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.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
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 (nfsd4_is_junction(dentry))
173 if (!(exp->ex_flags & NFSEXP_V4ROOT))
175 return dentry->d_inode != NULL;
179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
180 const char *name, unsigned int len,
181 struct svc_export **exp_ret, struct dentry **dentry_ret)
183 struct svc_export *exp;
184 struct dentry *dparent;
185 struct dentry *dentry;
188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
190 dparent = fhp->fh_dentry;
191 exp = fhp->fh_export;
194 /* Lookup the name, but don't follow links */
195 if (isdotent(name, len)) {
197 dentry = dget(dparent);
198 else if (dparent != exp->ex_path.dentry)
199 dentry = dget_parent(dparent);
200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
201 dentry = dget(dparent); /* .. == . just like at / */
203 /* checking mountpoint crossing is very different when stepping up */
204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
210 dentry = lookup_one_len(name, dparent, len);
211 host_err = PTR_ERR(dentry);
215 * check if we have crossed a mount point ...
217 if (nfsd_mountpoint(dentry, exp)) {
218 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
224 *dentry_ret = dentry;
230 return nfserrno(host_err);
234 * Look up one component of a pathname.
235 * N.B. After this call _both_ fhp and resfh need an fh_put
237 * If the lookup would cross a mountpoint, and the mounted filesystem
238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
239 * accepted as it stands and the mounted directory is
240 * returned. Otherwise the covered directory is returned.
241 * NOTE: this mountpoint crossing is not supported properly by all
242 * clients and is explicitly disallowed for NFSv3
243 * NeilBrown <neilb@cse.unsw.edu.au>
246 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
247 unsigned int len, struct svc_fh *resfh)
249 struct svc_export *exp;
250 struct dentry *dentry;
253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
259 err = check_nfsd_access(exp, rqstp);
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
275 static int nfsd_break_lease(struct inode *inode)
277 if (!S_ISREG(inode->i_mode))
279 return break_lease(inode, O_WRONLY | O_NONBLOCK);
283 * Commit metadata changes to stable storage.
286 commit_metadata(struct svc_fh *fhp)
288 struct inode *inode = fhp->fh_dentry->d_inode;
289 const struct export_operations *export_ops = inode->i_sb->s_export_op;
291 if (!EX_ISSYNC(fhp->fh_export))
294 if (export_ops->commit_metadata)
295 return export_ops->commit_metadata(inode);
296 return sync_inode_metadata(inode, 1);
300 * Go over the attributes and take care of the small differences between
301 * NFS semantics and what Linux expects.
304 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
307 * NFSv2 does not differentiate between "set-[ac]time-to-now"
308 * which only requires access, and "set-[ac]time-to-X" which
309 * requires ownership.
310 * So if it looks like it might be "set both to the same time which
311 * is close to now", and if inode_change_ok fails, then we
312 * convert to "set to now" instead of "set to explicit time"
314 * We only call inode_change_ok as the last test as technically
315 * it is not an interface that we should be using.
317 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
318 #define MAX_TOUCH_TIME_ERROR (30*60)
319 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
320 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
324 * Now just make sure time is in the right ballpark.
325 * Solaris, at least, doesn't seem to care what the time
326 * request is. We require it be within 30 minutes of now.
328 time_t delta = iap->ia_atime.tv_sec - get_seconds();
331 if (delta < MAX_TOUCH_TIME_ERROR &&
332 inode_change_ok(inode, iap) != 0) {
334 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
335 * This will cause notify_change to set these times
338 iap->ia_valid &= ~BOTH_TIME_SET;
342 /* sanitize the mode change */
343 if (iap->ia_valid & ATTR_MODE) {
344 iap->ia_mode &= S_IALLUGO;
345 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
348 /* Revoke setuid/setgid on chown */
349 if (!S_ISDIR(inode->i_mode) &&
350 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
351 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
352 iap->ia_valid |= ATTR_KILL_PRIV;
353 if (iap->ia_valid & ATTR_MODE) {
354 /* we're setting mode too, just clear the s*id bits */
355 iap->ia_mode &= ~S_ISUID;
356 if (iap->ia_mode & S_IXGRP)
357 iap->ia_mode &= ~S_ISGID;
359 /* set ATTR_KILL_* bits and let VFS handle it */
360 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
366 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
369 struct inode *inode = fhp->fh_dentry->d_inode;
372 if (iap->ia_size < inode->i_size) {
375 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
376 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
381 host_err = get_write_access(inode);
385 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
387 goto out_put_write_access;
390 out_put_write_access:
391 put_write_access(inode);
393 return nfserrno(host_err);
397 * Set various file attributes. After this call fhp needs an fh_put.
400 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
401 int check_guard, time_t guardtime)
403 struct dentry *dentry;
405 int accmode = NFSD_MAY_SATTR;
411 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
412 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
413 if (iap->ia_valid & ATTR_SIZE)
417 err = fh_verify(rqstp, fhp, ftype, accmode);
421 dentry = fhp->fh_dentry;
422 inode = dentry->d_inode;
424 /* Ignore any mode updates on symlinks */
425 if (S_ISLNK(inode->i_mode))
426 iap->ia_valid &= ~ATTR_MODE;
431 nfsd_sanitize_attrs(inode, iap);
434 * The size case is special, it changes the file in addition to the
437 if (iap->ia_valid & ATTR_SIZE) {
438 err = nfsd_get_write_access(rqstp, fhp, iap);
444 iap->ia_valid |= ATTR_CTIME;
446 err = nfserr_notsync;
447 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
448 host_err = nfsd_break_lease(inode);
453 host_err = notify_change(dentry, iap);
454 err = nfserrno(host_err);
458 put_write_access(inode);
460 commit_metadata(fhp);
465 err = nfserrno(host_err);
469 #if defined(CONFIG_NFSD_V2_ACL) || \
470 defined(CONFIG_NFSD_V3_ACL) || \
471 defined(CONFIG_NFSD_V4)
472 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
477 buflen = vfs_getxattr(dentry, key, NULL, 0);
481 *buf = kmalloc(buflen, GFP_KERNEL);
485 ret = vfs_getxattr(dentry, key, *buf, buflen);
492 #if defined(CONFIG_NFSD_V4)
494 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
501 buflen = posix_acl_xattr_size(pacl->a_count);
502 buf = kmalloc(buflen, GFP_KERNEL);
507 len = posix_acl_to_xattr(pacl, buf, buflen);
513 error = vfs_setxattr(dentry, key, buf, len, 0);
520 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
521 struct nfs4_acl *acl)
525 struct dentry *dentry;
527 struct posix_acl *pacl = NULL, *dpacl = NULL;
528 unsigned int flags = 0;
531 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
535 dentry = fhp->fh_dentry;
536 inode = dentry->d_inode;
537 if (S_ISDIR(inode->i_mode))
538 flags = NFS4_ACL_DIR;
540 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
541 if (host_error == -EINVAL) {
542 return nfserr_attrnotsupp;
543 } else if (host_error < 0)
546 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
550 if (S_ISDIR(inode->i_mode))
551 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
554 posix_acl_release(pacl);
555 posix_acl_release(dpacl);
557 if (host_error == -EOPNOTSUPP)
558 return nfserr_attrnotsupp;
560 return nfserrno(host_error);
563 static struct posix_acl *
564 _get_posix_acl(struct dentry *dentry, char *key)
567 struct posix_acl *pacl = NULL;
570 buflen = nfsd_getxattr(dentry, key, &buf);
574 return ERR_PTR(buflen);
576 pacl = posix_acl_from_xattr(buf, buflen);
582 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
584 struct inode *inode = dentry->d_inode;
586 struct posix_acl *pacl = NULL, *dpacl = NULL;
587 unsigned int flags = 0;
589 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
590 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
591 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
593 error = PTR_ERR(pacl);
598 if (S_ISDIR(inode->i_mode)) {
599 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
600 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
602 else if (IS_ERR(dpacl)) {
603 error = PTR_ERR(dpacl);
607 flags = NFS4_ACL_DIR;
610 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
612 error = PTR_ERR(*acl);
616 posix_acl_release(pacl);
617 posix_acl_release(dpacl);
621 #define NFSD_XATTR_JUNCTION_PREFIX XATTR_TRUSTED_PREFIX "junction."
622 #define NFSD_XATTR_JUNCTION_TYPE NFSD_XATTR_JUNCTION_PREFIX "type"
623 int nfsd4_is_junction(struct dentry *dentry)
625 struct inode *inode = dentry->d_inode;
629 if (inode->i_mode & S_IXUGO)
631 if (!(inode->i_mode & S_ISVTX))
633 if (vfs_getxattr(dentry, NFSD_XATTR_JUNCTION_TYPE, NULL, 0) <= 0)
637 #endif /* defined(CONFIG_NFSD_V4) */
639 #ifdef CONFIG_NFSD_V3
641 * Check server access rights to a file system object
647 static struct accessmap nfs3_regaccess[] = {
648 { NFS3_ACCESS_READ, NFSD_MAY_READ },
649 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
650 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
651 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
656 static struct accessmap nfs3_diraccess[] = {
657 { NFS3_ACCESS_READ, NFSD_MAY_READ },
658 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
659 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
660 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
661 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
666 static struct accessmap nfs3_anyaccess[] = {
667 /* Some clients - Solaris 2.6 at least, make an access call
668 * to the server to check for access for things like /dev/null
669 * (which really, the server doesn't care about). So
670 * We provide simple access checking for them, looking
671 * mainly at mode bits, and we make sure to ignore read-only
674 { NFS3_ACCESS_READ, NFSD_MAY_READ },
675 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
676 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
677 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
683 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
685 struct accessmap *map;
686 struct svc_export *export;
687 struct dentry *dentry;
688 u32 query, result = 0, sresult = 0;
691 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
695 export = fhp->fh_export;
696 dentry = fhp->fh_dentry;
698 if (S_ISREG(dentry->d_inode->i_mode))
699 map = nfs3_regaccess;
700 else if (S_ISDIR(dentry->d_inode->i_mode))
701 map = nfs3_diraccess;
703 map = nfs3_anyaccess;
707 for (; map->access; map++) {
708 if (map->access & query) {
711 sresult |= map->access;
713 err2 = nfsd_permission(rqstp, export, dentry, map->how);
716 result |= map->access;
719 /* the following error codes just mean the access was not allowed,
720 * rather than an error occurred */
724 /* simply don't "or" in the access bit. */
734 *supported = sresult;
739 #endif /* CONFIG_NFSD_V3 */
741 static int nfsd_open_break_lease(struct inode *inode, int access)
745 if (access & NFSD_MAY_NOT_BREAK_LEASE)
747 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
748 return break_lease(inode, mode | O_NONBLOCK);
752 * Open an existing file or directory.
753 * The may_flags argument indicates the type of open (read/write/lock)
754 * and additional flags.
755 * N.B. After this call fhp needs an fh_put
758 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
759 int may_flags, struct file **filp)
761 struct dentry *dentry;
763 int flags = O_RDONLY|O_LARGEFILE;
767 validate_process_creds();
770 * If we get here, then the client has already done an "open",
771 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
772 * in case a chmod has now revoked permission.
774 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE);
778 dentry = fhp->fh_dentry;
779 inode = dentry->d_inode;
781 /* Disallow write access to files with the append-only bit set
782 * or any access when mandatory locking enabled
785 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
788 * We must ignore files (but only files) which might have mandatory
789 * locks on them because there is no way to know if the accesser has
792 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
798 host_err = nfsd_open_break_lease(inode, may_flags);
799 if (host_err) /* NOMEM or WOULDBLOCK */
802 if (may_flags & NFSD_MAY_WRITE) {
803 if (may_flags & NFSD_MAY_READ)
804 flags = O_RDWR|O_LARGEFILE;
806 flags = O_WRONLY|O_LARGEFILE;
808 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
809 flags, current_cred());
811 host_err = PTR_ERR(*filp);
814 host_err = ima_file_check(*filp, may_flags);
816 if (may_flags & NFSD_MAY_64BIT_COOKIE)
817 (*filp)->f_mode |= FMODE_64BITHASH;
819 (*filp)->f_mode |= FMODE_32BITHASH;
823 err = nfserrno(host_err);
825 validate_process_creds();
833 nfsd_close(struct file *filp)
839 * Obtain the readahead parameters for the file
840 * specified by (dev, ino).
843 static inline struct raparms *
844 nfsd_get_raparms(dev_t dev, ino_t ino)
846 struct raparms *ra, **rap, **frap = NULL;
849 struct raparm_hbucket *rab;
851 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
852 rab = &raparm_hash[hash];
854 spin_lock(&rab->pb_lock);
855 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
856 if (ra->p_ino == ino && ra->p_dev == dev)
859 if (ra->p_count == 0)
862 depth = nfsdstats.ra_size;
864 spin_unlock(&rab->pb_lock);
874 if (rap != &rab->pb_head) {
876 ra->p_next = rab->pb_head;
880 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
881 spin_unlock(&rab->pb_lock);
886 * Grab and keep cached pages associated with a file in the svc_rqst
887 * so that they can be passed to the network sendmsg/sendpage routines
888 * directly. They will be released after the sending has completed.
891 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
892 struct splice_desc *sd)
894 struct svc_rqst *rqstp = sd->u.data;
895 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
896 struct page *page = buf->page;
901 if (rqstp->rq_res.page_len == 0) {
906 rqstp->rq_res.page_base = buf->offset;
907 rqstp->rq_res.page_len = size;
908 } else if (page != pp[-1]) {
914 rqstp->rq_res.page_len += size;
916 rqstp->rq_res.page_len += size;
921 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
922 struct splice_desc *sd)
924 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
928 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
929 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
937 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
938 struct splice_desc sd = {
945 rqstp->rq_resused = 1;
946 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
950 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
955 nfsdstats.io_read += host_err;
958 fsnotify_access(file);
960 err = nfserrno(host_err);
964 static void kill_suid(struct dentry *dentry)
967 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
969 mutex_lock(&dentry->d_inode->i_mutex);
970 notify_change(dentry, &ia);
971 mutex_unlock(&dentry->d_inode->i_mutex);
975 * Gathered writes: If another process is currently writing to the file,
976 * there's a high chance this is another nfsd (triggered by a bulk write
977 * from a client's biod). Rather than syncing the file with each write
978 * request, we sleep for 10 msec.
980 * I don't know if this roughly approximates C. Juszak's idea of
981 * gathered writes, but it's a nice and simple solution (IMHO), and it
984 * Note: we do this only in the NFSv2 case, since v3 and higher have a
985 * better tool (separate unstable writes and commits) for solving this
988 static int wait_for_concurrent_writes(struct file *file)
990 struct inode *inode = file->f_path.dentry->d_inode;
991 static ino_t last_ino;
992 static dev_t last_dev;
995 if (atomic_read(&inode->i_writecount) > 1
996 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
997 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
999 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1002 if (inode->i_state & I_DIRTY) {
1003 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1004 err = vfs_fsync(file, 0);
1006 last_ino = inode->i_ino;
1007 last_dev = inode->i_sb->s_dev;
1012 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1013 loff_t offset, struct kvec *vec, int vlen,
1014 unsigned long *cnt, int *stablep)
1016 struct svc_export *exp;
1017 struct dentry *dentry;
1018 struct inode *inode;
1022 int stable = *stablep;
1025 dentry = file->f_path.dentry;
1026 inode = dentry->d_inode;
1027 exp = fhp->fh_export;
1030 * Request sync writes if
1031 * - the sync export option has been set, or
1032 * - the client requested O_SYNC behavior (NFSv3 feature).
1033 * - The file system doesn't support fsync().
1034 * When NFSv2 gathered writes have been configured for this volume,
1035 * flushing the data to disk is handled separately below.
1037 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1039 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1041 *stablep = 2; /* FILE_SYNC */
1044 if (!EX_ISSYNC(exp))
1046 if (stable && !use_wgather) {
1047 spin_lock(&file->f_lock);
1048 file->f_flags |= O_SYNC;
1049 spin_unlock(&file->f_lock);
1052 /* Write the data. */
1053 oldfs = get_fs(); set_fs(KERNEL_DS);
1054 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1059 nfsdstats.io_write += host_err;
1060 fsnotify_modify(file);
1062 /* clear setuid/setgid flag after write */
1063 if (inode->i_mode & (S_ISUID | S_ISGID))
1066 if (stable && use_wgather)
1067 host_err = wait_for_concurrent_writes(file);
1070 dprintk("nfsd: write complete host_err=%d\n", host_err);
1074 err = nfserrno(host_err);
1079 * Read data from a file. count must contain the requested read count
1080 * on entry. On return, *count contains the number of bytes actually read.
1081 * N.B. After this call fhp needs an fh_put
1083 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1084 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1087 struct inode *inode;
1091 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1095 inode = file->f_path.dentry->d_inode;
1097 /* Get readahead parameters */
1098 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1100 if (ra && ra->p_set)
1101 file->f_ra = ra->p_ra;
1103 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1105 /* Write back readahead params */
1107 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1108 spin_lock(&rab->pb_lock);
1109 ra->p_ra = file->f_ra;
1112 spin_unlock(&rab->pb_lock);
1119 /* As above, but use the provided file descriptor. */
1121 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1122 loff_t offset, struct kvec *vec, int vlen,
1123 unsigned long *count)
1128 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1129 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1132 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1133 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1134 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1140 * Write data to a file.
1141 * The stable flag requests synchronous writes.
1142 * N.B. After this call fhp needs an fh_put
1145 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1146 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1152 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1153 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1156 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1159 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1164 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1172 #ifdef CONFIG_NFSD_V3
1174 * Commit all pending writes to stable storage.
1176 * Note: we only guarantee that data that lies within the range specified
1177 * by the 'offset' and 'count' parameters will be synced.
1179 * Unfortunately we cannot lock the file to make sure we return full WCC
1180 * data to the client, as locking happens lower down in the filesystem.
1183 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1184 loff_t offset, unsigned long count)
1187 loff_t end = LLONG_MAX;
1188 __be32 err = nfserr_inval;
1193 end = offset + (loff_t)count - 1;
1198 err = nfsd_open(rqstp, fhp, S_IFREG,
1199 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1202 if (EX_ISSYNC(fhp->fh_export)) {
1203 int err2 = vfs_fsync_range(file, offset, end, 0);
1205 if (err2 != -EINVAL)
1206 err = nfserrno(err2);
1208 err = nfserr_notsupp;
1215 #endif /* CONFIG_NFSD_V3 */
1218 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1222 * Mode has already been set earlier in create:
1224 iap->ia_valid &= ~ATTR_MODE;
1226 * Setting uid/gid works only for root. Irix appears to
1227 * send along the gid on create when it tries to implement
1228 * setgid directories via NFS:
1230 if (current_fsuid() != 0)
1231 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1233 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1237 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1238 * setting size to 0 may fail for some specific file systems by the permission
1239 * checking which requires WRITE permission but the mode is 000.
1240 * we ignore the resizing(to 0) on the just new created file, since the size is
1241 * 0 after file created.
1243 * call this only after vfs_create() is called.
1246 nfsd_check_ignore_resizing(struct iattr *iap)
1248 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1249 iap->ia_valid &= ~ATTR_SIZE;
1253 * Create a file (regular, directory, device, fifo); UNIX sockets
1254 * not yet implemented.
1255 * If the response fh has been verified, the parent directory should
1256 * already be locked. Note that the parent directory is left locked.
1258 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1261 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1262 char *fname, int flen, struct iattr *iap,
1263 int type, dev_t rdev, struct svc_fh *resfhp)
1265 struct dentry *dentry, *dchild = NULL;
1275 if (isdotent(fname, flen))
1278 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1282 dentry = fhp->fh_dentry;
1283 dirp = dentry->d_inode;
1285 err = nfserr_notdir;
1286 if (!dirp->i_op->lookup)
1289 * Check whether the response file handle has been verified yet.
1290 * If it has, the parent directory should already be locked.
1292 if (!resfhp->fh_dentry) {
1293 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1294 fh_lock_nested(fhp, I_MUTEX_PARENT);
1295 dchild = lookup_one_len(fname, dentry, flen);
1296 host_err = PTR_ERR(dchild);
1299 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1303 /* called from nfsd_proc_create */
1304 dchild = dget(resfhp->fh_dentry);
1305 if (!fhp->fh_locked) {
1306 /* not actually possible */
1308 "nfsd_create: parent %s/%s not locked!\n",
1309 dentry->d_parent->d_name.name,
1310 dentry->d_name.name);
1316 * Make sure the child dentry is still negative ...
1319 if (dchild->d_inode) {
1320 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1321 dentry->d_name.name, dchild->d_name.name);
1325 if (!(iap->ia_valid & ATTR_MODE))
1327 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1330 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1331 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1336 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1341 * Get the dir op function pointer.
1346 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1348 nfsd_check_ignore_resizing(iap);
1351 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1357 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1361 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1365 err = nfsd_create_setattr(rqstp, resfhp, iap);
1368 * nfsd_setattr already committed the child. Transactional filesystems
1369 * had a chance to commit changes for both parent and child
1370 * simultaneously making the following commit_metadata a noop.
1372 err2 = nfserrno(commit_metadata(fhp));
1375 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1377 * Update the file handle to get the new inode info.
1380 err = fh_update(resfhp);
1382 if (dchild && !IS_ERR(dchild))
1387 err = nfserrno(host_err);
1391 #ifdef CONFIG_NFSD_V3
1393 static inline int nfsd_create_is_exclusive(int createmode)
1395 return createmode == NFS3_CREATE_EXCLUSIVE
1396 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1400 * NFSv3 and NFSv4 version of nfsd_create
1403 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1404 char *fname, int flen, struct iattr *iap,
1405 struct svc_fh *resfhp, int createmode, u32 *verifier,
1406 bool *truncp, bool *created)
1408 struct dentry *dentry, *dchild = NULL;
1412 __u32 v_mtime=0, v_atime=0;
1418 if (isdotent(fname, flen))
1420 if (!(iap->ia_valid & ATTR_MODE))
1422 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1426 dentry = fhp->fh_dentry;
1427 dirp = dentry->d_inode;
1429 /* Get all the sanity checks out of the way before
1430 * we lock the parent. */
1431 err = nfserr_notdir;
1432 if (!dirp->i_op->lookup)
1434 fh_lock_nested(fhp, I_MUTEX_PARENT);
1437 * Compose the response file handle.
1439 dchild = lookup_one_len(fname, dentry, flen);
1440 host_err = PTR_ERR(dchild);
1444 /* If file doesn't exist, check for permissions to create one */
1445 if (!dchild->d_inode) {
1446 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1451 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1455 if (nfsd_create_is_exclusive(createmode)) {
1456 /* solaris7 gets confused (bugid 4218508) if these have
1457 * the high bit set, so just clear the high bits. If this is
1458 * ever changed to use different attrs for storing the
1459 * verifier, then do_open_lookup() will also need to be fixed
1462 v_mtime = verifier[0]&0x7fffffff;
1463 v_atime = verifier[1]&0x7fffffff;
1466 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1469 if (dchild->d_inode) {
1472 switch (createmode) {
1473 case NFS3_CREATE_UNCHECKED:
1474 if (! S_ISREG(dchild->d_inode->i_mode))
1477 /* in nfsv4, we need to treat this case a little
1478 * differently. we don't want to truncate the
1479 * file now; this would be wrong if the OPEN
1480 * fails for some other reason. furthermore,
1481 * if the size is nonzero, we should ignore it
1482 * according to spec!
1484 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1487 iap->ia_valid &= ATTR_SIZE;
1491 case NFS3_CREATE_EXCLUSIVE:
1492 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1493 && dchild->d_inode->i_atime.tv_sec == v_atime
1494 && dchild->d_inode->i_size == 0 ) {
1499 case NFS4_CREATE_EXCLUSIVE4_1:
1500 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1501 && dchild->d_inode->i_atime.tv_sec == v_atime
1502 && dchild->d_inode->i_size == 0 ) {
1508 case NFS3_CREATE_GUARDED:
1511 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1515 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1517 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1523 nfsd_check_ignore_resizing(iap);
1525 if (nfsd_create_is_exclusive(createmode)) {
1526 /* Cram the verifier into atime/mtime */
1527 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1528 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1529 /* XXX someone who knows this better please fix it for nsec */
1530 iap->ia_mtime.tv_sec = v_mtime;
1531 iap->ia_atime.tv_sec = v_atime;
1532 iap->ia_mtime.tv_nsec = 0;
1533 iap->ia_atime.tv_nsec = 0;
1537 err = nfsd_create_setattr(rqstp, resfhp, iap);
1540 * nfsd_setattr already committed the child (and possibly also the parent).
1543 err = nfserrno(commit_metadata(fhp));
1545 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1547 * Update the filehandle to get the new inode info.
1550 err = fh_update(resfhp);
1554 if (dchild && !IS_ERR(dchild))
1559 err = nfserrno(host_err);
1562 #endif /* CONFIG_NFSD_V3 */
1565 * Read a symlink. On entry, *lenp must contain the maximum path length that
1566 * fits into the buffer. On return, it contains the true length.
1567 * N.B. After this call fhp needs an fh_put
1570 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1572 struct dentry *dentry;
1573 struct inode *inode;
1578 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1582 dentry = fhp->fh_dentry;
1583 inode = dentry->d_inode;
1586 if (!inode->i_op->readlink)
1589 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1590 /* N.B. Why does this call need a get_fs()??
1591 * Remove the set_fs and watch the fireworks:-) --okir
1594 oldfs = get_fs(); set_fs(KERNEL_DS);
1595 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1606 err = nfserrno(host_err);
1611 * Create a symlink and look up its inode
1612 * N.B. After this call _both_ fhp and resfhp need an fh_put
1615 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1616 char *fname, int flen,
1617 char *path, int plen,
1618 struct svc_fh *resfhp,
1621 struct dentry *dentry, *dnew;
1629 if (isdotent(fname, flen))
1632 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1636 dentry = fhp->fh_dentry;
1637 dnew = lookup_one_len(fname, dentry, flen);
1638 host_err = PTR_ERR(dnew);
1642 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1646 if (unlikely(path[plen] != 0)) {
1647 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1648 if (path_alloced == NULL)
1651 strncpy(path_alloced, path, plen);
1652 path_alloced[plen] = 0;
1653 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1654 kfree(path_alloced);
1657 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1658 err = nfserrno(host_err);
1660 err = nfserrno(commit_metadata(fhp));
1663 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1665 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1667 if (err==0) err = cerr;
1672 err = nfserrno(host_err);
1678 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1681 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1682 char *name, int len, struct svc_fh *tfhp)
1684 struct dentry *ddir, *dnew, *dold;
1689 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1692 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1696 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1702 if (isdotent(name, len))
1705 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1706 ddir = ffhp->fh_dentry;
1707 dirp = ddir->d_inode;
1709 dnew = lookup_one_len(name, ddir, len);
1710 host_err = PTR_ERR(dnew);
1714 dold = tfhp->fh_dentry;
1716 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1718 err = nfserrno(host_err);
1723 goto out_drop_write;
1724 host_err = nfsd_break_lease(dold->d_inode);
1726 err = nfserrno(host_err);
1727 goto out_drop_write;
1729 host_err = vfs_link(dold, dirp, dnew);
1731 err = nfserrno(commit_metadata(ffhp));
1733 err = nfserrno(commit_metadata(tfhp));
1735 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1738 err = nfserrno(host_err);
1741 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1750 err = nfserrno(host_err);
1756 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1759 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1760 struct svc_fh *tfhp, char *tname, int tlen)
1762 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1763 struct inode *fdir, *tdir;
1767 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1770 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1774 fdentry = ffhp->fh_dentry;
1775 fdir = fdentry->d_inode;
1777 tdentry = tfhp->fh_dentry;
1778 tdir = tdentry->d_inode;
1780 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1781 if (ffhp->fh_export != tfhp->fh_export)
1785 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1788 /* cannot use fh_lock as we need deadlock protective ordering
1789 * so do it by hand */
1790 trap = lock_rename(tdentry, fdentry);
1791 ffhp->fh_locked = tfhp->fh_locked = 1;
1795 odentry = lookup_one_len(fname, fdentry, flen);
1796 host_err = PTR_ERR(odentry);
1797 if (IS_ERR(odentry))
1801 if (!odentry->d_inode)
1804 if (odentry == trap)
1807 ndentry = lookup_one_len(tname, tdentry, tlen);
1808 host_err = PTR_ERR(ndentry);
1809 if (IS_ERR(ndentry))
1811 host_err = -ENOTEMPTY;
1812 if (ndentry == trap)
1816 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1818 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1822 host_err = nfsd_break_lease(odentry->d_inode);
1824 goto out_drop_write;
1825 if (ndentry->d_inode) {
1826 host_err = nfsd_break_lease(ndentry->d_inode);
1828 goto out_drop_write;
1830 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1832 host_err = commit_metadata(tfhp);
1834 host_err = commit_metadata(ffhp);
1837 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1843 err = nfserrno(host_err);
1845 /* we cannot reply on fh_unlock on the two filehandles,
1846 * as that would do the wrong thing if the two directories
1847 * were the same, so again we do it by hand
1849 fill_post_wcc(ffhp);
1850 fill_post_wcc(tfhp);
1851 unlock_rename(tdentry, fdentry);
1852 ffhp->fh_locked = tfhp->fh_locked = 0;
1859 * Unlink a file or directory
1860 * N.B. After this call fhp needs an fh_put
1863 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1864 char *fname, int flen)
1866 struct dentry *dentry, *rdentry;
1872 if (!flen || isdotent(fname, flen))
1874 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1878 fh_lock_nested(fhp, I_MUTEX_PARENT);
1879 dentry = fhp->fh_dentry;
1880 dirp = dentry->d_inode;
1882 rdentry = lookup_one_len(fname, dentry, flen);
1883 host_err = PTR_ERR(rdentry);
1884 if (IS_ERR(rdentry))
1887 if (!rdentry->d_inode) {
1894 type = rdentry->d_inode->i_mode & S_IFMT;
1896 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1900 host_err = nfsd_break_lease(rdentry->d_inode);
1902 goto out_drop_write;
1903 if (type != S_IFDIR)
1904 host_err = vfs_unlink(dirp, rdentry);
1906 host_err = vfs_rmdir(dirp, rdentry);
1908 host_err = commit_metadata(fhp);
1910 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1915 err = nfserrno(host_err);
1921 * We do this buffering because we must not call back into the file
1922 * system's ->lookup() method from the filldir callback. That may well
1923 * deadlock a number of file systems.
1925 * This is based heavily on the implementation of same in XFS.
1927 struct buffered_dirent {
1931 unsigned int d_type;
1935 struct readdir_data {
1941 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1942 loff_t offset, u64 ino, unsigned int d_type)
1944 struct readdir_data *buf = __buf;
1945 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1946 unsigned int reclen;
1948 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1949 if (buf->used + reclen > PAGE_SIZE) {
1954 de->namlen = namlen;
1955 de->offset = offset;
1957 de->d_type = d_type;
1958 memcpy(de->name, name, namlen);
1959 buf->used += reclen;
1964 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1965 struct readdir_cd *cdp, loff_t *offsetp)
1967 struct readdir_data buf;
1968 struct buffered_dirent *de;
1973 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1975 return nfserrno(-ENOMEM);
1980 struct inode *dir_inode = file->f_path.dentry->d_inode;
1981 unsigned int reclen;
1983 cdp->err = nfserr_eof; /* will be cleared on successful read */
1987 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2000 * Various filldir functions may end up calling back into
2001 * lookup_one_len() and the file system's ->lookup() method.
2002 * These expect i_mutex to be held, as it would within readdir.
2004 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2008 de = (struct buffered_dirent *)buf.dirent;
2010 offset = de->offset;
2012 if (func(cdp, de->name, de->namlen, de->offset,
2013 de->ino, de->d_type))
2016 if (cdp->err != nfs_ok)
2019 reclen = ALIGN(sizeof(*de) + de->namlen,
2022 de = (struct buffered_dirent *)((char *)de + reclen);
2024 mutex_unlock(&dir_inode->i_mutex);
2025 if (size > 0) /* We bailed out early */
2028 offset = vfs_llseek(file, 0, SEEK_CUR);
2031 free_page((unsigned long)(buf.dirent));
2034 return nfserrno(host_err);
2041 * Read entries from a directory.
2042 * The NFSv3/4 verifier we ignore for now.
2045 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2046 struct readdir_cd *cdp, filldir_t func)
2050 loff_t offset = *offsetp;
2051 int may_flags = NFSD_MAY_READ;
2053 /* NFSv2 only supports 32 bit cookies */
2054 if (rqstp->rq_vers > 2)
2055 may_flags |= NFSD_MAY_64BIT_COOKIE;
2057 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2061 offset = vfs_llseek(file, offset, 0);
2063 err = nfserrno((int)offset);
2067 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2069 if (err == nfserr_eof || err == nfserr_toosmall)
2070 err = nfs_ok; /* can still be found in ->err */
2078 * Get file system stats
2079 * N.B. After this call fhp needs an fh_put
2082 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2086 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2088 struct path path = {
2089 .mnt = fhp->fh_export->ex_path.mnt,
2090 .dentry = fhp->fh_dentry,
2092 if (vfs_statfs(&path, stat))
2098 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2100 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2104 * Check for a user's access permissions to this inode.
2107 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2108 struct dentry *dentry, int acc)
2110 struct inode *inode = dentry->d_inode;
2113 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2116 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2118 (acc & NFSD_MAY_READ)? " read" : "",
2119 (acc & NFSD_MAY_WRITE)? " write" : "",
2120 (acc & NFSD_MAY_EXEC)? " exec" : "",
2121 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2122 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2123 (acc & NFSD_MAY_LOCK)? " lock" : "",
2124 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2126 IS_IMMUTABLE(inode)? " immut" : "",
2127 IS_APPEND(inode)? " append" : "",
2128 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2129 dprintk(" owner %d/%d user %d/%d\n",
2130 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2133 /* Normally we reject any write/sattr etc access on a read-only file
2134 * system. But if it is IRIX doing check on write-access for a
2135 * device special file, we ignore rofs.
2137 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2138 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2139 if (exp_rdonly(rqstp, exp) ||
2140 __mnt_is_readonly(exp->ex_path.mnt))
2142 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2145 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2148 if (acc & NFSD_MAY_LOCK) {
2149 /* If we cannot rely on authentication in NLM requests,
2150 * just allow locks, otherwise require read permission, or
2153 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2156 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2159 * The file owner always gets access permission for accesses that
2160 * would normally be checked at open time. This is to make
2161 * file access work even when the client has done a fchmod(fd, 0).
2163 * However, `cp foo bar' should fail nevertheless when bar is
2164 * readonly. A sensible way to do this might be to reject all
2165 * attempts to truncate a read-only file, because a creat() call
2166 * always implies file truncation.
2167 * ... but this isn't really fair. A process may reasonably call
2168 * ftruncate on an open file descriptor on a file with perm 000.
2169 * We must trust the client to do permission checking - using "ACCESS"
2172 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2173 inode->i_uid == current_fsuid())
2176 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2177 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2179 /* Allow read access to binaries even when mode 111 */
2180 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2181 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2182 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2183 err = inode_permission(inode, MAY_EXEC);
2185 return err? nfserrno(err) : 0;
2189 nfsd_racache_shutdown(void)
2191 struct raparms *raparm, *last_raparm;
2194 dprintk("nfsd: freeing readahead buffers.\n");
2196 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2197 raparm = raparm_hash[i].pb_head;
2199 last_raparm = raparm;
2200 raparm = raparm->p_next;
2203 raparm_hash[i].pb_head = NULL;
2207 * Initialize readahead param cache
2210 nfsd_racache_init(int cache_size)
2215 struct raparms **raparm = NULL;
2218 if (raparm_hash[0].pb_head)
2220 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2223 cache_size = nperbucket * RAPARM_HASH_SIZE;
2225 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2227 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2228 spin_lock_init(&raparm_hash[i].pb_lock);
2230 raparm = &raparm_hash[i].pb_head;
2231 for (j = 0; j < nperbucket; j++) {
2232 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2235 raparm = &(*raparm)->p_next;
2240 nfsdstats.ra_size = cache_size;
2244 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2245 nfsd_racache_shutdown();
2249 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2251 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2253 struct inode *inode = fhp->fh_dentry->d_inode;
2257 struct posix_acl *acl;
2259 if (!IS_POSIXACL(inode))
2260 return ERR_PTR(-EOPNOTSUPP);
2263 case ACL_TYPE_ACCESS:
2264 name = POSIX_ACL_XATTR_ACCESS;
2266 case ACL_TYPE_DEFAULT:
2267 name = POSIX_ACL_XATTR_DEFAULT;
2270 return ERR_PTR(-EOPNOTSUPP);
2273 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2275 return ERR_PTR(size);
2277 acl = posix_acl_from_xattr(value, size);
2283 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2285 struct inode *inode = fhp->fh_dentry->d_inode;
2291 if (!IS_POSIXACL(inode) ||
2292 !inode->i_op->setxattr || !inode->i_op->removexattr)
2295 case ACL_TYPE_ACCESS:
2296 name = POSIX_ACL_XATTR_ACCESS;
2298 case ACL_TYPE_DEFAULT:
2299 name = POSIX_ACL_XATTR_DEFAULT;
2305 if (acl && acl->a_count) {
2306 size = posix_acl_xattr_size(acl->a_count);
2307 value = kmalloc(size, GFP_KERNEL);
2310 error = posix_acl_to_xattr(acl, value, size);
2317 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2321 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2323 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2326 error = vfs_removexattr(fhp->fh_dentry, name);
2327 if (error == -ENODATA)
2331 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2337 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */