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>
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
65 * This is a cache of readahead params that help us choose the proper
66 * readahead strategy. Initially, we set all readahead parameters to 0
67 * and let the VFS handle things.
68 * If you increase the number of cached files very much, you'll need to
69 * add a hash table here.
72 struct raparms *p_next;
77 struct file_ra_state p_ra;
78 unsigned int p_hindex;
81 struct raparm_hbucket {
82 struct raparms *pb_head;
84 } ____cacheline_aligned_in_smp;
86 #define RAPARM_HASH_BITS 4
87 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
88 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
89 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
92 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
95 * or nfs_ok having possibly changed *dpp and *expp
98 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
99 struct svc_export **expp)
101 struct svc_export *exp = *expp, *exp2 = NULL;
102 struct dentry *dentry = *dpp;
103 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
104 struct dentry *mounts = dget(dentry);
107 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
109 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
111 if (PTR_ERR(exp2) != -ENOENT)
117 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
118 /* successfully crossed mount point */
120 * This is subtle: dentry is *not* under mnt at this point.
121 * The only reason we are safe is that original mnt is pinned
122 * down by exp, so we should dput before putting exp.
138 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
139 const char *name, unsigned int len,
140 struct svc_export **exp_ret, struct dentry **dentry_ret)
142 struct svc_export *exp;
143 struct dentry *dparent;
144 struct dentry *dentry;
148 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
150 /* Obtain dentry and export. */
151 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
155 dparent = fhp->fh_dentry;
156 exp = fhp->fh_export;
159 /* Lookup the name, but don't follow links */
160 if (isdotent(name, len)) {
162 dentry = dget(dparent);
163 else if (dparent != exp->ex_path.dentry)
164 dentry = dget_parent(dparent);
165 else if (!EX_NOHIDE(exp))
166 dentry = dget(dparent); /* .. == . just like at / */
168 /* checking mountpoint crossing is very different when stepping up */
169 struct svc_export *exp2 = NULL;
171 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
172 dentry = dget(dparent);
173 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
175 dp = dget_parent(dentry);
179 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
180 if (PTR_ERR(exp2) == -ENOENT) {
182 dentry = dget(dparent);
183 } else if (IS_ERR(exp2)) {
184 host_err = PTR_ERR(exp2);
196 dentry = lookup_one_len(name, dparent, len);
197 host_err = PTR_ERR(dentry);
201 * check if we have crossed a mount point ...
203 if (d_mountpoint(dentry)) {
204 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
210 *dentry_ret = dentry;
216 return nfserrno(host_err);
220 * Look up one component of a pathname.
221 * N.B. After this call _both_ fhp and resfh need an fh_put
223 * If the lookup would cross a mountpoint, and the mounted filesystem
224 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
225 * accepted as it stands and the mounted directory is
226 * returned. Otherwise the covered directory is returned.
227 * NOTE: this mountpoint crossing is not supported properly by all
228 * clients and is explicitly disallowed for NFSv3
229 * NeilBrown <neilb@cse.unsw.edu.au>
232 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
233 unsigned int len, struct svc_fh *resfh)
235 struct svc_export *exp;
236 struct dentry *dentry;
239 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
242 err = check_nfsd_access(exp, rqstp);
246 * Note: we compose the file handle now, but as the
247 * dentry may be negative, it may need to be updated.
249 err = fh_compose(resfh, exp, dentry, fhp);
250 if (!err && !dentry->d_inode)
260 * Set various file attributes.
261 * N.B. After this call fhp needs an fh_put
264 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
265 int check_guard, time_t guardtime)
267 struct dentry *dentry;
269 int accmode = NFSD_MAY_SATTR;
275 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
276 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
277 if (iap->ia_valid & ATTR_SIZE)
281 err = fh_verify(rqstp, fhp, ftype, accmode);
285 dentry = fhp->fh_dentry;
286 inode = dentry->d_inode;
288 /* Ignore any mode updates on symlinks */
289 if (S_ISLNK(inode->i_mode))
290 iap->ia_valid &= ~ATTR_MODE;
296 * NFSv2 does not differentiate between "set-[ac]time-to-now"
297 * which only requires access, and "set-[ac]time-to-X" which
298 * requires ownership.
299 * So if it looks like it might be "set both to the same time which
300 * is close to now", and if inode_change_ok fails, then we
301 * convert to "set to now" instead of "set to explicit time"
303 * We only call inode_change_ok as the last test as technically
304 * it is not an interface that we should be using. It is only
305 * valid if the filesystem does not define it's own i_op->setattr.
307 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
308 #define MAX_TOUCH_TIME_ERROR (30*60)
309 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
310 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
314 * Now just make sure time is in the right ballpark.
315 * Solaris, at least, doesn't seem to care what the time
316 * request is. We require it be within 30 minutes of now.
318 time_t delta = iap->ia_atime.tv_sec - get_seconds();
321 if (delta < MAX_TOUCH_TIME_ERROR &&
322 inode_change_ok(inode, iap) != 0) {
324 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
325 * This will cause notify_change to set these times
328 iap->ia_valid &= ~BOTH_TIME_SET;
333 * The size case is special.
334 * It changes the file as well as the attributes.
336 if (iap->ia_valid & ATTR_SIZE) {
337 if (iap->ia_size < inode->i_size) {
338 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
339 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
345 * If we are changing the size of the file, then
346 * we need to break all leases.
348 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
349 if (host_err == -EWOULDBLOCK)
350 host_err = -ETIMEDOUT;
351 if (host_err) /* ENOMEM or EWOULDBLOCK */
354 host_err = get_write_access(inode);
359 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
361 put_write_access(inode);
367 /* sanitize the mode change */
368 if (iap->ia_valid & ATTR_MODE) {
369 iap->ia_mode &= S_IALLUGO;
370 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
373 /* Revoke setuid/setgid on chown */
374 if (!S_ISDIR(inode->i_mode) &&
375 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
376 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
377 iap->ia_valid |= ATTR_KILL_PRIV;
378 if (iap->ia_valid & ATTR_MODE) {
379 /* we're setting mode too, just clear the s*id bits */
380 iap->ia_mode &= ~S_ISUID;
381 if (iap->ia_mode & S_IXGRP)
382 iap->ia_mode &= ~S_ISGID;
384 /* set ATTR_KILL_* bits and let VFS handle it */
385 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
389 /* Change the attributes. */
391 iap->ia_valid |= ATTR_CTIME;
393 err = nfserr_notsync;
394 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
396 host_err = notify_change(dentry, iap);
397 err = nfserrno(host_err);
401 put_write_access(inode);
403 if (EX_ISSYNC(fhp->fh_export))
404 write_inode_now(inode, 1);
409 err = nfserrno(host_err);
413 #if defined(CONFIG_NFSD_V2_ACL) || \
414 defined(CONFIG_NFSD_V3_ACL) || \
415 defined(CONFIG_NFSD_V4)
416 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
421 buflen = vfs_getxattr(dentry, key, NULL, 0);
425 *buf = kmalloc(buflen, GFP_KERNEL);
429 ret = vfs_getxattr(dentry, key, *buf, buflen);
436 #if defined(CONFIG_NFSD_V4)
438 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
445 buflen = posix_acl_xattr_size(pacl->a_count);
446 buf = kmalloc(buflen, GFP_KERNEL);
451 len = posix_acl_to_xattr(pacl, buf, buflen);
457 error = vfs_setxattr(dentry, key, buf, len, 0);
464 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
465 struct nfs4_acl *acl)
469 struct dentry *dentry;
471 struct posix_acl *pacl = NULL, *dpacl = NULL;
472 unsigned int flags = 0;
475 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
479 dentry = fhp->fh_dentry;
480 inode = dentry->d_inode;
481 if (S_ISDIR(inode->i_mode))
482 flags = NFS4_ACL_DIR;
484 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
485 if (host_error == -EINVAL) {
486 return nfserr_attrnotsupp;
487 } else if (host_error < 0)
490 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
494 if (S_ISDIR(inode->i_mode))
495 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
498 posix_acl_release(pacl);
499 posix_acl_release(dpacl);
501 if (host_error == -EOPNOTSUPP)
502 return nfserr_attrnotsupp;
504 return nfserrno(host_error);
507 static struct posix_acl *
508 _get_posix_acl(struct dentry *dentry, char *key)
511 struct posix_acl *pacl = NULL;
514 buflen = nfsd_getxattr(dentry, key, &buf);
518 return ERR_PTR(buflen);
520 pacl = posix_acl_from_xattr(buf, buflen);
526 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
528 struct inode *inode = dentry->d_inode;
530 struct posix_acl *pacl = NULL, *dpacl = NULL;
531 unsigned int flags = 0;
533 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
534 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
535 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
537 error = PTR_ERR(pacl);
542 if (S_ISDIR(inode->i_mode)) {
543 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
544 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
546 else if (IS_ERR(dpacl)) {
547 error = PTR_ERR(dpacl);
551 flags = NFS4_ACL_DIR;
554 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
556 error = PTR_ERR(*acl);
560 posix_acl_release(pacl);
561 posix_acl_release(dpacl);
565 #endif /* defined(CONFIG_NFS_V4) */
567 #ifdef CONFIG_NFSD_V3
569 * Check server access rights to a file system object
575 static struct accessmap nfs3_regaccess[] = {
576 { NFS3_ACCESS_READ, NFSD_MAY_READ },
577 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
578 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
579 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
584 static struct accessmap nfs3_diraccess[] = {
585 { NFS3_ACCESS_READ, NFSD_MAY_READ },
586 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
587 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
588 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
589 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
594 static struct accessmap nfs3_anyaccess[] = {
595 /* Some clients - Solaris 2.6 at least, make an access call
596 * to the server to check for access for things like /dev/null
597 * (which really, the server doesn't care about). So
598 * We provide simple access checking for them, looking
599 * mainly at mode bits, and we make sure to ignore read-only
602 { NFS3_ACCESS_READ, NFSD_MAY_READ },
603 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
604 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
605 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
611 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
613 struct accessmap *map;
614 struct svc_export *export;
615 struct dentry *dentry;
616 u32 query, result = 0, sresult = 0;
619 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
623 export = fhp->fh_export;
624 dentry = fhp->fh_dentry;
626 if (S_ISREG(dentry->d_inode->i_mode))
627 map = nfs3_regaccess;
628 else if (S_ISDIR(dentry->d_inode->i_mode))
629 map = nfs3_diraccess;
631 map = nfs3_anyaccess;
635 for (; map->access; map++) {
636 if (map->access & query) {
639 sresult |= map->access;
641 err2 = nfsd_permission(rqstp, export, dentry, map->how);
644 result |= map->access;
647 /* the following error codes just mean the access was not allowed,
648 * rather than an error occurred */
652 /* simply don't "or" in the access bit. */
662 *supported = sresult;
667 #endif /* CONFIG_NFSD_V3 */
672 * Open an existing file or directory.
673 * The access argument indicates the type of open (read/write/lock)
674 * N.B. After this call fhp needs an fh_put
677 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
678 int access, struct file **filp)
680 const struct cred *cred = current_cred();
681 struct dentry *dentry;
683 int flags = O_RDONLY|O_LARGEFILE;
688 * If we get here, then the client has already done an "open",
689 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
690 * in case a chmod has now revoked permission.
692 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
696 dentry = fhp->fh_dentry;
697 inode = dentry->d_inode;
699 /* Disallow write access to files with the append-only bit set
700 * or any access when mandatory locking enabled
703 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
706 * We must ignore files (but only files) which might have mandatory
707 * locks on them because there is no way to know if the accesser has
710 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
717 * Check to see if there are any leases on this file.
718 * This may block while leases are broken.
720 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
721 if (host_err == -EWOULDBLOCK)
722 host_err = -ETIMEDOUT;
723 if (host_err) /* NOMEM or WOULDBLOCK */
726 if (access & NFSD_MAY_WRITE) {
727 if (access & NFSD_MAY_READ)
728 flags = O_RDWR|O_LARGEFILE;
730 flags = O_WRONLY|O_LARGEFILE;
734 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
737 host_err = PTR_ERR(*filp);
739 err = nfserrno(host_err);
748 nfsd_close(struct file *filp)
755 * As this calls fsync (not fdatasync) there is no need for a write_inode
758 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
759 const struct file_operations *fop)
761 struct inode *inode = dp->d_inode;
762 int (*fsync) (struct file *, struct dentry *, int);
765 err = filemap_fdatawrite(inode->i_mapping);
766 if (err == 0 && fop && (fsync = fop->fsync))
767 err = fsync(filp, dp, 0);
769 err = filemap_fdatawait(inode->i_mapping);
775 nfsd_sync(struct file *filp)
778 struct inode *inode = filp->f_path.dentry->d_inode;
779 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
780 mutex_lock(&inode->i_mutex);
781 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
782 mutex_unlock(&inode->i_mutex);
788 nfsd_sync_dir(struct dentry *dp)
790 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
794 * Obtain the readahead parameters for the file
795 * specified by (dev, ino).
798 static inline struct raparms *
799 nfsd_get_raparms(dev_t dev, ino_t ino)
801 struct raparms *ra, **rap, **frap = NULL;
804 struct raparm_hbucket *rab;
806 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
807 rab = &raparm_hash[hash];
809 spin_lock(&rab->pb_lock);
810 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
811 if (ra->p_ino == ino && ra->p_dev == dev)
814 if (ra->p_count == 0)
817 depth = nfsdstats.ra_size*11/10;
819 spin_unlock(&rab->pb_lock);
829 if (rap != &rab->pb_head) {
831 ra->p_next = rab->pb_head;
835 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
836 spin_unlock(&rab->pb_lock);
841 * Grab and keep cached pages associated with a file in the svc_rqst
842 * so that they can be passed to the network sendmsg/sendpage routines
843 * directly. They will be released after the sending has completed.
846 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
847 struct splice_desc *sd)
849 struct svc_rqst *rqstp = sd->u.data;
850 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
851 struct page *page = buf->page;
855 ret = buf->ops->confirm(pipe, buf);
861 if (rqstp->rq_res.page_len == 0) {
866 rqstp->rq_res.page_base = buf->offset;
867 rqstp->rq_res.page_len = size;
868 } else if (page != pp[-1]) {
874 rqstp->rq_res.page_len += size;
876 rqstp->rq_res.page_len += size;
881 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
882 struct splice_desc *sd)
884 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
887 static inline int svc_msnfs(struct svc_fh *ffhp)
890 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
897 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
898 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
907 inode = file->f_path.dentry->d_inode;
909 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
912 /* Get readahead parameters */
913 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
916 file->f_ra = ra->p_ra;
918 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
919 struct splice_desc sd = {
926 rqstp->rq_resused = 1;
927 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
931 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
935 /* Write back readahead params */
937 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
938 spin_lock(&rab->pb_lock);
939 ra->p_ra = file->f_ra;
942 spin_unlock(&rab->pb_lock);
946 nfsdstats.io_read += host_err;
949 fsnotify_access(file->f_path.dentry);
951 err = nfserrno(host_err);
956 static void kill_suid(struct dentry *dentry)
959 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
961 mutex_lock(&dentry->d_inode->i_mutex);
962 notify_change(dentry, &ia);
963 mutex_unlock(&dentry->d_inode->i_mutex);
967 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
968 loff_t offset, struct kvec *vec, int vlen,
969 unsigned long *cnt, int *stablep)
971 struct svc_export *exp;
972 struct dentry *dentry;
977 int stable = *stablep;
983 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
984 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
988 dentry = file->f_path.dentry;
989 inode = dentry->d_inode;
990 exp = fhp->fh_export;
993 * Request sync writes if
994 * - the sync export option has been set, or
995 * - the client requested O_SYNC behavior (NFSv3 feature).
996 * - The file system doesn't support fsync().
997 * When NFSv2 gathered writes have been configured for this volume,
998 * flushing the data to disk is handled separately below.
1000 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1002 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1004 *stablep = 2; /* FILE_SYNC */
1007 if (!EX_ISSYNC(exp))
1009 if (stable && !use_wgather) {
1010 spin_lock(&file->f_lock);
1011 file->f_flags |= O_SYNC;
1012 spin_unlock(&file->f_lock);
1015 /* Write the data. */
1016 oldfs = get_fs(); set_fs(KERNEL_DS);
1017 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1019 if (host_err >= 0) {
1021 nfsdstats.io_write += host_err;
1022 fsnotify_modify(file->f_path.dentry);
1025 /* clear setuid/setgid flag after write */
1026 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
1029 if (host_err >= 0 && stable) {
1030 static ino_t last_ino;
1031 static dev_t last_dev;
1034 * Gathered writes: If another process is currently
1035 * writing to the file, there's a high chance
1036 * this is another nfsd (triggered by a bulk write
1037 * from a client's biod). Rather than syncing the
1038 * file with each write request, we sleep for 10 msec.
1040 * I don't know if this roughly approximates
1041 * C. Juszak's idea of gathered writes, but it's a
1042 * nice and simple solution (IMHO), and it seems to
1046 if (atomic_read(&inode->i_writecount) > 1
1047 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1048 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1050 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1053 if (inode->i_state & I_DIRTY) {
1054 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1055 host_err=nfsd_sync(file);
1058 wake_up(&inode->i_wait);
1061 last_ino = inode->i_ino;
1062 last_dev = inode->i_sb->s_dev;
1065 dprintk("nfsd: write complete host_err=%d\n", host_err);
1069 err = nfserrno(host_err);
1075 * Read data from a file. count must contain the requested read count
1076 * on entry. On return, *count contains the number of bytes actually read.
1077 * N.B. After this call fhp needs an fh_put
1080 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1081 loff_t offset, struct kvec *vec, int vlen,
1082 unsigned long *count)
1087 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1088 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1091 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1093 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1096 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1104 * Write data to a file.
1105 * The stable flag requests synchronous writes.
1106 * N.B. After this call fhp needs an fh_put
1109 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1110 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1116 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1117 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1120 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1123 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1128 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1136 #ifdef CONFIG_NFSD_V3
1138 * Commit all pending writes to stable storage.
1139 * Strictly speaking, we could sync just the indicated file region here,
1140 * but there's currently no way we can ask the VFS to do so.
1142 * Unfortunately we cannot lock the file to make sure we return full WCC
1143 * data to the client, as locking happens lower down in the filesystem.
1146 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1147 loff_t offset, unsigned long count)
1152 if ((u64)count > ~(u64)offset)
1153 return nfserr_inval;
1155 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1158 if (EX_ISSYNC(fhp->fh_export)) {
1159 if (file->f_op && file->f_op->fsync) {
1160 err = nfserrno(nfsd_sync(file));
1162 err = nfserr_notsupp;
1169 #endif /* CONFIG_NFSD_V3 */
1172 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1176 * Mode has already been set earlier in create:
1178 iap->ia_valid &= ~ATTR_MODE;
1180 * Setting uid/gid works only for root. Irix appears to
1181 * send along the gid on create when it tries to implement
1182 * setgid directories via NFS:
1184 if (current_fsuid() != 0)
1185 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1187 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1191 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1192 * setting size to 0 may fail for some specific file systems by the permission
1193 * checking which requires WRITE permission but the mode is 000.
1194 * we ignore the resizing(to 0) on the just new created file, since the size is
1195 * 0 after file created.
1197 * call this only after vfs_create() is called.
1200 nfsd_check_ignore_resizing(struct iattr *iap)
1202 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1203 iap->ia_valid &= ~ATTR_SIZE;
1207 * Create a file (regular, directory, device, fifo); UNIX sockets
1208 * not yet implemented.
1209 * If the response fh has been verified, the parent directory should
1210 * already be locked. Note that the parent directory is left locked.
1212 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1215 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1216 char *fname, int flen, struct iattr *iap,
1217 int type, dev_t rdev, struct svc_fh *resfhp)
1219 struct dentry *dentry, *dchild = NULL;
1229 if (isdotent(fname, flen))
1232 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1236 dentry = fhp->fh_dentry;
1237 dirp = dentry->d_inode;
1239 err = nfserr_notdir;
1240 if (!dirp->i_op->lookup)
1243 * Check whether the response file handle has been verified yet.
1244 * If it has, the parent directory should already be locked.
1246 if (!resfhp->fh_dentry) {
1247 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1248 fh_lock_nested(fhp, I_MUTEX_PARENT);
1249 dchild = lookup_one_len(fname, dentry, flen);
1250 host_err = PTR_ERR(dchild);
1253 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1257 /* called from nfsd_proc_create */
1258 dchild = dget(resfhp->fh_dentry);
1259 if (!fhp->fh_locked) {
1260 /* not actually possible */
1262 "nfsd_create: parent %s/%s not locked!\n",
1263 dentry->d_parent->d_name.name,
1264 dentry->d_name.name);
1270 * Make sure the child dentry is still negative ...
1273 if (dchild->d_inode) {
1274 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1275 dentry->d_name.name, dchild->d_name.name);
1279 if (!(iap->ia_valid & ATTR_MODE))
1281 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1284 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1285 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1290 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1295 * Get the dir op function pointer.
1300 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1302 nfsd_check_ignore_resizing(iap);
1305 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1311 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1315 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1319 if (EX_ISSYNC(fhp->fh_export)) {
1320 err = nfserrno(nfsd_sync_dir(dentry));
1321 write_inode_now(dchild->d_inode, 1);
1324 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1327 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1329 * Update the file handle to get the new inode info.
1332 err = fh_update(resfhp);
1334 if (dchild && !IS_ERR(dchild))
1339 err = nfserrno(host_err);
1343 #ifdef CONFIG_NFSD_V3
1345 * NFSv3 version of nfsd_create
1348 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1349 char *fname, int flen, struct iattr *iap,
1350 struct svc_fh *resfhp, int createmode, u32 *verifier,
1351 int *truncp, int *created)
1353 struct dentry *dentry, *dchild = NULL;
1358 __u32 v_mtime=0, v_atime=0;
1364 if (isdotent(fname, flen))
1366 if (!(iap->ia_valid & ATTR_MODE))
1368 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1372 dentry = fhp->fh_dentry;
1373 dirp = dentry->d_inode;
1375 /* Get all the sanity checks out of the way before
1376 * we lock the parent. */
1377 err = nfserr_notdir;
1378 if (!dirp->i_op->lookup)
1380 fh_lock_nested(fhp, I_MUTEX_PARENT);
1383 * Compose the response file handle.
1385 dchild = lookup_one_len(fname, dentry, flen);
1386 host_err = PTR_ERR(dchild);
1390 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1394 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1395 /* solaris7 gets confused (bugid 4218508) if these have
1396 * the high bit set, so just clear the high bits. If this is
1397 * ever changed to use different attrs for storing the
1398 * verifier, then do_open_lookup() will also need to be fixed
1401 v_mtime = verifier[0]&0x7fffffff;
1402 v_atime = verifier[1]&0x7fffffff;
1405 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1408 if (dchild->d_inode) {
1411 switch (createmode) {
1412 case NFS3_CREATE_UNCHECKED:
1413 if (! S_ISREG(dchild->d_inode->i_mode))
1416 /* in nfsv4, we need to treat this case a little
1417 * differently. we don't want to truncate the
1418 * file now; this would be wrong if the OPEN
1419 * fails for some other reason. furthermore,
1420 * if the size is nonzero, we should ignore it
1421 * according to spec!
1423 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1426 iap->ia_valid &= ATTR_SIZE;
1430 case NFS3_CREATE_EXCLUSIVE:
1431 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1432 && dchild->d_inode->i_atime.tv_sec == v_atime
1433 && dchild->d_inode->i_size == 0 )
1436 case NFS3_CREATE_GUARDED:
1439 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1443 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1445 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1451 if (EX_ISSYNC(fhp->fh_export)) {
1452 err = nfserrno(nfsd_sync_dir(dentry));
1453 /* setattr will sync the child (or not) */
1456 nfsd_check_ignore_resizing(iap);
1458 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1459 /* Cram the verifier into atime/mtime */
1460 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1461 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1462 /* XXX someone who knows this better please fix it for nsec */
1463 iap->ia_mtime.tv_sec = v_mtime;
1464 iap->ia_atime.tv_sec = v_atime;
1465 iap->ia_mtime.tv_nsec = 0;
1466 iap->ia_atime.tv_nsec = 0;
1470 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1474 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1476 * Update the filehandle to get the new inode info.
1479 err = fh_update(resfhp);
1483 if (dchild && !IS_ERR(dchild))
1488 err = nfserrno(host_err);
1491 #endif /* CONFIG_NFSD_V3 */
1494 * Read a symlink. On entry, *lenp must contain the maximum path length that
1495 * fits into the buffer. On return, it contains the true length.
1496 * N.B. After this call fhp needs an fh_put
1499 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1501 struct dentry *dentry;
1502 struct inode *inode;
1507 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1511 dentry = fhp->fh_dentry;
1512 inode = dentry->d_inode;
1515 if (!inode->i_op->readlink)
1518 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1519 /* N.B. Why does this call need a get_fs()??
1520 * Remove the set_fs and watch the fireworks:-) --okir
1523 oldfs = get_fs(); set_fs(KERNEL_DS);
1524 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1535 err = nfserrno(host_err);
1540 * Create a symlink and look up its inode
1541 * N.B. After this call _both_ fhp and resfhp need an fh_put
1544 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1545 char *fname, int flen,
1546 char *path, int plen,
1547 struct svc_fh *resfhp,
1550 struct dentry *dentry, *dnew;
1558 if (isdotent(fname, flen))
1561 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1565 dentry = fhp->fh_dentry;
1566 dnew = lookup_one_len(fname, dentry, flen);
1567 host_err = PTR_ERR(dnew);
1571 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1575 if (unlikely(path[plen] != 0)) {
1576 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1577 if (path_alloced == NULL)
1580 strncpy(path_alloced, path, plen);
1581 path_alloced[plen] = 0;
1582 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1583 kfree(path_alloced);
1586 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1589 if (EX_ISSYNC(fhp->fh_export))
1590 host_err = nfsd_sync_dir(dentry);
1592 err = nfserrno(host_err);
1595 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1597 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1599 if (err==0) err = cerr;
1604 err = nfserrno(host_err);
1610 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1613 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1614 char *name, int len, struct svc_fh *tfhp)
1616 struct dentry *ddir, *dnew, *dold;
1617 struct inode *dirp, *dest;
1621 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1624 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1632 if (isdotent(name, len))
1635 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1636 ddir = ffhp->fh_dentry;
1637 dirp = ddir->d_inode;
1639 dnew = lookup_one_len(name, ddir, len);
1640 host_err = PTR_ERR(dnew);
1644 dold = tfhp->fh_dentry;
1645 dest = dold->d_inode;
1647 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1649 err = nfserrno(host_err);
1652 host_err = vfs_link(dold, dirp, dnew);
1654 if (EX_ISSYNC(ffhp->fh_export)) {
1655 err = nfserrno(nfsd_sync_dir(ddir));
1656 write_inode_now(dest, 1);
1660 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1663 err = nfserrno(host_err);
1665 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1674 err = nfserrno(host_err);
1680 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1683 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1684 struct svc_fh *tfhp, char *tname, int tlen)
1686 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1687 struct inode *fdir, *tdir;
1691 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1694 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1698 fdentry = ffhp->fh_dentry;
1699 fdir = fdentry->d_inode;
1701 tdentry = tfhp->fh_dentry;
1702 tdir = tdentry->d_inode;
1704 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1705 if (ffhp->fh_export != tfhp->fh_export)
1709 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1712 /* cannot use fh_lock as we need deadlock protective ordering
1713 * so do it by hand */
1714 trap = lock_rename(tdentry, fdentry);
1715 ffhp->fh_locked = tfhp->fh_locked = 1;
1719 odentry = lookup_one_len(fname, fdentry, flen);
1720 host_err = PTR_ERR(odentry);
1721 if (IS_ERR(odentry))
1725 if (!odentry->d_inode)
1728 if (odentry == trap)
1731 ndentry = lookup_one_len(tname, tdentry, tlen);
1732 host_err = PTR_ERR(ndentry);
1733 if (IS_ERR(ndentry))
1735 host_err = -ENOTEMPTY;
1736 if (ndentry == trap)
1739 if (svc_msnfs(ffhp) &&
1740 ((atomic_read(&odentry->d_count) > 1)
1741 || (atomic_read(&ndentry->d_count) > 1))) {
1747 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1749 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1753 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1754 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1755 host_err = nfsd_sync_dir(tdentry);
1757 host_err = nfsd_sync_dir(fdentry);
1760 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1767 err = nfserrno(host_err);
1769 /* we cannot reply on fh_unlock on the two filehandles,
1770 * as that would do the wrong thing if the two directories
1771 * were the same, so again we do it by hand
1773 fill_post_wcc(ffhp);
1774 fill_post_wcc(tfhp);
1775 unlock_rename(tdentry, fdentry);
1776 ffhp->fh_locked = tfhp->fh_locked = 0;
1783 * Unlink a file or directory
1784 * N.B. After this call fhp needs an fh_put
1787 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1788 char *fname, int flen)
1790 struct dentry *dentry, *rdentry;
1796 if (!flen || isdotent(fname, flen))
1798 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1802 fh_lock_nested(fhp, I_MUTEX_PARENT);
1803 dentry = fhp->fh_dentry;
1804 dirp = dentry->d_inode;
1806 rdentry = lookup_one_len(fname, dentry, flen);
1807 host_err = PTR_ERR(rdentry);
1808 if (IS_ERR(rdentry))
1811 if (!rdentry->d_inode) {
1818 type = rdentry->d_inode->i_mode & S_IFMT;
1820 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1824 if (type != S_IFDIR) { /* It's UNLINK */
1826 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1827 (atomic_read(&rdentry->d_count) > 1)) {
1831 host_err = vfs_unlink(dirp, rdentry);
1832 } else { /* It's RMDIR */
1833 host_err = vfs_rmdir(dirp, rdentry);
1840 if (EX_ISSYNC(fhp->fh_export))
1841 host_err = nfsd_sync_dir(dentry);
1844 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1846 err = nfserrno(host_err);
1852 * We do this buffering because we must not call back into the file
1853 * system's ->lookup() method from the filldir callback. That may well
1854 * deadlock a number of file systems.
1856 * This is based heavily on the implementation of same in XFS.
1858 struct buffered_dirent {
1862 unsigned int d_type;
1866 struct readdir_data {
1872 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1873 loff_t offset, u64 ino, unsigned int d_type)
1875 struct readdir_data *buf = __buf;
1876 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1877 unsigned int reclen;
1879 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1880 if (buf->used + reclen > PAGE_SIZE) {
1885 de->namlen = namlen;
1886 de->offset = offset;
1888 de->d_type = d_type;
1889 memcpy(de->name, name, namlen);
1890 buf->used += reclen;
1895 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1896 struct readdir_cd *cdp, loff_t *offsetp)
1898 struct readdir_data buf;
1899 struct buffered_dirent *de;
1904 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1906 return nfserrno(-ENOMEM);
1911 struct inode *dir_inode = file->f_path.dentry->d_inode;
1912 unsigned int reclen;
1914 cdp->err = nfserr_eof; /* will be cleared on successful read */
1918 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1931 * Various filldir functions may end up calling back into
1932 * lookup_one_len() and the file system's ->lookup() method.
1933 * These expect i_mutex to be held, as it would within readdir.
1935 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1939 de = (struct buffered_dirent *)buf.dirent;
1941 offset = de->offset;
1943 if (func(cdp, de->name, de->namlen, de->offset,
1944 de->ino, de->d_type))
1947 if (cdp->err != nfs_ok)
1950 reclen = ALIGN(sizeof(*de) + de->namlen,
1953 de = (struct buffered_dirent *)((char *)de + reclen);
1955 mutex_unlock(&dir_inode->i_mutex);
1956 if (size > 0) /* We bailed out early */
1959 offset = vfs_llseek(file, 0, SEEK_CUR);
1962 free_page((unsigned long)(buf.dirent));
1965 return nfserrno(host_err);
1972 * Read entries from a directory.
1973 * The NFSv3/4 verifier we ignore for now.
1976 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1977 struct readdir_cd *cdp, filldir_t func)
1981 loff_t offset = *offsetp;
1983 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1987 offset = vfs_llseek(file, offset, 0);
1989 err = nfserrno((int)offset);
1993 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1995 if (err == nfserr_eof || err == nfserr_toosmall)
1996 err = nfs_ok; /* can still be found in ->err */
2004 * Get file system stats
2005 * N.B. After this call fhp needs an fh_put
2008 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2010 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2011 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2016 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2018 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2022 * Check for a user's access permissions to this inode.
2025 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2026 struct dentry *dentry, int acc)
2028 struct inode *inode = dentry->d_inode;
2031 if (acc == NFSD_MAY_NOP)
2034 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2036 (acc & NFSD_MAY_READ)? " read" : "",
2037 (acc & NFSD_MAY_WRITE)? " write" : "",
2038 (acc & NFSD_MAY_EXEC)? " exec" : "",
2039 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2040 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2041 (acc & NFSD_MAY_LOCK)? " lock" : "",
2042 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2044 IS_IMMUTABLE(inode)? " immut" : "",
2045 IS_APPEND(inode)? " append" : "",
2046 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2047 dprintk(" owner %d/%d user %d/%d\n",
2048 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2051 /* Normally we reject any write/sattr etc access on a read-only file
2052 * system. But if it is IRIX doing check on write-access for a
2053 * device special file, we ignore rofs.
2055 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2056 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2057 if (exp_rdonly(rqstp, exp) ||
2058 __mnt_is_readonly(exp->ex_path.mnt))
2060 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2063 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2066 if (acc & NFSD_MAY_LOCK) {
2067 /* If we cannot rely on authentication in NLM requests,
2068 * just allow locks, otherwise require read permission, or
2071 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2074 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2077 * The file owner always gets access permission for accesses that
2078 * would normally be checked at open time. This is to make
2079 * file access work even when the client has done a fchmod(fd, 0).
2081 * However, `cp foo bar' should fail nevertheless when bar is
2082 * readonly. A sensible way to do this might be to reject all
2083 * attempts to truncate a read-only file, because a creat() call
2084 * always implies file truncation.
2085 * ... but this isn't really fair. A process may reasonably call
2086 * ftruncate on an open file descriptor on a file with perm 000.
2087 * We must trust the client to do permission checking - using "ACCESS"
2090 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2091 inode->i_uid == current_fsuid())
2094 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2095 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2097 /* Allow read access to binaries even when mode 111 */
2098 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2099 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2100 err = inode_permission(inode, MAY_EXEC);
2102 return err? nfserrno(err) : 0;
2106 nfsd_racache_shutdown(void)
2108 struct raparms *raparm, *last_raparm;
2111 dprintk("nfsd: freeing readahead buffers.\n");
2113 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2114 raparm = raparm_hash[i].pb_head;
2116 last_raparm = raparm;
2117 raparm = raparm->p_next;
2120 raparm_hash[i].pb_head = NULL;
2124 * Initialize readahead param cache
2127 nfsd_racache_init(int cache_size)
2132 struct raparms **raparm = NULL;
2135 if (raparm_hash[0].pb_head)
2137 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2140 cache_size = nperbucket * RAPARM_HASH_SIZE;
2142 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2144 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2145 spin_lock_init(&raparm_hash[i].pb_lock);
2147 raparm = &raparm_hash[i].pb_head;
2148 for (j = 0; j < nperbucket; j++) {
2149 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2152 raparm = &(*raparm)->p_next;
2157 nfsdstats.ra_size = cache_size;
2161 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2162 nfsd_racache_shutdown();
2166 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2168 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2170 struct inode *inode = fhp->fh_dentry->d_inode;
2174 struct posix_acl *acl;
2176 if (!IS_POSIXACL(inode))
2177 return ERR_PTR(-EOPNOTSUPP);
2180 case ACL_TYPE_ACCESS:
2181 name = POSIX_ACL_XATTR_ACCESS;
2183 case ACL_TYPE_DEFAULT:
2184 name = POSIX_ACL_XATTR_DEFAULT;
2187 return ERR_PTR(-EOPNOTSUPP);
2190 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2192 return ERR_PTR(size);
2194 acl = posix_acl_from_xattr(value, size);
2200 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2202 struct inode *inode = fhp->fh_dentry->d_inode;
2208 if (!IS_POSIXACL(inode) ||
2209 !inode->i_op->setxattr || !inode->i_op->removexattr)
2212 case ACL_TYPE_ACCESS:
2213 name = POSIX_ACL_XATTR_ACCESS;
2215 case ACL_TYPE_DEFAULT:
2216 name = POSIX_ACL_XATTR_DEFAULT;
2222 if (acl && acl->a_count) {
2223 size = posix_acl_xattr_size(acl->a_count);
2224 value = kmalloc(size, GFP_KERNEL);
2227 error = posix_acl_to_xattr(acl, value, size);
2234 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2238 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2240 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2243 error = vfs_removexattr(fhp->fh_dentry, name);
2244 if (error == -ENODATA)
2248 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2254 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */