1 #define MSNFS /* HACK HACK */
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
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.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
20 #include <linux/file.h>
21 #include <linux/splice.h>
22 #include <linux/fcntl.h>
23 #include <linux/namei.h>
24 #include <linux/delay.h>
25 #include <linux/quotaops.h>
26 #include <linux/fsnotify.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/xattr.h>
29 #include <linux/jhash.h>
30 #include <linux/ima.h>
31 #include <asm/uaccess.h>
35 #endif /* CONFIG_NFSD_V3 */
38 #include <linux/nfs4_acl.h>
39 #include <linux/nfsd_idmap.h>
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms *p_next;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
68 } ____cacheline_aligned_in_smp;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
76 nfsd_v4client(struct svc_rqst *rq)
78 return rq->rq_prog == NFS_PROGRAM && rq->rq_vers == 4;
82 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
84 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
85 * or nfs_ok having possibly changed *dpp and *expp
88 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
89 struct svc_export **expp)
91 struct svc_export *exp = *expp, *exp2 = NULL;
92 struct dentry *dentry = *dpp;
93 struct path path = {.mnt = mntget(exp->ex_path.mnt),
94 .dentry = dget(dentry)};
97 while (d_mountpoint(path.dentry) && follow_down(&path))
100 exp2 = rqst_exp_get_by_name(rqstp, &path);
102 if (PTR_ERR(exp2) != -ENOENT)
107 if (nfsd_v4client(rqstp) ||
108 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
109 /* successfully crossed mount point */
111 * This is subtle: path.dentry is *not* on path.mnt
112 * at this point. The only reason we are safe is that
113 * original mnt is pinned down by exp, so we should
114 * put path *before* putting exp
117 path.dentry = dentry;
127 static void follow_to_parent(struct path *path)
131 while (path->dentry == path->mnt->mnt_root && follow_up(path))
133 dp = dget_parent(path->dentry);
138 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
140 struct svc_export *exp2;
141 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
142 .dentry = dget(dparent)};
144 follow_to_parent(&path);
146 exp2 = rqst_exp_parent(rqstp, &path);
147 if (PTR_ERR(exp2) == -ENOENT) {
148 *dentryp = dget(dparent);
149 } else if (IS_ERR(exp2)) {
151 return PTR_ERR(exp2);
153 *dentryp = dget(path.dentry);
162 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
163 const char *name, unsigned int len,
164 struct svc_export **exp_ret, struct dentry **dentry_ret)
166 struct svc_export *exp;
167 struct dentry *dparent;
168 struct dentry *dentry;
172 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
174 /* Obtain dentry and export. */
175 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
179 dparent = fhp->fh_dentry;
180 exp = fhp->fh_export;
183 /* Lookup the name, but don't follow links */
184 if (isdotent(name, len)) {
186 dentry = dget(dparent);
187 else if (dparent != exp->ex_path.dentry)
188 dentry = dget_parent(dparent);
189 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
190 dentry = dget(dparent); /* .. == . just like at / */
192 /* checking mountpoint crossing is very different when stepping up */
193 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
199 dentry = lookup_one_len(name, dparent, len);
200 host_err = PTR_ERR(dentry);
204 * check if we have crossed a mount point ...
206 if (d_mountpoint(dentry)) {
207 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
213 *dentry_ret = dentry;
219 return nfserrno(host_err);
223 * Look up one component of a pathname.
224 * N.B. After this call _both_ fhp and resfh need an fh_put
226 * If the lookup would cross a mountpoint, and the mounted filesystem
227 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
228 * accepted as it stands and the mounted directory is
229 * returned. Otherwise the covered directory is returned.
230 * NOTE: this mountpoint crossing is not supported properly by all
231 * clients and is explicitly disallowed for NFSv3
232 * NeilBrown <neilb@cse.unsw.edu.au>
235 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
236 unsigned int len, struct svc_fh *resfh)
238 struct svc_export *exp;
239 struct dentry *dentry;
242 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
245 err = check_nfsd_access(exp, rqstp);
249 * Note: we compose the file handle now, but as the
250 * dentry may be negative, it may need to be updated.
252 err = fh_compose(resfh, exp, dentry, fhp);
253 if (!err && !dentry->d_inode)
263 * Set various file attributes.
264 * N.B. After this call fhp needs an fh_put
267 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
268 int check_guard, time_t guardtime)
270 struct dentry *dentry;
272 int accmode = NFSD_MAY_SATTR;
278 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
279 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
280 if (iap->ia_valid & ATTR_SIZE)
284 err = fh_verify(rqstp, fhp, ftype, accmode);
288 dentry = fhp->fh_dentry;
289 inode = dentry->d_inode;
291 /* Ignore any mode updates on symlinks */
292 if (S_ISLNK(inode->i_mode))
293 iap->ia_valid &= ~ATTR_MODE;
299 * NFSv2 does not differentiate between "set-[ac]time-to-now"
300 * which only requires access, and "set-[ac]time-to-X" which
301 * requires ownership.
302 * So if it looks like it might be "set both to the same time which
303 * is close to now", and if inode_change_ok fails, then we
304 * convert to "set to now" instead of "set to explicit time"
306 * We only call inode_change_ok as the last test as technically
307 * it is not an interface that we should be using. It is only
308 * valid if the filesystem does not define it's own i_op->setattr.
310 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
311 #define MAX_TOUCH_TIME_ERROR (30*60)
312 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
313 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
317 * Now just make sure time is in the right ballpark.
318 * Solaris, at least, doesn't seem to care what the time
319 * request is. We require it be within 30 minutes of now.
321 time_t delta = iap->ia_atime.tv_sec - get_seconds();
324 if (delta < MAX_TOUCH_TIME_ERROR &&
325 inode_change_ok(inode, iap) != 0) {
327 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
328 * This will cause notify_change to set these times
331 iap->ia_valid &= ~BOTH_TIME_SET;
336 * The size case is special.
337 * It changes the file as well as the attributes.
339 if (iap->ia_valid & ATTR_SIZE) {
340 if (iap->ia_size < inode->i_size) {
341 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
342 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
348 * If we are changing the size of the file, then
349 * we need to break all leases.
351 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
352 if (host_err == -EWOULDBLOCK)
353 host_err = -ETIMEDOUT;
354 if (host_err) /* ENOMEM or EWOULDBLOCK */
357 host_err = get_write_access(inode);
362 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
364 put_write_access(inode);
370 /* sanitize the mode change */
371 if (iap->ia_valid & ATTR_MODE) {
372 iap->ia_mode &= S_IALLUGO;
373 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
376 /* Revoke setuid/setgid on chown */
377 if (!S_ISDIR(inode->i_mode) &&
378 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
379 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
380 iap->ia_valid |= ATTR_KILL_PRIV;
381 if (iap->ia_valid & ATTR_MODE) {
382 /* we're setting mode too, just clear the s*id bits */
383 iap->ia_mode &= ~S_ISUID;
384 if (iap->ia_mode & S_IXGRP)
385 iap->ia_mode &= ~S_ISGID;
387 /* set ATTR_KILL_* bits and let VFS handle it */
388 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
392 /* Change the attributes. */
394 iap->ia_valid |= ATTR_CTIME;
396 err = nfserr_notsync;
397 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
399 host_err = notify_change(dentry, iap);
400 err = nfserrno(host_err);
404 put_write_access(inode);
406 if (EX_ISSYNC(fhp->fh_export))
407 write_inode_now(inode, 1);
412 err = nfserrno(host_err);
416 #if defined(CONFIG_NFSD_V2_ACL) || \
417 defined(CONFIG_NFSD_V3_ACL) || \
418 defined(CONFIG_NFSD_V4)
419 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
424 buflen = vfs_getxattr(dentry, key, NULL, 0);
428 *buf = kmalloc(buflen, GFP_KERNEL);
432 ret = vfs_getxattr(dentry, key, *buf, buflen);
439 #if defined(CONFIG_NFSD_V4)
441 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
448 buflen = posix_acl_xattr_size(pacl->a_count);
449 buf = kmalloc(buflen, GFP_KERNEL);
454 len = posix_acl_to_xattr(pacl, buf, buflen);
460 error = vfs_setxattr(dentry, key, buf, len, 0);
467 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
468 struct nfs4_acl *acl)
472 struct dentry *dentry;
474 struct posix_acl *pacl = NULL, *dpacl = NULL;
475 unsigned int flags = 0;
478 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
482 dentry = fhp->fh_dentry;
483 inode = dentry->d_inode;
484 if (S_ISDIR(inode->i_mode))
485 flags = NFS4_ACL_DIR;
487 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
488 if (host_error == -EINVAL) {
489 return nfserr_attrnotsupp;
490 } else if (host_error < 0)
493 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
497 if (S_ISDIR(inode->i_mode))
498 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
501 posix_acl_release(pacl);
502 posix_acl_release(dpacl);
504 if (host_error == -EOPNOTSUPP)
505 return nfserr_attrnotsupp;
507 return nfserrno(host_error);
510 static struct posix_acl *
511 _get_posix_acl(struct dentry *dentry, char *key)
514 struct posix_acl *pacl = NULL;
517 buflen = nfsd_getxattr(dentry, key, &buf);
521 return ERR_PTR(buflen);
523 pacl = posix_acl_from_xattr(buf, buflen);
529 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
531 struct inode *inode = dentry->d_inode;
533 struct posix_acl *pacl = NULL, *dpacl = NULL;
534 unsigned int flags = 0;
536 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
537 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
538 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
540 error = PTR_ERR(pacl);
545 if (S_ISDIR(inode->i_mode)) {
546 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
547 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
549 else if (IS_ERR(dpacl)) {
550 error = PTR_ERR(dpacl);
554 flags = NFS4_ACL_DIR;
557 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
559 error = PTR_ERR(*acl);
563 posix_acl_release(pacl);
564 posix_acl_release(dpacl);
568 #endif /* defined(CONFIG_NFS_V4) */
570 #ifdef CONFIG_NFSD_V3
572 * Check server access rights to a file system object
578 static struct accessmap nfs3_regaccess[] = {
579 { NFS3_ACCESS_READ, NFSD_MAY_READ },
580 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
581 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
582 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
587 static struct accessmap nfs3_diraccess[] = {
588 { NFS3_ACCESS_READ, NFSD_MAY_READ },
589 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
590 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
591 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
592 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
597 static struct accessmap nfs3_anyaccess[] = {
598 /* Some clients - Solaris 2.6 at least, make an access call
599 * to the server to check for access for things like /dev/null
600 * (which really, the server doesn't care about). So
601 * We provide simple access checking for them, looking
602 * mainly at mode bits, and we make sure to ignore read-only
605 { NFS3_ACCESS_READ, NFSD_MAY_READ },
606 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
607 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
608 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
614 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
616 struct accessmap *map;
617 struct svc_export *export;
618 struct dentry *dentry;
619 u32 query, result = 0, sresult = 0;
622 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
626 export = fhp->fh_export;
627 dentry = fhp->fh_dentry;
629 if (S_ISREG(dentry->d_inode->i_mode))
630 map = nfs3_regaccess;
631 else if (S_ISDIR(dentry->d_inode->i_mode))
632 map = nfs3_diraccess;
634 map = nfs3_anyaccess;
638 for (; map->access; map++) {
639 if (map->access & query) {
642 sresult |= map->access;
644 err2 = nfsd_permission(rqstp, export, dentry, map->how);
647 result |= map->access;
650 /* the following error codes just mean the access was not allowed,
651 * rather than an error occurred */
655 /* simply don't "or" in the access bit. */
665 *supported = sresult;
670 #endif /* CONFIG_NFSD_V3 */
675 * Open an existing file or directory.
676 * The access argument indicates the type of open (read/write/lock)
677 * N.B. After this call fhp needs an fh_put
680 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
681 int access, struct file **filp)
683 struct dentry *dentry;
685 int flags = O_RDONLY|O_LARGEFILE;
689 validate_process_creds();
692 * If we get here, then the client has already done an "open",
693 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
694 * in case a chmod has now revoked permission.
696 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
700 dentry = fhp->fh_dentry;
701 inode = dentry->d_inode;
703 /* Disallow write access to files with the append-only bit set
704 * or any access when mandatory locking enabled
707 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
710 * We must ignore files (but only files) which might have mandatory
711 * locks on them because there is no way to know if the accesser has
714 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
721 * Check to see if there are any leases on this file.
722 * This may block while leases are broken.
724 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
725 if (host_err == -EWOULDBLOCK)
726 host_err = -ETIMEDOUT;
727 if (host_err) /* NOMEM or WOULDBLOCK */
730 if (access & NFSD_MAY_WRITE) {
731 if (access & NFSD_MAY_READ)
732 flags = O_RDWR|O_LARGEFILE;
734 flags = O_WRONLY|O_LARGEFILE;
738 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
739 flags, current_cred());
741 host_err = PTR_ERR(*filp);
743 ima_counts_get(*filp);
745 err = nfserrno(host_err);
747 validate_process_creds();
755 nfsd_close(struct file *filp)
762 * As this calls fsync (not fdatasync) there is no need for a write_inode
765 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
766 const struct file_operations *fop)
768 struct inode *inode = dp->d_inode;
769 int (*fsync) (struct file *, struct dentry *, int);
772 err = filemap_fdatawrite(inode->i_mapping);
773 if (err == 0 && fop && (fsync = fop->fsync))
774 err = fsync(filp, dp, 0);
776 err = filemap_fdatawait(inode->i_mapping);
782 nfsd_sync(struct file *filp)
785 struct inode *inode = filp->f_path.dentry->d_inode;
786 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
787 mutex_lock(&inode->i_mutex);
788 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
789 mutex_unlock(&inode->i_mutex);
795 nfsd_sync_dir(struct dentry *dp)
797 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
801 * Obtain the readahead parameters for the file
802 * specified by (dev, ino).
805 static inline struct raparms *
806 nfsd_get_raparms(dev_t dev, ino_t ino)
808 struct raparms *ra, **rap, **frap = NULL;
811 struct raparm_hbucket *rab;
813 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
814 rab = &raparm_hash[hash];
816 spin_lock(&rab->pb_lock);
817 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
818 if (ra->p_ino == ino && ra->p_dev == dev)
821 if (ra->p_count == 0)
824 depth = nfsdstats.ra_size*11/10;
826 spin_unlock(&rab->pb_lock);
836 if (rap != &rab->pb_head) {
838 ra->p_next = rab->pb_head;
842 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
843 spin_unlock(&rab->pb_lock);
848 * Grab and keep cached pages associated with a file in the svc_rqst
849 * so that they can be passed to the network sendmsg/sendpage routines
850 * directly. They will be released after the sending has completed.
853 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
854 struct splice_desc *sd)
856 struct svc_rqst *rqstp = sd->u.data;
857 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
858 struct page *page = buf->page;
862 ret = buf->ops->confirm(pipe, buf);
868 if (rqstp->rq_res.page_len == 0) {
873 rqstp->rq_res.page_base = buf->offset;
874 rqstp->rq_res.page_len = size;
875 } else if (page != pp[-1]) {
881 rqstp->rq_res.page_len += size;
883 rqstp->rq_res.page_len += size;
888 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
889 struct splice_desc *sd)
891 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
894 static inline int svc_msnfs(struct svc_fh *ffhp)
897 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
904 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
905 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
914 inode = file->f_path.dentry->d_inode;
916 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
919 /* Get readahead parameters */
920 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
923 file->f_ra = ra->p_ra;
925 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
926 struct splice_desc sd = {
933 rqstp->rq_resused = 1;
934 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
938 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
942 /* Write back readahead params */
944 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
945 spin_lock(&rab->pb_lock);
946 ra->p_ra = file->f_ra;
949 spin_unlock(&rab->pb_lock);
953 nfsdstats.io_read += host_err;
956 fsnotify_access(file->f_path.dentry);
958 err = nfserrno(host_err);
963 static void kill_suid(struct dentry *dentry)
966 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
968 mutex_lock(&dentry->d_inode->i_mutex);
969 notify_change(dentry, &ia);
970 mutex_unlock(&dentry->d_inode->i_mutex);
974 * Gathered writes: If another process is currently writing to the file,
975 * there's a high chance this is another nfsd (triggered by a bulk write
976 * from a client's biod). Rather than syncing the file with each write
977 * request, we sleep for 10 msec.
979 * I don't know if this roughly approximates C. Juszak's idea of
980 * gathered writes, but it's a nice and simple solution (IMHO), and it
983 * Note: we do this only in the NFSv2 case, since v3 and higher have a
984 * better tool (separate unstable writes and commits) for solving this
987 static int wait_for_concurrent_writes(struct file *file)
989 struct inode *inode = file->f_path.dentry->d_inode;
990 static ino_t last_ino;
991 static dev_t last_dev;
994 if (atomic_read(&inode->i_writecount) > 1
995 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
996 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
998 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1001 if (inode->i_state & I_DIRTY) {
1002 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1003 err = nfsd_sync(file);
1005 last_ino = inode->i_ino;
1006 last_dev = inode->i_sb->s_dev;
1011 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1012 loff_t offset, struct kvec *vec, int vlen,
1013 unsigned long *cnt, int *stablep)
1015 struct svc_export *exp;
1016 struct dentry *dentry;
1017 struct inode *inode;
1021 int stable = *stablep;
1027 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1028 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1032 dentry = file->f_path.dentry;
1033 inode = dentry->d_inode;
1034 exp = fhp->fh_export;
1037 * Request sync writes if
1038 * - the sync export option has been set, or
1039 * - the client requested O_SYNC behavior (NFSv3 feature).
1040 * - The file system doesn't support fsync().
1041 * When NFSv2 gathered writes have been configured for this volume,
1042 * flushing the data to disk is handled separately below.
1044 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1046 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1048 *stablep = 2; /* FILE_SYNC */
1051 if (!EX_ISSYNC(exp))
1053 if (stable && !use_wgather) {
1054 spin_lock(&file->f_lock);
1055 file->f_flags |= O_SYNC;
1056 spin_unlock(&file->f_lock);
1059 /* Write the data. */
1060 oldfs = get_fs(); set_fs(KERNEL_DS);
1061 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1066 nfsdstats.io_write += host_err;
1067 fsnotify_modify(file->f_path.dentry);
1069 /* clear setuid/setgid flag after write */
1070 if (inode->i_mode & (S_ISUID | S_ISGID))
1073 if (stable && use_wgather)
1074 host_err = wait_for_concurrent_writes(file);
1077 dprintk("nfsd: write complete host_err=%d\n", host_err);
1081 err = nfserrno(host_err);
1087 * Read data from a file. count must contain the requested read count
1088 * on entry. On return, *count contains the number of bytes actually read.
1089 * N.B. After this call fhp needs an fh_put
1092 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1093 loff_t offset, struct kvec *vec, int vlen,
1094 unsigned long *count)
1099 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1100 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1103 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1105 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1108 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1116 * Write data to a file.
1117 * The stable flag requests synchronous writes.
1118 * N.B. After this call fhp needs an fh_put
1121 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1122 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1128 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1129 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1132 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1135 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1140 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1148 #ifdef CONFIG_NFSD_V3
1150 * Commit all pending writes to stable storage.
1151 * Strictly speaking, we could sync just the indicated file region here,
1152 * but there's currently no way we can ask the VFS to do so.
1154 * Unfortunately we cannot lock the file to make sure we return full WCC
1155 * data to the client, as locking happens lower down in the filesystem.
1158 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1159 loff_t offset, unsigned long count)
1164 if ((u64)count > ~(u64)offset)
1165 return nfserr_inval;
1167 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1170 if (EX_ISSYNC(fhp->fh_export)) {
1171 if (file->f_op && file->f_op->fsync) {
1172 err = nfserrno(nfsd_sync(file));
1174 err = nfserr_notsupp;
1181 #endif /* CONFIG_NFSD_V3 */
1184 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1188 * Mode has already been set earlier in create:
1190 iap->ia_valid &= ~ATTR_MODE;
1192 * Setting uid/gid works only for root. Irix appears to
1193 * send along the gid on create when it tries to implement
1194 * setgid directories via NFS:
1196 if (current_fsuid() != 0)
1197 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1199 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1203 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1204 * setting size to 0 may fail for some specific file systems by the permission
1205 * checking which requires WRITE permission but the mode is 000.
1206 * we ignore the resizing(to 0) on the just new created file, since the size is
1207 * 0 after file created.
1209 * call this only after vfs_create() is called.
1212 nfsd_check_ignore_resizing(struct iattr *iap)
1214 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1215 iap->ia_valid &= ~ATTR_SIZE;
1219 * Create a file (regular, directory, device, fifo); UNIX sockets
1220 * not yet implemented.
1221 * If the response fh has been verified, the parent directory should
1222 * already be locked. Note that the parent directory is left locked.
1224 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1227 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1228 char *fname, int flen, struct iattr *iap,
1229 int type, dev_t rdev, struct svc_fh *resfhp)
1231 struct dentry *dentry, *dchild = NULL;
1241 if (isdotent(fname, flen))
1244 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1248 dentry = fhp->fh_dentry;
1249 dirp = dentry->d_inode;
1251 err = nfserr_notdir;
1252 if (!dirp->i_op->lookup)
1255 * Check whether the response file handle has been verified yet.
1256 * If it has, the parent directory should already be locked.
1258 if (!resfhp->fh_dentry) {
1259 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1260 fh_lock_nested(fhp, I_MUTEX_PARENT);
1261 dchild = lookup_one_len(fname, dentry, flen);
1262 host_err = PTR_ERR(dchild);
1265 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1269 /* called from nfsd_proc_create */
1270 dchild = dget(resfhp->fh_dentry);
1271 if (!fhp->fh_locked) {
1272 /* not actually possible */
1274 "nfsd_create: parent %s/%s not locked!\n",
1275 dentry->d_parent->d_name.name,
1276 dentry->d_name.name);
1282 * Make sure the child dentry is still negative ...
1285 if (dchild->d_inode) {
1286 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1287 dentry->d_name.name, dchild->d_name.name);
1291 if (!(iap->ia_valid & ATTR_MODE))
1293 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1296 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1297 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1302 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1307 * Get the dir op function pointer.
1312 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1314 nfsd_check_ignore_resizing(iap);
1317 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1323 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1327 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1331 if (EX_ISSYNC(fhp->fh_export)) {
1332 err = nfserrno(nfsd_sync_dir(dentry));
1333 write_inode_now(dchild->d_inode, 1);
1336 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1339 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1341 * Update the file handle to get the new inode info.
1344 err = fh_update(resfhp);
1346 if (dchild && !IS_ERR(dchild))
1351 err = nfserrno(host_err);
1355 #ifdef CONFIG_NFSD_V3
1357 * NFSv3 version of nfsd_create
1360 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1361 char *fname, int flen, struct iattr *iap,
1362 struct svc_fh *resfhp, int createmode, u32 *verifier,
1363 int *truncp, int *created)
1365 struct dentry *dentry, *dchild = NULL;
1370 __u32 v_mtime=0, v_atime=0;
1376 if (isdotent(fname, flen))
1378 if (!(iap->ia_valid & ATTR_MODE))
1380 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1384 dentry = fhp->fh_dentry;
1385 dirp = dentry->d_inode;
1387 /* Get all the sanity checks out of the way before
1388 * we lock the parent. */
1389 err = nfserr_notdir;
1390 if (!dirp->i_op->lookup)
1392 fh_lock_nested(fhp, I_MUTEX_PARENT);
1395 * Compose the response file handle.
1397 dchild = lookup_one_len(fname, dentry, flen);
1398 host_err = PTR_ERR(dchild);
1402 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1406 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1407 /* solaris7 gets confused (bugid 4218508) if these have
1408 * the high bit set, so just clear the high bits. If this is
1409 * ever changed to use different attrs for storing the
1410 * verifier, then do_open_lookup() will also need to be fixed
1413 v_mtime = verifier[0]&0x7fffffff;
1414 v_atime = verifier[1]&0x7fffffff;
1417 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1420 if (dchild->d_inode) {
1423 switch (createmode) {
1424 case NFS3_CREATE_UNCHECKED:
1425 if (! S_ISREG(dchild->d_inode->i_mode))
1428 /* in nfsv4, we need to treat this case a little
1429 * differently. we don't want to truncate the
1430 * file now; this would be wrong if the OPEN
1431 * fails for some other reason. furthermore,
1432 * if the size is nonzero, we should ignore it
1433 * according to spec!
1435 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1438 iap->ia_valid &= ATTR_SIZE;
1442 case NFS3_CREATE_EXCLUSIVE:
1443 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1444 && dchild->d_inode->i_atime.tv_sec == v_atime
1445 && dchild->d_inode->i_size == 0 )
1448 case NFS3_CREATE_GUARDED:
1451 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1455 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1457 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1463 if (EX_ISSYNC(fhp->fh_export)) {
1464 err = nfserrno(nfsd_sync_dir(dentry));
1465 /* setattr will sync the child (or not) */
1468 nfsd_check_ignore_resizing(iap);
1470 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1471 /* Cram the verifier into atime/mtime */
1472 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1473 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1474 /* XXX someone who knows this better please fix it for nsec */
1475 iap->ia_mtime.tv_sec = v_mtime;
1476 iap->ia_atime.tv_sec = v_atime;
1477 iap->ia_mtime.tv_nsec = 0;
1478 iap->ia_atime.tv_nsec = 0;
1482 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1486 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1488 * Update the filehandle to get the new inode info.
1491 err = fh_update(resfhp);
1495 if (dchild && !IS_ERR(dchild))
1500 err = nfserrno(host_err);
1503 #endif /* CONFIG_NFSD_V3 */
1506 * Read a symlink. On entry, *lenp must contain the maximum path length that
1507 * fits into the buffer. On return, it contains the true length.
1508 * N.B. After this call fhp needs an fh_put
1511 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1513 struct dentry *dentry;
1514 struct inode *inode;
1519 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1523 dentry = fhp->fh_dentry;
1524 inode = dentry->d_inode;
1527 if (!inode->i_op->readlink)
1530 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1531 /* N.B. Why does this call need a get_fs()??
1532 * Remove the set_fs and watch the fireworks:-) --okir
1535 oldfs = get_fs(); set_fs(KERNEL_DS);
1536 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1547 err = nfserrno(host_err);
1552 * Create a symlink and look up its inode
1553 * N.B. After this call _both_ fhp and resfhp need an fh_put
1556 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1557 char *fname, int flen,
1558 char *path, int plen,
1559 struct svc_fh *resfhp,
1562 struct dentry *dentry, *dnew;
1570 if (isdotent(fname, flen))
1573 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1577 dentry = fhp->fh_dentry;
1578 dnew = lookup_one_len(fname, dentry, flen);
1579 host_err = PTR_ERR(dnew);
1583 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1587 if (unlikely(path[plen] != 0)) {
1588 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1589 if (path_alloced == NULL)
1592 strncpy(path_alloced, path, plen);
1593 path_alloced[plen] = 0;
1594 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1595 kfree(path_alloced);
1598 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1601 if (EX_ISSYNC(fhp->fh_export))
1602 host_err = nfsd_sync_dir(dentry);
1604 err = nfserrno(host_err);
1607 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1609 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1611 if (err==0) err = cerr;
1616 err = nfserrno(host_err);
1622 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1625 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1626 char *name, int len, struct svc_fh *tfhp)
1628 struct dentry *ddir, *dnew, *dold;
1629 struct inode *dirp, *dest;
1633 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1636 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1644 if (isdotent(name, len))
1647 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1648 ddir = ffhp->fh_dentry;
1649 dirp = ddir->d_inode;
1651 dnew = lookup_one_len(name, ddir, len);
1652 host_err = PTR_ERR(dnew);
1656 dold = tfhp->fh_dentry;
1657 dest = dold->d_inode;
1659 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1661 err = nfserrno(host_err);
1664 host_err = vfs_link(dold, dirp, dnew);
1666 if (EX_ISSYNC(ffhp->fh_export)) {
1667 err = nfserrno(nfsd_sync_dir(ddir));
1668 write_inode_now(dest, 1);
1672 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1675 err = nfserrno(host_err);
1677 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1686 err = nfserrno(host_err);
1692 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1695 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1696 struct svc_fh *tfhp, char *tname, int tlen)
1698 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1699 struct inode *fdir, *tdir;
1703 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1706 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1710 fdentry = ffhp->fh_dentry;
1711 fdir = fdentry->d_inode;
1713 tdentry = tfhp->fh_dentry;
1714 tdir = tdentry->d_inode;
1716 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1717 if (ffhp->fh_export != tfhp->fh_export)
1721 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1724 /* cannot use fh_lock as we need deadlock protective ordering
1725 * so do it by hand */
1726 trap = lock_rename(tdentry, fdentry);
1727 ffhp->fh_locked = tfhp->fh_locked = 1;
1731 odentry = lookup_one_len(fname, fdentry, flen);
1732 host_err = PTR_ERR(odentry);
1733 if (IS_ERR(odentry))
1737 if (!odentry->d_inode)
1740 if (odentry == trap)
1743 ndentry = lookup_one_len(tname, tdentry, tlen);
1744 host_err = PTR_ERR(ndentry);
1745 if (IS_ERR(ndentry))
1747 host_err = -ENOTEMPTY;
1748 if (ndentry == trap)
1751 if (svc_msnfs(ffhp) &&
1752 ((atomic_read(&odentry->d_count) > 1)
1753 || (atomic_read(&ndentry->d_count) > 1))) {
1759 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1761 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1765 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1766 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1767 host_err = nfsd_sync_dir(tdentry);
1769 host_err = nfsd_sync_dir(fdentry);
1772 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1779 err = nfserrno(host_err);
1781 /* we cannot reply on fh_unlock on the two filehandles,
1782 * as that would do the wrong thing if the two directories
1783 * were the same, so again we do it by hand
1785 fill_post_wcc(ffhp);
1786 fill_post_wcc(tfhp);
1787 unlock_rename(tdentry, fdentry);
1788 ffhp->fh_locked = tfhp->fh_locked = 0;
1795 * Unlink a file or directory
1796 * N.B. After this call fhp needs an fh_put
1799 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1800 char *fname, int flen)
1802 struct dentry *dentry, *rdentry;
1808 if (!flen || isdotent(fname, flen))
1810 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1814 fh_lock_nested(fhp, I_MUTEX_PARENT);
1815 dentry = fhp->fh_dentry;
1816 dirp = dentry->d_inode;
1818 rdentry = lookup_one_len(fname, dentry, flen);
1819 host_err = PTR_ERR(rdentry);
1820 if (IS_ERR(rdentry))
1823 if (!rdentry->d_inode) {
1830 type = rdentry->d_inode->i_mode & S_IFMT;
1832 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1836 if (type != S_IFDIR) { /* It's UNLINK */
1838 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1839 (atomic_read(&rdentry->d_count) > 1)) {
1843 host_err = vfs_unlink(dirp, rdentry);
1844 } else { /* It's RMDIR */
1845 host_err = vfs_rmdir(dirp, rdentry);
1852 if (EX_ISSYNC(fhp->fh_export))
1853 host_err = nfsd_sync_dir(dentry);
1856 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1858 err = nfserrno(host_err);
1864 * We do this buffering because we must not call back into the file
1865 * system's ->lookup() method from the filldir callback. That may well
1866 * deadlock a number of file systems.
1868 * This is based heavily on the implementation of same in XFS.
1870 struct buffered_dirent {
1874 unsigned int d_type;
1878 struct readdir_data {
1884 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1885 loff_t offset, u64 ino, unsigned int d_type)
1887 struct readdir_data *buf = __buf;
1888 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1889 unsigned int reclen;
1891 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1892 if (buf->used + reclen > PAGE_SIZE) {
1897 de->namlen = namlen;
1898 de->offset = offset;
1900 de->d_type = d_type;
1901 memcpy(de->name, name, namlen);
1902 buf->used += reclen;
1907 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1908 struct readdir_cd *cdp, loff_t *offsetp)
1910 struct readdir_data buf;
1911 struct buffered_dirent *de;
1916 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1918 return nfserrno(-ENOMEM);
1923 struct inode *dir_inode = file->f_path.dentry->d_inode;
1924 unsigned int reclen;
1926 cdp->err = nfserr_eof; /* will be cleared on successful read */
1930 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1943 * Various filldir functions may end up calling back into
1944 * lookup_one_len() and the file system's ->lookup() method.
1945 * These expect i_mutex to be held, as it would within readdir.
1947 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1951 de = (struct buffered_dirent *)buf.dirent;
1953 offset = de->offset;
1955 if (func(cdp, de->name, de->namlen, de->offset,
1956 de->ino, de->d_type))
1959 if (cdp->err != nfs_ok)
1962 reclen = ALIGN(sizeof(*de) + de->namlen,
1965 de = (struct buffered_dirent *)((char *)de + reclen);
1967 mutex_unlock(&dir_inode->i_mutex);
1968 if (size > 0) /* We bailed out early */
1971 offset = vfs_llseek(file, 0, SEEK_CUR);
1974 free_page((unsigned long)(buf.dirent));
1977 return nfserrno(host_err);
1984 * Read entries from a directory.
1985 * The NFSv3/4 verifier we ignore for now.
1988 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1989 struct readdir_cd *cdp, filldir_t func)
1993 loff_t offset = *offsetp;
1995 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1999 offset = vfs_llseek(file, offset, 0);
2001 err = nfserrno((int)offset);
2005 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2007 if (err == nfserr_eof || err == nfserr_toosmall)
2008 err = nfs_ok; /* can still be found in ->err */
2016 * Get file system stats
2017 * N.B. After this call fhp needs an fh_put
2020 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2022 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2023 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2028 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2030 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2034 * Check for a user's access permissions to this inode.
2037 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2038 struct dentry *dentry, int acc)
2040 struct inode *inode = dentry->d_inode;
2044 if (acc == NFSD_MAY_NOP)
2047 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2049 (acc & NFSD_MAY_READ)? " read" : "",
2050 (acc & NFSD_MAY_WRITE)? " write" : "",
2051 (acc & NFSD_MAY_EXEC)? " exec" : "",
2052 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2053 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2054 (acc & NFSD_MAY_LOCK)? " lock" : "",
2055 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2057 IS_IMMUTABLE(inode)? " immut" : "",
2058 IS_APPEND(inode)? " append" : "",
2059 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2060 dprintk(" owner %d/%d user %d/%d\n",
2061 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2064 /* Normally we reject any write/sattr etc access on a read-only file
2065 * system. But if it is IRIX doing check on write-access for a
2066 * device special file, we ignore rofs.
2068 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2069 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2070 if (exp_rdonly(rqstp, exp) ||
2071 __mnt_is_readonly(exp->ex_path.mnt))
2073 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2076 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2079 if (acc & NFSD_MAY_LOCK) {
2080 /* If we cannot rely on authentication in NLM requests,
2081 * just allow locks, otherwise require read permission, or
2084 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2087 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2090 * The file owner always gets access permission for accesses that
2091 * would normally be checked at open time. This is to make
2092 * file access work even when the client has done a fchmod(fd, 0).
2094 * However, `cp foo bar' should fail nevertheless when bar is
2095 * readonly. A sensible way to do this might be to reject all
2096 * attempts to truncate a read-only file, because a creat() call
2097 * always implies file truncation.
2098 * ... but this isn't really fair. A process may reasonably call
2099 * ftruncate on an open file descriptor on a file with perm 000.
2100 * We must trust the client to do permission checking - using "ACCESS"
2103 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2104 inode->i_uid == current_fsuid())
2107 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2108 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2110 /* Allow read access to binaries even when mode 111 */
2111 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2112 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2113 err = inode_permission(inode, MAY_EXEC);
2117 /* Do integrity (permission) checking now, but defer incrementing
2118 * IMA counts to the actual file open.
2120 path.mnt = exp->ex_path.mnt;
2121 path.dentry = dentry;
2122 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2125 return err? nfserrno(err) : 0;
2129 nfsd_racache_shutdown(void)
2131 struct raparms *raparm, *last_raparm;
2134 dprintk("nfsd: freeing readahead buffers.\n");
2136 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2137 raparm = raparm_hash[i].pb_head;
2139 last_raparm = raparm;
2140 raparm = raparm->p_next;
2143 raparm_hash[i].pb_head = NULL;
2147 * Initialize readahead param cache
2150 nfsd_racache_init(int cache_size)
2155 struct raparms **raparm = NULL;
2158 if (raparm_hash[0].pb_head)
2160 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2163 cache_size = nperbucket * RAPARM_HASH_SIZE;
2165 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2167 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2168 spin_lock_init(&raparm_hash[i].pb_lock);
2170 raparm = &raparm_hash[i].pb_head;
2171 for (j = 0; j < nperbucket; j++) {
2172 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2175 raparm = &(*raparm)->p_next;
2180 nfsdstats.ra_size = cache_size;
2184 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2185 nfsd_racache_shutdown();
2189 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2191 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2193 struct inode *inode = fhp->fh_dentry->d_inode;
2197 struct posix_acl *acl;
2199 if (!IS_POSIXACL(inode))
2200 return ERR_PTR(-EOPNOTSUPP);
2203 case ACL_TYPE_ACCESS:
2204 name = POSIX_ACL_XATTR_ACCESS;
2206 case ACL_TYPE_DEFAULT:
2207 name = POSIX_ACL_XATTR_DEFAULT;
2210 return ERR_PTR(-EOPNOTSUPP);
2213 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2215 return ERR_PTR(size);
2217 acl = posix_acl_from_xattr(value, size);
2223 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2225 struct inode *inode = fhp->fh_dentry->d_inode;
2231 if (!IS_POSIXACL(inode) ||
2232 !inode->i_op->setxattr || !inode->i_op->removexattr)
2235 case ACL_TYPE_ACCESS:
2236 name = POSIX_ACL_XATTR_ACCESS;
2238 case ACL_TYPE_DEFAULT:
2239 name = POSIX_ACL_XATTR_DEFAULT;
2245 if (acl && acl->a_count) {
2246 size = posix_acl_xattr_size(acl->a_count);
2247 value = kmalloc(size, GFP_KERNEL);
2250 error = posix_acl_to_xattr(acl, value, size);
2257 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2261 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2263 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2266 error = vfs_removexattr(fhp->fh_dentry, name);
2267 if (error == -ENODATA)
2271 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2277 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */