~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~ [ freetext search ] ~ [ file search ] ~

Linux Cross Reference
Linux/fs/nfsd/vfs.c

Version: ~ [ 2.4.0 ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 #define MSNFS   /* HACK HACK */
  2 /*
  3  * linux/fs/nfsd/vfs.c
  4  *
  5  * File operations used by nfsd. Some of these have been ripped from
  6  * other parts of the kernel because they weren't in ksyms.c, others
  7  * are partial duplicates with added or changed functionality.
  8  *
  9  * Note that several functions dget() the dentry upon which they want
 10  * to act, most notably those that create directory entries. Response
 11  * dentry's are dput()'d if necessary in the release callback.
 12  * So if you notice code paths that apparently fail to dput() the
 13  * dentry, don't worry--they have been taken care of.
 14  *
 15  * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
 16  */
 17 
 18 #include <linux/config.h>
 19 #include <linux/version.h>
 20 #include <linux/string.h>
 21 #include <linux/sched.h>
 22 #include <linux/errno.h>
 23 #include <linux/locks.h>
 24 #include <linux/fs.h>
 25 #include <linux/major.h>
 26 #include <linux/ext2_fs.h>
 27 #include <linux/proc_fs.h>
 28 #include <linux/stat.h>
 29 #include <linux/fcntl.h>
 30 #include <linux/net.h>
 31 #include <linux/unistd.h>
 32 #include <linux/malloc.h>
 33 #include <linux/in.h>
 34 #define __NO_VERSION__
 35 #include <linux/module.h>
 36 
 37 #include <linux/sunrpc/svc.h>
 38 #include <linux/nfsd/nfsd.h>
 39 #ifdef CONFIG_NFSD_V3
 40 #include <linux/nfs3.h>
 41 #include <linux/nfsd/xdr3.h>
 42 #endif /* CONFIG_NFSD_V3 */
 43 #include <linux/nfsd/nfsfh.h>
 44 #include <linux/quotaops.h>
 45 
 46 #include <asm/uaccess.h>
 47 
 48 #define NFSDDBG_FACILITY                NFSDDBG_FILEOP
 49 #define NFSD_PARANOIA
 50 
 51 
 52 /* We must ignore files (but only files) which might have mandatory
 53  * locks on them because there is no way to know if the accesser has
 54  * the lock.
 55  */
 56 #define IS_ISMNDLK(i)   (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
 57 
 58 /*
 59  * This is a cache of readahead params that help us choose the proper
 60  * readahead strategy. Initially, we set all readahead parameters to 0
 61  * and let the VFS handle things.
 62  * If you increase the number of cached files very much, you'll need to
 63  * add a hash table here.
 64  */
 65 struct raparms {
 66         struct raparms          *p_next;
 67         unsigned int            p_count;
 68         ino_t                   p_ino;
 69         dev_t                   p_dev;
 70         unsigned long           p_reada,
 71                                 p_ramax,
 72                                 p_raend,
 73                                 p_ralen,
 74                                 p_rawin;
 75 };
 76 
 77 static struct raparms *         raparml;
 78 static struct raparms *         raparm_cache;
 79 
 80 /*
 81  * Look up one component of a pathname.
 82  * N.B. After this call _both_ fhp and resfh need an fh_put
 83  *
 84  * If the lookup would cross a mountpoint, and the mounted filesystem
 85  * is exported to the client with NFSEXP_CROSSMNT, then the lookup is
 86  * accepted as it stands and the mounted directory is
 87  * returned. Otherwise the covered directory is returned.
 88  * NOTE: this mountpoint crossing is not supported properly by all
 89  *   clients and is explicitly disallowed for NFSv3
 90  *      NeilBrown <neilb@cse.unsw.edu.au>
 91  */
 92 int
 93 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
 94                                         int len, struct svc_fh *resfh)
 95 {
 96         struct svc_export       *exp;
 97         struct dentry           *dparent;
 98         struct dentry           *dentry;
 99         int                     err;
100 
101         dprintk("nfsd: nfsd_lookup(fh %s, %s)\n", SVCFH_fmt(fhp), name);
102 
103         /* Obtain dentry and export. */
104         err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
105         if (err)
106                 goto out;
107 
108         dparent = fhp->fh_dentry;
109         exp  = fhp->fh_export;
110 
111         err = nfserr_acces;
112 
113         /* Lookup the name, but don't follow links */
114         if (strcmp(name, ".")==0) {
115                 dentry = dget(dparent);
116         } else if (strcmp(name, "..")==0) {
117                 /* checking mountpoint crossing is very different when stepping up */
118                 if (dparent == exp->ex_dentry) {
119                         if (!EX_CROSSMNT(exp))
120                                 dentry = dget(dparent); /* .. == . just like at / */
121                         else
122                         {
123                                 struct svc_export *exp2 = NULL;
124                                 struct dentry *dp;
125                                 struct vfsmount *mnt = mntget(exp->ex_mnt);
126                                 dentry = dget(dparent);
127                                 while(follow_up(&mnt, &dentry))
128                                         ;
129                                 dp = dget(dentry->d_parent);
130                                 dput(dentry);
131                                 dentry = dp;
132                                 for ( ; exp2 == NULL && dp->d_parent != dp;
133                                      dp=dp->d_parent)
134                                         exp2 = exp_get(exp->ex_client, dp->d_inode->i_dev, dp->d_inode->i_ino);
135                                 if (exp2==NULL) {
136                                         dput(dentry);
137                                         dentry = dget(dparent);
138                                 } else {
139                                         exp = exp2;
140                                 }
141                                 mntput(mnt);
142                         }
143                 } else
144                         dentry = dget(dparent->d_parent);
145         } else {
146                 fh_lock(fhp);
147                 dentry = lookup_one(name, dparent);
148                 err = PTR_ERR(dentry);
149                 if (IS_ERR(dentry))
150                         goto out_nfserr;
151                 /*
152                  * check if we have crossed a mount point ...
153                  */
154                 if (d_mountpoint(dentry)) {
155                         struct svc_export *exp2 = NULL;
156                         struct vfsmount *mnt = mntget(exp->ex_mnt);
157                         struct dentry *mounts = dget(dentry);
158                         while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts))
159                                 ;
160                         exp2 = exp_get(rqstp->rq_client,
161                                        mounts->d_inode->i_dev,
162                                        mounts->d_inode->i_ino);
163                         if (exp2 && EX_CROSSMNT(exp2)) {
164                                 /* successfully crossed mount point */
165                                 exp = exp2;
166                                 dput(dentry);
167                                 dentry = mounts;
168                         } else
169                                 dput(mounts);
170                         mntput(mnt);
171                 }
172         }
173         /*
174          * Note: we compose the file handle now, but as the
175          * dentry may be negative, it may need to be updated.
176          */
177         err = fh_compose(resfh, exp, dentry);
178         if (!err && !dentry->d_inode)
179                 err = nfserr_noent;
180 out:
181         return err;
182 
183 out_nfserr:
184         err = nfserrno(err);
185         goto out;
186 }
187 
188 /*
189  * Set various file attributes.
190  * N.B. After this call fhp needs an fh_put
191  */
192 int
193 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap)
194 {
195         struct dentry   *dentry;
196         struct inode    *inode;
197         int             accmode = MAY_SATTR;
198         int             ftype = 0;
199         int             imode;
200         int             err;
201         int             size_change = 0;
202 
203         if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
204                 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
205         if (iap->ia_valid & ATTR_SIZE)
206                 ftype = S_IFREG;
207 
208         /* Get inode */
209         err = fh_verify(rqstp, fhp, ftype, accmode);
210         if (err || !iap->ia_valid)
211                 goto out;
212 
213         dentry = fhp->fh_dentry;
214         inode = dentry->d_inode;
215 
216         err = inode_change_ok(inode, iap);
217         /* could be a "touch" (utimes) request where the user is not the owner but does
218          * have write permission. In this case the user should be allowed to set
219          * both times to the current time.  We could just assume any such SETATTR
220          * is intended to set the times to "now", but we do a couple of simple tests
221          * to increase our confidence.
222          */
223 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
224 #define MAX_TOUCH_TIME_ERROR (30*60)
225         if (err
226             && (iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
227             && iap->ia_mtime == iap->ia_ctime
228             ) {
229             /* looks good.  now just make sure time is in the right ballpark.
230              * solaris, at least, doesn't seem to care what the time request is
231              */
232             time_t delta = iap->ia_atime - CURRENT_TIME;
233             if (delta<0) delta = -delta;
234             if (delta < MAX_TOUCH_TIME_ERROR) {
235                 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
236                  * this will cause notify_change to set these times to "now"
237                  */
238                 iap->ia_valid &= ~BOTH_TIME_SET;
239                 err = inode_change_ok(inode, iap);
240             }
241         }
242             
243         if (err)
244                 goto out_nfserr;
245 
246         /* The size case is special. It changes the file as well as the attributes.  */
247         if (iap->ia_valid & ATTR_SIZE) {
248                 if (iap->ia_size < inode->i_size) {
249                         err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
250                         if (err)
251                                 goto out;
252                 }
253 
254                 /*
255                  * If we are changing the size of the file, then
256                  * we need to break all leases.
257                  */
258                 err = get_lease(inode, FMODE_WRITE);
259                 if (err)
260                         goto out_nfserr;
261 
262                 err = get_write_access(inode);
263                 if (err)
264                         goto out_nfserr;
265 
266                 err = locks_verify_truncate(inode, NULL, iap->ia_size);
267                 if (err) {
268                         put_write_access(inode);
269                         goto out_nfserr;
270                 }
271                 DQUOT_INIT(inode);
272         }
273 
274         imode = inode->i_mode;
275         if (iap->ia_valid & ATTR_MODE) {
276                 iap->ia_mode &= S_IALLUGO;
277                 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
278         }
279 
280         /* Revoke setuid/setgid bit on chown/chgrp */
281         if ((iap->ia_valid & ATTR_UID) && (imode & S_ISUID)
282          && iap->ia_uid != inode->i_uid) {
283                 iap->ia_valid |= ATTR_MODE;
284                 iap->ia_mode = imode &= ~S_ISUID;
285         }
286         if ((iap->ia_valid & ATTR_GID) && (imode & S_ISGID)
287          && iap->ia_gid != inode->i_gid) {
288                 iap->ia_valid |= ATTR_MODE;
289                 iap->ia_mode = imode &= ~S_ISGID;
290         }
291 
292         /* Change the attributes. */
293 
294 
295         iap->ia_valid |= ATTR_CTIME;
296 #ifdef CONFIG_QUOTA
297         /* DQUOT_TRANSFER needs both ia_uid and ia_gid defined */
298         if (iap->ia_valid & (ATTR_UID|ATTR_GID)) {
299                 if (! (iap->ia_valid & ATTR_UID))
300                         iap->ia_uid = inode->i_uid;
301                 if (! (iap->ia_valid & ATTR_GID))
302                         iap->ia_gid = inode->i_gid;
303                 iap->ia_valid |= ATTR_UID|ATTR_GID;
304         }
305 #endif /* CONFIG_QUOTA */
306 
307         if (iap->ia_valid & ATTR_SIZE) {
308                 fh_lock(fhp);
309                 size_change = 1;
310         }
311 #ifdef CONFIG_QUOTA
312         if (iap->ia_valid & (ATTR_UID|ATTR_GID)) 
313                 err = DQUOT_TRANSFER(dentry, iap);
314         else
315 #endif
316                 err = notify_change(dentry, iap);
317         if (size_change) {
318                 fh_unlock(fhp);
319                 put_write_access(inode);
320         }
321         if (err)
322                 goto out_nfserr;
323         if (EX_ISSYNC(fhp->fh_export))
324                 write_inode_now(inode, 1);
325         err = 0;
326 out:
327         return err;
328 
329 out_nfserr:
330         err = nfserrno(err);
331         goto out;
332 }
333 
334 #ifdef CONFIG_NFSD_V3
335 /*
336  * Check server access rights to a file system object
337  */
338 struct accessmap {
339         u32             access;
340         int             how;
341 };
342 static struct accessmap nfs3_regaccess[] = {
343     {   NFS3_ACCESS_READ,       MAY_READ                        },
344     {   NFS3_ACCESS_EXECUTE,    MAY_EXEC                        },
345     {   NFS3_ACCESS_MODIFY,     MAY_WRITE|MAY_TRUNC             },
346     {   NFS3_ACCESS_EXTEND,     MAY_WRITE                       },
347 
348     {   0,                      0                               }
349 };
350 
351 static struct accessmap nfs3_diraccess[] = {
352     {   NFS3_ACCESS_READ,       MAY_READ                        },
353     {   NFS3_ACCESS_LOOKUP,     MAY_EXEC                        },
354     {   NFS3_ACCESS_MODIFY,     MAY_EXEC|MAY_WRITE|MAY_TRUNC    },
355     {   NFS3_ACCESS_EXTEND,     MAY_EXEC|MAY_WRITE              },
356     {   NFS3_ACCESS_DELETE,     MAY_REMOVE                      },
357 
358     {   0,                      0                               }
359 };
360 
361 static struct accessmap nfs3_anyaccess[] = {
362         /* Some clients - Solaris 2.6 at least, make an access call
363          * to the server to check for access for things like /dev/null
364          * (which really, the server doesn't care about).  So
365          * We provide simple access checking for them, looking
366          * mainly at mode bits
367          */
368     {   NFS3_ACCESS_READ,       MAY_READ                        },
369     {   NFS3_ACCESS_EXECUTE,    MAY_EXEC                        },
370     {   NFS3_ACCESS_MODIFY,     MAY_WRITE                       },
371     {   NFS3_ACCESS_EXTEND,     MAY_WRITE                       },
372 
373     {   0,                      0                               }
374 };
375 
376 int
377 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access)
378 {
379         struct accessmap        *map;
380         struct svc_export       *export;
381         struct dentry           *dentry;
382         u32                     query, result = 0;
383         unsigned int            error;
384 
385         error = fh_verify(rqstp, fhp, 0, MAY_NOP);
386         if (error)
387                 goto out;
388 
389         export = fhp->fh_export;
390         dentry = fhp->fh_dentry;
391 
392         if (S_ISREG(dentry->d_inode->i_mode))
393                 map = nfs3_regaccess;
394         else if (S_ISDIR(dentry->d_inode->i_mode))
395                 map = nfs3_diraccess;
396         else
397                 map = nfs3_anyaccess;
398 
399 
400         query = *access;
401         for  (; map->access; map++) {
402                 if (map->access & query) {
403                         unsigned int err2;
404                         err2 = nfsd_permission(export, dentry, map->how);
405                         switch (err2) {
406                         case nfs_ok:
407                                 result |= map->access;
408                                 break;
409                                 
410                         /* the following error codes just mean the access was not allowed,
411                          * rather than an error occurred */
412                         case nfserr_rofs:
413                         case nfserr_acces:
414                         case nfserr_perm:
415                                 /* simply don't "or" in the access bit. */
416                                 break;
417                         default:
418                                 error = err2;
419                                 goto out;
420                         }
421                 }
422         }
423         *access = result;
424 
425  out:
426         return error;
427 }
428 #endif /* CONFIG_NFSD_V3 */
429 
430 
431 
432 /*
433  * Open an existing file or directory.
434  * The access argument indicates the type of open (read/write/lock)
435  * N.B. After this call fhp needs an fh_put
436  */
437 int
438 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
439                         int access, struct file *filp)
440 {
441         struct dentry   *dentry;
442         struct inode    *inode;
443         int             err;
444 
445         /* If we get here, then the client has already done an "open", and (hopefully)
446          * checked permission - so allow OWNER_OVERRIDE in case a chmod has now revoked
447          * permission */
448         err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
449         if (err)
450                 goto out;
451 
452         dentry = fhp->fh_dentry;
453         inode = dentry->d_inode;
454 
455         /* Disallow access to files with the append-only bit set or
456          * with mandatory locking enabled
457          */
458         err = nfserr_perm;
459         if (IS_APPEND(inode) || IS_ISMNDLK(inode))
460                 goto out;
461         if (!inode->i_fop)
462                 goto out;
463 
464         /*
465          * Check to see if there are any leases on this file.
466          * This may block while leases are broken.
467          */
468         err = get_lease(inode, (access & MAY_WRITE) ? FMODE_WRITE : 0);
469         if (err)
470                 goto out_nfserr;
471 
472         if ((access & MAY_WRITE) && (err = get_write_access(inode)) != 0)
473                 goto out_nfserr;
474 
475         memset(filp, 0, sizeof(*filp));
476         filp->f_op    = fops_get(inode->i_fop);
477         atomic_set(&filp->f_count, 1);
478         filp->f_dentry = dentry;
479         if (access & MAY_WRITE) {
480                 filp->f_flags = O_WRONLY|O_LARGEFILE;
481                 filp->f_mode  = FMODE_WRITE;
482                 DQUOT_INIT(inode);
483         } else {
484                 filp->f_flags = O_RDONLY|O_LARGEFILE;
485                 filp->f_mode  = FMODE_READ;
486         }
487 
488         err = 0;
489         if (filp->f_op && filp->f_op->open) {
490                 err = filp->f_op->open(inode, filp);
491                 if (err) {
492                         fops_put(filp->f_op);
493                         if (access & MAY_WRITE)
494                                 put_write_access(inode);
495 
496                         /* I nearly added put_filp() call here, but this filp
497                          * is really on callers stack frame. -DaveM
498                          */
499                         atomic_dec(&filp->f_count);
500                 }
501         }
502 out_nfserr:
503         if (err)
504                 err = nfserrno(err);
505 out:
506         return err;
507 }
508 
509 /*
510  * Close a file.
511  */
512 void
513 nfsd_close(struct file *filp)
514 {
515         struct dentry   *dentry = filp->f_dentry;
516         struct inode    *inode = dentry->d_inode;
517 
518         if (filp->f_op && filp->f_op->release)
519                 filp->f_op->release(inode, filp);
520         fops_put(filp->f_op);
521         if (filp->f_mode & FMODE_WRITE)
522                 put_write_access(inode);
523 }
524 
525 /*
526  * Sync a file
527  * As this calls fsync (not fdatasync) there is no need for a write_inode
528  * after it.
529  */
530 void
531 nfsd_sync(struct file *filp)
532 {
533         dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
534         down(&filp->f_dentry->d_inode->i_sem);
535         filp->f_op->fsync(filp, filp->f_dentry, 0);
536         up(&filp->f_dentry->d_inode->i_sem);
537 }
538 
539 void
540 nfsd_sync_dir(struct dentry *dp)
541 {
542         struct inode *inode = dp->d_inode;
543         int (*fsync) (struct file *, struct dentry *, int);
544         
545         if (inode->i_fop && (fsync = inode->i_fop->fsync)) {
546                 fsync(NULL, dp, 0);
547         }
548 }
549 
550 /*
551  * Obtain the readahead parameters for the file
552  * specified by (dev, ino).
553  */
554 static inline struct raparms *
555 nfsd_get_raparms(dev_t dev, ino_t ino)
556 {
557         struct raparms  *ra, **rap, **frap = NULL;
558         int depth = 0;
559         
560         for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
561                 if (ra->p_ino == ino && ra->p_dev == dev)
562                         goto found;
563                 depth++;
564                 if (ra->p_count == 0)
565                         frap = rap;
566         }
567         depth = nfsdstats.ra_size*11/10;
568         if (!frap)
569                 return NULL;
570         rap = frap;
571         ra = *frap;
572         memset(ra, 0, sizeof(*ra));
573         ra->p_dev = dev;
574         ra->p_ino = ino;
575 found:
576         if (rap != &raparm_cache) {
577                 *rap = ra->p_next;
578                 ra->p_next   = raparm_cache;
579                 raparm_cache = ra;
580         }
581         ra->p_count++;
582         nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
583         return ra;
584 }
585 
586 /*
587  * Read data from a file. count must contain the requested read count
588  * on entry. On return, *count contains the number of bytes actually read.
589  * N.B. After this call fhp needs an fh_put
590  */
591 int
592 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
593           char *buf, unsigned long *count)
594 {
595         struct raparms  *ra;
596         mm_segment_t    oldfs;
597         int             err;
598         struct file     file;
599 
600         err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
601         if (err)
602                 goto out;
603         err = nfserr_perm;
604         if (!file.f_op->read)
605                 goto out_close;
606 #ifdef MSNFS
607         if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
608                 (!lock_may_read(file.f_dentry->d_inode, offset, *count)))
609                 goto out_close;
610 #endif
611 
612         /* Get readahead parameters */
613         ra = nfsd_get_raparms(fhp->fh_export->ex_dev, fhp->fh_dentry->d_inode->i_ino);
614         if (ra) {
615                 file.f_reada = ra->p_reada;
616                 file.f_ramax = ra->p_ramax;
617                 file.f_raend = ra->p_raend;
618                 file.f_ralen = ra->p_ralen;
619                 file.f_rawin = ra->p_rawin;
620         }
621         file.f_pos = offset;
622 
623         oldfs = get_fs(); set_fs(KERNEL_DS);
624         err = file.f_op->read(&file, buf, *count, &file.f_pos);
625         set_fs(oldfs);
626 
627         /* Write back readahead params */
628         if (ra != NULL) {
629                 dprintk("nfsd: raparms %ld %ld %ld %ld %ld\n",
630                         file.f_reada, file.f_ramax, file.f_raend,
631                         file.f_ralen, file.f_rawin);
632                 ra->p_reada = file.f_reada;
633                 ra->p_ramax = file.f_ramax;
634                 ra->p_raend = file.f_raend;
635                 ra->p_ralen = file.f_ralen;
636                 ra->p_rawin = file.f_rawin;
637                 ra->p_count -= 1;
638         }
639 
640         if (err >= 0) {
641                 nfsdstats.io_read += err;
642                 *count = err;
643                 err = 0;
644         } else 
645                 err = nfserrno(err);
646 out_close:
647         nfsd_close(&file);
648 out:
649         return err;
650 }
651 
652 /*
653  * Write data to a file.
654  * The stable flag requests synchronous writes.
655  * N.B. After this call fhp needs an fh_put
656  */
657 int
658 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
659                                 char *buf, unsigned long cnt, int *stablep)
660 {
661         struct svc_export       *exp;
662         struct file             file;
663         struct dentry           *dentry;
664         struct inode            *inode;
665         mm_segment_t            oldfs;
666         int                     err = 0;
667         int                     stable = *stablep;
668 
669         err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
670         if (err)
671                 goto out;
672         if (!cnt)
673                 goto out_close;
674         err = nfserr_perm;
675         if (!file.f_op->write)
676                 goto out_close;
677 #ifdef MSNFS
678         if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
679                 (!lock_may_write(file.f_dentry->d_inode, offset, cnt)))
680                 goto out_close;
681 #endif
682 
683         dentry = file.f_dentry;
684         inode = dentry->d_inode;
685         exp   = fhp->fh_export;
686 
687         /*
688          * Request sync writes if
689          *  -   the sync export option has been set, or
690          *  -   the client requested O_SYNC behavior (NFSv3 feature).
691          *  -   The file system doesn't support fsync().
692          * When gathered writes have been configured for this volume,
693          * flushing the data to disk is handled separately below.
694          */
695 
696         if (file.f_op->fsync == 0) {/* COMMIT3 cannot work */
697                stable = 2;
698                *stablep = 2; /* FILE_SYNC */
699         }
700 
701         if (!EX_ISSYNC(exp))
702                 stable = 0;
703         if (stable && !EX_WGATHER(exp))
704                 file.f_flags |= O_SYNC;
705 
706         file.f_pos = offset;            /* set write offset */
707 
708         /* Write the data. */
709         oldfs = get_fs(); set_fs(KERNEL_DS);
710         err = file.f_op->write(&file, buf, cnt, &file.f_pos);
711         if (err >= 0)
712                 nfsdstats.io_write += cnt;
713         set_fs(oldfs);
714 
715         /* clear setuid/setgid flag after write */
716         if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
717                 struct iattr    ia;
718 
719                 ia.ia_valid = ATTR_MODE;
720                 ia.ia_mode  = inode->i_mode & ~(S_ISUID | S_ISGID);
721                 notify_change(dentry, &ia);
722         }
723 
724         if (err >= 0 && stable) {
725                 static unsigned long    last_ino;
726                 static kdev_t           last_dev = NODEV;
727 
728                 /*
729                  * Gathered writes: If another process is currently
730                  * writing to the file, there's a high chance
731                  * this is another nfsd (triggered by a bulk write
732                  * from a client's biod). Rather than syncing the
733                  * file with each write request, we sleep for 10 msec.
734                  *
735                  * I don't know if this roughly approximates
736                  * C. Juszak's idea of gathered writes, but it's a
737                  * nice and simple solution (IMHO), and it seems to
738                  * work:-)
739                  */
740                 if (EX_WGATHER(exp) && (atomic_read(&inode->i_writecount) > 1
741                  || (last_ino == inode->i_ino && last_dev == inode->i_dev))) {
742 #if 0
743                         interruptible_sleep_on_timeout(&inode->i_wait, 10 * HZ / 1000);
744 #else
745                         dprintk("nfsd: write defer %d\n", current->pid);
746 /* FIXME: Olaf commented this out [gam3] */
747                         set_current_state(TASK_UNINTERRUPTIBLE);
748                         schedule_timeout((HZ+99)/100);
749                         current->state = TASK_RUNNING;
750                         dprintk("nfsd: write resume %d\n", current->pid);
751 #endif
752                 }
753 
754                 if (inode->i_state & I_DIRTY) {
755                         dprintk("nfsd: write sync %d\n", current->pid);
756                         nfsd_sync(&file);
757                 }
758 #if 0
759                 wake_up(&inode->i_wait);
760 #endif
761                 last_ino = inode->i_ino;
762                 last_dev = inode->i_dev;
763         }
764 
765         dprintk("nfsd: write complete err=%d\n", err);
766         if (err >= 0)
767                 err = 0;
768         else 
769                 err = nfserrno(err);
770 out_close:
771         nfsd_close(&file);
772 out:
773         return err;
774 }
775 
776 
777 #ifdef CONFIG_NFSD_V3
778 /*
779  * Commit all pending writes to stable storage.
780  * Strictly speaking, we could sync just the indicated file region here,
781  * but there's currently no way we can ask the VFS to do so.
782  *
783  * Unfortunately we cannot lock the file to make sure we return full WCC
784  * data to the client, as locking happens lower down in the filesystem.
785  */
786 int
787 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
788                off_t offset, unsigned long count)
789 {
790         struct file     file;
791         int             err;
792 
793         if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
794                 return err;
795         if (EX_ISSYNC(fhp->fh_export)) {
796                 if (file.f_op && file.f_op->fsync) {
797                         nfsd_sync(&file);
798                 } else {
799                         err = nfserr_notsupp;
800                 }
801         }
802 
803         nfsd_close(&file);
804         return err;
805 }
806 #endif /* CONFIG_NFSD_V3 */
807 
808 /*
809  * Create a file (regular, directory, device, fifo); UNIX sockets 
810  * not yet implemented.
811  * If the response fh has been verified, the parent directory should
812  * already be locked. Note that the parent directory is left locked.
813  *
814  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
815  */
816 int
817 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
818                 char *fname, int flen, struct iattr *iap,
819                 int type, dev_t rdev, struct svc_fh *resfhp)
820 {
821         struct dentry   *dentry, *dchild;
822         struct inode    *dirp;
823         int             err;
824 
825         err = nfserr_perm;
826         if (!flen)
827                 goto out;
828         err = nfserr_exist;
829         if (isdotent(fname, flen))
830                 goto out;
831 
832         err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
833         if (err)
834                 goto out;
835 
836         dentry = fhp->fh_dentry;
837         dirp = dentry->d_inode;
838 
839         err = nfserr_notdir;
840         if(!dirp->i_op || !dirp->i_op->lookup)
841                 goto out;
842         /*
843          * Check whether the response file handle has been verified yet.
844          * If it has, the parent directory should already be locked.
845          */
846         if (!resfhp->fh_dentry) {
847                 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
848                 fh_lock(fhp);
849                 dchild = lookup_one(fname, dentry);
850                 err = PTR_ERR(dchild);
851                 if (IS_ERR(dchild))
852                         goto out_nfserr;
853                 err = fh_compose(resfhp, fhp->fh_export, dchild);
854                 if (err)
855                         goto out;
856         } else {
857                 /* called from nfsd_proc_create */
858                 dchild = resfhp->fh_dentry;
859                 if (!fhp->fh_locked) {
860                         /* not actually possible */
861                         printk(KERN_ERR
862                                 "nfsd_create: parent %s/%s not locked!\n",
863                                 dentry->d_parent->d_name.name,
864                                 dentry->d_name.name);
865                         err = -EIO;
866                         goto out;
867                 }
868         }
869         /*
870          * Make sure the child dentry is still negative ...
871          */
872         err = nfserr_exist;
873         if (dchild->d_inode) {
874                 dprintk("nfsd_create: dentry %s/%s not negative!\n",
875                         dentry->d_name.name, dchild->d_name.name);
876                 goto out; 
877         }
878 
879         if (!(iap->ia_valid & ATTR_MODE))
880                 iap->ia_mode = 0;
881         iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
882 
883         /*
884          * Get the dir op function pointer.
885          */
886         err = nfserr_perm;
887         switch (type) {
888         case S_IFREG:
889                 err = vfs_create(dirp, dchild, iap->ia_mode);
890                 break;
891         case S_IFDIR:
892                 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
893                 break;
894         case S_IFCHR:
895         case S_IFBLK:
896         case S_IFIFO:
897         case S_IFSOCK:
898                 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
899                 break;
900         default:
901                 printk("nfsd: bad file type %o in nfsd_create\n", type);
902                 err = -EINVAL;
903         }
904         if (err < 0)
905                 goto out_nfserr;
906 
907         if (EX_ISSYNC(fhp->fh_export)) {
908                 nfsd_sync_dir(dentry);
909                 write_inode_now(dchild->d_inode, 1);
910         }
911 
912 
913         /* Set file attributes. Mode has already been set and
914          * setting uid/gid works only for root. Irix appears to
915          * send along the gid when it tries to implement setgid
916          * directories via NFS.
917          */
918         err = 0;
919         if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
920                 err = nfsd_setattr(rqstp, resfhp, iap);
921         /*
922          * Update the file handle to get the new inode info.
923          */
924         if (!err)
925                 err = fh_update(resfhp);
926 out:
927         return err;
928 
929 out_nfserr:
930         err = nfserrno(err);
931         goto out;
932 }
933 
934 #ifdef CONFIG_NFSD_V3
935 /*
936  * NFSv3 version of nfsd_create
937  */
938 int
939 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
940                 char *fname, int flen, struct iattr *iap,
941                 struct svc_fh *resfhp, int createmode, u32 *verifier)
942 {
943         struct dentry   *dentry, *dchild;
944         struct inode    *dirp;
945         int             err;
946         __u32           v_mtime=0, v_atime=0;
947         int             v_mode=0;
948 
949         err = nfserr_perm;
950         if (!flen)
951                 goto out;
952         err = nfserr_exist;
953         if (isdotent(fname, flen))
954                 goto out;
955         if (!(iap->ia_valid & ATTR_MODE))
956                 iap->ia_mode = 0;
957         err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
958         if (err)
959                 goto out;
960 
961         dentry = fhp->fh_dentry;
962         dirp = dentry->d_inode;
963 
964         /* Get all the sanity checks out of the way before
965          * we lock the parent. */
966         err = nfserr_notdir;
967         if(!dirp->i_op || !dirp->i_op->lookup)
968                 goto out;
969         fh_lock(fhp);
970 
971         /*
972          * Compose the response file handle.
973          */
974         dchild = lookup_one(fname, dentry);
975         err = PTR_ERR(dchild);
976         if (IS_ERR(dchild))
977                 goto out_nfserr;
978 
979         err = fh_compose(resfhp, fhp->fh_export, dchild);
980         if (err)
981                 goto out;
982 
983         if (createmode == NFS3_CREATE_EXCLUSIVE) {
984                 /* while the verifier would fit in mtime+atime,
985                  * solaris7 gets confused (bugid 4218508) if these have
986                  * the high bit set, so we use the mode as well
987                  */
988                 v_mtime = verifier[0]&0x7fffffff;
989                 v_atime = verifier[1]&0x7fffffff;
990                 v_mode  = S_IFREG
991                         | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
992                         | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
993                         ;
994         }
995         
996         if (dchild->d_inode) {
997                 err = 0;
998 
999                 switch (createmode) {
1000                 case NFS3_CREATE_UNCHECKED:
1001                         if (! S_ISREG(dchild->d_inode->i_mode))
1002                                 err = nfserr_exist;
1003                         else {
1004                                 iap->ia_valid &= ATTR_SIZE;
1005                                 goto set_attr;
1006                         }
1007                         break;
1008                 case NFS3_CREATE_EXCLUSIVE:
1009                         if (   dchild->d_inode->i_mtime == v_mtime
1010                             && dchild->d_inode->i_atime == v_atime
1011                             && dchild->d_inode->i_mode  == v_mode
1012                             && dchild->d_inode->i_size  == 0 )
1013                                 break;
1014                          /* fallthru */
1015                 case NFS3_CREATE_GUARDED:
1016                         err = nfserr_exist;
1017                 }
1018                 goto out;
1019         }
1020 
1021         err = vfs_create(dirp, dchild, iap->ia_mode);
1022         if (err < 0)
1023                 goto out_nfserr;
1024 
1025         if (EX_ISSYNC(fhp->fh_export)) {
1026                 nfsd_sync_dir(dentry);
1027                 /* setattr will sync the child (or not) */
1028         }
1029 
1030         /*
1031          * Update the filehandle to get the new inode info.
1032          */
1033         err = fh_update(resfhp);
1034         if (err)
1035                 goto out;
1036 
1037         if (createmode == NFS3_CREATE_EXCLUSIVE) {
1038                 /* Cram the verifier into atime/mtime/mode */
1039                 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1040                         | ATTR_MTIME_SET|ATTR_ATIME_SET
1041                         | ATTR_MODE;
1042                 iap->ia_mtime = v_mtime;
1043                 iap->ia_atime = v_atime;
1044                 iap->ia_mode  = v_mode;
1045         }
1046 
1047         /* Set file attributes.
1048          * Mode has already been set but we might need to reset it
1049          * for CREATE_EXCLUSIVE
1050          * Irix appears to send along the gid when it tries to
1051          * implement setgid directories via NFS. Clear out all that cruft.
1052          */
1053  set_attr:
1054         if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1055                 err = nfsd_setattr(rqstp, resfhp, iap);
1056 
1057  out:
1058         fh_unlock(fhp);
1059         return err;
1060  
1061  out_nfserr:
1062         err = nfserrno(err);
1063         goto out;
1064 }
1065 #endif /* CONFIG_NFSD_V3 */
1066 
1067 /*
1068  * Read a symlink. On entry, *lenp must contain the maximum path length that
1069  * fits into the buffer. On return, it contains the true length.
1070  * N.B. After this call fhp needs an fh_put
1071  */
1072 int
1073 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1074 {
1075         struct dentry   *dentry;
1076         struct inode    *inode;
1077         mm_segment_t    oldfs;
1078         int             err;
1079 
1080         err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1081         if (err)
1082                 goto out;
1083 
1084         dentry = fhp->fh_dentry;
1085         inode = dentry->d_inode;
1086 
1087         err = nfserr_inval;
1088         if (!inode->i_op || !inode->i_op->readlink)
1089                 goto out;
1090 
1091         UPDATE_ATIME(inode);
1092         /* N.B. Why does this call need a get_fs()??
1093          * Remove the set_fs and watch the fireworks:-) --okir
1094          */
1095 
1096         oldfs = get_fs(); set_fs(KERNEL_DS);
1097         err = inode->i_op->readlink(dentry, buf, *lenp);
1098         set_fs(oldfs);
1099 
1100         if (err < 0)
1101                 goto out_nfserr;
1102         *lenp = err;
1103         err = 0;
1104 out:
1105         return err;
1106 
1107 out_nfserr:
1108         err = nfserrno(err);
1109         goto out;
1110 }
1111 
1112 /*
1113  * Create a symlink and look up its inode
1114  * N.B. After this call _both_ fhp and resfhp need an fh_put
1115  */
1116 int
1117 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1118                                 char *fname, int flen,
1119                                 char *path,  int plen,
1120                                 struct svc_fh *resfhp,
1121                                 struct iattr *iap)
1122 {
1123         struct dentry   *dentry, *dnew;
1124         int             err, cerr;
1125 
1126         err = nfserr_noent;
1127         if (!flen || !plen)
1128                 goto out;
1129         err = nfserr_exist;
1130         if (isdotent(fname, flen))
1131                 goto out;
1132 
1133         err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1134         if (err)
1135                 goto out;
1136         fh_lock(fhp);
1137         dentry = fhp->fh_dentry;
1138         dnew = lookup_one(fname, dentry);
1139         err = PTR_ERR(dnew);
1140         if (IS_ERR(dnew))
1141                 goto out_nfserr;
1142 
1143         err = vfs_symlink(dentry->d_inode, dnew, path);
1144         if (!err) {
1145                 if (EX_ISSYNC(fhp->fh_export))
1146                         nfsd_sync_dir(dentry);
1147                 if (iap) {
1148                         iap->ia_valid &= ATTR_MODE /* ~(ATTR_MODE|ATTR_UID|ATTR_GID)*/;
1149                         if (iap->ia_valid) {
1150                                 iap->ia_valid |= ATTR_CTIME;
1151                                 iap->ia_mode = (iap->ia_mode&S_IALLUGO)
1152                                         | S_IFLNK;
1153                                 err = notify_change(dnew, iap);
1154                                 if (!err && EX_ISSYNC(fhp->fh_export))
1155                                         write_inode_now(dentry->d_inode, 1);
1156                        }
1157                 }
1158         } else
1159                 err = nfserrno(err);
1160         fh_unlock(fhp);
1161 
1162         /* Compose the fh so the dentry will be freed ... */
1163         cerr = fh_compose(resfhp, fhp->fh_export, dnew);
1164         if (err==0) err = cerr;
1165 out:
1166         return err;
1167 
1168 out_nfserr:
1169         err = nfserrno(err);
1170         goto out;
1171 }
1172 
1173 /*
1174  * Create a hardlink
1175  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1176  */
1177 int
1178 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1179                                 char *fname, int len, struct svc_fh *tfhp)
1180 {
1181         struct dentry   *ddir, *dnew, *dold;
1182         struct inode    *dirp, *dest;
1183         int             err;
1184 
1185         err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1186         if (err)
1187                 goto out;
1188         err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1189         if (err)
1190                 goto out;
1191 
1192         err = nfserr_perm;
1193         if (!len)
1194                 goto out;
1195         err = nfserr_exist;
1196         if (isdotent(fname, len))
1197                 goto out;
1198 
1199         fh_lock(ffhp);
1200         ddir = ffhp->fh_dentry;
1201         dirp = ddir->d_inode;
1202 
1203         dnew = lookup_one(fname, ddir);
1204         err = PTR_ERR(dnew);
1205         if (IS_ERR(dnew))
1206                 goto out_nfserr;
1207 
1208         dold = tfhp->fh_dentry;
1209         dest = dold->d_inode;
1210 
1211         err = vfs_link(dold, dirp, dnew);
1212         if (!err) {
1213                 if (EX_ISSYNC(ffhp->fh_export)) {
1214                         nfsd_sync_dir(ddir);
1215                         write_inode_now(dest, 1);
1216                 }
1217         } else {
1218                 if (err == -EXDEV && rqstp->rq_vers == 2)
1219                         err = nfserr_acces;
1220                 else
1221                         err = nfserrno(err);
1222         }
1223 
1224         fh_unlock(ffhp);
1225         dput(dnew);
1226 out:
1227         return err;
1228 
1229 out_nfserr:
1230         err = nfserrno(err);
1231         goto out;
1232 }
1233 
1234 /*
1235  * Rename a file
1236  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1237  */
1238 int
1239 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1240                             struct svc_fh *tfhp, char *tname, int tlen)
1241 {
1242         struct dentry   *fdentry, *tdentry, *odentry, *ndentry;
1243         struct inode    *fdir, *tdir;
1244         int             err;
1245 
1246         err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1247         if (err)
1248                 goto out;
1249         err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1250         if (err)
1251                 goto out;
1252 
1253         fdentry = ffhp->fh_dentry;
1254         fdir = fdentry->d_inode;
1255 
1256         tdentry = tfhp->fh_dentry;
1257         tdir = tdentry->d_inode;
1258 
1259         err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1260         if (fdir->i_dev != tdir->i_dev)
1261                 goto out;
1262 
1263         err = nfserr_perm;
1264         if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1265                 goto out;
1266 
1267         /* cannot use fh_lock as we need deadlock protective ordering
1268          * so do it by hand */
1269         double_down(&tdir->i_sem, &fdir->i_sem);
1270         ffhp->fh_locked = tfhp->fh_locked = 1;
1271         fill_pre_wcc(ffhp);
1272         fill_pre_wcc(tfhp);
1273 
1274         odentry = lookup_one(fname, fdentry);
1275         err = PTR_ERR(odentry);
1276         if (IS_ERR(odentry))
1277                 goto out_nfserr;
1278 
1279         err = -ENOENT;
1280         if (!odentry->d_inode)
1281                 goto out_dput_old;
1282 
1283         ndentry = lookup_one(tname, tdentry);
1284         err = PTR_ERR(ndentry);
1285         if (IS_ERR(ndentry))
1286                 goto out_dput_old;
1287 
1288 
1289 #ifdef MSNFS
1290         if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1291                 ((atomic_read(&odentry->d_count) > 1)
1292                  || (atomic_read(&ndentry->d_count) > 1))) {
1293                         err = nfserr_perm;
1294         } else
1295 #endif
1296         err = vfs_rename(fdir, odentry, tdir, ndentry);
1297         if (!err && EX_ISSYNC(tfhp->fh_export)) {
1298                 nfsd_sync_dir(tdentry);
1299                 nfsd_sync_dir(fdentry);
1300         }
1301         dput(ndentry);
1302 
1303  out_dput_old:
1304         dput(odentry);
1305  out_nfserr:
1306         if (err)
1307                 err = nfserrno(err);
1308 
1309         /* we cannot reply on fh_unlock on the two filehandles,
1310          * as that would do the wrong thing if the two directories
1311          * were the same, so again we do it by hand
1312          */
1313         fill_post_wcc(ffhp);
1314         fill_post_wcc(tfhp);
1315         double_up(&tdir->i_sem, &fdir->i_sem);
1316         ffhp->fh_locked = tfhp->fh_locked = 0;
1317         
1318 out:
1319         return err;
1320 }
1321 
1322 /*
1323  * Unlink a file or directory
1324  * N.B. After this call fhp needs an fh_put
1325  */
1326 int
1327 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1328                                 char *fname, int flen)
1329 {
1330         struct dentry   *dentry, *rdentry;
1331         struct inode    *dirp;
1332         int             err;
1333 
1334         err = nfserr_acces;
1335         if (!flen || isdotent(fname, flen))
1336                 goto out;
1337         err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1338         if (err)
1339                 goto out;
1340 
1341         fh_lock(fhp);
1342         dentry = fhp->fh_dentry;
1343         dirp = dentry->d_inode;
1344 
1345         rdentry = lookup_one(fname, dentry);
1346         err = PTR_ERR(rdentry);
1347         if (IS_ERR(rdentry))
1348                 goto out_nfserr;
1349 
1350         if (!rdentry->d_inode) {
1351                 dput(rdentry);
1352                 err = nfserr_noent;
1353                 goto out;
1354         }
1355 
1356         if (type != S_IFDIR) { /* It's UNLINK */
1357 #ifdef MSNFS
1358                 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1359                         (atomic_read(&rdentry->d_count) > 1)) {
1360                         err = nfserr_perm;
1361                 } else
1362 #endif
1363                 err = vfs_unlink(dirp, rdentry);
1364         } else { /* It's RMDIR */
1365                 err = vfs_rmdir(dirp, rdentry);
1366         }
1367 
1368         dput(rdentry);
1369 
1370         if (err)
1371                 goto out_nfserr;
1372         if (EX_ISSYNC(fhp->fh_export)) 
1373                 nfsd_sync_dir(dentry);
1374 
1375 out:
1376         return err;
1377 
1378 out_nfserr:
1379         err = nfserrno(err);
1380         goto out;
1381 }
1382 
1383 /*
1384  * Read entries from a directory.
1385  * The verifier is an NFSv3 thing we ignore for now.
1386  */
1387 int
1388 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset, 
1389              encode_dent_fn func, u32 *buffer, int *countp, u32 *verf)
1390 {
1391         struct inode    *inode;
1392         u32             *p;
1393         int             oldlen, eof, err;
1394         struct file     file;
1395         struct readdir_cd cd;
1396 
1397         err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1398         if (err)
1399                 goto out;
1400         if (offset > ~(u32) 0)
1401                 goto out_close;
1402 
1403         err = nfserr_notdir;
1404         if (!file.f_op->readdir)
1405                 goto out_close;
1406         file.f_pos = offset;
1407 
1408         /* Set up the readdir context */
1409         memset(&cd, 0, sizeof(cd));
1410         cd.rqstp  = rqstp;
1411         cd.buffer = buffer;
1412         cd.buflen = *countp; /* count of words */
1413         cd.dirfh  = fhp;
1414 
1415         /*
1416          * Read the directory entries. This silly loop is necessary because
1417          * readdir() is not guaranteed to fill up the entire buffer, but
1418          * may choose to do less.
1419          */
1420         inode = file.f_dentry->d_inode;
1421         down(&inode->i_sem);
1422         while (1) {
1423                 oldlen = cd.buflen;
1424 
1425                 /*
1426                 dprintk("nfsd: f_op->readdir(%x/%ld @ %d) buflen = %d (%d)\n",
1427                         file.f_inode->i_dev, file.f_inode->i_ino,
1428                         (int) file.f_pos, (int) oldlen, (int) cd.buflen);
1429                  */
1430                 err = file.f_op->readdir(&file, &cd, (filldir_t) func);
1431                 if (err < 0)
1432                         goto out_nfserr;
1433                 if (oldlen == cd.buflen)
1434                         break;
1435                 if (cd.eob)
1436                         break;
1437         }
1438         up(&inode->i_sem);
1439 
1440         /* If we didn't fill the buffer completely, we're at EOF */
1441         eof = !cd.eob;
1442 
1443         if (cd.offset) {
1444                 if (rqstp->rq_vers == 3)
1445                         (void)xdr_encode_hyper(cd.offset, file.f_pos);
1446                 else
1447                         *cd.offset = htonl(file.f_pos);
1448         }
1449 
1450         p = cd.buffer;
1451         *p++ = 0;                       /* no more entries */
1452         *p++ = htonl(eof);              /* end of directory */
1453         *countp = (caddr_t) p - (caddr_t) buffer;
1454 
1455         dprintk("nfsd: readdir result %d bytes, eof %d offset %d\n",
1456                                 *countp, eof,
1457                                 cd.offset? ntohl(*cd.offset) : -1);
1458         err = 0;
1459 out_close:
1460         nfsd_close(&file);
1461 out:
1462         return err;
1463 
1464 out_nfserr:
1465         up(&inode->i_sem);
1466         err = nfserrno(err);
1467         goto out_close;
1468 }
1469 
1470 /*
1471  * Get file system stats
1472  * N.B. After this call fhp needs an fh_put
1473  */
1474 int
1475 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct statfs *stat)
1476 {
1477         int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1478         if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1479                 err = nfserr_io;
1480         return err;
1481 }
1482 
1483 /*
1484  * Check for a user's access permissions to this inode.
1485  */
1486 int
1487 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1488 {
1489         struct inode    *inode = dentry->d_inode;
1490         int             err;
1491 
1492         if (acc == MAY_NOP)
1493                 return 0;
1494 #if 0
1495         dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1496                 acc,
1497                 (acc & MAY_READ)?       " read"  : "",
1498                 (acc & MAY_WRITE)?      " write" : "",
1499                 (acc & MAY_EXEC)?       " exec"  : "",
1500                 (acc & MAY_SATTR)?      " sattr" : "",
1501                 (acc & MAY_TRUNC)?      " trunc" : "",
1502                 (acc & MAY_LOCK)?       " lock"  : "",
1503                 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1504                 inode->i_mode,
1505                 IS_IMMUTABLE(inode)?    " immut" : "",
1506                 IS_APPEND(inode)?       " append" : "",
1507                 IS_RDONLY(inode)?       " ro" : "");
1508         dprintk("      owner %d/%d user %d/%d\n",
1509                 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1510 #endif
1511 
1512         /* only care about readonly exports for files and
1513          * directories. links don't have meaningful write access,
1514          * and all else is local to the client
1515          */
1516         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) 
1517                 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1518                         if (EX_RDONLY(exp) || IS_RDONLY(inode))
1519                                 return nfserr_rofs;
1520                         if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1521                                 return nfserr_perm;
1522                 }
1523         if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1524                 return nfserr_perm;
1525 
1526         if (acc & MAY_LOCK) {
1527                 /* If we cannot rely on authentication in NLM requests,
1528                  * just allow locks, otherwise require read permission, or
1529                  * ownership
1530                  */
1531                 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1532                         return 0;
1533                 else
1534                         acc = MAY_READ | MAY_OWNER_OVERRIDE;
1535         }
1536         /*
1537          * The file owner always gets access permission for accesses that
1538          * would normally be checked at open time. This is to make
1539          * file access work even when the client has done a fchmod(fd, 0).
1540          *
1541          * However, `cp foo bar' should fail nevertheless when bar is
1542          * readonly. A sensible way to do this might be to reject all
1543          * attempts to truncate a read-only file, because a creat() call
1544          * always implies file truncation.
1545          * ... but this isn't really fair.  A process may reasonably call
1546          * ftruncate on an open file descriptor on a file with perm 000.
1547          * We must trust the client to do permission checking - using "ACCESS"
1548          * with NFSv3.
1549          */
1550         if ((acc & MAY_OWNER_OVERRIDE) &&
1551             inode->i_uid == current->fsuid)
1552                 return 0;
1553 
1554         acc &= ~ MAY_OWNER_OVERRIDE; /* This bit is no longer needed,
1555                                         and gets in the way later */
1556 
1557         err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
1558 
1559         /* Allow read access to binaries even when mode 111 */
1560         if (err == -EACCES && S_ISREG(inode->i_mode) && acc == MAY_READ)
1561                 err = permission(inode, MAY_EXEC);
1562 
1563         return err? nfserrno(err) : 0;
1564 }
1565 
1566 void
1567 nfsd_racache_shutdown(void)
1568 {
1569         if (!raparm_cache)
1570                 return;
1571         dprintk("nfsd: freeing readahead buffers.\n");
1572         kfree(raparml);
1573         raparm_cache = raparml = NULL;
1574 }
1575 /*
1576  * Initialize readahead param cache
1577  */
1578 int
1579 nfsd_racache_init(int cache_size)
1580 {
1581         int     i;
1582 
1583         if (raparm_cache)
1584                 return 0;
1585         raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1586 
1587         if (raparml != NULL) {
1588                 dprintk("nfsd: allocating %d readahead buffers.\n",
1589                         cache_size);
1590                 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1591                 for (i = 0; i < cache_size - 1; i++) {
1592                         raparml[i].p_next = raparml + i + 1;
1593                 }
1594                 raparm_cache = raparml;
1595         } else {
1596                 printk(KERN_WARNING
1597                        "nfsd: Could not allocate memory read-ahead cache.\n");
1598                 return -ENOMEM;
1599         }
1600         nfsdstats.ra_size = cache_size;
1601         return 0;
1602 }
1603 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~ [ freetext search ] ~ [ file search ] ~

This page was automatically generated by the LXR engine.
Visit the LXR main site for more information.