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

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

Version: ~ [ linux-5.9 ] ~ [ linux-5.8.14 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.70 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.150 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.200 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.238 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.238 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * File operations used by nfsd. Some of these have been ripped from
  3  * other parts of the kernel because they weren't exported, others
  4  * are partial duplicates with added or changed functionality.
  5  *
  6  * Note that several functions dget() the dentry upon which they want
  7  * to act, most notably those that create directory entries. Response
  8  * dentry's are dput()'d if necessary in the release callback.
  9  * So if you notice code paths that apparently fail to dput() the
 10  * dentry, don't worry--they have been taken care of.
 11  *
 12  * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
 13  * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
 14  */
 15 
 16 #include <linux/fs.h>
 17 #include <linux/file.h>
 18 #include <linux/splice.h>
 19 #include <linux/fcntl.h>
 20 #include <linux/namei.h>
 21 #include <linux/delay.h>
 22 #include <linux/fsnotify.h>
 23 #include <linux/posix_acl_xattr.h>
 24 #include <linux/xattr.h>
 25 #include <linux/jhash.h>
 26 #include <linux/ima.h>
 27 #include <linux/slab.h>
 28 #include <asm/uaccess.h>
 29 #include <linux/exportfs.h>
 30 #include <linux/writeback.h>
 31 
 32 #ifdef CONFIG_NFSD_V3
 33 #include "xdr3.h"
 34 #endif /* CONFIG_NFSD_V3 */
 35 
 36 #ifdef CONFIG_NFSD_V4
 37 #include "acl.h"
 38 #include "idmap.h"
 39 #endif /* CONFIG_NFSD_V4 */
 40 
 41 #include "nfsd.h"
 42 #include "vfs.h"
 43 
 44 #define NFSDDBG_FACILITY                NFSDDBG_FILEOP
 45 
 46 
 47 /*
 48  * This is a cache of readahead params that help us choose the proper
 49  * readahead strategy. Initially, we set all readahead parameters to 0
 50  * and let the VFS handle things.
 51  * If you increase the number of cached files very much, you'll need to
 52  * add a hash table here.
 53  */
 54 struct raparms {
 55         struct raparms          *p_next;
 56         unsigned int            p_count;
 57         ino_t                   p_ino;
 58         dev_t                   p_dev;
 59         int                     p_set;
 60         struct file_ra_state    p_ra;
 61         unsigned int            p_hindex;
 62 };
 63 
 64 struct raparm_hbucket {
 65         struct raparms          *pb_head;
 66         spinlock_t              pb_lock;
 67 } ____cacheline_aligned_in_smp;
 68 
 69 #define RAPARM_HASH_BITS        4
 70 #define RAPARM_HASH_SIZE        (1<<RAPARM_HASH_BITS)
 71 #define RAPARM_HASH_MASK        (RAPARM_HASH_SIZE-1)
 72 static struct raparm_hbucket    raparm_hash[RAPARM_HASH_SIZE];
 73 
 74 /* 
 75  * Called from nfsd_lookup and encode_dirent. Check if we have crossed 
 76  * a mount point.
 77  * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
 78  *  or nfs_ok having possibly changed *dpp and *expp
 79  */
 80 int
 81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 
 82                         struct svc_export **expp)
 83 {
 84         struct svc_export *exp = *expp, *exp2 = NULL;
 85         struct dentry *dentry = *dpp;
 86         struct path path = {.mnt = mntget(exp->ex_path.mnt),
 87                             .dentry = dget(dentry)};
 88         int err = 0;
 89 
 90         err = follow_down(&path);
 91         if (err < 0)
 92                 goto out;
 93 
 94         exp2 = rqst_exp_get_by_name(rqstp, &path);
 95         if (IS_ERR(exp2)) {
 96                 err = PTR_ERR(exp2);
 97                 /*
 98                  * We normally allow NFS clients to continue
 99                  * "underneath" a mountpoint that is not exported.
100                  * The exception is V4ROOT, where no traversal is ever
101                  * allowed without an explicit export of the new
102                  * directory.
103                  */
104                 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
105                         err = 0;
106                 path_put(&path);
107                 goto out;
108         }
109         if (nfsd_v4client(rqstp) ||
110                 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111                 /* successfully crossed mount point */
112                 /*
113                  * This is subtle: path.dentry is *not* on path.mnt
114                  * at this point.  The only reason we are safe is that
115                  * original mnt is pinned down by exp, so we should
116                  * put path *before* putting exp
117                  */
118                 *dpp = path.dentry;
119                 path.dentry = dentry;
120                 *expp = exp2;
121                 exp2 = exp;
122         }
123         path_put(&path);
124         exp_put(exp2);
125 out:
126         return err;
127 }
128 
129 static void follow_to_parent(struct path *path)
130 {
131         struct dentry *dp;
132 
133         while (path->dentry == path->mnt->mnt_root && follow_up(path))
134                 ;
135         dp = dget_parent(path->dentry);
136         dput(path->dentry);
137         path->dentry = dp;
138 }
139 
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
141 {
142         struct svc_export *exp2;
143         struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144                             .dentry = dget(dparent)};
145 
146         follow_to_parent(&path);
147 
148         exp2 = rqst_exp_parent(rqstp, &path);
149         if (PTR_ERR(exp2) == -ENOENT) {
150                 *dentryp = dget(dparent);
151         } else if (IS_ERR(exp2)) {
152                 path_put(&path);
153                 return PTR_ERR(exp2);
154         } else {
155                 *dentryp = dget(path.dentry);
156                 exp_put(*exp);
157                 *exp = exp2;
158         }
159         path_put(&path);
160         return 0;
161 }
162 
163 /*
164  * For nfsd purposes, we treat V4ROOT exports as though there was an
165  * export at *every* directory.
166  */
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
168 {
169         if (d_mountpoint(dentry))
170                 return 1;
171         if (nfsd4_is_junction(dentry))
172                 return 1;
173         if (!(exp->ex_flags & NFSEXP_V4ROOT))
174                 return 0;
175         return dentry->d_inode != NULL;
176 }
177 
178 __be32
179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
180                    const char *name, unsigned int len,
181                    struct svc_export **exp_ret, struct dentry **dentry_ret)
182 {
183         struct svc_export       *exp;
184         struct dentry           *dparent;
185         struct dentry           *dentry;
186         int                     host_err;
187 
188         dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
189 
190         dparent = fhp->fh_dentry;
191         exp  = fhp->fh_export;
192         exp_get(exp);
193 
194         /* Lookup the name, but don't follow links */
195         if (isdotent(name, len)) {
196                 if (len==1)
197                         dentry = dget(dparent);
198                 else if (dparent != exp->ex_path.dentry)
199                         dentry = dget_parent(dparent);
200                 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
201                         dentry = dget(dparent); /* .. == . just like at / */
202                 else {
203                         /* checking mountpoint crossing is very different when stepping up */
204                         host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
205                         if (host_err)
206                                 goto out_nfserr;
207                 }
208         } else {
209                 fh_lock(fhp);
210                 dentry = lookup_one_len(name, dparent, len);
211                 host_err = PTR_ERR(dentry);
212                 if (IS_ERR(dentry))
213                         goto out_nfserr;
214                 /*
215                  * check if we have crossed a mount point ...
216                  */
217                 if (nfsd_mountpoint(dentry, exp)) {
218                         if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
219                                 dput(dentry);
220                                 goto out_nfserr;
221                         }
222                 }
223         }
224         *dentry_ret = dentry;
225         *exp_ret = exp;
226         return 0;
227 
228 out_nfserr:
229         exp_put(exp);
230         return nfserrno(host_err);
231 }
232 
233 /*
234  * Look up one component of a pathname.
235  * N.B. After this call _both_ fhp and resfh need an fh_put
236  *
237  * If the lookup would cross a mountpoint, and the mounted filesystem
238  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
239  * accepted as it stands and the mounted directory is
240  * returned. Otherwise the covered directory is returned.
241  * NOTE: this mountpoint crossing is not supported properly by all
242  *   clients and is explicitly disallowed for NFSv3
243  *      NeilBrown <neilb@cse.unsw.edu.au>
244  */
245 __be32
246 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
247                                 unsigned int len, struct svc_fh *resfh)
248 {
249         struct svc_export       *exp;
250         struct dentry           *dentry;
251         __be32 err;
252 
253         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
254         if (err)
255                 return err;
256         err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
257         if (err)
258                 return err;
259         err = check_nfsd_access(exp, rqstp);
260         if (err)
261                 goto out;
262         /*
263          * Note: we compose the file handle now, but as the
264          * dentry may be negative, it may need to be updated.
265          */
266         err = fh_compose(resfh, exp, dentry, fhp);
267         if (!err && !dentry->d_inode)
268                 err = nfserr_noent;
269 out:
270         dput(dentry);
271         exp_put(exp);
272         return err;
273 }
274 
275 static int nfsd_break_lease(struct inode *inode)
276 {
277         if (!S_ISREG(inode->i_mode))
278                 return 0;
279         return break_lease(inode, O_WRONLY | O_NONBLOCK);
280 }
281 
282 /*
283  * Commit metadata changes to stable storage.
284  */
285 static int
286 commit_metadata(struct svc_fh *fhp)
287 {
288         struct inode *inode = fhp->fh_dentry->d_inode;
289         const struct export_operations *export_ops = inode->i_sb->s_export_op;
290 
291         if (!EX_ISSYNC(fhp->fh_export))
292                 return 0;
293 
294         if (export_ops->commit_metadata)
295                 return export_ops->commit_metadata(inode);
296         return sync_inode_metadata(inode, 1);
297 }
298 
299 /*
300  * Go over the attributes and take care of the small differences between
301  * NFS semantics and what Linux expects.
302  */
303 static void
304 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
305 {
306         /*
307          * NFSv2 does not differentiate between "set-[ac]time-to-now"
308          * which only requires access, and "set-[ac]time-to-X" which
309          * requires ownership.
310          * So if it looks like it might be "set both to the same time which
311          * is close to now", and if inode_change_ok fails, then we
312          * convert to "set to now" instead of "set to explicit time"
313          *
314          * We only call inode_change_ok as the last test as technically
315          * it is not an interface that we should be using.
316          */
317 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
318 #define MAX_TOUCH_TIME_ERROR (30*60)
319         if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
320             iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
321                 /*
322                  * Looks probable.
323                  *
324                  * Now just make sure time is in the right ballpark.
325                  * Solaris, at least, doesn't seem to care what the time
326                  * request is.  We require it be within 30 minutes of now.
327                  */
328                 time_t delta = iap->ia_atime.tv_sec - get_seconds();
329                 if (delta < 0)
330                         delta = -delta;
331                 if (delta < MAX_TOUCH_TIME_ERROR &&
332                     inode_change_ok(inode, iap) != 0) {
333                         /*
334                          * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
335                          * This will cause notify_change to set these times
336                          * to "now"
337                          */
338                         iap->ia_valid &= ~BOTH_TIME_SET;
339                 }
340         }
341 
342         /* sanitize the mode change */
343         if (iap->ia_valid & ATTR_MODE) {
344                 iap->ia_mode &= S_IALLUGO;
345                 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
346         }
347 
348         /* Revoke setuid/setgid on chown */
349         if (!S_ISDIR(inode->i_mode) &&
350             (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
351              ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) {
352                 iap->ia_valid |= ATTR_KILL_PRIV;
353                 if (iap->ia_valid & ATTR_MODE) {
354                         /* we're setting mode too, just clear the s*id bits */
355                         iap->ia_mode &= ~S_ISUID;
356                         if (iap->ia_mode & S_IXGRP)
357                                 iap->ia_mode &= ~S_ISGID;
358                 } else {
359                         /* set ATTR_KILL_* bits and let VFS handle it */
360                         iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
361                 }
362         }
363 }
364 
365 static __be32
366 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
367                 struct iattr *iap)
368 {
369         struct inode *inode = fhp->fh_dentry->d_inode;
370         int host_err;
371 
372         if (iap->ia_size < inode->i_size) {
373                 __be32 err;
374 
375                 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
376                                 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
377                 if (err)
378                         return err;
379         }
380 
381         host_err = get_write_access(inode);
382         if (host_err)
383                 goto out_nfserrno;
384 
385         host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
386         if (host_err)
387                 goto out_put_write_access;
388         return 0;
389 
390 out_put_write_access:
391         put_write_access(inode);
392 out_nfserrno:
393         return nfserrno(host_err);
394 }
395 
396 /*
397  * Set various file attributes.  After this call fhp needs an fh_put.
398  */
399 __be32
400 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
401              int check_guard, time_t guardtime)
402 {
403         struct dentry   *dentry;
404         struct inode    *inode;
405         int             accmode = NFSD_MAY_SATTR;
406         umode_t         ftype = 0;
407         __be32          err;
408         int             host_err;
409         bool            get_write_count;
410         int             size_change = 0;
411 
412         if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
413                 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
414         if (iap->ia_valid & ATTR_SIZE)
415                 ftype = S_IFREG;
416 
417         /* Callers that do fh_verify should do the fh_want_write: */
418         get_write_count = !fhp->fh_dentry;
419 
420         /* Get inode */
421         err = fh_verify(rqstp, fhp, ftype, accmode);
422         if (err)
423                 goto out;
424         if (get_write_count) {
425                 host_err = fh_want_write(fhp);
426                 if (host_err)
427                         return nfserrno(host_err);
428         }
429 
430         dentry = fhp->fh_dentry;
431         inode = dentry->d_inode;
432 
433         /* Ignore any mode updates on symlinks */
434         if (S_ISLNK(inode->i_mode))
435                 iap->ia_valid &= ~ATTR_MODE;
436 
437         if (!iap->ia_valid)
438                 goto out;
439 
440         nfsd_sanitize_attrs(inode, iap);
441 
442         /*
443          * The size case is special, it changes the file in addition to the
444          * attributes.
445          */
446         if (iap->ia_valid & ATTR_SIZE) {
447                 err = nfsd_get_write_access(rqstp, fhp, iap);
448                 if (err)
449                         goto out;
450                 size_change = 1;
451         }
452 
453         iap->ia_valid |= ATTR_CTIME;
454 
455         if (check_guard && guardtime != inode->i_ctime.tv_sec) {
456                 err = nfserr_notsync;
457                 goto out_put_write_access;
458         }
459 
460         host_err = nfsd_break_lease(inode);
461         if (host_err)
462                 goto out_put_write_access_nfserror;
463 
464         fh_lock(fhp);
465         host_err = notify_change(dentry, iap);
466         fh_unlock(fhp);
467 
468 out_put_write_access_nfserror:
469         err = nfserrno(host_err);
470 out_put_write_access:
471         if (size_change)
472                 put_write_access(inode);
473         if (!err)
474                 commit_metadata(fhp);
475 out:
476         return err;
477 }
478 
479 #if defined(CONFIG_NFSD_V2_ACL) || \
480     defined(CONFIG_NFSD_V3_ACL) || \
481     defined(CONFIG_NFSD_V4)
482 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
483 {
484         ssize_t buflen;
485         ssize_t ret;
486 
487         buflen = vfs_getxattr(dentry, key, NULL, 0);
488         if (buflen <= 0)
489                 return buflen;
490 
491         *buf = kmalloc(buflen, GFP_KERNEL);
492         if (!*buf)
493                 return -ENOMEM;
494 
495         ret = vfs_getxattr(dentry, key, *buf, buflen);
496         if (ret < 0)
497                 kfree(*buf);
498         return ret;
499 }
500 #endif
501 
502 #if defined(CONFIG_NFSD_V4)
503 static int
504 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
505 {
506         int len;
507         size_t buflen;
508         char *buf = NULL;
509         int error = 0;
510 
511         if (!pacl)
512                 return vfs_setxattr(dentry, key, NULL, 0, 0);
513 
514         buflen = posix_acl_xattr_size(pacl->a_count);
515         buf = kmalloc(buflen, GFP_KERNEL);
516         error = -ENOMEM;
517         if (buf == NULL)
518                 goto out;
519 
520         len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
521         if (len < 0) {
522                 error = len;
523                 goto out;
524         }
525 
526         error = vfs_setxattr(dentry, key, buf, len, 0);
527 out:
528         kfree(buf);
529         return error;
530 }
531 
532 __be32
533 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
534     struct nfs4_acl *acl)
535 {
536         __be32 error;
537         int host_error;
538         struct dentry *dentry;
539         struct inode *inode;
540         struct posix_acl *pacl = NULL, *dpacl = NULL;
541         unsigned int flags = 0;
542 
543         /* Get inode */
544         error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
545         if (error)
546                 return error;
547 
548         dentry = fhp->fh_dentry;
549         inode = dentry->d_inode;
550         if (S_ISDIR(inode->i_mode))
551                 flags = NFS4_ACL_DIR;
552 
553         host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
554         if (host_error == -EINVAL) {
555                 return nfserr_attrnotsupp;
556         } else if (host_error < 0)
557                 goto out_nfserr;
558 
559         host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
560         if (host_error < 0)
561                 goto out_release;
562 
563         if (S_ISDIR(inode->i_mode))
564                 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
565 
566 out_release:
567         posix_acl_release(pacl);
568         posix_acl_release(dpacl);
569 out_nfserr:
570         if (host_error == -EOPNOTSUPP)
571                 return nfserr_attrnotsupp;
572         else
573                 return nfserrno(host_error);
574 }
575 
576 static struct posix_acl *
577 _get_posix_acl(struct dentry *dentry, char *key)
578 {
579         void *buf = NULL;
580         struct posix_acl *pacl = NULL;
581         int buflen;
582 
583         buflen = nfsd_getxattr(dentry, key, &buf);
584         if (!buflen)
585                 buflen = -ENODATA;
586         if (buflen <= 0)
587                 return ERR_PTR(buflen);
588 
589         pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
590         kfree(buf);
591         return pacl;
592 }
593 
594 int
595 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
596 {
597         struct inode *inode = dentry->d_inode;
598         int error = 0;
599         struct posix_acl *pacl = NULL, *dpacl = NULL;
600         unsigned int flags = 0;
601 
602         pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
603         if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
604                 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
605         if (IS_ERR(pacl)) {
606                 error = PTR_ERR(pacl);
607                 pacl = NULL;
608                 goto out;
609         }
610 
611         if (S_ISDIR(inode->i_mode)) {
612                 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
613                 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
614                         dpacl = NULL;
615                 else if (IS_ERR(dpacl)) {
616                         error = PTR_ERR(dpacl);
617                         dpacl = NULL;
618                         goto out;
619                 }
620                 flags = NFS4_ACL_DIR;
621         }
622 
623         *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
624         if (IS_ERR(*acl)) {
625                 error = PTR_ERR(*acl);
626                 *acl = NULL;
627         }
628  out:
629         posix_acl_release(pacl);
630         posix_acl_release(dpacl);
631         return error;
632 }
633 
634 /*
635  * NFS junction information is stored in an extended attribute.
636  */
637 #define NFSD_JUNCTION_XATTR_NAME        XATTR_TRUSTED_PREFIX "junction.nfs"
638 
639 /**
640  * nfsd4_is_junction - Test if an object could be an NFS junction
641  *
642  * @dentry: object to test
643  *
644  * Returns 1 if "dentry" appears to contain NFS junction information.
645  * Otherwise 0 is returned.
646  */
647 int nfsd4_is_junction(struct dentry *dentry)
648 {
649         struct inode *inode = dentry->d_inode;
650 
651         if (inode == NULL)
652                 return 0;
653         if (inode->i_mode & S_IXUGO)
654                 return 0;
655         if (!(inode->i_mode & S_ISVTX))
656                 return 0;
657         if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
658                 return 0;
659         return 1;
660 }
661 #endif /* defined(CONFIG_NFSD_V4) */
662 
663 #ifdef CONFIG_NFSD_V3
664 /*
665  * Check server access rights to a file system object
666  */
667 struct accessmap {
668         u32             access;
669         int             how;
670 };
671 static struct accessmap nfs3_regaccess[] = {
672     {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
673     {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
674     {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_TRUNC   },
675     {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE                  },
676 
677     {   0,                      0                               }
678 };
679 
680 static struct accessmap nfs3_diraccess[] = {
681     {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
682     {   NFS3_ACCESS_LOOKUP,     NFSD_MAY_EXEC                   },
683     {   NFS3_ACCESS_MODIFY,     NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
684     {   NFS3_ACCESS_EXTEND,     NFSD_MAY_EXEC|NFSD_MAY_WRITE    },
685     {   NFS3_ACCESS_DELETE,     NFSD_MAY_REMOVE                 },
686 
687     {   0,                      0                               }
688 };
689 
690 static struct accessmap nfs3_anyaccess[] = {
691         /* Some clients - Solaris 2.6 at least, make an access call
692          * to the server to check for access for things like /dev/null
693          * (which really, the server doesn't care about).  So
694          * We provide simple access checking for them, looking
695          * mainly at mode bits, and we make sure to ignore read-only
696          * filesystem checks
697          */
698     {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
699     {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
700     {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
701     {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
702 
703     {   0,                      0                               }
704 };
705 
706 __be32
707 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
708 {
709         struct accessmap        *map;
710         struct svc_export       *export;
711         struct dentry           *dentry;
712         u32                     query, result = 0, sresult = 0;
713         __be32                  error;
714 
715         error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
716         if (error)
717                 goto out;
718 
719         export = fhp->fh_export;
720         dentry = fhp->fh_dentry;
721 
722         if (S_ISREG(dentry->d_inode->i_mode))
723                 map = nfs3_regaccess;
724         else if (S_ISDIR(dentry->d_inode->i_mode))
725                 map = nfs3_diraccess;
726         else
727                 map = nfs3_anyaccess;
728 
729 
730         query = *access;
731         for  (; map->access; map++) {
732                 if (map->access & query) {
733                         __be32 err2;
734 
735                         sresult |= map->access;
736 
737                         err2 = nfsd_permission(rqstp, export, dentry, map->how);
738                         switch (err2) {
739                         case nfs_ok:
740                                 result |= map->access;
741                                 break;
742                                 
743                         /* the following error codes just mean the access was not allowed,
744                          * rather than an error occurred */
745                         case nfserr_rofs:
746                         case nfserr_acces:
747                         case nfserr_perm:
748                                 /* simply don't "or" in the access bit. */
749                                 break;
750                         default:
751                                 error = err2;
752                                 goto out;
753                         }
754                 }
755         }
756         *access = result;
757         if (supported)
758                 *supported = sresult;
759 
760  out:
761         return error;
762 }
763 #endif /* CONFIG_NFSD_V3 */
764 
765 static int nfsd_open_break_lease(struct inode *inode, int access)
766 {
767         unsigned int mode;
768 
769         if (access & NFSD_MAY_NOT_BREAK_LEASE)
770                 return 0;
771         mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
772         return break_lease(inode, mode | O_NONBLOCK);
773 }
774 
775 /*
776  * Open an existing file or directory.
777  * The may_flags argument indicates the type of open (read/write/lock)
778  * and additional flags.
779  * N.B. After this call fhp needs an fh_put
780  */
781 __be32
782 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
783                         int may_flags, struct file **filp)
784 {
785         struct path     path;
786         struct inode    *inode;
787         int             flags = O_RDONLY|O_LARGEFILE;
788         __be32          err;
789         int             host_err = 0;
790 
791         validate_process_creds();
792 
793         /*
794          * If we get here, then the client has already done an "open",
795          * and (hopefully) checked permission - so allow OWNER_OVERRIDE
796          * in case a chmod has now revoked permission.
797          *
798          * Arguably we should also allow the owner override for
799          * directories, but we never have and it doesn't seem to have
800          * caused anyone a problem.  If we were to change this, note
801          * also that our filldir callbacks would need a variant of
802          * lookup_one_len that doesn't check permissions.
803          */
804         if (type == S_IFREG)
805                 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
806         err = fh_verify(rqstp, fhp, type, may_flags);
807         if (err)
808                 goto out;
809 
810         path.mnt = fhp->fh_export->ex_path.mnt;
811         path.dentry = fhp->fh_dentry;
812         inode = path.dentry->d_inode;
813 
814         /* Disallow write access to files with the append-only bit set
815          * or any access when mandatory locking enabled
816          */
817         err = nfserr_perm;
818         if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
819                 goto out;
820         /*
821          * We must ignore files (but only files) which might have mandatory
822          * locks on them because there is no way to know if the accesser has
823          * the lock.
824          */
825         if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
826                 goto out;
827 
828         if (!inode->i_fop)
829                 goto out;
830 
831         host_err = nfsd_open_break_lease(inode, may_flags);
832         if (host_err) /* NOMEM or WOULDBLOCK */
833                 goto out_nfserr;
834 
835         if (may_flags & NFSD_MAY_WRITE) {
836                 if (may_flags & NFSD_MAY_READ)
837                         flags = O_RDWR|O_LARGEFILE;
838                 else
839                         flags = O_WRONLY|O_LARGEFILE;
840         }
841         *filp = dentry_open(&path, flags, current_cred());
842         if (IS_ERR(*filp)) {
843                 host_err = PTR_ERR(*filp);
844                 *filp = NULL;
845         } else {
846                 host_err = ima_file_check(*filp, may_flags);
847 
848                 if (may_flags & NFSD_MAY_64BIT_COOKIE)
849                         (*filp)->f_mode |= FMODE_64BITHASH;
850                 else
851                         (*filp)->f_mode |= FMODE_32BITHASH;
852         }
853 
854 out_nfserr:
855         err = nfserrno(host_err);
856 out:
857         validate_process_creds();
858         return err;
859 }
860 
861 /*
862  * Close a file.
863  */
864 void
865 nfsd_close(struct file *filp)
866 {
867         fput(filp);
868 }
869 
870 /*
871  * Obtain the readahead parameters for the file
872  * specified by (dev, ino).
873  */
874 
875 static inline struct raparms *
876 nfsd_get_raparms(dev_t dev, ino_t ino)
877 {
878         struct raparms  *ra, **rap, **frap = NULL;
879         int depth = 0;
880         unsigned int hash;
881         struct raparm_hbucket *rab;
882 
883         hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
884         rab = &raparm_hash[hash];
885 
886         spin_lock(&rab->pb_lock);
887         for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
888                 if (ra->p_ino == ino && ra->p_dev == dev)
889                         goto found;
890                 depth++;
891                 if (ra->p_count == 0)
892                         frap = rap;
893         }
894         depth = nfsdstats.ra_size;
895         if (!frap) {    
896                 spin_unlock(&rab->pb_lock);
897                 return NULL;
898         }
899         rap = frap;
900         ra = *frap;
901         ra->p_dev = dev;
902         ra->p_ino = ino;
903         ra->p_set = 0;
904         ra->p_hindex = hash;
905 found:
906         if (rap != &rab->pb_head) {
907                 *rap = ra->p_next;
908                 ra->p_next   = rab->pb_head;
909                 rab->pb_head = ra;
910         }
911         ra->p_count++;
912         nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
913         spin_unlock(&rab->pb_lock);
914         return ra;
915 }
916 
917 /*
918  * Grab and keep cached pages associated with a file in the svc_rqst
919  * so that they can be passed to the network sendmsg/sendpage routines
920  * directly. They will be released after the sending has completed.
921  */
922 static int
923 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
924                   struct splice_desc *sd)
925 {
926         struct svc_rqst *rqstp = sd->u.data;
927         struct page **pp = rqstp->rq_next_page;
928         struct page *page = buf->page;
929         size_t size;
930 
931         size = sd->len;
932 
933         if (rqstp->rq_res.page_len == 0) {
934                 get_page(page);
935                 put_page(*rqstp->rq_next_page);
936                 *(rqstp->rq_next_page++) = page;
937                 rqstp->rq_res.page_base = buf->offset;
938                 rqstp->rq_res.page_len = size;
939         } else if (page != pp[-1]) {
940                 get_page(page);
941                 if (*rqstp->rq_next_page)
942                         put_page(*rqstp->rq_next_page);
943                 *(rqstp->rq_next_page++) = page;
944                 rqstp->rq_res.page_len += size;
945         } else
946                 rqstp->rq_res.page_len += size;
947 
948         return size;
949 }
950 
951 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
952                                     struct splice_desc *sd)
953 {
954         return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
955 }
956 
957 static __be32
958 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
959               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
960 {
961         mm_segment_t    oldfs;
962         __be32          err;
963         int             host_err;
964 
965         err = nfserr_perm;
966 
967         if (file->f_op->splice_read && rqstp->rq_splice_ok) {
968                 struct splice_desc sd = {
969                         .len            = 0,
970                         .total_len      = *count,
971                         .pos            = offset,
972                         .u.data         = rqstp,
973                 };
974 
975                 rqstp->rq_next_page = rqstp->rq_respages + 1;
976                 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
977         } else {
978                 oldfs = get_fs();
979                 set_fs(KERNEL_DS);
980                 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
981                 set_fs(oldfs);
982         }
983 
984         if (host_err >= 0) {
985                 nfsdstats.io_read += host_err;
986                 *count = host_err;
987                 err = 0;
988                 fsnotify_access(file);
989         } else 
990                 err = nfserrno(host_err);
991         return err;
992 }
993 
994 static void kill_suid(struct dentry *dentry)
995 {
996         struct iattr    ia;
997         ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
998 
999         mutex_lock(&dentry->d_inode->i_mutex);
1000         notify_change(dentry, &ia);
1001         mutex_unlock(&dentry->d_inode->i_mutex);
1002 }
1003 
1004 /*
1005  * Gathered writes: If another process is currently writing to the file,
1006  * there's a high chance this is another nfsd (triggered by a bulk write
1007  * from a client's biod). Rather than syncing the file with each write
1008  * request, we sleep for 10 msec.
1009  *
1010  * I don't know if this roughly approximates C. Juszak's idea of
1011  * gathered writes, but it's a nice and simple solution (IMHO), and it
1012  * seems to work:-)
1013  *
1014  * Note: we do this only in the NFSv2 case, since v3 and higher have a
1015  * better tool (separate unstable writes and commits) for solving this
1016  * problem.
1017  */
1018 static int wait_for_concurrent_writes(struct file *file)
1019 {
1020         struct inode *inode = file_inode(file);
1021         static ino_t last_ino;
1022         static dev_t last_dev;
1023         int err = 0;
1024 
1025         if (atomic_read(&inode->i_writecount) > 1
1026             || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1027                 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1028                 msleep(10);
1029                 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1030         }
1031 
1032         if (inode->i_state & I_DIRTY) {
1033                 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1034                 err = vfs_fsync(file, 0);
1035         }
1036         last_ino = inode->i_ino;
1037         last_dev = inode->i_sb->s_dev;
1038         return err;
1039 }
1040 
1041 static __be32
1042 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1043                                 loff_t offset, struct kvec *vec, int vlen,
1044                                 unsigned long *cnt, int *stablep)
1045 {
1046         struct svc_export       *exp;
1047         struct dentry           *dentry;
1048         struct inode            *inode;
1049         mm_segment_t            oldfs;
1050         __be32                  err = 0;
1051         int                     host_err;
1052         int                     stable = *stablep;
1053         int                     use_wgather;
1054         loff_t                  pos = offset;
1055 
1056         dentry = file->f_path.dentry;
1057         inode = dentry->d_inode;
1058         exp   = fhp->fh_export;
1059 
1060         use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1061 
1062         if (!EX_ISSYNC(exp))
1063                 stable = 0;
1064 
1065         /* Write the data. */
1066         oldfs = get_fs(); set_fs(KERNEL_DS);
1067         host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1068         set_fs(oldfs);
1069         if (host_err < 0)
1070                 goto out_nfserr;
1071         *cnt = host_err;
1072         nfsdstats.io_write += host_err;
1073         fsnotify_modify(file);
1074 
1075         /* clear setuid/setgid flag after write */
1076         if (inode->i_mode & (S_ISUID | S_ISGID))
1077                 kill_suid(dentry);
1078 
1079         if (stable) {
1080                 if (use_wgather)
1081                         host_err = wait_for_concurrent_writes(file);
1082                 else
1083                         host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1084         }
1085 
1086 out_nfserr:
1087         dprintk("nfsd: write complete host_err=%d\n", host_err);
1088         if (host_err >= 0)
1089                 err = 0;
1090         else
1091                 err = nfserrno(host_err);
1092         return err;
1093 }
1094 
1095 /*
1096  * Read data from a file. count must contain the requested read count
1097  * on entry. On return, *count contains the number of bytes actually read.
1098  * N.B. After this call fhp needs an fh_put
1099  */
1100 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1101         loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1102 {
1103         struct file *file;
1104         struct inode *inode;
1105         struct raparms  *ra;
1106         __be32 err;
1107 
1108         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1109         if (err)
1110                 return err;
1111 
1112         inode = file_inode(file);
1113 
1114         /* Get readahead parameters */
1115         ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1116 
1117         if (ra && ra->p_set)
1118                 file->f_ra = ra->p_ra;
1119 
1120         err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1121 
1122         /* Write back readahead params */
1123         if (ra) {
1124                 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1125                 spin_lock(&rab->pb_lock);
1126                 ra->p_ra = file->f_ra;
1127                 ra->p_set = 1;
1128                 ra->p_count--;
1129                 spin_unlock(&rab->pb_lock);
1130         }
1131 
1132         nfsd_close(file);
1133         return err;
1134 }
1135 
1136 /* As above, but use the provided file descriptor. */
1137 __be32
1138 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1139                 loff_t offset, struct kvec *vec, int vlen,
1140                 unsigned long *count)
1141 {
1142         __be32          err;
1143 
1144         if (file) {
1145                 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1146                                 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1147                 if (err)
1148                         goto out;
1149                 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1150         } else /* Note file may still be NULL in NFSv4 special stateid case: */
1151                 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1152 out:
1153         return err;
1154 }
1155 
1156 /*
1157  * Write data to a file.
1158  * The stable flag requests synchronous writes.
1159  * N.B. After this call fhp needs an fh_put
1160  */
1161 __be32
1162 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1163                 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1164                 int *stablep)
1165 {
1166         __be32                  err = 0;
1167 
1168         if (file) {
1169                 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1170                                 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1171                 if (err)
1172                         goto out;
1173                 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1174                                 stablep);
1175         } else {
1176                 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1177                 if (err)
1178                         goto out;
1179 
1180                 if (cnt)
1181                         err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1182                                              cnt, stablep);
1183                 nfsd_close(file);
1184         }
1185 out:
1186         return err;
1187 }
1188 
1189 #ifdef CONFIG_NFSD_V3
1190 /*
1191  * Commit all pending writes to stable storage.
1192  *
1193  * Note: we only guarantee that data that lies within the range specified
1194  * by the 'offset' and 'count' parameters will be synced.
1195  *
1196  * Unfortunately we cannot lock the file to make sure we return full WCC
1197  * data to the client, as locking happens lower down in the filesystem.
1198  */
1199 __be32
1200 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1201                loff_t offset, unsigned long count)
1202 {
1203         struct file     *file;
1204         loff_t          end = LLONG_MAX;
1205         __be32          err = nfserr_inval;
1206 
1207         if (offset < 0)
1208                 goto out;
1209         if (count != 0) {
1210                 end = offset + (loff_t)count - 1;
1211                 if (end < offset)
1212                         goto out;
1213         }
1214 
1215         err = nfsd_open(rqstp, fhp, S_IFREG,
1216                         NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1217         if (err)
1218                 goto out;
1219         if (EX_ISSYNC(fhp->fh_export)) {
1220                 int err2 = vfs_fsync_range(file, offset, end, 0);
1221 
1222                 if (err2 != -EINVAL)
1223                         err = nfserrno(err2);
1224                 else
1225                         err = nfserr_notsupp;
1226         }
1227 
1228         nfsd_close(file);
1229 out:
1230         return err;
1231 }
1232 #endif /* CONFIG_NFSD_V3 */
1233 
1234 static __be32
1235 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1236                         struct iattr *iap)
1237 {
1238         /*
1239          * Mode has already been set earlier in create:
1240          */
1241         iap->ia_valid &= ~ATTR_MODE;
1242         /*
1243          * Setting uid/gid works only for root.  Irix appears to
1244          * send along the gid on create when it tries to implement
1245          * setgid directories via NFS:
1246          */
1247         if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1248                 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1249         if (iap->ia_valid)
1250                 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1251         return 0;
1252 }
1253 
1254 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1255  * setting size to 0 may fail for some specific file systems by the permission
1256  * checking which requires WRITE permission but the mode is 000.
1257  * we ignore the resizing(to 0) on the just new created file, since the size is
1258  * 0 after file created.
1259  *
1260  * call this only after vfs_create() is called.
1261  * */
1262 static void
1263 nfsd_check_ignore_resizing(struct iattr *iap)
1264 {
1265         if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1266                 iap->ia_valid &= ~ATTR_SIZE;
1267 }
1268 
1269 /*
1270  * Create a file (regular, directory, device, fifo); UNIX sockets 
1271  * not yet implemented.
1272  * If the response fh has been verified, the parent directory should
1273  * already be locked. Note that the parent directory is left locked.
1274  *
1275  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1276  */
1277 __be32
1278 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1279                 char *fname, int flen, struct iattr *iap,
1280                 int type, dev_t rdev, struct svc_fh *resfhp)
1281 {
1282         struct dentry   *dentry, *dchild = NULL;
1283         struct inode    *dirp;
1284         __be32          err;
1285         __be32          err2;
1286         int             host_err;
1287 
1288         err = nfserr_perm;
1289         if (!flen)
1290                 goto out;
1291         err = nfserr_exist;
1292         if (isdotent(fname, flen))
1293                 goto out;
1294 
1295         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1296         if (err)
1297                 goto out;
1298 
1299         dentry = fhp->fh_dentry;
1300         dirp = dentry->d_inode;
1301 
1302         err = nfserr_notdir;
1303         if (!dirp->i_op->lookup)
1304                 goto out;
1305         /*
1306          * Check whether the response file handle has been verified yet.
1307          * If it has, the parent directory should already be locked.
1308          */
1309         if (!resfhp->fh_dentry) {
1310                 host_err = fh_want_write(fhp);
1311                 if (host_err)
1312                         goto out_nfserr;
1313 
1314                 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1315                 fh_lock_nested(fhp, I_MUTEX_PARENT);
1316                 dchild = lookup_one_len(fname, dentry, flen);
1317                 host_err = PTR_ERR(dchild);
1318                 if (IS_ERR(dchild))
1319                         goto out_nfserr;
1320                 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1321                 if (err)
1322                         goto out;
1323         } else {
1324                 /* called from nfsd_proc_create */
1325                 dchild = dget(resfhp->fh_dentry);
1326                 if (!fhp->fh_locked) {
1327                         /* not actually possible */
1328                         printk(KERN_ERR
1329                                 "nfsd_create: parent %s/%s not locked!\n",
1330                                 dentry->d_parent->d_name.name,
1331                                 dentry->d_name.name);
1332                         err = nfserr_io;
1333                         goto out;
1334                 }
1335         }
1336         /*
1337          * Make sure the child dentry is still negative ...
1338          */
1339         err = nfserr_exist;
1340         if (dchild->d_inode) {
1341                 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1342                         dentry->d_name.name, dchild->d_name.name);
1343                 goto out; 
1344         }
1345 
1346         if (!(iap->ia_valid & ATTR_MODE))
1347                 iap->ia_mode = 0;
1348         iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1349 
1350         err = nfserr_inval;
1351         if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1352                 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1353                        type);
1354                 goto out;
1355         }
1356 
1357         /*
1358          * Get the dir op function pointer.
1359          */
1360         err = 0;
1361         host_err = 0;
1362         switch (type) {
1363         case S_IFREG:
1364                 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1365                 if (!host_err)
1366                         nfsd_check_ignore_resizing(iap);
1367                 break;
1368         case S_IFDIR:
1369                 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1370                 break;
1371         case S_IFCHR:
1372         case S_IFBLK:
1373         case S_IFIFO:
1374         case S_IFSOCK:
1375                 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1376                 break;
1377         }
1378         if (host_err < 0)
1379                 goto out_nfserr;
1380 
1381         err = nfsd_create_setattr(rqstp, resfhp, iap);
1382 
1383         /*
1384          * nfsd_setattr already committed the child.  Transactional filesystems
1385          * had a chance to commit changes for both parent and child
1386          * simultaneously making the following commit_metadata a noop.
1387          */
1388         err2 = nfserrno(commit_metadata(fhp));
1389         if (err2)
1390                 err = err2;
1391         /*
1392          * Update the file handle to get the new inode info.
1393          */
1394         if (!err)
1395                 err = fh_update(resfhp);
1396 out:
1397         if (dchild && !IS_ERR(dchild))
1398                 dput(dchild);
1399         return err;
1400 
1401 out_nfserr:
1402         err = nfserrno(host_err);
1403         goto out;
1404 }
1405 
1406 #ifdef CONFIG_NFSD_V3
1407 
1408 static inline int nfsd_create_is_exclusive(int createmode)
1409 {
1410         return createmode == NFS3_CREATE_EXCLUSIVE
1411                || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1412 }
1413 
1414 /*
1415  * NFSv3 and NFSv4 version of nfsd_create
1416  */
1417 __be32
1418 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1419                 char *fname, int flen, struct iattr *iap,
1420                 struct svc_fh *resfhp, int createmode, u32 *verifier,
1421                 bool *truncp, bool *created)
1422 {
1423         struct dentry   *dentry, *dchild = NULL;
1424         struct inode    *dirp;
1425         __be32          err;
1426         int             host_err;
1427         __u32           v_mtime=0, v_atime=0;
1428 
1429         err = nfserr_perm;
1430         if (!flen)
1431                 goto out;
1432         err = nfserr_exist;
1433         if (isdotent(fname, flen))
1434                 goto out;
1435         if (!(iap->ia_valid & ATTR_MODE))
1436                 iap->ia_mode = 0;
1437         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1438         if (err)
1439                 goto out;
1440 
1441         dentry = fhp->fh_dentry;
1442         dirp = dentry->d_inode;
1443 
1444         /* Get all the sanity checks out of the way before
1445          * we lock the parent. */
1446         err = nfserr_notdir;
1447         if (!dirp->i_op->lookup)
1448                 goto out;
1449 
1450         host_err = fh_want_write(fhp);
1451         if (host_err)
1452                 goto out_nfserr;
1453 
1454         fh_lock_nested(fhp, I_MUTEX_PARENT);
1455 
1456         /*
1457          * Compose the response file handle.
1458          */
1459         dchild = lookup_one_len(fname, dentry, flen);
1460         host_err = PTR_ERR(dchild);
1461         if (IS_ERR(dchild))
1462                 goto out_nfserr;
1463 
1464         /* If file doesn't exist, check for permissions to create one */
1465         if (!dchild->d_inode) {
1466                 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1467                 if (err)
1468                         goto out;
1469         }
1470 
1471         err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1472         if (err)
1473                 goto out;
1474 
1475         if (nfsd_create_is_exclusive(createmode)) {
1476                 /* solaris7 gets confused (bugid 4218508) if these have
1477                  * the high bit set, so just clear the high bits. If this is
1478                  * ever changed to use different attrs for storing the
1479                  * verifier, then do_open_lookup() will also need to be fixed
1480                  * accordingly.
1481                  */
1482                 v_mtime = verifier[0]&0x7fffffff;
1483                 v_atime = verifier[1]&0x7fffffff;
1484         }
1485         
1486         if (dchild->d_inode) {
1487                 err = 0;
1488 
1489                 switch (createmode) {
1490                 case NFS3_CREATE_UNCHECKED:
1491                         if (! S_ISREG(dchild->d_inode->i_mode))
1492                                 goto out;
1493                         else if (truncp) {
1494                                 /* in nfsv4, we need to treat this case a little
1495                                  * differently.  we don't want to truncate the
1496                                  * file now; this would be wrong if the OPEN
1497                                  * fails for some other reason.  furthermore,
1498                                  * if the size is nonzero, we should ignore it
1499                                  * according to spec!
1500                                  */
1501                                 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1502                         }
1503                         else {
1504                                 iap->ia_valid &= ATTR_SIZE;
1505                                 goto set_attr;
1506                         }
1507                         break;
1508                 case NFS3_CREATE_EXCLUSIVE:
1509                         if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1510                             && dchild->d_inode->i_atime.tv_sec == v_atime
1511                             && dchild->d_inode->i_size  == 0 ) {
1512                                 if (created)
1513                                         *created = 1;
1514                                 break;
1515                         }
1516                 case NFS4_CREATE_EXCLUSIVE4_1:
1517                         if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1518                             && dchild->d_inode->i_atime.tv_sec == v_atime
1519                             && dchild->d_inode->i_size  == 0 ) {
1520                                 if (created)
1521                                         *created = 1;
1522                                 goto set_attr;
1523                         }
1524                          /* fallthru */
1525                 case NFS3_CREATE_GUARDED:
1526                         err = nfserr_exist;
1527                 }
1528                 fh_drop_write(fhp);
1529                 goto out;
1530         }
1531 
1532         host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1533         if (host_err < 0) {
1534                 fh_drop_write(fhp);
1535                 goto out_nfserr;
1536         }
1537         if (created)
1538                 *created = 1;
1539 
1540         nfsd_check_ignore_resizing(iap);
1541 
1542         if (nfsd_create_is_exclusive(createmode)) {
1543                 /* Cram the verifier into atime/mtime */
1544                 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1545                         | ATTR_MTIME_SET|ATTR_ATIME_SET;
1546                 /* XXX someone who knows this better please fix it for nsec */ 
1547                 iap->ia_mtime.tv_sec = v_mtime;
1548                 iap->ia_atime.tv_sec = v_atime;
1549                 iap->ia_mtime.tv_nsec = 0;
1550                 iap->ia_atime.tv_nsec = 0;
1551         }
1552 
1553  set_attr:
1554         err = nfsd_create_setattr(rqstp, resfhp, iap);
1555 
1556         /*
1557          * nfsd_setattr already committed the child (and possibly also the parent).
1558          */
1559         if (!err)
1560                 err = nfserrno(commit_metadata(fhp));
1561 
1562         /*
1563          * Update the filehandle to get the new inode info.
1564          */
1565         if (!err)
1566                 err = fh_update(resfhp);
1567 
1568  out:
1569         fh_unlock(fhp);
1570         if (dchild && !IS_ERR(dchild))
1571                 dput(dchild);
1572         fh_drop_write(fhp);
1573         return err;
1574  
1575  out_nfserr:
1576         err = nfserrno(host_err);
1577         goto out;
1578 }
1579 #endif /* CONFIG_NFSD_V3 */
1580 
1581 /*
1582  * Read a symlink. On entry, *lenp must contain the maximum path length that
1583  * fits into the buffer. On return, it contains the true length.
1584  * N.B. After this call fhp needs an fh_put
1585  */
1586 __be32
1587 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1588 {
1589         struct inode    *inode;
1590         mm_segment_t    oldfs;
1591         __be32          err;
1592         int             host_err;
1593         struct path path;
1594 
1595         err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1596         if (err)
1597                 goto out;
1598 
1599         path.mnt = fhp->fh_export->ex_path.mnt;
1600         path.dentry = fhp->fh_dentry;
1601         inode = path.dentry->d_inode;
1602 
1603         err = nfserr_inval;
1604         if (!inode->i_op->readlink)
1605                 goto out;
1606 
1607         touch_atime(&path);
1608         /* N.B. Why does this call need a get_fs()??
1609          * Remove the set_fs and watch the fireworks:-) --okir
1610          */
1611 
1612         oldfs = get_fs(); set_fs(KERNEL_DS);
1613         host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1614         set_fs(oldfs);
1615 
1616         if (host_err < 0)
1617                 goto out_nfserr;
1618         *lenp = host_err;
1619         err = 0;
1620 out:
1621         return err;
1622 
1623 out_nfserr:
1624         err = nfserrno(host_err);
1625         goto out;
1626 }
1627 
1628 /*
1629  * Create a symlink and look up its inode
1630  * N.B. After this call _both_ fhp and resfhp need an fh_put
1631  */
1632 __be32
1633 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1634                                 char *fname, int flen,
1635                                 char *path,  int plen,
1636                                 struct svc_fh *resfhp,
1637                                 struct iattr *iap)
1638 {
1639         struct dentry   *dentry, *dnew;
1640         __be32          err, cerr;
1641         int             host_err;
1642 
1643         err = nfserr_noent;
1644         if (!flen || !plen)
1645                 goto out;
1646         err = nfserr_exist;
1647         if (isdotent(fname, flen))
1648                 goto out;
1649 
1650         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1651         if (err)
1652                 goto out;
1653 
1654         host_err = fh_want_write(fhp);
1655         if (host_err)
1656                 goto out_nfserr;
1657 
1658         fh_lock(fhp);
1659         dentry = fhp->fh_dentry;
1660         dnew = lookup_one_len(fname, dentry, flen);
1661         host_err = PTR_ERR(dnew);
1662         if (IS_ERR(dnew))
1663                 goto out_nfserr;
1664 
1665         if (unlikely(path[plen] != 0)) {
1666                 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1667                 if (path_alloced == NULL)
1668                         host_err = -ENOMEM;
1669                 else {
1670                         strncpy(path_alloced, path, plen);
1671                         path_alloced[plen] = 0;
1672                         host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1673                         kfree(path_alloced);
1674                 }
1675         } else
1676                 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1677         err = nfserrno(host_err);
1678         if (!err)
1679                 err = nfserrno(commit_metadata(fhp));
1680         fh_unlock(fhp);
1681 
1682         fh_drop_write(fhp);
1683 
1684         cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1685         dput(dnew);
1686         if (err==0) err = cerr;
1687 out:
1688         return err;
1689 
1690 out_nfserr:
1691         err = nfserrno(host_err);
1692         goto out;
1693 }
1694 
1695 /*
1696  * Create a hardlink
1697  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1698  */
1699 __be32
1700 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1701                                 char *name, int len, struct svc_fh *tfhp)
1702 {
1703         struct dentry   *ddir, *dnew, *dold;
1704         struct inode    *dirp;
1705         __be32          err;
1706         int             host_err;
1707 
1708         err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1709         if (err)
1710                 goto out;
1711         err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1712         if (err)
1713                 goto out;
1714         err = nfserr_isdir;
1715         if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1716                 goto out;
1717         err = nfserr_perm;
1718         if (!len)
1719                 goto out;
1720         err = nfserr_exist;
1721         if (isdotent(name, len))
1722                 goto out;
1723 
1724         host_err = fh_want_write(tfhp);
1725         if (host_err) {
1726                 err = nfserrno(host_err);
1727                 goto out;
1728         }
1729 
1730         fh_lock_nested(ffhp, I_MUTEX_PARENT);
1731         ddir = ffhp->fh_dentry;
1732         dirp = ddir->d_inode;
1733 
1734         dnew = lookup_one_len(name, ddir, len);
1735         host_err = PTR_ERR(dnew);
1736         if (IS_ERR(dnew))
1737                 goto out_nfserr;
1738 
1739         dold = tfhp->fh_dentry;
1740 
1741         err = nfserr_noent;
1742         if (!dold->d_inode)
1743                 goto out_dput;
1744         host_err = nfsd_break_lease(dold->d_inode);
1745         if (host_err) {
1746                 err = nfserrno(host_err);
1747                 goto out_dput;
1748         }
1749         host_err = vfs_link(dold, dirp, dnew);
1750         if (!host_err) {
1751                 err = nfserrno(commit_metadata(ffhp));
1752                 if (!err)
1753                         err = nfserrno(commit_metadata(tfhp));
1754         } else {
1755                 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1756                         err = nfserr_acces;
1757                 else
1758                         err = nfserrno(host_err);
1759         }
1760 out_dput:
1761         dput(dnew);
1762 out_unlock:
1763         fh_unlock(ffhp);
1764         fh_drop_write(tfhp);
1765 out:
1766         return err;
1767 
1768 out_nfserr:
1769         err = nfserrno(host_err);
1770         goto out_unlock;
1771 }
1772 
1773 /*
1774  * Rename a file
1775  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1776  */
1777 __be32
1778 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1779                             struct svc_fh *tfhp, char *tname, int tlen)
1780 {
1781         struct dentry   *fdentry, *tdentry, *odentry, *ndentry, *trap;
1782         struct inode    *fdir, *tdir;
1783         __be32          err;
1784         int             host_err;
1785 
1786         err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1787         if (err)
1788                 goto out;
1789         err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1790         if (err)
1791                 goto out;
1792 
1793         fdentry = ffhp->fh_dentry;
1794         fdir = fdentry->d_inode;
1795 
1796         tdentry = tfhp->fh_dentry;
1797         tdir = tdentry->d_inode;
1798 
1799         err = nfserr_perm;
1800         if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1801                 goto out;
1802 
1803         host_err = fh_want_write(ffhp);
1804         if (host_err) {
1805                 err = nfserrno(host_err);
1806                 goto out;
1807         }
1808 
1809         /* cannot use fh_lock as we need deadlock protective ordering
1810          * so do it by hand */
1811         trap = lock_rename(tdentry, fdentry);
1812         ffhp->fh_locked = tfhp->fh_locked = 1;
1813         fill_pre_wcc(ffhp);
1814         fill_pre_wcc(tfhp);
1815 
1816         odentry = lookup_one_len(fname, fdentry, flen);
1817         host_err = PTR_ERR(odentry);
1818         if (IS_ERR(odentry))
1819                 goto out_nfserr;
1820 
1821         host_err = -ENOENT;
1822         if (!odentry->d_inode)
1823                 goto out_dput_old;
1824         host_err = -EINVAL;
1825         if (odentry == trap)
1826                 goto out_dput_old;
1827 
1828         ndentry = lookup_one_len(tname, tdentry, tlen);
1829         host_err = PTR_ERR(ndentry);
1830         if (IS_ERR(ndentry))
1831                 goto out_dput_old;
1832         host_err = -ENOTEMPTY;
1833         if (ndentry == trap)
1834                 goto out_dput_new;
1835 
1836         host_err = -EXDEV;
1837         if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1838                 goto out_dput_new;
1839         if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1840                 goto out_dput_new;
1841 
1842         host_err = nfsd_break_lease(odentry->d_inode);
1843         if (host_err)
1844                 goto out_dput_new;
1845         if (ndentry->d_inode) {
1846                 host_err = nfsd_break_lease(ndentry->d_inode);
1847                 if (host_err)
1848                         goto out_dput_new;
1849         }
1850         host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1851         if (!host_err) {
1852                 host_err = commit_metadata(tfhp);
1853                 if (!host_err)
1854                         host_err = commit_metadata(ffhp);
1855         }
1856  out_dput_new:
1857         dput(ndentry);
1858  out_dput_old:
1859         dput(odentry);
1860  out_nfserr:
1861         err = nfserrno(host_err);
1862 
1863         /* we cannot reply on fh_unlock on the two filehandles,
1864          * as that would do the wrong thing if the two directories
1865          * were the same, so again we do it by hand
1866          */
1867         fill_post_wcc(ffhp);
1868         fill_post_wcc(tfhp);
1869         unlock_rename(tdentry, fdentry);
1870         ffhp->fh_locked = tfhp->fh_locked = 0;
1871         fh_drop_write(ffhp);
1872 
1873 out:
1874         return err;
1875 }
1876 
1877 /*
1878  * Unlink a file or directory
1879  * N.B. After this call fhp needs an fh_put
1880  */
1881 __be32
1882 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1883                                 char *fname, int flen)
1884 {
1885         struct dentry   *dentry, *rdentry;
1886         struct inode    *dirp;
1887         __be32          err;
1888         int             host_err;
1889 
1890         err = nfserr_acces;
1891         if (!flen || isdotent(fname, flen))
1892                 goto out;
1893         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1894         if (err)
1895                 goto out;
1896 
1897         host_err = fh_want_write(fhp);
1898         if (host_err)
1899                 goto out_nfserr;
1900 
1901         fh_lock_nested(fhp, I_MUTEX_PARENT);
1902         dentry = fhp->fh_dentry;
1903         dirp = dentry->d_inode;
1904 
1905         rdentry = lookup_one_len(fname, dentry, flen);
1906         host_err = PTR_ERR(rdentry);
1907         if (IS_ERR(rdentry))
1908                 goto out_nfserr;
1909 
1910         if (!rdentry->d_inode) {
1911                 dput(rdentry);
1912                 err = nfserr_noent;
1913                 goto out;
1914         }
1915 
1916         if (!type)
1917                 type = rdentry->d_inode->i_mode & S_IFMT;
1918 
1919         host_err = nfsd_break_lease(rdentry->d_inode);
1920         if (host_err)
1921                 goto out_put;
1922         if (type != S_IFDIR)
1923                 host_err = vfs_unlink(dirp, rdentry);
1924         else
1925                 host_err = vfs_rmdir(dirp, rdentry);
1926         if (!host_err)
1927                 host_err = commit_metadata(fhp);
1928 out_put:
1929         dput(rdentry);
1930 
1931 out_nfserr:
1932         err = nfserrno(host_err);
1933 out:
1934         return err;
1935 }
1936 
1937 /*
1938  * We do this buffering because we must not call back into the file
1939  * system's ->lookup() method from the filldir callback. That may well
1940  * deadlock a number of file systems.
1941  *
1942  * This is based heavily on the implementation of same in XFS.
1943  */
1944 struct buffered_dirent {
1945         u64             ino;
1946         loff_t          offset;
1947         int             namlen;
1948         unsigned int    d_type;
1949         char            name[];
1950 };
1951 
1952 struct readdir_data {
1953         char            *dirent;
1954         size_t          used;
1955         int             full;
1956 };
1957 
1958 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1959                                  loff_t offset, u64 ino, unsigned int d_type)
1960 {
1961         struct readdir_data *buf = __buf;
1962         struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1963         unsigned int reclen;
1964 
1965         reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1966         if (buf->used + reclen > PAGE_SIZE) {
1967                 buf->full = 1;
1968                 return -EINVAL;
1969         }
1970 
1971         de->namlen = namlen;
1972         de->offset = offset;
1973         de->ino = ino;
1974         de->d_type = d_type;
1975         memcpy(de->name, name, namlen);
1976         buf->used += reclen;
1977 
1978         return 0;
1979 }
1980 
1981 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1982                                     struct readdir_cd *cdp, loff_t *offsetp)
1983 {
1984         struct readdir_data buf;
1985         struct buffered_dirent *de;
1986         int host_err;
1987         int size;
1988         loff_t offset;
1989 
1990         buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1991         if (!buf.dirent)
1992                 return nfserrno(-ENOMEM);
1993 
1994         offset = *offsetp;
1995 
1996         while (1) {
1997                 struct inode *dir_inode = file_inode(file);
1998                 unsigned int reclen;
1999 
2000                 cdp->err = nfserr_eof; /* will be cleared on successful read */
2001                 buf.used = 0;
2002                 buf.full = 0;
2003 
2004                 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2005                 if (buf.full)
2006                         host_err = 0;
2007 
2008                 if (host_err < 0)
2009                         break;
2010 
2011                 size = buf.used;
2012 
2013                 if (!size)
2014                         break;
2015 
2016                 /*
2017                  * Various filldir functions may end up calling back into
2018                  * lookup_one_len() and the file system's ->lookup() method.
2019                  * These expect i_mutex to be held, as it would within readdir.
2020                  */
2021                 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2022                 if (host_err)
2023                         break;
2024 
2025                 de = (struct buffered_dirent *)buf.dirent;
2026                 while (size > 0) {
2027                         offset = de->offset;
2028 
2029                         if (func(cdp, de->name, de->namlen, de->offset,
2030                                  de->ino, de->d_type))
2031                                 break;
2032 
2033                         if (cdp->err != nfs_ok)
2034                                 break;
2035 
2036                         reclen = ALIGN(sizeof(*de) + de->namlen,
2037                                        sizeof(u64));
2038                         size -= reclen;
2039                         de = (struct buffered_dirent *)((char *)de + reclen);
2040                 }
2041                 mutex_unlock(&dir_inode->i_mutex);
2042                 if (size > 0) /* We bailed out early */
2043                         break;
2044 
2045                 offset = vfs_llseek(file, 0, SEEK_CUR);
2046         }
2047 
2048         free_page((unsigned long)(buf.dirent));
2049 
2050         if (host_err)
2051                 return nfserrno(host_err);
2052 
2053         *offsetp = offset;
2054         return cdp->err;
2055 }
2056 
2057 /*
2058  * Read entries from a directory.
2059  * The  NFSv3/4 verifier we ignore for now.
2060  */
2061 __be32
2062 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 
2063              struct readdir_cd *cdp, filldir_t func)
2064 {
2065         __be32          err;
2066         struct file     *file;
2067         loff_t          offset = *offsetp;
2068         int             may_flags = NFSD_MAY_READ;
2069 
2070         /* NFSv2 only supports 32 bit cookies */
2071         if (rqstp->rq_vers > 2)
2072                 may_flags |= NFSD_MAY_64BIT_COOKIE;
2073 
2074         err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2075         if (err)
2076                 goto out;
2077 
2078         offset = vfs_llseek(file, offset, SEEK_SET);
2079         if (offset < 0) {
2080                 err = nfserrno((int)offset);
2081                 goto out_close;
2082         }
2083 
2084         err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2085 
2086         if (err == nfserr_eof || err == nfserr_toosmall)
2087                 err = nfs_ok; /* can still be found in ->err */
2088 out_close:
2089         nfsd_close(file);
2090 out:
2091         return err;
2092 }
2093 
2094 /*
2095  * Get file system stats
2096  * N.B. After this call fhp needs an fh_put
2097  */
2098 __be32
2099 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2100 {
2101         __be32 err;
2102 
2103         err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2104         if (!err) {
2105                 struct path path = {
2106                         .mnt    = fhp->fh_export->ex_path.mnt,
2107                         .dentry = fhp->fh_dentry,
2108                 };
2109                 if (vfs_statfs(&path, stat))
2110                         err = nfserr_io;
2111         }
2112         return err;
2113 }
2114 
2115 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2116 {
2117         return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2118 }
2119 
2120 /*
2121  * Check for a user's access permissions to this inode.
2122  */
2123 __be32
2124 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2125                                         struct dentry *dentry, int acc)
2126 {
2127         struct inode    *inode = dentry->d_inode;
2128         int             err;
2129 
2130         if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2131                 return 0;
2132 #if 0
2133         dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2134                 acc,
2135                 (acc & NFSD_MAY_READ)?  " read"  : "",
2136                 (acc & NFSD_MAY_WRITE)? " write" : "",
2137                 (acc & NFSD_MAY_EXEC)?  " exec"  : "",
2138                 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2139                 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2140                 (acc & NFSD_MAY_LOCK)?  " lock"  : "",
2141                 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2142                 inode->i_mode,
2143                 IS_IMMUTABLE(inode)?    " immut" : "",
2144                 IS_APPEND(inode)?       " append" : "",
2145                 __mnt_is_readonly(exp->ex_path.mnt)?    " ro" : "");
2146         dprintk("      owner %d/%d user %d/%d\n",
2147                 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2148 #endif
2149 
2150         /* Normally we reject any write/sattr etc access on a read-only file
2151          * system.  But if it is IRIX doing check on write-access for a 
2152          * device special file, we ignore rofs.
2153          */
2154         if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2155                 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2156                         if (exp_rdonly(rqstp, exp) ||
2157                             __mnt_is_readonly(exp->ex_path.mnt))
2158                                 return nfserr_rofs;
2159                         if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2160                                 return nfserr_perm;
2161                 }
2162         if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2163                 return nfserr_perm;
2164 
2165         if (acc & NFSD_MAY_LOCK) {
2166                 /* If we cannot rely on authentication in NLM requests,
2167                  * just allow locks, otherwise require read permission, or
2168                  * ownership
2169                  */
2170                 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2171                         return 0;
2172                 else
2173                         acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2174         }
2175         /*
2176          * The file owner always gets access permission for accesses that
2177          * would normally be checked at open time. This is to make
2178          * file access work even when the client has done a fchmod(fd, 0).
2179          *
2180          * However, `cp foo bar' should fail nevertheless when bar is
2181          * readonly. A sensible way to do this might be to reject all
2182          * attempts to truncate a read-only file, because a creat() call
2183          * always implies file truncation.
2184          * ... but this isn't really fair.  A process may reasonably call
2185          * ftruncate on an open file descriptor on a file with perm 000.
2186          * We must trust the client to do permission checking - using "ACCESS"
2187          * with NFSv3.
2188          */
2189         if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2190             uid_eq(inode->i_uid, current_fsuid()))
2191                 return 0;
2192 
2193         /* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2194         err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2195 
2196         /* Allow read access to binaries even when mode 111 */
2197         if (err == -EACCES && S_ISREG(inode->i_mode) &&
2198              (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2199               acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2200                 err = inode_permission(inode, MAY_EXEC);
2201 
2202         return err? nfserrno(err) : 0;
2203 }
2204 
2205 void
2206 nfsd_racache_shutdown(void)
2207 {
2208         struct raparms *raparm, *last_raparm;
2209         unsigned int i;
2210 
2211         dprintk("nfsd: freeing readahead buffers.\n");
2212 
2213         for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2214                 raparm = raparm_hash[i].pb_head;
2215                 while(raparm) {
2216                         last_raparm = raparm;
2217                         raparm = raparm->p_next;
2218                         kfree(last_raparm);
2219                 }
2220                 raparm_hash[i].pb_head = NULL;
2221         }
2222 }
2223 /*
2224  * Initialize readahead param cache
2225  */
2226 int
2227 nfsd_racache_init(int cache_size)
2228 {
2229         int     i;
2230         int     j = 0;
2231         int     nperbucket;
2232         struct raparms **raparm = NULL;
2233 
2234 
2235         if (raparm_hash[0].pb_head)
2236                 return 0;
2237         nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2238         if (nperbucket < 2)
2239                 nperbucket = 2;
2240         cache_size = nperbucket * RAPARM_HASH_SIZE;
2241 
2242         dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2243 
2244         for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2245                 spin_lock_init(&raparm_hash[i].pb_lock);
2246 
2247                 raparm = &raparm_hash[i].pb_head;
2248                 for (j = 0; j < nperbucket; j++) {
2249                         *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2250                         if (!*raparm)
2251                                 goto out_nomem;
2252                         raparm = &(*raparm)->p_next;
2253                 }
2254                 *raparm = NULL;
2255         }
2256 
2257         nfsdstats.ra_size = cache_size;
2258         return 0;
2259 
2260 out_nomem:
2261         dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2262         nfsd_racache_shutdown();
2263         return -ENOMEM;
2264 }
2265 
2266 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2267 struct posix_acl *
2268 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2269 {
2270         struct inode *inode = fhp->fh_dentry->d_inode;
2271         char *name;
2272         void *value = NULL;
2273         ssize_t size;
2274         struct posix_acl *acl;
2275 
2276         if (!IS_POSIXACL(inode))
2277                 return ERR_PTR(-EOPNOTSUPP);
2278 
2279         switch (type) {
2280         case ACL_TYPE_ACCESS:
2281                 name = POSIX_ACL_XATTR_ACCESS;
2282                 break;
2283         case ACL_TYPE_DEFAULT:
2284                 name = POSIX_ACL_XATTR_DEFAULT;
2285                 break;
2286         default:
2287                 return ERR_PTR(-EOPNOTSUPP);
2288         }
2289 
2290         size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2291         if (size < 0)
2292                 return ERR_PTR(size);
2293 
2294         acl = posix_acl_from_xattr(&init_user_ns, value, size);
2295         kfree(value);
2296         return acl;
2297 }
2298 
2299 int
2300 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2301 {
2302         struct inode *inode = fhp->fh_dentry->d_inode;
2303         char *name;
2304         void *value = NULL;
2305         size_t size;
2306         int error;
2307 
2308         if (!IS_POSIXACL(inode) ||
2309             !inode->i_op->setxattr || !inode->i_op->removexattr)
2310                 return -EOPNOTSUPP;
2311         switch(type) {
2312                 case ACL_TYPE_ACCESS:
2313                         name = POSIX_ACL_XATTR_ACCESS;
2314                         break;
2315                 case ACL_TYPE_DEFAULT:
2316                         name = POSIX_ACL_XATTR_DEFAULT;
2317                         break;
2318                 default:
2319                         return -EOPNOTSUPP;
2320         }
2321 
2322         if (acl && acl->a_count) {
2323                 size = posix_acl_xattr_size(acl->a_count);
2324                 value = kmalloc(size, GFP_KERNEL);
2325                 if (!value)
2326                         return -ENOMEM;
2327                 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2328                 if (error < 0)
2329                         goto getout;
2330                 size = error;
2331         } else
2332                 size = 0;
2333 
2334         error = fh_want_write(fhp);
2335         if (error)
2336                 goto getout;
2337         if (size)
2338                 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2339         else {
2340                 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2341                         error = 0;
2342                 else {
2343                         error = vfs_removexattr(fhp->fh_dentry, name);
2344                         if (error == -ENODATA)
2345                                 error = 0;
2346                 }
2347         }
2348         fh_drop_write(fhp);
2349 
2350 getout:
2351         kfree(value);
2352         return error;
2353 }
2354 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2355 

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

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp