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 int  nfs_file_mmap(struct file *, struct vm_area_struct *);
  41 static ssize_t nfs_file_sendfile(struct file *, loff_t *, size_t, read_actor_t, void *);
  42 static ssize_t nfs_file_read(struct kiocb *, char __user *, size_t, loff_t);
  43 static ssize_t nfs_file_write(struct kiocb *, const char __user *, size_t, loff_t);
  44 static int  nfs_file_flush(struct file *);
  45 static int  nfs_fsync(struct file *, struct dentry *dentry, int datasync);
  46 static int nfs_check_flags(int flags);
  47
  48 struct file_operations nfs_file_operations = {
  49         .llseek         = remote_llseek,
  50         .read           = do_sync_read,
  51         .write          = do_sync_write,
  52         .aio_read               = nfs_file_read,
  53         .aio_write              = nfs_file_write,
  54         .mmap           = nfs_file_mmap,
  55         .open           = nfs_file_open,
  56         .flush          = nfs_file_flush,
  57         .release        = nfs_file_release,
  58         .fsync          = nfs_fsync,
  59         .lock           = nfs_lock,
  60         .sendfile       = nfs_file_sendfile,
  61         .check_flags    = nfs_check_flags,
  62 };
  63
  64 struct inode_operations nfs_file_inode_operations = {
  65         .permission     = nfs_permission,
  66         .getattr        = nfs_getattr,
  67         .setattr        = nfs_setattr,
  68 };
  69
  70 /* Hack for future NFS swap support */
  71 #ifndef IS_SWAPFILE
  72 # define IS_SWAPFILE(inode)     (0)
  73 #endif
  74
  75 static int nfs_check_flags(int flags)
  76 {
  77         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
  78                 return -EINVAL;
  79
  80         return 0;
  81 }
  82
  83 /*
  84  * Open file
  85  */
  86 static int
  87 nfs_file_open(struct inode *inode, struct file *filp)
  88 {
  89         struct nfs_server *server = NFS_SERVER(inode);
  90         int (*open)(struct inode *, struct file *);
  91         int res;
  92
  93         res = nfs_check_flags(filp->f_flags);
  94         if (res)
  95                 return res;
  96
  97         lock_kernel();
  98         /* Do NFSv4 open() call */
  99         if ((open = server->rpc_ops->file_open) != NULL)
 100                 res = open(inode, filp);
 101         unlock_kernel();
 102         return res;
 103 }
 104
 105 static int
 106 nfs_file_release(struct inode *inode, struct file *filp)
 107 {
 108         return NFS_PROTO(inode)->file_release(inode, filp);
 109 }
 110
 111 /*
 112  * Flush all dirty pages, and check for write errors.
 113  *
 114  */
 115 static int
 116 nfs_file_flush(struct file *file)
 117 {
 118         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 119         struct inode    *inode = file->f_dentry->d_inode;
 120         int             status;
 121
 122         dfprintk(VFS, "nfs: flush(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 123
 124         if ((file->f_mode & FMODE_WRITE) == 0)
 125                 return 0;
 126         lock_kernel();
 127         /* Ensure that data+attribute caches are up to date after close() */
 128         status = nfs_wb_all(inode);
 129         if (!status) {
 130                 status = ctx->error;
 131                 ctx->error = 0;
 132                 if (!status && !nfs_have_delegation(inode, FMODE_READ))
 133                         __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 134         }
 135         unlock_kernel();
 136         return status;
 137 }
 138
 139 static ssize_t
 140 nfs_file_read(struct kiocb *iocb, char __user * buf, size_t count, loff_t pos)
 141 {
 142         struct dentry * dentry = iocb->ki_filp->f_dentry;
 143         struct inode * inode = dentry->d_inode;
 144         ssize_t result;
 145
 146 #ifdef CONFIG_NFS_DIRECTIO
 147         if (iocb->ki_filp->f_flags & O_DIRECT)
 148                 return nfs_file_direct_read(iocb, buf, count, pos);
 149 #endif
 150
 151         dfprintk(VFS, "nfs: read(%s/%s, %lu@%lu)\n",
 152                 dentry->d_parent->d_name.name, dentry->d_name.name,
 153                 (unsigned long) count, (unsigned long) pos);
 154
 155         result = nfs_revalidate_inode(NFS_SERVER(inode), inode);
 156         if (!result)
 157                 result = generic_file_aio_read(iocb, buf, count, pos);
 158         return result;
 159 }
 160
 161 static ssize_t
 162 nfs_file_sendfile(struct file *filp, loff_t *ppos, size_t count,
 163                 read_actor_t actor, void *target)
 164 {
 165         struct dentry *dentry = filp->f_dentry;
 166         struct inode *inode = dentry->d_inode;
 167         ssize_t res;
 168
 169         dfprintk(VFS, "nfs: sendfile(%s/%s, %lu@%Lu)\n",
 170                 dentry->d_parent->d_name.name, dentry->d_name.name,
 171                 (unsigned long) count, (unsigned long long) *ppos);
 172
 173         res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
 174         if (!res)
 175                 res = generic_file_sendfile(filp, ppos, count, actor, target);
 176         return res;
 177 }
 178
 179 static int
 180 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
 181 {
 182         struct dentry *dentry = file->f_dentry;
 183         struct inode *inode = dentry->d_inode;
 184         int     status;
 185
 186         dfprintk(VFS, "nfs: mmap(%s/%s)\n",
 187                 dentry->d_parent->d_name.name, dentry->d_name.name);
 188
 189         status = nfs_revalidate_inode(NFS_SERVER(inode), inode);
 190         if (!status)
 191                 status = generic_file_mmap(file, vma);
 192         return status;
 193 }
 194
 195 /*
 196  * Flush any dirty pages for this process, and check for write errors.
 197  * The return status from this call provides a reliable indication of
 198  * whether any write errors occurred for this process.
 199  */
 200 static int
 201 nfs_fsync(struct file *file, struct dentry *dentry, int datasync)
 202 {
 203         struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
 204         struct inode *inode = dentry->d_inode;
 205         int status;
 206
 207         dfprintk(VFS, "nfs: fsync(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 208
 209         lock_kernel();
 210         status = nfs_wb_all(inode);
 211         if (!status) {
 212                 status = ctx->error;
 213                 ctx->error = 0;
 214         }
 215         unlock_kernel();
 216         return status;
 217 }
 218
 219 /*
 220  * This does the "real" work of the write. The generic routine has
 221  * allocated the page, locked it, done all the page alignment stuff
 222  * calculations etc. Now we should just copy the data from user
 223  * space and write it back to the real medium..
 224  *
 225  * If the writer ends up delaying the write, the writer needs to
 226  * increment the page use counts until he is done with the page.
 227  */
 228 static int nfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 229 {
 230         return nfs_flush_incompatible(file, page);
 231 }
 232
 233 static int nfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
 234 {
 235         long status;
 236
 237         lock_kernel();
 238         status = nfs_updatepage(file, page, offset, to-offset);
 239         unlock_kernel();
 240         return status;
 241 }
 242
 243 struct address_space_operations nfs_file_aops = {
 244         .readpage = nfs_readpage,
 245         .readpages = nfs_readpages,
 246         .set_page_dirty = __set_page_dirty_nobuffers,
 247         .writepage = nfs_writepage,
 248         .writepages = nfs_writepages,
 249         .prepare_write = nfs_prepare_write,
 250         .commit_write = nfs_commit_write,
 251 #ifdef CONFIG_NFS_DIRECTIO
 252         .direct_IO = nfs_direct_IO,
 253 #endif
 254 };
 255
 256 /*
 257  * Write to a file (through the page cache).
 258  */
 259 static ssize_t
 260 nfs_file_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
 261 {
 262         struct dentry * dentry = iocb->ki_filp->f_dentry;
 263         struct inode * inode = dentry->d_inode;
 264         ssize_t result;
 265
 266 #ifdef CONFIG_NFS_DIRECTIO
 267         if (iocb->ki_filp->f_flags & O_DIRECT)
 268                 return nfs_file_direct_write(iocb, buf, count, pos);
 269 #endif
 270
 271         dfprintk(VFS, "nfs: write(%s/%s(%ld), %lu@%lu)\n",
 272                 dentry->d_parent->d_name.name, dentry->d_name.name,
 273                 inode->i_ino, (unsigned long) count, (unsigned long) pos);
 274
 275         result = -EBUSY;
 276         if (IS_SWAPFILE(inode))
 277                 goto out_swapfile;
 278         result = nfs_revalidate_inode(NFS_SERVER(inode), inode);
 279         if (result)
 280                 goto out;
 281
 282         result = count;
 283         if (!count)
 284                 goto out;
 285
 286         result = generic_file_aio_write(iocb, buf, count, pos);
 287 out:
 288         return result;
 289
 290 out_swapfile:
 291         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
 292         goto out;
 293 }
 294
 295 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
 296 {
 297         struct inode *inode = filp->f_mapping->host;
 298         int status;
 299
 300         lock_kernel();
 301         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 302         unlock_kernel();
 303         return status;
 304 }
 305
 306 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
 307 {
 308         struct inode *inode = filp->f_mapping->host;
 309         sigset_t oldset;
 310         int status;
 311
 312         rpc_clnt_sigmask(NFS_CLIENT(inode), &oldset);
 313         /*
 314          * Flush all pending writes before doing anything
 315          * with locks..
 316          */
 317         filemap_fdatawrite(filp->f_mapping);
 318         down(&inode->i_sem);
 319         nfs_wb_all(inode);
 320         up(&inode->i_sem);
 321         filemap_fdatawait(filp->f_mapping);
 322
 323         /* NOTE: special case
 324          *      If we're signalled while cleaning up locks on process exit, we
 325          *      still need to complete the unlock.
 326          */
 327         lock_kernel();
 328         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 329         rpc_clnt_sigunmask(NFS_CLIENT(inode), &oldset);
 330         return status;
 331 }
 332
 333 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
 334 {
 335         struct inode *inode = filp->f_mapping->host;
 336         int status;
 337
 338         /*
 339          * Flush all pending writes before doing anything
 340          * with locks..
 341          */
 342         status = filemap_fdatawrite(filp->f_mapping);
 343         if (status == 0) {
 344                 down(&inode->i_sem);
 345                 status = nfs_wb_all(inode);
 346                 up(&inode->i_sem);
 347                 if (status == 0)
 348                         status = filemap_fdatawait(filp->f_mapping);
 349         }
 350         if (status < 0)
 351                 return status;
 352
 353         lock_kernel();
 354         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
 355         /* If we were signalled we still need to ensure that
 356          * we clean up any state on the server. We therefore
 357          * record the lock call as having succeeded in order to
 358          * ensure that locks_remove_posix() cleans it out when
 359          * the process exits.
 360          */
 361         if (status == -EINTR || status == -ERESTARTSYS)
 362                 posix_lock_file(filp, fl);
 363         unlock_kernel();
 364         if (status < 0)
 365                 return status;
 366         /*
 367          * Make sure we clear the cache whenever we try to get the lock.
 368          * This makes locking act as a cache coherency point.
 369          */
 370         filemap_fdatawrite(filp->f_mapping);
 371         down(&inode->i_sem);
 372         nfs_wb_all(inode);      /* we may have slept */
 373         up(&inode->i_sem);
 374         filemap_fdatawait(filp->f_mapping);
 375         nfs_zap_caches(inode);
 376         return 0;
 377 }
 378
 379 /*
 380  * Lock a (portion of) a file
 381  */
 382 int
 383 nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
 384 {
 385         struct inode * inode = filp->f_mapping->host;
 386
 387         dprintk("NFS: nfs_lock(f=%s/%ld, t=%x, fl=%x, r=%Ld:%Ld)\n",
 388                         inode->i_sb->s_id, inode->i_ino,
 389                         fl->fl_type, fl->fl_flags,
 390                         (long long)fl->fl_start, (long long)fl->fl_end);
 391
 392         if (!inode)
 393                 return -EINVAL;
 394
 395         /* No mandatory locks over NFS */
 396         if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
 397                 return -ENOLCK;
 398
 399         if (NFS_PROTO(inode)->version != 4) {
 400                 /* Fake OK code if mounted without NLM support */
 401                 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM) {
 402                         if (IS_GETLK(cmd))
 403                                 return LOCK_USE_CLNT;
 404                         return 0;
 405                 }
 406         }
 407
 408         /*
 409          * No BSD flocks over NFS allowed.
 410          * Note: we could try to fake a POSIX lock request here by
 411          * using ((u32) filp | 0x80000000) or some such as the pid.
 412          * Not sure whether that would be unique, though, or whether
 413          * that would break in other places.
 414          */
 415         if (!fl->fl_owner || !(fl->fl_flags & FL_POSIX))
 416                 return -ENOLCK;
 417
 418         if (IS_GETLK(cmd))
 419                 return do_getlk(filp, cmd, fl);
 420         if (fl->fl_type == F_UNLCK)
 421                 return do_unlk(filp, cmd, fl);
 422         return do_setlk(filp, cmd, fl);
 423 }