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 dentry *dentry, struct iattr *iap)
306 struct inode *inode = dentry->d_inode;
309 * NFSv2 does not differentiate between "set-[ac]time-to-now"
310 * which only requires access, and "set-[ac]time-to-X" which
311 * requires ownership.
312 * So if it looks like it might be "set both to the same time which
313 * is close to now", and if setattr_prepare fails, then we
314 * convert to "set to now" instead of "set to explicit time"
316 * We only call setattr_prepare as the last test as technically
317 * it is not an interface that we should be using.
319 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
320 #define MAX_TOUCH_TIME_ERROR (30*60)
321 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
322 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
326 * Now just make sure time is in the right ballpark.
327 * Solaris, at least, doesn't seem to care what the time
328 * request is. We require it be within 30 minutes of now.
330 time_t delta = iap->ia_atime.tv_sec - get_seconds();
333 if (delta < MAX_TOUCH_TIME_ERROR &&
334 setattr_prepare(dentry, iap) != 0) {
336 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
337 * This will cause notify_change to set these times
340 iap->ia_valid &= ~BOTH_TIME_SET;
344 /* sanitize the mode change */
345 if (iap->ia_valid & ATTR_MODE) {
346 iap->ia_mode &= S_IALLUGO;
347 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
350 /* Revoke setuid/setgid on chown */
351 if (!S_ISDIR(inode->i_mode) &&
352 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
353 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
354 iap->ia_valid |= ATTR_KILL_PRIV;
355 if (iap->ia_valid & ATTR_MODE) {
356 /* we're setting mode too, just clear the s*id bits */
357 iap->ia_mode &= ~S_ISUID;
358 if (iap->ia_mode & S_IXGRP)
359 iap->ia_mode &= ~S_ISGID;
361 /* set ATTR_KILL_* bits and let VFS handle it */
362 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
368 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
371 struct inode *inode = fhp->fh_dentry->d_inode;
374 if (iap->ia_size < inode->i_size) {
377 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
378 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
383 host_err = get_write_access(inode);
387 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
389 goto out_put_write_access;
392 out_put_write_access:
393 put_write_access(inode);
395 return nfserrno(host_err);
399 * Set various file attributes. After this call fhp needs an fh_put.
402 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
403 int check_guard, time_t guardtime)
405 struct dentry *dentry;
407 int accmode = NFSD_MAY_SATTR;
411 bool get_write_count;
412 bool size_change = (iap->ia_valid & ATTR_SIZE);
414 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
415 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
416 if (iap->ia_valid & ATTR_SIZE)
419 /* Callers that do fh_verify should do the fh_want_write: */
420 get_write_count = !fhp->fh_dentry;
423 err = fh_verify(rqstp, fhp, ftype, accmode);
426 if (get_write_count) {
427 host_err = fh_want_write(fhp);
432 dentry = fhp->fh_dentry;
433 inode = dentry->d_inode;
435 /* Ignore any mode updates on symlinks */
436 if (S_ISLNK(inode->i_mode))
437 iap->ia_valid &= ~ATTR_MODE;
442 nfsd_sanitize_attrs(dentry, iap);
444 if (check_guard && guardtime != inode->i_ctime.tv_sec)
445 return nfserr_notsync;
448 * The size case is special, it changes the file in addition to the
452 err = nfsd_get_write_access(rqstp, fhp, iap);
457 * RFC5661, Section 18.30.4:
458 * Changing the size of a file with SETATTR indirectly
459 * changes the time_modify and change attributes.
461 * (and similar for the older RFCs)
463 if (iap->ia_size != i_size_read(inode))
464 iap->ia_valid |= ATTR_MTIME;
467 iap->ia_valid |= ATTR_CTIME;
469 host_err = nfsd_break_lease(inode);
471 goto out_put_write_access_nfserror;
474 host_err = notify_change(dentry, iap);
477 out_put_write_access_nfserror:
479 put_write_access(inode);
482 commit_metadata(fhp);
483 return nfserrno(host_err);
486 #if defined(CONFIG_NFSD_V2_ACL) || \
487 defined(CONFIG_NFSD_V3_ACL) || \
488 defined(CONFIG_NFSD_V4)
489 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
494 buflen = vfs_getxattr(dentry, key, NULL, 0);
498 *buf = kmalloc(buflen, GFP_KERNEL);
502 ret = vfs_getxattr(dentry, key, *buf, buflen);
509 #if defined(CONFIG_NFSD_V4)
511 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
519 return vfs_setxattr(dentry, key, NULL, 0, 0);
521 buflen = posix_acl_xattr_size(pacl->a_count);
522 buf = kmalloc(buflen, GFP_KERNEL);
527 len = posix_acl_to_xattr(pacl, buf, buflen);
533 error = vfs_setxattr(dentry, key, buf, len, 0);
540 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
541 struct nfs4_acl *acl)
545 struct dentry *dentry;
547 struct posix_acl *pacl = NULL, *dpacl = NULL;
548 unsigned int flags = 0;
551 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
555 dentry = fhp->fh_dentry;
556 inode = dentry->d_inode;
557 if (S_ISDIR(inode->i_mode))
558 flags = NFS4_ACL_DIR;
560 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
561 if (host_error == -EINVAL) {
562 return nfserr_attrnotsupp;
563 } else if (host_error < 0)
566 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
570 if (S_ISDIR(inode->i_mode))
571 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
574 posix_acl_release(pacl);
575 posix_acl_release(dpacl);
577 if (host_error == -EOPNOTSUPP)
578 return nfserr_attrnotsupp;
580 return nfserrno(host_error);
583 static struct posix_acl *
584 _get_posix_acl(struct dentry *dentry, char *key)
587 struct posix_acl *pacl = NULL;
590 buflen = nfsd_getxattr(dentry, key, &buf);
594 return ERR_PTR(buflen);
596 pacl = posix_acl_from_xattr(buf, buflen);
602 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
604 struct inode *inode = dentry->d_inode;
606 struct posix_acl *pacl = NULL, *dpacl = NULL;
607 unsigned int flags = 0;
609 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
610 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
611 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
613 error = PTR_ERR(pacl);
618 if (S_ISDIR(inode->i_mode)) {
619 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
620 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
622 else if (IS_ERR(dpacl)) {
623 error = PTR_ERR(dpacl);
627 flags = NFS4_ACL_DIR;
630 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
632 error = PTR_ERR(*acl);
636 posix_acl_release(pacl);
637 posix_acl_release(dpacl);
641 #define NFSD_XATTR_JUNCTION_PREFIX XATTR_TRUSTED_PREFIX "junction."
642 #define NFSD_XATTR_JUNCTION_TYPE NFSD_XATTR_JUNCTION_PREFIX "type"
643 int nfsd4_is_junction(struct dentry *dentry)
645 struct inode *inode = dentry->d_inode;
649 if (inode->i_mode & S_IXUGO)
651 if (!(inode->i_mode & S_ISVTX))
653 if (vfs_getxattr(dentry, NFSD_XATTR_JUNCTION_TYPE, NULL, 0) <= 0)
657 #endif /* defined(CONFIG_NFSD_V4) */
659 #ifdef CONFIG_NFSD_V3
661 * Check server access rights to a file system object
667 static struct accessmap nfs3_regaccess[] = {
668 { NFS3_ACCESS_READ, NFSD_MAY_READ },
669 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
670 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
671 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
676 static struct accessmap nfs3_diraccess[] = {
677 { NFS3_ACCESS_READ, NFSD_MAY_READ },
678 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
679 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
680 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
681 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
686 static struct accessmap nfs3_anyaccess[] = {
687 /* Some clients - Solaris 2.6 at least, make an access call
688 * to the server to check for access for things like /dev/null
689 * (which really, the server doesn't care about). So
690 * We provide simple access checking for them, looking
691 * mainly at mode bits, and we make sure to ignore read-only
694 { NFS3_ACCESS_READ, NFSD_MAY_READ },
695 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
696 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
697 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
703 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
705 struct accessmap *map;
706 struct svc_export *export;
707 struct dentry *dentry;
708 u32 query, result = 0, sresult = 0;
711 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
715 export = fhp->fh_export;
716 dentry = fhp->fh_dentry;
718 if (S_ISREG(dentry->d_inode->i_mode))
719 map = nfs3_regaccess;
720 else if (S_ISDIR(dentry->d_inode->i_mode))
721 map = nfs3_diraccess;
723 map = nfs3_anyaccess;
727 for (; map->access; map++) {
728 if (map->access & query) {
731 sresult |= map->access;
733 err2 = nfsd_permission(rqstp, export, dentry, map->how);
736 result |= map->access;
739 /* the following error codes just mean the access was not allowed,
740 * rather than an error occurred */
744 /* simply don't "or" in the access bit. */
754 *supported = sresult;
759 #endif /* CONFIG_NFSD_V3 */
761 static int nfsd_open_break_lease(struct inode *inode, int access)
765 if (access & NFSD_MAY_NOT_BREAK_LEASE)
767 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
768 return break_lease(inode, mode | O_NONBLOCK);
772 * Open an existing file or directory.
773 * The may_flags argument indicates the type of open (read/write/lock)
774 * and additional flags.
775 * N.B. After this call fhp needs an fh_put
778 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
779 int may_flags, struct file **filp)
781 struct dentry *dentry;
783 int flags = O_RDONLY|O_LARGEFILE;
787 validate_process_creds();
790 * If we get here, then the client has already done an "open",
791 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
792 * in case a chmod has now revoked permission.
794 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE);
798 dentry = fhp->fh_dentry;
799 inode = dentry->d_inode;
801 /* Disallow write access to files with the append-only bit set
802 * or any access when mandatory locking enabled
805 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
808 * We must ignore files (but only files) which might have mandatory
809 * locks on them because there is no way to know if the accesser has
812 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
818 host_err = nfsd_open_break_lease(inode, may_flags);
819 if (host_err) /* NOMEM or WOULDBLOCK */
822 if (may_flags & NFSD_MAY_WRITE) {
823 if (may_flags & NFSD_MAY_READ)
824 flags = O_RDWR|O_LARGEFILE;
826 flags = O_WRONLY|O_LARGEFILE;
828 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
829 flags, current_cred());
831 host_err = PTR_ERR(*filp);
834 host_err = ima_file_check(*filp, may_flags);
836 if (may_flags & NFSD_MAY_64BIT_COOKIE)
837 (*filp)->f_mode |= FMODE_64BITHASH;
839 (*filp)->f_mode |= FMODE_32BITHASH;
843 err = nfserrno(host_err);
845 validate_process_creds();
853 nfsd_close(struct file *filp)
859 * Obtain the readahead parameters for the file
860 * specified by (dev, ino).
863 static inline struct raparms *
864 nfsd_get_raparms(dev_t dev, ino_t ino)
866 struct raparms *ra, **rap, **frap = NULL;
869 struct raparm_hbucket *rab;
871 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
872 rab = &raparm_hash[hash];
874 spin_lock(&rab->pb_lock);
875 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
876 if (ra->p_ino == ino && ra->p_dev == dev)
879 if (ra->p_count == 0)
882 depth = nfsdstats.ra_size;
884 spin_unlock(&rab->pb_lock);
894 if (rap != &rab->pb_head) {
896 ra->p_next = rab->pb_head;
900 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
901 spin_unlock(&rab->pb_lock);
906 * Grab and keep cached pages associated with a file in the svc_rqst
907 * so that they can be passed to the network sendmsg/sendpage routines
908 * directly. They will be released after the sending has completed.
911 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
912 struct splice_desc *sd)
914 struct svc_rqst *rqstp = sd->u.data;
915 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
916 struct page *page = buf->page;
921 if (rqstp->rq_res.page_len == 0) {
926 rqstp->rq_res.page_base = buf->offset;
927 rqstp->rq_res.page_len = size;
928 } else if (page != pp[-1]) {
934 rqstp->rq_res.page_len += size;
936 rqstp->rq_res.page_len += size;
941 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
942 struct splice_desc *sd)
944 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
948 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
949 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
957 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
958 struct splice_desc sd = {
965 rqstp->rq_resused = 1;
966 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
970 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
975 nfsdstats.io_read += host_err;
978 fsnotify_access(file);
980 err = nfserrno(host_err);
984 static void kill_suid(struct dentry *dentry)
987 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
989 mutex_lock(&dentry->d_inode->i_mutex);
990 notify_change(dentry, &ia);
991 mutex_unlock(&dentry->d_inode->i_mutex);
995 * Gathered writes: If another process is currently writing to the file,
996 * there's a high chance this is another nfsd (triggered by a bulk write
997 * from a client's biod). Rather than syncing the file with each write
998 * request, we sleep for 10 msec.
1000 * I don't know if this roughly approximates C. Juszak's idea of
1001 * gathered writes, but it's a nice and simple solution (IMHO), and it
1004 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1005 * better tool (separate unstable writes and commits) for solving this
1008 static int wait_for_concurrent_writes(struct file *file)
1010 struct inode *inode = file->f_path.dentry->d_inode;
1011 static ino_t last_ino;
1012 static dev_t last_dev;
1015 if (atomic_read(&inode->i_writecount) > 1
1016 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1017 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1019 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1022 if (inode->i_state & I_DIRTY) {
1023 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1024 err = vfs_fsync(file, 0);
1026 last_ino = inode->i_ino;
1027 last_dev = inode->i_sb->s_dev;
1032 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1033 loff_t offset, struct kvec *vec, int vlen,
1034 unsigned long *cnt, int *stablep)
1036 struct svc_export *exp;
1037 struct dentry *dentry;
1038 struct inode *inode;
1042 int stable = *stablep;
1045 dentry = file->f_path.dentry;
1046 inode = dentry->d_inode;
1047 exp = fhp->fh_export;
1050 * Request sync writes if
1051 * - the sync export option has been set, or
1052 * - the client requested O_SYNC behavior (NFSv3 feature).
1053 * - The file system doesn't support fsync().
1054 * When NFSv2 gathered writes have been configured for this volume,
1055 * flushing the data to disk is handled separately below.
1057 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1059 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1061 *stablep = 2; /* FILE_SYNC */
1064 if (!EX_ISSYNC(exp))
1066 if (stable && !use_wgather) {
1067 spin_lock(&file->f_lock);
1068 file->f_flags |= O_SYNC;
1069 spin_unlock(&file->f_lock);
1072 /* Write the data. */
1073 oldfs = get_fs(); set_fs(KERNEL_DS);
1074 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1079 nfsdstats.io_write += host_err;
1080 fsnotify_modify(file);
1082 /* clear setuid/setgid flag after write */
1083 if (inode->i_mode & (S_ISUID | S_ISGID))
1086 if (stable && use_wgather)
1087 host_err = wait_for_concurrent_writes(file);
1090 dprintk("nfsd: write complete host_err=%d\n", host_err);
1094 err = nfserrno(host_err);
1099 * Read data from a file. count must contain the requested read count
1100 * on entry. On return, *count contains the number of bytes actually read.
1101 * N.B. After this call fhp needs an fh_put
1103 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1104 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1107 struct inode *inode;
1111 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1115 inode = file->f_path.dentry->d_inode;
1117 /* Get readahead parameters */
1118 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1120 if (ra && ra->p_set)
1121 file->f_ra = ra->p_ra;
1123 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1125 /* Write back readahead params */
1127 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1128 spin_lock(&rab->pb_lock);
1129 ra->p_ra = file->f_ra;
1132 spin_unlock(&rab->pb_lock);
1139 /* As above, but use the provided file descriptor. */
1141 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1142 loff_t offset, struct kvec *vec, int vlen,
1143 unsigned long *count)
1148 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1149 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1152 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1153 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1154 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1160 * Write data to a file.
1161 * The stable flag requests synchronous writes.
1162 * N.B. After this call fhp needs an fh_put
1165 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1166 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1172 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1173 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1176 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1179 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1184 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1192 #ifdef CONFIG_NFSD_V3
1194 * Commit all pending writes to stable storage.
1196 * Note: we only guarantee that data that lies within the range specified
1197 * by the 'offset' and 'count' parameters will be synced.
1199 * Unfortunately we cannot lock the file to make sure we return full WCC
1200 * data to the client, as locking happens lower down in the filesystem.
1203 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1204 loff_t offset, unsigned long count)
1207 loff_t end = LLONG_MAX;
1208 __be32 err = nfserr_inval;
1213 end = offset + (loff_t)count - 1;
1218 err = nfsd_open(rqstp, fhp, S_IFREG,
1219 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1222 if (EX_ISSYNC(fhp->fh_export)) {
1223 int err2 = vfs_fsync_range(file, offset, end, 0);
1225 if (err2 != -EINVAL)
1226 err = nfserrno(err2);
1228 err = nfserr_notsupp;
1235 #endif /* CONFIG_NFSD_V3 */
1238 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1242 * Mode has already been set earlier in create:
1244 iap->ia_valid &= ~ATTR_MODE;
1246 * Setting uid/gid works only for root. Irix appears to
1247 * send along the gid on create when it tries to implement
1248 * setgid directories via NFS:
1250 if (current_fsuid() != 0)
1251 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1253 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1257 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1258 * setting size to 0 may fail for some specific file systems by the permission
1259 * checking which requires WRITE permission but the mode is 000.
1260 * we ignore the resizing(to 0) on the just new created file, since the size is
1261 * 0 after file created.
1263 * call this only after vfs_create() is called.
1266 nfsd_check_ignore_resizing(struct iattr *iap)
1268 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1269 iap->ia_valid &= ~ATTR_SIZE;
1273 * Create a file (regular, directory, device, fifo); UNIX sockets
1274 * not yet implemented.
1275 * If the response fh has been verified, the parent directory should
1276 * already be locked. Note that the parent directory is left locked.
1278 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1281 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1282 char *fname, int flen, struct iattr *iap,
1283 int type, dev_t rdev, struct svc_fh *resfhp)
1285 struct dentry *dentry, *dchild = NULL;
1295 if (isdotent(fname, flen))
1298 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1302 dentry = fhp->fh_dentry;
1303 dirp = dentry->d_inode;
1305 err = nfserr_notdir;
1306 if (!dirp->i_op->lookup)
1309 * Check whether the response file handle has been verified yet.
1310 * If it has, the parent directory should already be locked.
1312 if (!resfhp->fh_dentry) {
1313 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1314 fh_lock_nested(fhp, I_MUTEX_PARENT);
1315 dchild = lookup_one_len(fname, dentry, flen);
1316 host_err = PTR_ERR(dchild);
1319 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1323 /* called from nfsd_proc_create */
1324 dchild = dget(resfhp->fh_dentry);
1325 if (!fhp->fh_locked) {
1326 /* not actually possible */
1328 "nfsd_create: parent %s/%s not locked!\n",
1329 dentry->d_parent->d_name.name,
1330 dentry->d_name.name);
1336 * Make sure the child dentry is still negative ...
1339 if (dchild->d_inode) {
1340 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1341 dentry->d_name.name, dchild->d_name.name);
1345 if (!(iap->ia_valid & ATTR_MODE))
1347 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1350 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1351 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1356 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1361 * Get the dir op function pointer.
1366 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1368 nfsd_check_ignore_resizing(iap);
1371 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1377 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1381 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1385 err = nfsd_create_setattr(rqstp, resfhp, iap);
1388 * nfsd_setattr already committed the child. Transactional filesystems
1389 * had a chance to commit changes for both parent and child
1390 * simultaneously making the following commit_metadata a noop.
1392 err2 = nfserrno(commit_metadata(fhp));
1395 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1397 * Update the file handle to get the new inode info.
1400 err = fh_update(resfhp);
1402 if (dchild && !IS_ERR(dchild))
1407 err = nfserrno(host_err);
1411 #ifdef CONFIG_NFSD_V3
1413 static inline int nfsd_create_is_exclusive(int createmode)
1415 return createmode == NFS3_CREATE_EXCLUSIVE
1416 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1420 * NFSv3 and NFSv4 version of nfsd_create
1423 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1424 char *fname, int flen, struct iattr *iap,
1425 struct svc_fh *resfhp, int createmode, u32 *verifier,
1426 bool *truncp, bool *created)
1428 struct dentry *dentry, *dchild = NULL;
1432 __u32 v_mtime=0, v_atime=0;
1438 if (isdotent(fname, flen))
1440 if (!(iap->ia_valid & ATTR_MODE))
1442 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1446 dentry = fhp->fh_dentry;
1447 dirp = dentry->d_inode;
1449 /* Get all the sanity checks out of the way before
1450 * we lock the parent. */
1451 err = nfserr_notdir;
1452 if (!dirp->i_op->lookup)
1454 fh_lock_nested(fhp, I_MUTEX_PARENT);
1457 * Compose the response file handle.
1459 dchild = lookup_one_len(fname, dentry, flen);
1460 host_err = PTR_ERR(dchild);
1464 /* If file doesn't exist, check for permissions to create one */
1465 if (!dchild->d_inode) {
1466 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1471 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1475 if (nfsd_create_is_exclusive(createmode)) {
1476 /* solaris7 gets confused (bugid 4218508) if these have
1477 * the high bit set, so just clear the high bits. If this is
1478 * ever changed to use different attrs for storing the
1479 * verifier, then do_open_lookup() will also need to be fixed
1482 v_mtime = verifier[0]&0x7fffffff;
1483 v_atime = verifier[1]&0x7fffffff;
1486 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1489 if (dchild->d_inode) {
1492 switch (createmode) {
1493 case NFS3_CREATE_UNCHECKED:
1494 if (! S_ISREG(dchild->d_inode->i_mode))
1497 /* in nfsv4, we need to treat this case a little
1498 * differently. we don't want to truncate the
1499 * file now; this would be wrong if the OPEN
1500 * fails for some other reason. furthermore,
1501 * if the size is nonzero, we should ignore it
1502 * according to spec!
1504 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1507 iap->ia_valid &= ATTR_SIZE;
1511 case NFS3_CREATE_EXCLUSIVE:
1512 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1513 && dchild->d_inode->i_atime.tv_sec == v_atime
1514 && dchild->d_inode->i_size == 0 ) {
1519 case NFS4_CREATE_EXCLUSIVE4_1:
1520 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1521 && dchild->d_inode->i_atime.tv_sec == v_atime
1522 && dchild->d_inode->i_size == 0 ) {
1528 case NFS3_CREATE_GUARDED:
1531 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1535 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1537 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1543 nfsd_check_ignore_resizing(iap);
1545 if (nfsd_create_is_exclusive(createmode)) {
1546 /* Cram the verifier into atime/mtime */
1547 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1548 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1549 /* XXX someone who knows this better please fix it for nsec */
1550 iap->ia_mtime.tv_sec = v_mtime;
1551 iap->ia_atime.tv_sec = v_atime;
1552 iap->ia_mtime.tv_nsec = 0;
1553 iap->ia_atime.tv_nsec = 0;
1557 err = nfsd_create_setattr(rqstp, resfhp, iap);
1560 * nfsd_setattr already committed the child (and possibly also the parent).
1563 err = nfserrno(commit_metadata(fhp));
1565 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1567 * Update the filehandle to get the new inode info.
1570 err = fh_update(resfhp);
1574 if (dchild && !IS_ERR(dchild))
1579 err = nfserrno(host_err);
1582 #endif /* CONFIG_NFSD_V3 */
1585 * Read a symlink. On entry, *lenp must contain the maximum path length that
1586 * fits into the buffer. On return, it contains the true length.
1587 * N.B. After this call fhp needs an fh_put
1590 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1592 struct dentry *dentry;
1593 struct inode *inode;
1598 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1602 dentry = fhp->fh_dentry;
1603 inode = dentry->d_inode;
1606 if (!inode->i_op->readlink)
1609 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1610 /* N.B. Why does this call need a get_fs()??
1611 * Remove the set_fs and watch the fireworks:-) --okir
1614 oldfs = get_fs(); set_fs(KERNEL_DS);
1615 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1626 err = nfserrno(host_err);
1631 * Create a symlink and look up its inode
1632 * N.B. After this call _both_ fhp and resfhp need an fh_put
1635 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1636 char *fname, int flen,
1637 char *path, int plen,
1638 struct svc_fh *resfhp,
1641 struct dentry *dentry, *dnew;
1649 if (isdotent(fname, flen))
1652 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1656 dentry = fhp->fh_dentry;
1657 dnew = lookup_one_len(fname, dentry, flen);
1658 host_err = PTR_ERR(dnew);
1662 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1666 if (unlikely(path[plen] != 0)) {
1667 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1668 if (path_alloced == NULL)
1671 strncpy(path_alloced, path, plen);
1672 path_alloced[plen] = 0;
1673 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1674 kfree(path_alloced);
1677 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1678 err = nfserrno(host_err);
1680 err = nfserrno(commit_metadata(fhp));
1683 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1685 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1687 if (err==0) err = cerr;
1692 err = nfserrno(host_err);
1698 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1701 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1702 char *name, int len, struct svc_fh *tfhp)
1704 struct dentry *ddir, *dnew, *dold;
1709 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1712 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1716 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1722 if (isdotent(name, len))
1725 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1726 ddir = ffhp->fh_dentry;
1727 dirp = ddir->d_inode;
1729 dnew = lookup_one_len(name, ddir, len);
1730 host_err = PTR_ERR(dnew);
1734 dold = tfhp->fh_dentry;
1736 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1738 err = nfserrno(host_err);
1743 goto out_drop_write;
1744 host_err = nfsd_break_lease(dold->d_inode);
1746 err = nfserrno(host_err);
1747 goto out_drop_write;
1749 host_err = vfs_link(dold, dirp, dnew);
1751 err = nfserrno(commit_metadata(ffhp));
1753 err = nfserrno(commit_metadata(tfhp));
1755 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1758 err = nfserrno(host_err);
1761 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1770 err = nfserrno(host_err);
1776 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1779 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1780 struct svc_fh *tfhp, char *tname, int tlen)
1782 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1783 struct inode *fdir, *tdir;
1787 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1790 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1794 fdentry = ffhp->fh_dentry;
1795 fdir = fdentry->d_inode;
1797 tdentry = tfhp->fh_dentry;
1798 tdir = tdentry->d_inode;
1800 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1801 if (ffhp->fh_export != tfhp->fh_export)
1805 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1808 /* cannot use fh_lock as we need deadlock protective ordering
1809 * so do it by hand */
1810 trap = lock_rename(tdentry, fdentry);
1811 ffhp->fh_locked = tfhp->fh_locked = 1;
1815 odentry = lookup_one_len(fname, fdentry, flen);
1816 host_err = PTR_ERR(odentry);
1817 if (IS_ERR(odentry))
1821 if (!odentry->d_inode)
1824 if (odentry == trap)
1827 ndentry = lookup_one_len(tname, tdentry, tlen);
1828 host_err = PTR_ERR(ndentry);
1829 if (IS_ERR(ndentry))
1831 host_err = -ENOTEMPTY;
1832 if (ndentry == trap)
1836 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1838 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1842 host_err = nfsd_break_lease(odentry->d_inode);
1844 goto out_drop_write;
1845 if (ndentry->d_inode) {
1846 host_err = nfsd_break_lease(ndentry->d_inode);
1848 goto out_drop_write;
1850 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1852 host_err = commit_metadata(tfhp);
1854 host_err = commit_metadata(ffhp);
1857 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1863 err = nfserrno(host_err);
1865 /* we cannot reply on fh_unlock on the two filehandles,
1866 * as that would do the wrong thing if the two directories
1867 * were the same, so again we do it by hand
1869 fill_post_wcc(ffhp);
1870 fill_post_wcc(tfhp);
1871 unlock_rename(tdentry, fdentry);
1872 ffhp->fh_locked = tfhp->fh_locked = 0;
1879 * Unlink a file or directory
1880 * N.B. After this call fhp needs an fh_put
1883 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1884 char *fname, int flen)
1886 struct dentry *dentry, *rdentry;
1892 if (!flen || isdotent(fname, flen))
1894 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1898 fh_lock_nested(fhp, I_MUTEX_PARENT);
1899 dentry = fhp->fh_dentry;
1900 dirp = dentry->d_inode;
1902 rdentry = lookup_one_len(fname, dentry, flen);
1903 host_err = PTR_ERR(rdentry);
1904 if (IS_ERR(rdentry))
1907 if (!rdentry->d_inode) {
1914 type = rdentry->d_inode->i_mode & S_IFMT;
1916 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1920 host_err = nfsd_break_lease(rdentry->d_inode);
1922 goto out_drop_write;
1923 if (type != S_IFDIR)
1924 host_err = vfs_unlink(dirp, rdentry);
1926 host_err = vfs_rmdir(dirp, rdentry);
1928 host_err = commit_metadata(fhp);
1930 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1935 err = nfserrno(host_err);
1941 * We do this buffering because we must not call back into the file
1942 * system's ->lookup() method from the filldir callback. That may well
1943 * deadlock a number of file systems.
1945 * This is based heavily on the implementation of same in XFS.
1947 struct buffered_dirent {
1951 unsigned int d_type;
1955 struct readdir_data {
1961 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1962 loff_t offset, u64 ino, unsigned int d_type)
1964 struct readdir_data *buf = __buf;
1965 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1966 unsigned int reclen;
1968 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1969 if (buf->used + reclen > PAGE_SIZE) {
1974 de->namlen = namlen;
1975 de->offset = offset;
1977 de->d_type = d_type;
1978 memcpy(de->name, name, namlen);
1979 buf->used += reclen;
1984 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1985 struct readdir_cd *cdp, loff_t *offsetp)
1987 struct readdir_data buf;
1988 struct buffered_dirent *de;
1993 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1995 return nfserrno(-ENOMEM);
2000 struct inode *dir_inode = file->f_path.dentry->d_inode;
2001 unsigned int reclen;
2003 cdp->err = nfserr_eof; /* will be cleared on successful read */
2007 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2020 * Various filldir functions may end up calling back into
2021 * lookup_one_len() and the file system's ->lookup() method.
2022 * These expect i_mutex to be held, as it would within readdir.
2024 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2028 de = (struct buffered_dirent *)buf.dirent;
2030 offset = de->offset;
2032 if (func(cdp, de->name, de->namlen, de->offset,
2033 de->ino, de->d_type))
2036 if (cdp->err != nfs_ok)
2039 reclen = ALIGN(sizeof(*de) + de->namlen,
2042 de = (struct buffered_dirent *)((char *)de + reclen);
2044 mutex_unlock(&dir_inode->i_mutex);
2045 if (size > 0) /* We bailed out early */
2048 offset = vfs_llseek(file, 0, SEEK_CUR);
2051 free_page((unsigned long)(buf.dirent));
2054 return nfserrno(host_err);
2061 * Read entries from a directory.
2062 * The NFSv3/4 verifier we ignore for now.
2065 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2066 struct readdir_cd *cdp, filldir_t func)
2070 loff_t offset = *offsetp;
2071 int may_flags = NFSD_MAY_READ;
2073 /* NFSv2 only supports 32 bit cookies */
2074 if (rqstp->rq_vers > 2)
2075 may_flags |= NFSD_MAY_64BIT_COOKIE;
2077 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2081 offset = vfs_llseek(file, offset, 0);
2083 err = nfserrno((int)offset);
2087 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2089 if (err == nfserr_eof || err == nfserr_toosmall)
2090 err = nfs_ok; /* can still be found in ->err */
2098 * Get file system stats
2099 * N.B. After this call fhp needs an fh_put
2102 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2106 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2108 struct path path = {
2109 .mnt = fhp->fh_export->ex_path.mnt,
2110 .dentry = fhp->fh_dentry,
2112 if (vfs_statfs(&path, stat))
2118 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2120 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2124 * Check for a user's access permissions to this inode.
2127 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2128 struct dentry *dentry, int acc)
2130 struct inode *inode = dentry->d_inode;
2133 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2136 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2138 (acc & NFSD_MAY_READ)? " read" : "",
2139 (acc & NFSD_MAY_WRITE)? " write" : "",
2140 (acc & NFSD_MAY_EXEC)? " exec" : "",
2141 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2142 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2143 (acc & NFSD_MAY_LOCK)? " lock" : "",
2144 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2146 IS_IMMUTABLE(inode)? " immut" : "",
2147 IS_APPEND(inode)? " append" : "",
2148 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2149 dprintk(" owner %d/%d user %d/%d\n",
2150 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2153 /* Normally we reject any write/sattr etc access on a read-only file
2154 * system. But if it is IRIX doing check on write-access for a
2155 * device special file, we ignore rofs.
2157 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2158 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2159 if (exp_rdonly(rqstp, exp) ||
2160 __mnt_is_readonly(exp->ex_path.mnt))
2162 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2165 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2168 if (acc & NFSD_MAY_LOCK) {
2169 /* If we cannot rely on authentication in NLM requests,
2170 * just allow locks, otherwise require read permission, or
2173 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2176 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2179 * The file owner always gets access permission for accesses that
2180 * would normally be checked at open time. This is to make
2181 * file access work even when the client has done a fchmod(fd, 0).
2183 * However, `cp foo bar' should fail nevertheless when bar is
2184 * readonly. A sensible way to do this might be to reject all
2185 * attempts to truncate a read-only file, because a creat() call
2186 * always implies file truncation.
2187 * ... but this isn't really fair. A process may reasonably call
2188 * ftruncate on an open file descriptor on a file with perm 000.
2189 * We must trust the client to do permission checking - using "ACCESS"
2192 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2193 inode->i_uid == current_fsuid())
2196 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2197 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2199 /* Allow read access to binaries even when mode 111 */
2200 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2201 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2202 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2203 err = inode_permission(inode, MAY_EXEC);
2205 return err? nfserrno(err) : 0;
2209 nfsd_racache_shutdown(void)
2211 struct raparms *raparm, *last_raparm;
2214 dprintk("nfsd: freeing readahead buffers.\n");
2216 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2217 raparm = raparm_hash[i].pb_head;
2219 last_raparm = raparm;
2220 raparm = raparm->p_next;
2223 raparm_hash[i].pb_head = NULL;
2227 * Initialize readahead param cache
2230 nfsd_racache_init(int cache_size)
2235 struct raparms **raparm = NULL;
2238 if (raparm_hash[0].pb_head)
2240 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2243 cache_size = nperbucket * RAPARM_HASH_SIZE;
2245 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2247 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2248 spin_lock_init(&raparm_hash[i].pb_lock);
2250 raparm = &raparm_hash[i].pb_head;
2251 for (j = 0; j < nperbucket; j++) {
2252 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2255 raparm = &(*raparm)->p_next;
2260 nfsdstats.ra_size = cache_size;
2264 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2265 nfsd_racache_shutdown();
2269 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2271 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2273 struct inode *inode = fhp->fh_dentry->d_inode;
2277 struct posix_acl *acl;
2279 if (!IS_POSIXACL(inode))
2280 return ERR_PTR(-EOPNOTSUPP);
2283 case ACL_TYPE_ACCESS:
2284 name = POSIX_ACL_XATTR_ACCESS;
2286 case ACL_TYPE_DEFAULT:
2287 name = POSIX_ACL_XATTR_DEFAULT;
2290 return ERR_PTR(-EOPNOTSUPP);
2293 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2295 return ERR_PTR(size);
2297 acl = posix_acl_from_xattr(value, size);
2303 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2305 struct inode *inode = fhp->fh_dentry->d_inode;
2311 if (!IS_POSIXACL(inode) ||
2312 !inode->i_op->setxattr || !inode->i_op->removexattr)
2315 case ACL_TYPE_ACCESS:
2316 name = POSIX_ACL_XATTR_ACCESS;
2318 case ACL_TYPE_DEFAULT:
2319 name = POSIX_ACL_XATTR_DEFAULT;
2325 if (acl && acl->a_count) {
2326 size = posix_acl_xattr_size(acl->a_count);
2327 value = kmalloc(size, GFP_KERNEL);
2330 error = posix_acl_to_xattr(acl, value, size);
2337 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2341 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2343 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2346 error = vfs_removexattr(fhp->fh_dentry, name);
2347 if (error == -ENODATA)
2351 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2357 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */