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 static struct raparms * raparml;
87 #define RAPARM_HASH_BITS 4
88 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
89 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
90 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
93 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
95 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
96 * or nfs_ok having possibly changed *dpp and *expp
99 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
100 struct svc_export **expp)
102 struct svc_export *exp = *expp, *exp2 = NULL;
103 struct dentry *dentry = *dpp;
104 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
105 struct dentry *mounts = dget(dentry);
108 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
110 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
112 if (PTR_ERR(exp2) != -ENOENT)
118 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
119 /* successfully crossed mount point */
134 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
135 const char *name, unsigned int len,
136 struct svc_export **exp_ret, struct dentry **dentry_ret)
138 struct svc_export *exp;
139 struct dentry *dparent;
140 struct dentry *dentry;
144 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
146 /* Obtain dentry and export. */
147 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
151 dparent = fhp->fh_dentry;
152 exp = fhp->fh_export;
155 /* Lookup the name, but don't follow links */
156 if (isdotent(name, len)) {
158 dentry = dget(dparent);
159 else if (dparent != exp->ex_path.dentry)
160 dentry = dget_parent(dparent);
161 else if (!EX_NOHIDE(exp))
162 dentry = dget(dparent); /* .. == . just like at / */
164 /* checking mountpoint crossing is very different when stepping up */
165 struct svc_export *exp2 = NULL;
167 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
168 dentry = dget(dparent);
169 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
171 dp = dget_parent(dentry);
175 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
176 if (PTR_ERR(exp2) == -ENOENT) {
178 dentry = dget(dparent);
179 } else if (IS_ERR(exp2)) {
180 host_err = PTR_ERR(exp2);
192 dentry = lookup_one_len(name, dparent, len);
193 host_err = PTR_ERR(dentry);
197 * check if we have crossed a mount point ...
199 if (d_mountpoint(dentry)) {
200 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
206 *dentry_ret = dentry;
212 return nfserrno(host_err);
216 * Look up one component of a pathname.
217 * N.B. After this call _both_ fhp and resfh need an fh_put
219 * If the lookup would cross a mountpoint, and the mounted filesystem
220 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
221 * accepted as it stands and the mounted directory is
222 * returned. Otherwise the covered directory is returned.
223 * NOTE: this mountpoint crossing is not supported properly by all
224 * clients and is explicitly disallowed for NFSv3
225 * NeilBrown <neilb@cse.unsw.edu.au>
228 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
229 unsigned int len, struct svc_fh *resfh)
231 struct svc_export *exp;
232 struct dentry *dentry;
235 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
238 err = check_nfsd_access(exp, rqstp);
242 * Note: we compose the file handle now, but as the
243 * dentry may be negative, it may need to be updated.
245 err = fh_compose(resfh, exp, dentry, fhp);
246 if (!err && !dentry->d_inode)
256 * Set various file attributes.
257 * N.B. After this call fhp needs an fh_put
260 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
261 int check_guard, time_t guardtime)
263 struct dentry *dentry;
265 int accmode = NFSD_MAY_SATTR;
271 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
272 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
273 if (iap->ia_valid & ATTR_SIZE)
277 err = fh_verify(rqstp, fhp, ftype, accmode);
281 dentry = fhp->fh_dentry;
282 inode = dentry->d_inode;
284 /* Ignore any mode updates on symlinks */
285 if (S_ISLNK(inode->i_mode))
286 iap->ia_valid &= ~ATTR_MODE;
292 * NFSv2 does not differentiate between "set-[ac]time-to-now"
293 * which only requires access, and "set-[ac]time-to-X" which
294 * requires ownership.
295 * So if it looks like it might be "set both to the same time which
296 * is close to now", and if inode_change_ok fails, then we
297 * convert to "set to now" instead of "set to explicit time"
299 * We only call inode_change_ok as the last test as technically
300 * it is not an interface that we should be using. It is only
301 * valid if the filesystem does not define it's own i_op->setattr.
303 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
304 #define MAX_TOUCH_TIME_ERROR (30*60)
305 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
306 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
310 * Now just make sure time is in the right ballpark.
311 * Solaris, at least, doesn't seem to care what the time
312 * request is. We require it be within 30 minutes of now.
314 time_t delta = iap->ia_atime.tv_sec - get_seconds();
317 if (delta < MAX_TOUCH_TIME_ERROR &&
318 inode_change_ok(inode, iap) != 0) {
320 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
321 * This will cause notify_change to set these times
324 iap->ia_valid &= ~BOTH_TIME_SET;
329 * The size case is special.
330 * It changes the file as well as the attributes.
332 if (iap->ia_valid & ATTR_SIZE) {
333 if (iap->ia_size < inode->i_size) {
334 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
335 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
341 * If we are changing the size of the file, then
342 * we need to break all leases.
344 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
345 if (host_err == -EWOULDBLOCK)
346 host_err = -ETIMEDOUT;
347 if (host_err) /* ENOMEM or EWOULDBLOCK */
350 host_err = get_write_access(inode);
355 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
357 put_write_access(inode);
363 /* sanitize the mode change */
364 if (iap->ia_valid & ATTR_MODE) {
365 iap->ia_mode &= S_IALLUGO;
366 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
369 /* Revoke setuid/setgid on chown */
370 if (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
371 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)) {
372 iap->ia_valid |= ATTR_KILL_PRIV;
373 if (iap->ia_valid & ATTR_MODE) {
374 /* we're setting mode too, just clear the s*id bits */
375 iap->ia_mode &= ~S_ISUID;
376 if (iap->ia_mode & S_IXGRP)
377 iap->ia_mode &= ~S_ISGID;
379 /* set ATTR_KILL_* bits and let VFS handle it */
380 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
384 /* Change the attributes. */
386 iap->ia_valid |= ATTR_CTIME;
388 err = nfserr_notsync;
389 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
391 host_err = notify_change(dentry, iap);
392 err = nfserrno(host_err);
396 put_write_access(inode);
398 if (EX_ISSYNC(fhp->fh_export))
399 write_inode_now(inode, 1);
404 err = nfserrno(host_err);
408 #if defined(CONFIG_NFSD_V2_ACL) || \
409 defined(CONFIG_NFSD_V3_ACL) || \
410 defined(CONFIG_NFSD_V4)
411 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
415 buflen = vfs_getxattr(dentry, key, NULL, 0);
419 *buf = kmalloc(buflen, GFP_KERNEL);
423 return vfs_getxattr(dentry, key, *buf, buflen);
427 #if defined(CONFIG_NFSD_V4)
429 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
436 buflen = posix_acl_xattr_size(pacl->a_count);
437 buf = kmalloc(buflen, GFP_KERNEL);
442 len = posix_acl_to_xattr(pacl, buf, buflen);
448 error = vfs_setxattr(dentry, key, buf, len, 0);
455 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
456 struct nfs4_acl *acl)
460 struct dentry *dentry;
462 struct posix_acl *pacl = NULL, *dpacl = NULL;
463 unsigned int flags = 0;
466 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
470 dentry = fhp->fh_dentry;
471 inode = dentry->d_inode;
472 if (S_ISDIR(inode->i_mode))
473 flags = NFS4_ACL_DIR;
475 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
476 if (host_error == -EINVAL) {
477 return nfserr_attrnotsupp;
478 } else if (host_error < 0)
481 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
485 if (S_ISDIR(inode->i_mode))
486 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
489 posix_acl_release(pacl);
490 posix_acl_release(dpacl);
492 if (host_error == -EOPNOTSUPP)
493 return nfserr_attrnotsupp;
495 return nfserrno(host_error);
498 static struct posix_acl *
499 _get_posix_acl(struct dentry *dentry, char *key)
502 struct posix_acl *pacl = NULL;
505 buflen = nfsd_getxattr(dentry, key, &buf);
509 return ERR_PTR(buflen);
511 pacl = posix_acl_from_xattr(buf, buflen);
517 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
519 struct inode *inode = dentry->d_inode;
521 struct posix_acl *pacl = NULL, *dpacl = NULL;
522 unsigned int flags = 0;
524 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
525 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
526 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
528 error = PTR_ERR(pacl);
533 if (S_ISDIR(inode->i_mode)) {
534 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
535 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
537 else if (IS_ERR(dpacl)) {
538 error = PTR_ERR(dpacl);
542 flags = NFS4_ACL_DIR;
545 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
547 error = PTR_ERR(*acl);
551 posix_acl_release(pacl);
552 posix_acl_release(dpacl);
556 #endif /* defined(CONFIG_NFS_V4) */
558 #ifdef CONFIG_NFSD_V3
560 * Check server access rights to a file system object
566 static struct accessmap nfs3_regaccess[] = {
567 { NFS3_ACCESS_READ, NFSD_MAY_READ },
568 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
569 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
570 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
575 static struct accessmap nfs3_diraccess[] = {
576 { NFS3_ACCESS_READ, NFSD_MAY_READ },
577 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
578 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
579 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
580 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
585 static struct accessmap nfs3_anyaccess[] = {
586 /* Some clients - Solaris 2.6 at least, make an access call
587 * to the server to check for access for things like /dev/null
588 * (which really, the server doesn't care about). So
589 * We provide simple access checking for them, looking
590 * mainly at mode bits, and we make sure to ignore read-only
593 { NFS3_ACCESS_READ, NFSD_MAY_READ },
594 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
595 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
596 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
602 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
604 struct accessmap *map;
605 struct svc_export *export;
606 struct dentry *dentry;
607 u32 query, result = 0, sresult = 0;
610 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
614 export = fhp->fh_export;
615 dentry = fhp->fh_dentry;
617 if (S_ISREG(dentry->d_inode->i_mode))
618 map = nfs3_regaccess;
619 else if (S_ISDIR(dentry->d_inode->i_mode))
620 map = nfs3_diraccess;
622 map = nfs3_anyaccess;
626 for (; map->access; map++) {
627 if (map->access & query) {
630 sresult |= map->access;
632 err2 = nfsd_permission(rqstp, export, dentry, map->how);
635 result |= map->access;
638 /* the following error codes just mean the access was not allowed,
639 * rather than an error occurred */
643 /* simply don't "or" in the access bit. */
653 *supported = sresult;
658 #endif /* CONFIG_NFSD_V3 */
663 * Open an existing file or directory.
664 * The access argument indicates the type of open (read/write/lock)
665 * N.B. After this call fhp needs an fh_put
668 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
669 int access, struct file **filp)
671 struct dentry *dentry;
673 int flags = O_RDONLY|O_LARGEFILE;
678 * If we get here, then the client has already done an "open",
679 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
680 * in case a chmod has now revoked permission.
682 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
686 dentry = fhp->fh_dentry;
687 inode = dentry->d_inode;
689 /* Disallow write access to files with the append-only bit set
690 * or any access when mandatory locking enabled
693 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
696 * We must ignore files (but only files) which might have mandatory
697 * locks on them because there is no way to know if the accesser has
700 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
707 * Check to see if there are any leases on this file.
708 * This may block while leases are broken.
710 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
711 if (host_err == -EWOULDBLOCK)
712 host_err = -ETIMEDOUT;
713 if (host_err) /* NOMEM or WOULDBLOCK */
716 if (access & NFSD_MAY_WRITE) {
717 if (access & NFSD_MAY_READ)
718 flags = O_RDWR|O_LARGEFILE;
720 flags = O_WRONLY|O_LARGEFILE;
724 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
727 host_err = PTR_ERR(*filp);
729 err = nfserrno(host_err);
738 nfsd_close(struct file *filp)
745 * As this calls fsync (not fdatasync) there is no need for a write_inode
748 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
749 const struct file_operations *fop)
751 struct inode *inode = dp->d_inode;
752 int (*fsync) (struct file *, struct dentry *, int);
755 err = filemap_fdatawrite(inode->i_mapping);
756 if (err == 0 && fop && (fsync = fop->fsync))
757 err = fsync(filp, dp, 0);
759 err = filemap_fdatawait(inode->i_mapping);
766 nfsd_sync(struct file *filp)
769 struct inode *inode = filp->f_path.dentry->d_inode;
770 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
771 mutex_lock(&inode->i_mutex);
772 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
773 mutex_unlock(&inode->i_mutex);
779 nfsd_sync_dir(struct dentry *dp)
781 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
785 * Obtain the readahead parameters for the file
786 * specified by (dev, ino).
789 static inline struct raparms *
790 nfsd_get_raparms(dev_t dev, ino_t ino)
792 struct raparms *ra, **rap, **frap = NULL;
795 struct raparm_hbucket *rab;
797 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
798 rab = &raparm_hash[hash];
800 spin_lock(&rab->pb_lock);
801 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
802 if (ra->p_ino == ino && ra->p_dev == dev)
805 if (ra->p_count == 0)
808 depth = nfsdstats.ra_size;
810 spin_unlock(&rab->pb_lock);
820 if (rap != &rab->pb_head) {
822 ra->p_next = rab->pb_head;
826 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
827 spin_unlock(&rab->pb_lock);
832 * Grab and keep cached pages associated with a file in the svc_rqst
833 * so that they can be passed to the network sendmsg/sendpage routines
834 * directly. They will be released after the sending has completed.
837 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
838 struct splice_desc *sd)
840 struct svc_rqst *rqstp = sd->u.data;
841 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
842 struct page *page = buf->page;
846 ret = buf->ops->confirm(pipe, buf);
852 if (rqstp->rq_res.page_len == 0) {
857 rqstp->rq_res.page_base = buf->offset;
858 rqstp->rq_res.page_len = size;
859 } else if (page != pp[-1]) {
865 rqstp->rq_res.page_len += size;
867 rqstp->rq_res.page_len += size;
872 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
873 struct splice_desc *sd)
875 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
878 static inline int svc_msnfs(struct svc_fh *ffhp)
881 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
888 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
889 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
898 inode = file->f_path.dentry->d_inode;
900 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
903 /* Get readahead parameters */
904 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
907 file->f_ra = ra->p_ra;
909 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
910 struct splice_desc sd = {
917 rqstp->rq_resused = 1;
918 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
922 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
926 /* Write back readahead params */
928 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
929 spin_lock(&rab->pb_lock);
930 ra->p_ra = file->f_ra;
933 spin_unlock(&rab->pb_lock);
937 nfsdstats.io_read += host_err;
940 fsnotify_access(file->f_path.dentry);
942 err = nfserrno(host_err);
947 static void kill_suid(struct dentry *dentry)
950 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
952 mutex_lock(&dentry->d_inode->i_mutex);
953 notify_change(dentry, &ia);
954 mutex_unlock(&dentry->d_inode->i_mutex);
958 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
959 loff_t offset, struct kvec *vec, int vlen,
960 unsigned long *cnt, int *stablep)
962 struct svc_export *exp;
963 struct dentry *dentry;
968 int stable = *stablep;
973 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
974 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
978 dentry = file->f_path.dentry;
979 inode = dentry->d_inode;
980 exp = fhp->fh_export;
983 * Request sync writes if
984 * - the sync export option has been set, or
985 * - the client requested O_SYNC behavior (NFSv3 feature).
986 * - The file system doesn't support fsync().
987 * When gathered writes have been configured for this volume,
988 * flushing the data to disk is handled separately below.
991 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
993 *stablep = 2; /* FILE_SYNC */
998 if (stable && !EX_WGATHER(exp))
999 file->f_flags |= O_SYNC;
1001 /* Write the data. */
1002 oldfs = get_fs(); set_fs(KERNEL_DS);
1003 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1005 if (host_err >= 0) {
1007 nfsdstats.io_write += host_err;
1008 fsnotify_modify(file->f_path.dentry);
1011 /* clear setuid/setgid flag after write */
1012 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
1015 if (host_err >= 0 && stable) {
1016 static ino_t last_ino;
1017 static dev_t last_dev;
1020 * Gathered writes: If another process is currently
1021 * writing to the file, there's a high chance
1022 * this is another nfsd (triggered by a bulk write
1023 * from a client's biod). Rather than syncing the
1024 * file with each write request, we sleep for 10 msec.
1026 * I don't know if this roughly approximates
1027 * C. Juszak's idea of gathered writes, but it's a
1028 * nice and simple solution (IMHO), and it seems to
1031 if (EX_WGATHER(exp)) {
1032 if (atomic_read(&inode->i_writecount) > 1
1033 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1034 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1036 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1039 if (inode->i_state & I_DIRTY) {
1040 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1041 host_err=nfsd_sync(file);
1044 wake_up(&inode->i_wait);
1047 last_ino = inode->i_ino;
1048 last_dev = inode->i_sb->s_dev;
1051 dprintk("nfsd: write complete host_err=%d\n", host_err);
1055 err = nfserrno(host_err);
1061 * Read data from a file. count must contain the requested read count
1062 * on entry. On return, *count contains the number of bytes actually read.
1063 * N.B. After this call fhp needs an fh_put
1066 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1067 loff_t offset, struct kvec *vec, int vlen,
1068 unsigned long *count)
1073 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1074 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1077 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1079 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1082 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1090 * Write data to a file.
1091 * The stable flag requests synchronous writes.
1092 * N.B. After this call fhp needs an fh_put
1095 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1096 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1102 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1103 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1106 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1109 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1114 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1122 #ifdef CONFIG_NFSD_V3
1124 * Commit all pending writes to stable storage.
1125 * Strictly speaking, we could sync just the indicated file region here,
1126 * but there's currently no way we can ask the VFS to do so.
1128 * Unfortunately we cannot lock the file to make sure we return full WCC
1129 * data to the client, as locking happens lower down in the filesystem.
1132 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1133 loff_t offset, unsigned long count)
1138 if ((u64)count > ~(u64)offset)
1139 return nfserr_inval;
1141 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1144 if (EX_ISSYNC(fhp->fh_export)) {
1145 if (file->f_op && file->f_op->fsync) {
1146 err = nfserrno(nfsd_sync(file));
1148 err = nfserr_notsupp;
1155 #endif /* CONFIG_NFSD_V3 */
1158 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1162 * Mode has already been set earlier in create:
1164 iap->ia_valid &= ~ATTR_MODE;
1166 * Setting uid/gid works only for root. Irix appears to
1167 * send along the gid on create when it tries to implement
1168 * setgid directories via NFS:
1170 if (current->fsuid != 0)
1171 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1173 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1178 * Create a file (regular, directory, device, fifo); UNIX sockets
1179 * not yet implemented.
1180 * If the response fh has been verified, the parent directory should
1181 * already be locked. Note that the parent directory is left locked.
1183 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1186 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1187 char *fname, int flen, struct iattr *iap,
1188 int type, dev_t rdev, struct svc_fh *resfhp)
1190 struct dentry *dentry, *dchild = NULL;
1200 if (isdotent(fname, flen))
1203 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1207 dentry = fhp->fh_dentry;
1208 dirp = dentry->d_inode;
1210 err = nfserr_notdir;
1211 if(!dirp->i_op || !dirp->i_op->lookup)
1214 * Check whether the response file handle has been verified yet.
1215 * If it has, the parent directory should already be locked.
1217 if (!resfhp->fh_dentry) {
1218 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1219 fh_lock_nested(fhp, I_MUTEX_PARENT);
1220 dchild = lookup_one_len(fname, dentry, flen);
1221 host_err = PTR_ERR(dchild);
1224 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1228 /* called from nfsd_proc_create */
1229 dchild = dget(resfhp->fh_dentry);
1230 if (!fhp->fh_locked) {
1231 /* not actually possible */
1233 "nfsd_create: parent %s/%s not locked!\n",
1234 dentry->d_parent->d_name.name,
1235 dentry->d_name.name);
1241 * Make sure the child dentry is still negative ...
1244 if (dchild->d_inode) {
1245 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1246 dentry->d_name.name, dchild->d_name.name);
1250 if (!(iap->ia_valid & ATTR_MODE))
1252 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1255 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1256 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1261 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1266 * Get the dir op function pointer.
1271 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1274 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1280 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1284 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1288 if (EX_ISSYNC(fhp->fh_export)) {
1289 err = nfserrno(nfsd_sync_dir(dentry));
1290 write_inode_now(dchild->d_inode, 1);
1293 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1296 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1298 * Update the file handle to get the new inode info.
1301 err = fh_update(resfhp);
1303 if (dchild && !IS_ERR(dchild))
1308 err = nfserrno(host_err);
1312 #ifdef CONFIG_NFSD_V3
1314 * NFSv3 version of nfsd_create
1317 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1318 char *fname, int flen, struct iattr *iap,
1319 struct svc_fh *resfhp, int createmode, u32 *verifier,
1320 int *truncp, int *created)
1322 struct dentry *dentry, *dchild = NULL;
1327 __u32 v_mtime=0, v_atime=0;
1333 if (isdotent(fname, flen))
1335 if (!(iap->ia_valid & ATTR_MODE))
1337 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1341 dentry = fhp->fh_dentry;
1342 dirp = dentry->d_inode;
1344 /* Get all the sanity checks out of the way before
1345 * we lock the parent. */
1346 err = nfserr_notdir;
1347 if(!dirp->i_op || !dirp->i_op->lookup)
1349 fh_lock_nested(fhp, I_MUTEX_PARENT);
1352 * Compose the response file handle.
1354 dchild = lookup_one_len(fname, dentry, flen);
1355 host_err = PTR_ERR(dchild);
1359 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1363 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1364 /* solaris7 gets confused (bugid 4218508) if these have
1365 * the high bit set, so just clear the high bits. If this is
1366 * ever changed to use different attrs for storing the
1367 * verifier, then do_open_lookup() will also need to be fixed
1370 v_mtime = verifier[0]&0x7fffffff;
1371 v_atime = verifier[1]&0x7fffffff;
1374 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1377 if (dchild->d_inode) {
1380 switch (createmode) {
1381 case NFS3_CREATE_UNCHECKED:
1382 if (! S_ISREG(dchild->d_inode->i_mode))
1385 /* in nfsv4, we need to treat this case a little
1386 * differently. we don't want to truncate the
1387 * file now; this would be wrong if the OPEN
1388 * fails for some other reason. furthermore,
1389 * if the size is nonzero, we should ignore it
1390 * according to spec!
1392 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1395 iap->ia_valid &= ATTR_SIZE;
1399 case NFS3_CREATE_EXCLUSIVE:
1400 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1401 && dchild->d_inode->i_atime.tv_sec == v_atime
1402 && dchild->d_inode->i_size == 0 )
1405 case NFS3_CREATE_GUARDED:
1408 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1412 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1414 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1420 if (EX_ISSYNC(fhp->fh_export)) {
1421 err = nfserrno(nfsd_sync_dir(dentry));
1422 /* setattr will sync the child (or not) */
1425 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1426 /* Cram the verifier into atime/mtime */
1427 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1428 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1429 /* XXX someone who knows this better please fix it for nsec */
1430 iap->ia_mtime.tv_sec = v_mtime;
1431 iap->ia_atime.tv_sec = v_atime;
1432 iap->ia_mtime.tv_nsec = 0;
1433 iap->ia_atime.tv_nsec = 0;
1437 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1441 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1443 * Update the filehandle to get the new inode info.
1446 err = fh_update(resfhp);
1450 if (dchild && !IS_ERR(dchild))
1455 err = nfserrno(host_err);
1458 #endif /* CONFIG_NFSD_V3 */
1461 * Read a symlink. On entry, *lenp must contain the maximum path length that
1462 * fits into the buffer. On return, it contains the true length.
1463 * N.B. After this call fhp needs an fh_put
1466 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1468 struct dentry *dentry;
1469 struct inode *inode;
1474 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1478 dentry = fhp->fh_dentry;
1479 inode = dentry->d_inode;
1482 if (!inode->i_op || !inode->i_op->readlink)
1485 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1486 /* N.B. Why does this call need a get_fs()??
1487 * Remove the set_fs and watch the fireworks:-) --okir
1490 oldfs = get_fs(); set_fs(KERNEL_DS);
1491 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1502 err = nfserrno(host_err);
1507 * Create a symlink and look up its inode
1508 * N.B. After this call _both_ fhp and resfhp need an fh_put
1511 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1512 char *fname, int flen,
1513 char *path, int plen,
1514 struct svc_fh *resfhp,
1517 struct dentry *dentry, *dnew;
1525 if (isdotent(fname, flen))
1528 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1532 dentry = fhp->fh_dentry;
1533 dnew = lookup_one_len(fname, dentry, flen);
1534 host_err = PTR_ERR(dnew);
1538 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1542 if (unlikely(path[plen] != 0)) {
1543 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1544 if (path_alloced == NULL)
1547 strncpy(path_alloced, path, plen);
1548 path_alloced[plen] = 0;
1549 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1550 kfree(path_alloced);
1553 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1556 if (EX_ISSYNC(fhp->fh_export))
1557 host_err = nfsd_sync_dir(dentry);
1559 err = nfserrno(host_err);
1562 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1564 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1566 if (err==0) err = cerr;
1571 err = nfserrno(host_err);
1577 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1580 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1581 char *name, int len, struct svc_fh *tfhp)
1583 struct dentry *ddir, *dnew, *dold;
1584 struct inode *dirp, *dest;
1588 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1591 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1599 if (isdotent(name, len))
1602 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1603 ddir = ffhp->fh_dentry;
1604 dirp = ddir->d_inode;
1606 dnew = lookup_one_len(name, ddir, len);
1607 host_err = PTR_ERR(dnew);
1611 dold = tfhp->fh_dentry;
1612 dest = dold->d_inode;
1614 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1616 err = nfserrno(host_err);
1619 host_err = vfs_link(dold, dirp, dnew);
1621 if (EX_ISSYNC(ffhp->fh_export)) {
1622 err = nfserrno(nfsd_sync_dir(ddir));
1623 write_inode_now(dest, 1);
1627 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1630 err = nfserrno(host_err);
1632 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1641 err = nfserrno(host_err);
1647 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1650 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1651 struct svc_fh *tfhp, char *tname, int tlen)
1653 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1654 struct inode *fdir, *tdir;
1658 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1661 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1665 fdentry = ffhp->fh_dentry;
1666 fdir = fdentry->d_inode;
1668 tdentry = tfhp->fh_dentry;
1669 tdir = tdentry->d_inode;
1671 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1672 if (ffhp->fh_export != tfhp->fh_export)
1676 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1679 /* cannot use fh_lock as we need deadlock protective ordering
1680 * so do it by hand */
1681 trap = lock_rename(tdentry, fdentry);
1682 ffhp->fh_locked = tfhp->fh_locked = 1;
1686 odentry = lookup_one_len(fname, fdentry, flen);
1687 host_err = PTR_ERR(odentry);
1688 if (IS_ERR(odentry))
1692 if (!odentry->d_inode)
1695 if (odentry == trap)
1698 ndentry = lookup_one_len(tname, tdentry, tlen);
1699 host_err = PTR_ERR(ndentry);
1700 if (IS_ERR(ndentry))
1702 host_err = -ENOTEMPTY;
1703 if (ndentry == trap)
1706 if (svc_msnfs(ffhp) &&
1707 ((atomic_read(&odentry->d_count) > 1)
1708 || (atomic_read(&ndentry->d_count) > 1))) {
1714 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1716 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1720 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1721 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1722 host_err = nfsd_sync_dir(tdentry);
1724 host_err = nfsd_sync_dir(fdentry);
1727 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1734 err = nfserrno(host_err);
1736 /* we cannot reply on fh_unlock on the two filehandles,
1737 * as that would do the wrong thing if the two directories
1738 * were the same, so again we do it by hand
1740 fill_post_wcc(ffhp);
1741 fill_post_wcc(tfhp);
1742 unlock_rename(tdentry, fdentry);
1743 ffhp->fh_locked = tfhp->fh_locked = 0;
1750 * Unlink a file or directory
1751 * N.B. After this call fhp needs an fh_put
1754 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1755 char *fname, int flen)
1757 struct dentry *dentry, *rdentry;
1763 if (!flen || isdotent(fname, flen))
1765 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1769 fh_lock_nested(fhp, I_MUTEX_PARENT);
1770 dentry = fhp->fh_dentry;
1771 dirp = dentry->d_inode;
1773 rdentry = lookup_one_len(fname, dentry, flen);
1774 host_err = PTR_ERR(rdentry);
1775 if (IS_ERR(rdentry))
1778 if (!rdentry->d_inode) {
1785 type = rdentry->d_inode->i_mode & S_IFMT;
1787 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1791 if (type != S_IFDIR) { /* It's UNLINK */
1793 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1794 (atomic_read(&rdentry->d_count) > 1)) {
1798 host_err = vfs_unlink(dirp, rdentry);
1799 } else { /* It's RMDIR */
1800 host_err = vfs_rmdir(dirp, rdentry);
1807 if (EX_ISSYNC(fhp->fh_export))
1808 host_err = nfsd_sync_dir(dentry);
1811 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1813 err = nfserrno(host_err);
1819 * Read entries from a directory.
1820 * The NFSv3/4 verifier we ignore for now.
1823 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1824 struct readdir_cd *cdp, filldir_t func)
1829 loff_t offset = *offsetp;
1831 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1835 offset = vfs_llseek(file, offset, 0);
1837 err = nfserrno((int)offset);
1842 * Read the directory entries. This silly loop is necessary because
1843 * readdir() is not guaranteed to fill up the entire buffer, but
1844 * may choose to do less.
1848 cdp->err = nfserr_eof; /* will be cleared on successful read */
1849 host_err = vfs_readdir(file, func, cdp);
1850 } while (host_err >=0 && cdp->err == nfs_ok);
1852 err = nfserrno(host_err);
1855 *offsetp = vfs_llseek(file, 0, 1);
1857 if (err == nfserr_eof || err == nfserr_toosmall)
1858 err = nfs_ok; /* can still be found in ->err */
1866 * Get file system stats
1867 * N.B. After this call fhp needs an fh_put
1870 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1872 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
1873 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1878 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1880 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1884 * Check for a user's access permissions to this inode.
1887 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1888 struct dentry *dentry, int acc)
1890 struct inode *inode = dentry->d_inode;
1893 if (acc == NFSD_MAY_NOP)
1896 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1898 (acc & NFSD_MAY_READ)? " read" : "",
1899 (acc & NFSD_MAY_WRITE)? " write" : "",
1900 (acc & NFSD_MAY_EXEC)? " exec" : "",
1901 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1902 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1903 (acc & NFSD_MAY_LOCK)? " lock" : "",
1904 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1906 IS_IMMUTABLE(inode)? " immut" : "",
1907 IS_APPEND(inode)? " append" : "",
1908 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1909 dprintk(" owner %d/%d user %d/%d\n",
1910 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1913 /* Normally we reject any write/sattr etc access on a read-only file
1914 * system. But if it is IRIX doing check on write-access for a
1915 * device special file, we ignore rofs.
1917 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
1918 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
1919 if (exp_rdonly(rqstp, exp) ||
1920 __mnt_is_readonly(exp->ex_path.mnt))
1922 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1925 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1928 if (acc & NFSD_MAY_LOCK) {
1929 /* If we cannot rely on authentication in NLM requests,
1930 * just allow locks, otherwise require read permission, or
1933 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1936 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
1939 * The file owner always gets access permission for accesses that
1940 * would normally be checked at open time. This is to make
1941 * file access work even when the client has done a fchmod(fd, 0).
1943 * However, `cp foo bar' should fail nevertheless when bar is
1944 * readonly. A sensible way to do this might be to reject all
1945 * attempts to truncate a read-only file, because a creat() call
1946 * always implies file truncation.
1947 * ... but this isn't really fair. A process may reasonably call
1948 * ftruncate on an open file descriptor on a file with perm 000.
1949 * We must trust the client to do permission checking - using "ACCESS"
1952 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
1953 inode->i_uid == current->fsuid)
1956 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
1957 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
1959 /* Allow read access to binaries even when mode 111 */
1960 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1961 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
1962 err = inode_permission(inode, MAY_EXEC);
1964 return err? nfserrno(err) : 0;
1968 nfsd_racache_shutdown(void)
1972 dprintk("nfsd: freeing readahead buffers.\n");
1977 * Initialize readahead param cache
1980 nfsd_racache_init(int cache_size)
1989 if (cache_size < 2*RAPARM_HASH_SIZE)
1990 cache_size = 2*RAPARM_HASH_SIZE;
1991 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1995 "nfsd: Could not allocate memory read-ahead cache.\n");
1999 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2000 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
2001 raparm_hash[i].pb_head = NULL;
2002 spin_lock_init(&raparm_hash[i].pb_lock);
2004 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2005 for (i = 0; i < cache_size - 1; i++) {
2006 if (i % nperbucket == 0)
2007 raparm_hash[j++].pb_head = raparml + i;
2008 if (i % nperbucket < nperbucket-1)
2009 raparml[i].p_next = raparml + i + 1;
2012 nfsdstats.ra_size = cache_size;
2016 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2018 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2020 struct inode *inode = fhp->fh_dentry->d_inode;
2024 struct posix_acl *acl;
2026 if (!IS_POSIXACL(inode))
2027 return ERR_PTR(-EOPNOTSUPP);
2030 case ACL_TYPE_ACCESS:
2031 name = POSIX_ACL_XATTR_ACCESS;
2033 case ACL_TYPE_DEFAULT:
2034 name = POSIX_ACL_XATTR_DEFAULT;
2037 return ERR_PTR(-EOPNOTSUPP);
2040 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2042 return ERR_PTR(size);
2044 acl = posix_acl_from_xattr(value, size);
2050 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2052 struct inode *inode = fhp->fh_dentry->d_inode;
2058 if (!IS_POSIXACL(inode) || !inode->i_op ||
2059 !inode->i_op->setxattr || !inode->i_op->removexattr)
2062 case ACL_TYPE_ACCESS:
2063 name = POSIX_ACL_XATTR_ACCESS;
2065 case ACL_TYPE_DEFAULT:
2066 name = POSIX_ACL_XATTR_DEFAULT;
2072 if (acl && acl->a_count) {
2073 size = posix_acl_xattr_size(acl->a_count);
2074 value = kmalloc(size, GFP_KERNEL);
2077 error = posix_acl_to_xattr(acl, value, size);
2084 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2088 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2090 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2093 error = vfs_removexattr(fhp->fh_dentry, name);
2094 if (error == -ENODATA)
2098 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2104 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */