2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
39 #endif /* CONFIG_NFSD_V4 */
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
55 struct raparms *p_next;
60 struct file_ra_state p_ra;
61 unsigned int p_hindex;
64 struct raparm_hbucket {
65 struct raparms *pb_head;
67 } ____cacheline_aligned_in_smp;
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
90 err = follow_down(&path);
94 exp2 = rqst_exp_get_by_name(rqstp, &path);
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
104 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
109 if (nfsd_v4client(rqstp) ||
110 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111 /* successfully crossed mount point */
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
119 path.dentry = dentry;
129 static void follow_to_parent(struct path *path)
133 while (path->dentry == path->mnt->mnt_root && follow_up(path))
135 dp = dget_parent(path->dentry);
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
142 struct svc_export *exp2;
143 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144 .dentry = dget(dparent)};
146 follow_to_parent(&path);
148 exp2 = rqst_exp_parent(rqstp, &path);
149 if (PTR_ERR(exp2) == -ENOENT) {
150 *dentryp = dget(dparent);
151 } else if (IS_ERR(exp2)) {
153 return PTR_ERR(exp2);
155 *dentryp = dget(path.dentry);
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
169 if (d_mountpoint(dentry))
171 if (!(exp->ex_flags & NFSEXP_V4ROOT))
173 return dentry->d_inode != NULL;
177 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
178 const char *name, unsigned int len,
179 struct svc_export **exp_ret, struct dentry **dentry_ret)
181 struct svc_export *exp;
182 struct dentry *dparent;
183 struct dentry *dentry;
187 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
189 /* Obtain dentry and export. */
190 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
194 dparent = fhp->fh_dentry;
195 exp = fhp->fh_export;
198 /* Lookup the name, but don't follow links */
199 if (isdotent(name, len)) {
201 dentry = dget(dparent);
202 else if (dparent != exp->ex_path.dentry)
203 dentry = dget_parent(dparent);
204 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
205 dentry = dget(dparent); /* .. == . just like at / */
207 /* checking mountpoint crossing is very different when stepping up */
208 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
214 dentry = lookup_one_len(name, dparent, len);
215 host_err = PTR_ERR(dentry);
219 * check if we have crossed a mount point ...
221 if (nfsd_mountpoint(dentry, exp)) {
222 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
228 *dentry_ret = dentry;
234 return nfserrno(host_err);
238 * Look up one component of a pathname.
239 * N.B. After this call _both_ fhp and resfh need an fh_put
241 * If the lookup would cross a mountpoint, and the mounted filesystem
242 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
243 * accepted as it stands and the mounted directory is
244 * returned. Otherwise the covered directory is returned.
245 * NOTE: this mountpoint crossing is not supported properly by all
246 * clients and is explicitly disallowed for NFSv3
250 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
251 unsigned int len, struct svc_fh *resfh)
253 struct svc_export *exp;
254 struct dentry *dentry;
257 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
260 err = check_nfsd_access(exp, rqstp);
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err = fh_compose(resfh, exp, dentry, fhp);
268 if (!err && !dentry->d_inode)
276 static int nfsd_break_lease(struct inode *inode)
278 if (!S_ISREG(inode->i_mode))
280 return break_lease(inode, O_WRONLY | O_NONBLOCK);
284 * Commit metadata changes to stable storage.
287 commit_metadata(struct svc_fh *fhp)
289 struct inode *inode = fhp->fh_dentry->d_inode;
290 const struct export_operations *export_ops = inode->i_sb->s_export_op;
292 if (!EX_ISSYNC(fhp->fh_export))
295 if (export_ops->commit_metadata)
296 return export_ops->commit_metadata(inode);
297 return sync_inode_metadata(inode, 1);
301 * Set various file attributes.
302 * N.B. After this call fhp needs an fh_put
305 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
306 int check_guard, time_t guardtime)
308 struct dentry *dentry;
310 int accmode = NFSD_MAY_SATTR;
316 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
317 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
318 if (iap->ia_valid & ATTR_SIZE)
322 err = fh_verify(rqstp, fhp, ftype, accmode);
326 dentry = fhp->fh_dentry;
327 inode = dentry->d_inode;
329 /* Ignore any mode updates on symlinks */
330 if (S_ISLNK(inode->i_mode))
331 iap->ia_valid &= ~ATTR_MODE;
337 * NFSv2 does not differentiate between "set-[ac]time-to-now"
338 * which only requires access, and "set-[ac]time-to-X" which
339 * requires ownership.
340 * So if it looks like it might be "set both to the same time which
341 * is close to now", and if inode_change_ok fails, then we
342 * convert to "set to now" instead of "set to explicit time"
344 * We only call inode_change_ok as the last test as technically
345 * it is not an interface that we should be using. It is only
346 * valid if the filesystem does not define it's own i_op->setattr.
348 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
349 #define MAX_TOUCH_TIME_ERROR (30*60)
350 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
351 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
355 * Now just make sure time is in the right ballpark.
356 * Solaris, at least, doesn't seem to care what the time
357 * request is. We require it be within 30 minutes of now.
359 time_t delta = iap->ia_atime.tv_sec - get_seconds();
362 if (delta < MAX_TOUCH_TIME_ERROR &&
363 inode_change_ok(inode, iap) != 0) {
365 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
366 * This will cause notify_change to set these times
369 iap->ia_valid &= ~BOTH_TIME_SET;
374 * The size case is special.
375 * It changes the file as well as the attributes.
377 if (iap->ia_valid & ATTR_SIZE) {
378 if (iap->ia_size < inode->i_size) {
379 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
380 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
385 host_err = get_write_access(inode);
390 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
392 put_write_access(inode);
397 /* sanitize the mode change */
398 if (iap->ia_valid & ATTR_MODE) {
399 iap->ia_mode &= S_IALLUGO;
400 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
403 /* Revoke setuid/setgid on chown */
404 if (!S_ISDIR(inode->i_mode) &&
405 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
406 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
407 iap->ia_valid |= ATTR_KILL_PRIV;
408 if (iap->ia_valid & ATTR_MODE) {
409 /* we're setting mode too, just clear the s*id bits */
410 iap->ia_mode &= ~S_ISUID;
411 if (iap->ia_mode & S_IXGRP)
412 iap->ia_mode &= ~S_ISGID;
414 /* set ATTR_KILL_* bits and let VFS handle it */
415 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
419 /* Change the attributes. */
421 iap->ia_valid |= ATTR_CTIME;
423 err = nfserr_notsync;
424 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
425 host_err = nfsd_break_lease(inode);
430 host_err = notify_change(dentry, iap);
431 err = nfserrno(host_err);
435 put_write_access(inode);
437 commit_metadata(fhp);
442 err = nfserrno(host_err);
446 #if defined(CONFIG_NFSD_V2_ACL) || \
447 defined(CONFIG_NFSD_V3_ACL) || \
448 defined(CONFIG_NFSD_V4)
449 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
454 buflen = vfs_getxattr(dentry, key, NULL, 0);
458 *buf = kmalloc(buflen, GFP_KERNEL);
462 ret = vfs_getxattr(dentry, key, *buf, buflen);
469 #if defined(CONFIG_NFSD_V4)
471 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
478 buflen = posix_acl_xattr_size(pacl->a_count);
479 buf = kmalloc(buflen, GFP_KERNEL);
484 len = posix_acl_to_xattr(pacl, buf, buflen);
490 error = vfs_setxattr(dentry, key, buf, len, 0);
497 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
498 struct nfs4_acl *acl)
502 struct dentry *dentry;
504 struct posix_acl *pacl = NULL, *dpacl = NULL;
505 unsigned int flags = 0;
508 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
512 dentry = fhp->fh_dentry;
513 inode = dentry->d_inode;
514 if (S_ISDIR(inode->i_mode))
515 flags = NFS4_ACL_DIR;
517 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
518 if (host_error == -EINVAL) {
519 return nfserr_attrnotsupp;
520 } else if (host_error < 0)
523 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
527 if (S_ISDIR(inode->i_mode))
528 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
531 posix_acl_release(pacl);
532 posix_acl_release(dpacl);
534 if (host_error == -EOPNOTSUPP)
535 return nfserr_attrnotsupp;
537 return nfserrno(host_error);
540 static struct posix_acl *
541 _get_posix_acl(struct dentry *dentry, char *key)
544 struct posix_acl *pacl = NULL;
547 buflen = nfsd_getxattr(dentry, key, &buf);
551 return ERR_PTR(buflen);
553 pacl = posix_acl_from_xattr(buf, buflen);
559 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
561 struct inode *inode = dentry->d_inode;
563 struct posix_acl *pacl = NULL, *dpacl = NULL;
564 unsigned int flags = 0;
566 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
567 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
568 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
570 error = PTR_ERR(pacl);
575 if (S_ISDIR(inode->i_mode)) {
576 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
577 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
579 else if (IS_ERR(dpacl)) {
580 error = PTR_ERR(dpacl);
584 flags = NFS4_ACL_DIR;
587 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
589 error = PTR_ERR(*acl);
593 posix_acl_release(pacl);
594 posix_acl_release(dpacl);
598 #endif /* defined(CONFIG_NFSD_V4) */
600 #ifdef CONFIG_NFSD_V3
602 * Check server access rights to a file system object
608 static struct accessmap nfs3_regaccess[] = {
609 { NFS3_ACCESS_READ, NFSD_MAY_READ },
610 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
611 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
612 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
617 static struct accessmap nfs3_diraccess[] = {
618 { NFS3_ACCESS_READ, NFSD_MAY_READ },
619 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
620 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
621 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
622 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
627 static struct accessmap nfs3_anyaccess[] = {
628 /* Some clients - Solaris 2.6 at least, make an access call
629 * to the server to check for access for things like /dev/null
630 * (which really, the server doesn't care about). So
631 * We provide simple access checking for them, looking
632 * mainly at mode bits, and we make sure to ignore read-only
635 { NFS3_ACCESS_READ, NFSD_MAY_READ },
636 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
637 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
638 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
644 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
646 struct accessmap *map;
647 struct svc_export *export;
648 struct dentry *dentry;
649 u32 query, result = 0, sresult = 0;
652 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
656 export = fhp->fh_export;
657 dentry = fhp->fh_dentry;
659 if (S_ISREG(dentry->d_inode->i_mode))
660 map = nfs3_regaccess;
661 else if (S_ISDIR(dentry->d_inode->i_mode))
662 map = nfs3_diraccess;
664 map = nfs3_anyaccess;
668 for (; map->access; map++) {
669 if (map->access & query) {
672 sresult |= map->access;
674 err2 = nfsd_permission(rqstp, export, dentry, map->how);
677 result |= map->access;
680 /* the following error codes just mean the access was not allowed,
681 * rather than an error occurred */
685 /* simply don't "or" in the access bit. */
695 *supported = sresult;
700 #endif /* CONFIG_NFSD_V3 */
705 * Open an existing file or directory.
706 * The access argument indicates the type of open (read/write/lock)
707 * N.B. After this call fhp needs an fh_put
710 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
711 int access, struct file **filp)
713 struct dentry *dentry;
715 int flags = O_RDONLY|O_LARGEFILE;
719 validate_process_creds();
722 * If we get here, then the client has already done an "open",
723 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
724 * in case a chmod has now revoked permission.
726 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
730 dentry = fhp->fh_dentry;
731 inode = dentry->d_inode;
733 /* Disallow write access to files with the append-only bit set
734 * or any access when mandatory locking enabled
737 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
740 * We must ignore files (but only files) which might have mandatory
741 * locks on them because there is no way to know if the accesser has
744 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
751 * Check to see if there are any leases on this file.
752 * This may block while leases are broken.
754 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
755 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
756 if (host_err) /* NOMEM or WOULDBLOCK */
759 if (access & NFSD_MAY_WRITE) {
760 if (access & NFSD_MAY_READ)
761 flags = O_RDWR|O_LARGEFILE;
763 flags = O_WRONLY|O_LARGEFILE;
765 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
766 flags, current_cred());
768 host_err = PTR_ERR(*filp);
770 host_err = ima_file_check(*filp, access);
772 err = nfserrno(host_err);
774 validate_process_creds();
782 nfsd_close(struct file *filp)
788 * Obtain the readahead parameters for the file
789 * specified by (dev, ino).
792 static inline struct raparms *
793 nfsd_get_raparms(dev_t dev, ino_t ino)
795 struct raparms *ra, **rap, **frap = NULL;
798 struct raparm_hbucket *rab;
800 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
801 rab = &raparm_hash[hash];
803 spin_lock(&rab->pb_lock);
804 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
805 if (ra->p_ino == ino && ra->p_dev == dev)
808 if (ra->p_count == 0)
811 depth = nfsdstats.ra_size;
813 spin_unlock(&rab->pb_lock);
823 if (rap != &rab->pb_head) {
825 ra->p_next = rab->pb_head;
829 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
830 spin_unlock(&rab->pb_lock);
835 * Grab and keep cached pages associated with a file in the svc_rqst
836 * so that they can be passed to the network sendmsg/sendpage routines
837 * directly. They will be released after the sending has completed.
840 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
841 struct splice_desc *sd)
843 struct svc_rqst *rqstp = sd->u.data;
844 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
845 struct page *page = buf->page;
850 if (rqstp->rq_res.page_len == 0) {
855 rqstp->rq_res.page_base = buf->offset;
856 rqstp->rq_res.page_len = size;
857 } else if (page != pp[-1]) {
863 rqstp->rq_res.page_len += size;
865 rqstp->rq_res.page_len += size;
870 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
871 struct splice_desc *sd)
873 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
877 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
878 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
886 inode = file->f_path.dentry->d_inode;
888 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
889 struct splice_desc sd = {
896 rqstp->rq_resused = 1;
897 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
901 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
906 nfsdstats.io_read += host_err;
909 fsnotify_access(file);
911 err = nfserrno(host_err);
915 static void kill_suid(struct dentry *dentry)
918 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
920 mutex_lock(&dentry->d_inode->i_mutex);
921 notify_change(dentry, &ia);
922 mutex_unlock(&dentry->d_inode->i_mutex);
926 * Gathered writes: If another process is currently writing to the file,
927 * there's a high chance this is another nfsd (triggered by a bulk write
928 * from a client's biod). Rather than syncing the file with each write
929 * request, we sleep for 10 msec.
931 * I don't know if this roughly approximates C. Juszak's idea of
932 * gathered writes, but it's a nice and simple solution (IMHO), and it
935 * Note: we do this only in the NFSv2 case, since v3 and higher have a
936 * better tool (separate unstable writes and commits) for solving this
939 static int wait_for_concurrent_writes(struct file *file)
941 struct inode *inode = file->f_path.dentry->d_inode;
942 static ino_t last_ino;
943 static dev_t last_dev;
946 if (atomic_read(&inode->i_writecount) > 1
947 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
948 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
950 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
953 if (inode->i_state & I_DIRTY) {
954 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
955 err = vfs_fsync(file, 0);
957 last_ino = inode->i_ino;
958 last_dev = inode->i_sb->s_dev;
963 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
964 loff_t offset, struct kvec *vec, int vlen,
965 unsigned long *cnt, int *stablep)
967 struct svc_export *exp;
968 struct dentry *dentry;
973 int stable = *stablep;
976 dentry = file->f_path.dentry;
977 inode = dentry->d_inode;
978 exp = fhp->fh_export;
981 * Request sync writes if
982 * - the sync export option has been set, or
983 * - the client requested O_SYNC behavior (NFSv3 feature).
984 * - The file system doesn't support fsync().
985 * When NFSv2 gathered writes have been configured for this volume,
986 * flushing the data to disk is handled separately below.
988 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
990 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
992 *stablep = 2; /* FILE_SYNC */
997 if (stable && !use_wgather) {
998 spin_lock(&file->f_lock);
999 file->f_flags |= O_SYNC;
1000 spin_unlock(&file->f_lock);
1003 /* Write the data. */
1004 oldfs = get_fs(); set_fs(KERNEL_DS);
1005 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1010 nfsdstats.io_write += host_err;
1011 fsnotify_modify(file);
1013 /* clear setuid/setgid flag after write */
1014 if (inode->i_mode & (S_ISUID | S_ISGID))
1017 if (stable && use_wgather)
1018 host_err = wait_for_concurrent_writes(file);
1021 dprintk("nfsd: write complete host_err=%d\n", host_err);
1025 err = nfserrno(host_err);
1030 * Read data from a file. count must contain the requested read count
1031 * on entry. On return, *count contains the number of bytes actually read.
1032 * N.B. After this call fhp needs an fh_put
1034 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1035 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1038 struct inode *inode;
1042 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1046 inode = file->f_path.dentry->d_inode;
1048 /* Get readahead parameters */
1049 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1051 if (ra && ra->p_set)
1052 file->f_ra = ra->p_ra;
1054 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1056 /* Write back readahead params */
1058 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1059 spin_lock(&rab->pb_lock);
1060 ra->p_ra = file->f_ra;
1063 spin_unlock(&rab->pb_lock);
1070 /* As above, but use the provided file descriptor. */
1072 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1073 loff_t offset, struct kvec *vec, int vlen,
1074 unsigned long *count)
1079 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1080 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1083 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1084 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1085 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1091 * Write data to a file.
1092 * The stable flag requests synchronous writes.
1093 * N.B. After this call fhp needs an fh_put
1096 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1097 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1103 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1104 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1107 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1110 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1115 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1123 #ifdef CONFIG_NFSD_V3
1125 * Commit all pending writes to stable storage.
1127 * Note: we only guarantee that data that lies within the range specified
1128 * by the 'offset' and 'count' parameters will be synced.
1130 * Unfortunately we cannot lock the file to make sure we return full WCC
1131 * data to the client, as locking happens lower down in the filesystem.
1134 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1135 loff_t offset, unsigned long count)
1138 loff_t end = LLONG_MAX;
1139 __be32 err = nfserr_inval;
1144 end = offset + (loff_t)count - 1;
1149 err = nfsd_open(rqstp, fhp, S_IFREG,
1150 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1153 if (EX_ISSYNC(fhp->fh_export)) {
1154 int err2 = vfs_fsync_range(file, offset, end, 0);
1156 if (err2 != -EINVAL)
1157 err = nfserrno(err2);
1159 err = nfserr_notsupp;
1166 #endif /* CONFIG_NFSD_V3 */
1169 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1173 * Mode has already been set earlier in create:
1175 iap->ia_valid &= ~ATTR_MODE;
1177 * Setting uid/gid works only for root. Irix appears to
1178 * send along the gid on create when it tries to implement
1179 * setgid directories via NFS:
1181 if (current_fsuid() != 0)
1182 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1184 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1188 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1189 * setting size to 0 may fail for some specific file systems by the permission
1190 * checking which requires WRITE permission but the mode is 000.
1191 * we ignore the resizing(to 0) on the just new created file, since the size is
1192 * 0 after file created.
1194 * call this only after vfs_create() is called.
1197 nfsd_check_ignore_resizing(struct iattr *iap)
1199 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1200 iap->ia_valid &= ~ATTR_SIZE;
1204 * Create a file (regular, directory, device, fifo); UNIX sockets
1205 * not yet implemented.
1206 * If the response fh has been verified, the parent directory should
1207 * already be locked. Note that the parent directory is left locked.
1209 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1212 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1213 char *fname, int flen, struct iattr *iap,
1214 int type, dev_t rdev, struct svc_fh *resfhp)
1216 struct dentry *dentry, *dchild = NULL;
1226 if (isdotent(fname, flen))
1229 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1233 dentry = fhp->fh_dentry;
1234 dirp = dentry->d_inode;
1236 err = nfserr_notdir;
1237 if (!dirp->i_op->lookup)
1240 * Check whether the response file handle has been verified yet.
1241 * If it has, the parent directory should already be locked.
1243 if (!resfhp->fh_dentry) {
1244 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1245 fh_lock_nested(fhp, I_MUTEX_PARENT);
1246 dchild = lookup_one_len(fname, dentry, flen);
1247 host_err = PTR_ERR(dchild);
1250 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1254 /* called from nfsd_proc_create */
1255 dchild = dget(resfhp->fh_dentry);
1256 if (!fhp->fh_locked) {
1257 /* not actually possible */
1259 "nfsd_create: parent %s/%s not locked!\n",
1260 dentry->d_parent->d_name.name,
1261 dentry->d_name.name);
1267 * Make sure the child dentry is still negative ...
1270 if (dchild->d_inode) {
1271 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1272 dentry->d_name.name, dchild->d_name.name);
1276 if (!(iap->ia_valid & ATTR_MODE))
1278 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1281 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1282 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1287 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1292 * Get the dir op function pointer.
1297 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1299 nfsd_check_ignore_resizing(iap);
1302 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1308 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1312 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1316 err = nfsd_create_setattr(rqstp, resfhp, iap);
1319 * nfsd_setattr already committed the child. Transactional filesystems
1320 * had a chance to commit changes for both parent and child
1321 * simultaneously making the following commit_metadata a noop.
1323 err2 = nfserrno(commit_metadata(fhp));
1326 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1328 * Update the file handle to get the new inode info.
1331 err = fh_update(resfhp);
1333 if (dchild && !IS_ERR(dchild))
1338 err = nfserrno(host_err);
1342 #ifdef CONFIG_NFSD_V3
1344 * NFSv3 version of nfsd_create
1347 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1348 char *fname, int flen, struct iattr *iap,
1349 struct svc_fh *resfhp, int createmode, u32 *verifier,
1350 int *truncp, int *created)
1352 struct dentry *dentry, *dchild = NULL;
1356 __u32 v_mtime=0, v_atime=0;
1362 if (isdotent(fname, flen))
1364 if (!(iap->ia_valid & ATTR_MODE))
1366 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1370 dentry = fhp->fh_dentry;
1371 dirp = dentry->d_inode;
1373 /* Get all the sanity checks out of the way before
1374 * we lock the parent. */
1375 err = nfserr_notdir;
1376 if (!dirp->i_op->lookup)
1378 fh_lock_nested(fhp, I_MUTEX_PARENT);
1381 * Compose the response file handle.
1383 dchild = lookup_one_len(fname, dentry, flen);
1384 host_err = PTR_ERR(dchild);
1388 /* If file doesn't exist, check for permissions to create one */
1389 if (!dchild->d_inode) {
1390 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1395 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1399 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1400 /* solaris7 gets confused (bugid 4218508) if these have
1401 * the high bit set, so just clear the high bits. If this is
1402 * ever changed to use different attrs for storing the
1403 * verifier, then do_open_lookup() will also need to be fixed
1406 v_mtime = verifier[0]&0x7fffffff;
1407 v_atime = verifier[1]&0x7fffffff;
1410 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1413 if (dchild->d_inode) {
1416 switch (createmode) {
1417 case NFS3_CREATE_UNCHECKED:
1418 if (! S_ISREG(dchild->d_inode->i_mode))
1421 /* in nfsv4, we need to treat this case a little
1422 * differently. we don't want to truncate the
1423 * file now; this would be wrong if the OPEN
1424 * fails for some other reason. furthermore,
1425 * if the size is nonzero, we should ignore it
1426 * according to spec!
1428 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1431 iap->ia_valid &= ATTR_SIZE;
1435 case NFS3_CREATE_EXCLUSIVE:
1436 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1437 && dchild->d_inode->i_atime.tv_sec == v_atime
1438 && dchild->d_inode->i_size == 0 )
1441 case NFS3_CREATE_GUARDED:
1444 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1448 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1450 mnt_drop_write(fhp->fh_export->ex_path.mnt);
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 err = nfsd_create_setattr(rqstp, resfhp, iap);
1473 * nfsd_setattr already committed the child (and possibly also the parent).
1476 err = nfserrno(commit_metadata(fhp));
1478 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1480 * Update the filehandle to get the new inode info.
1483 err = fh_update(resfhp);
1487 if (dchild && !IS_ERR(dchild))
1492 err = nfserrno(host_err);
1495 #endif /* CONFIG_NFSD_V3 */
1498 * Read a symlink. On entry, *lenp must contain the maximum path length that
1499 * fits into the buffer. On return, it contains the true length.
1500 * N.B. After this call fhp needs an fh_put
1503 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1505 struct dentry *dentry;
1506 struct inode *inode;
1511 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1515 dentry = fhp->fh_dentry;
1516 inode = dentry->d_inode;
1519 if (!inode->i_op->readlink)
1522 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1523 /* N.B. Why does this call need a get_fs()??
1524 * Remove the set_fs and watch the fireworks:-) --okir
1527 oldfs = get_fs(); set_fs(KERNEL_DS);
1528 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1539 err = nfserrno(host_err);
1544 * Create a symlink and look up its inode
1545 * N.B. After this call _both_ fhp and resfhp need an fh_put
1548 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1549 char *fname, int flen,
1550 char *path, int plen,
1551 struct svc_fh *resfhp,
1554 struct dentry *dentry, *dnew;
1562 if (isdotent(fname, flen))
1565 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1569 dentry = fhp->fh_dentry;
1570 dnew = lookup_one_len(fname, dentry, flen);
1571 host_err = PTR_ERR(dnew);
1575 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1579 if (unlikely(path[plen] != 0)) {
1580 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1581 if (path_alloced == NULL)
1584 strncpy(path_alloced, path, plen);
1585 path_alloced[plen] = 0;
1586 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1587 kfree(path_alloced);
1590 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1591 err = nfserrno(host_err);
1593 err = nfserrno(commit_metadata(fhp));
1596 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1598 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1600 if (err==0) err = cerr;
1605 err = nfserrno(host_err);
1611 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1614 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1615 char *name, int len, struct svc_fh *tfhp)
1617 struct dentry *ddir, *dnew, *dold;
1622 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1625 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1633 if (isdotent(name, len))
1636 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1637 ddir = ffhp->fh_dentry;
1638 dirp = ddir->d_inode;
1640 dnew = lookup_one_len(name, ddir, len);
1641 host_err = PTR_ERR(dnew);
1645 dold = tfhp->fh_dentry;
1647 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1649 err = nfserrno(host_err);
1654 goto out_drop_write;
1655 host_err = nfsd_break_lease(dold->d_inode);
1657 goto out_drop_write;
1658 host_err = vfs_link(dold, dirp, dnew);
1660 err = nfserrno(commit_metadata(ffhp));
1662 err = nfserrno(commit_metadata(tfhp));
1664 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1667 err = nfserrno(host_err);
1670 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1679 err = nfserrno(host_err);
1685 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1688 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1689 struct svc_fh *tfhp, char *tname, int tlen)
1691 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1692 struct inode *fdir, *tdir;
1696 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1699 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1703 fdentry = ffhp->fh_dentry;
1704 fdir = fdentry->d_inode;
1706 tdentry = tfhp->fh_dentry;
1707 tdir = tdentry->d_inode;
1709 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1710 if (ffhp->fh_export != tfhp->fh_export)
1714 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1717 /* cannot use fh_lock as we need deadlock protective ordering
1718 * so do it by hand */
1719 trap = lock_rename(tdentry, fdentry);
1720 ffhp->fh_locked = tfhp->fh_locked = 1;
1724 odentry = lookup_one_len(fname, fdentry, flen);
1725 host_err = PTR_ERR(odentry);
1726 if (IS_ERR(odentry))
1730 if (!odentry->d_inode)
1733 if (odentry == trap)
1736 ndentry = lookup_one_len(tname, tdentry, tlen);
1737 host_err = PTR_ERR(ndentry);
1738 if (IS_ERR(ndentry))
1740 host_err = -ENOTEMPTY;
1741 if (ndentry == trap)
1745 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1747 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1751 host_err = nfsd_break_lease(odentry->d_inode);
1753 goto out_drop_write;
1754 if (ndentry->d_inode) {
1755 host_err = nfsd_break_lease(ndentry->d_inode);
1757 goto out_drop_write;
1759 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1761 host_err = commit_metadata(tfhp);
1763 host_err = commit_metadata(ffhp);
1766 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1772 err = nfserrno(host_err);
1774 /* we cannot reply on fh_unlock on the two filehandles,
1775 * as that would do the wrong thing if the two directories
1776 * were the same, so again we do it by hand
1778 fill_post_wcc(ffhp);
1779 fill_post_wcc(tfhp);
1780 unlock_rename(tdentry, fdentry);
1781 ffhp->fh_locked = tfhp->fh_locked = 0;
1788 * Unlink a file or directory
1789 * N.B. After this call fhp needs an fh_put
1792 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1793 char *fname, int flen)
1795 struct dentry *dentry, *rdentry;
1801 if (!flen || isdotent(fname, flen))
1803 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1807 fh_lock_nested(fhp, I_MUTEX_PARENT);
1808 dentry = fhp->fh_dentry;
1809 dirp = dentry->d_inode;
1811 rdentry = lookup_one_len(fname, dentry, flen);
1812 host_err = PTR_ERR(rdentry);
1813 if (IS_ERR(rdentry))
1816 if (!rdentry->d_inode) {
1823 type = rdentry->d_inode->i_mode & S_IFMT;
1825 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1829 host_err = nfsd_break_lease(rdentry->d_inode);
1831 goto out_drop_write;
1832 if (type != S_IFDIR)
1833 host_err = vfs_unlink(dirp, rdentry);
1835 host_err = vfs_rmdir(dirp, rdentry);
1837 host_err = commit_metadata(fhp);
1839 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1844 err = nfserrno(host_err);
1850 * We do this buffering because we must not call back into the file
1851 * system's ->lookup() method from the filldir callback. That may well
1852 * deadlock a number of file systems.
1854 * This is based heavily on the implementation of same in XFS.
1856 struct buffered_dirent {
1860 unsigned int d_type;
1864 struct readdir_data {
1870 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1871 loff_t offset, u64 ino, unsigned int d_type)
1873 struct readdir_data *buf = __buf;
1874 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1875 unsigned int reclen;
1877 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1878 if (buf->used + reclen > PAGE_SIZE) {
1883 de->namlen = namlen;
1884 de->offset = offset;
1886 de->d_type = d_type;
1887 memcpy(de->name, name, namlen);
1888 buf->used += reclen;
1893 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1894 struct readdir_cd *cdp, loff_t *offsetp)
1896 struct readdir_data buf;
1897 struct buffered_dirent *de;
1902 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1904 return nfserrno(-ENOMEM);
1909 struct inode *dir_inode = file->f_path.dentry->d_inode;
1910 unsigned int reclen;
1912 cdp->err = nfserr_eof; /* will be cleared on successful read */
1916 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1929 * Various filldir functions may end up calling back into
1930 * lookup_one_len() and the file system's ->lookup() method.
1931 * These expect i_mutex to be held, as it would within readdir.
1933 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1937 de = (struct buffered_dirent *)buf.dirent;
1939 offset = de->offset;
1941 if (func(cdp, de->name, de->namlen, de->offset,
1942 de->ino, de->d_type))
1945 if (cdp->err != nfs_ok)
1948 reclen = ALIGN(sizeof(*de) + de->namlen,
1951 de = (struct buffered_dirent *)((char *)de + reclen);
1953 mutex_unlock(&dir_inode->i_mutex);
1954 if (size > 0) /* We bailed out early */
1957 offset = vfs_llseek(file, 0, SEEK_CUR);
1960 free_page((unsigned long)(buf.dirent));
1963 return nfserrno(host_err);
1970 * Read entries from a directory.
1971 * The NFSv3/4 verifier we ignore for now.
1974 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1975 struct readdir_cd *cdp, filldir_t func)
1979 loff_t offset = *offsetp;
1981 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1985 offset = vfs_llseek(file, offset, 0);
1987 err = nfserrno((int)offset);
1991 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1993 if (err == nfserr_eof || err == nfserr_toosmall)
1994 err = nfs_ok; /* can still be found in ->err */
2002 * Get file system stats
2003 * N.B. After this call fhp needs an fh_put
2006 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2010 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2012 struct path path = {
2013 .mnt = fhp->fh_export->ex_path.mnt,
2014 .dentry = fhp->fh_dentry,
2016 if (vfs_statfs(&path, stat))
2022 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2024 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2028 * Check for a user's access permissions to this inode.
2031 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2032 struct dentry *dentry, int acc)
2034 struct inode *inode = dentry->d_inode;
2037 if (acc == NFSD_MAY_NOP)
2040 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2042 (acc & NFSD_MAY_READ)? " read" : "",
2043 (acc & NFSD_MAY_WRITE)? " write" : "",
2044 (acc & NFSD_MAY_EXEC)? " exec" : "",
2045 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2046 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2047 (acc & NFSD_MAY_LOCK)? " lock" : "",
2048 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2050 IS_IMMUTABLE(inode)? " immut" : "",
2051 IS_APPEND(inode)? " append" : "",
2052 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2053 dprintk(" owner %d/%d user %d/%d\n",
2054 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2057 /* Normally we reject any write/sattr etc access on a read-only file
2058 * system. But if it is IRIX doing check on write-access for a
2059 * device special file, we ignore rofs.
2061 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2062 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2063 if (exp_rdonly(rqstp, exp) ||
2064 __mnt_is_readonly(exp->ex_path.mnt))
2066 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2069 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2072 if (acc & NFSD_MAY_LOCK) {
2073 /* If we cannot rely on authentication in NLM requests,
2074 * just allow locks, otherwise require read permission, or
2077 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2080 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2083 * The file owner always gets access permission for accesses that
2084 * would normally be checked at open time. This is to make
2085 * file access work even when the client has done a fchmod(fd, 0).
2087 * However, `cp foo bar' should fail nevertheless when bar is
2088 * readonly. A sensible way to do this might be to reject all
2089 * attempts to truncate a read-only file, because a creat() call
2090 * always implies file truncation.
2091 * ... but this isn't really fair. A process may reasonably call
2092 * ftruncate on an open file descriptor on a file with perm 000.
2093 * We must trust the client to do permission checking - using "ACCESS"
2096 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2097 inode->i_uid == current_fsuid())
2100 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2101 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2103 /* Allow read access to binaries even when mode 111 */
2104 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2105 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2106 err = inode_permission(inode, MAY_EXEC);
2108 return err? nfserrno(err) : 0;
2112 nfsd_racache_shutdown(void)
2114 struct raparms *raparm, *last_raparm;
2117 dprintk("nfsd: freeing readahead buffers.\n");
2119 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2120 raparm = raparm_hash[i].pb_head;
2122 last_raparm = raparm;
2123 raparm = raparm->p_next;
2126 raparm_hash[i].pb_head = NULL;
2130 * Initialize readahead param cache
2133 nfsd_racache_init(int cache_size)
2138 struct raparms **raparm = NULL;
2141 if (raparm_hash[0].pb_head)
2143 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2146 cache_size = nperbucket * RAPARM_HASH_SIZE;
2148 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2150 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2151 spin_lock_init(&raparm_hash[i].pb_lock);
2153 raparm = &raparm_hash[i].pb_head;
2154 for (j = 0; j < nperbucket; j++) {
2155 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2158 raparm = &(*raparm)->p_next;
2163 nfsdstats.ra_size = cache_size;
2167 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2168 nfsd_racache_shutdown();
2172 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2174 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2176 struct inode *inode = fhp->fh_dentry->d_inode;
2180 struct posix_acl *acl;
2182 if (!IS_POSIXACL(inode))
2183 return ERR_PTR(-EOPNOTSUPP);
2186 case ACL_TYPE_ACCESS:
2187 name = POSIX_ACL_XATTR_ACCESS;
2189 case ACL_TYPE_DEFAULT:
2190 name = POSIX_ACL_XATTR_DEFAULT;
2193 return ERR_PTR(-EOPNOTSUPP);
2196 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2198 return ERR_PTR(size);
2200 acl = posix_acl_from_xattr(value, size);
2206 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2208 struct inode *inode = fhp->fh_dentry->d_inode;
2214 if (!IS_POSIXACL(inode) ||
2215 !inode->i_op->setxattr || !inode->i_op->removexattr)
2218 case ACL_TYPE_ACCESS:
2219 name = POSIX_ACL_XATTR_ACCESS;
2221 case ACL_TYPE_DEFAULT:
2222 name = POSIX_ACL_XATTR_DEFAULT;
2228 if (acl && acl->a_count) {
2229 size = posix_acl_xattr_size(acl->a_count);
2230 value = kmalloc(size, GFP_KERNEL);
2233 error = posix_acl_to_xattr(acl, value, size);
2240 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2244 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2246 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2249 error = vfs_removexattr(fhp->fh_dentry, name);
2250 if (error == -ENODATA)
2254 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2260 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */