs_wb_all(inode);      /* we may have slept */
 493         up(&inode->i_sem);
 494         filemap_fdatawait(filp->f_mapping);
 495         nfs_zap_caches(inode);
 496 out:
 497         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 498         return status;
 499 }
 500
 501 /*
 502  * Lock a (portion of) a file
 503  */
 504 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 505 {
 506         struct inode * inode = filp->f_mapping->host;
 507
 508         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 509                         inode->i_sb->s_id, inode->i_ino,
 510                         fl->fl_type, fl->fl_flags,
 511                         (long long)fl->fl_start, (long long)fl->fl_end);
 512
 513         if (!inode)
 514                 return -EINVAL;
 515
 516         /* No mandatory locks over NFS */
 517         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
 518                 return -ENOLCK;
 519
 520         if (IS_GETLK(cmd))
 521                 return do_getlk(filp, cmd, fl);
 522         if (fl->fl_type == F_UNLCK)
 523                 return do_unlk(filp, cmd, fl);
 524         return do_setlk(filp, cmd, fl);
 525 }
 526
 527 /*
 528  * Lock a (portion of) a file
 529  */
 530 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
 531 {
 532         struct inode * inode = filp->f_mapping->host;
 533
 534         dprintk("NFS: nfs_flock(f=%s/%ld, t=%x, fl=%x)\n",
 535                         inode->i_sb->s_id, inode->i_ino,
 536                         fl->fl_type, fl->fl_flags);
 537
 538         if (!inode)
 539                 return -EINVAL;
 540
 541         /*
 542          * No BSD flocks over NFS allowed.
 543          * Note: we could try to fake a POSIX lock request here by
 544          * using ((u32) filp | 0x80000000) or some such as the pid.
 545          * Not sure whether that would be unique, though, or whether
 546          * that would break in other places.
 547          */
 548         if (!(fl->fl_flags & FL_FLOCK))
 549                 return -ENOLCK;
 550
 551         /* We're simulating flock() locks using posix locks on the server */
 552         fl->fl_owner = (fl_owner_t)filp;
 553         fl->fl_start = 0;
 554         fl->fl_end = OFFSET_MAX;
 555
 556         if (fl->fl_type == F_UNLCK)
 557                 return do_unlk(filp, cmd, fl);
 558         return do_setlk(filp, cmd, fl);
 559 }