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;
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);
429 return nfserrno(host_err);
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);
445 * The size case is special, it changes the file in addition to the
448 if (iap->ia_valid & ATTR_SIZE) {
449 err = nfsd_get_write_access(rqstp, fhp, iap);
455 iap->ia_valid |= ATTR_CTIME;
457 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
458 err = nfserr_notsync;
459 goto out_put_write_access;
462 host_err = nfsd_break_lease(inode);
464 goto out_put_write_access_nfserror;
467 host_err = notify_change(dentry, iap);
470 out_put_write_access_nfserror:
471 err = nfserrno(host_err);
472 out_put_write_access:
474 put_write_access(inode);
476 commit_metadata(fhp);
481 #if defined(CONFIG_NFSD_V2_ACL) || \
482 defined(CONFIG_NFSD_V3_ACL) || \
483 defined(CONFIG_NFSD_V4)
484 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
489 buflen = vfs_getxattr(dentry, key, NULL, 0);
493 *buf = kmalloc(buflen, GFP_KERNEL);
497 ret = vfs_getxattr(dentry, key, *buf, buflen);
504 #if defined(CONFIG_NFSD_V4)
506 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
514 return vfs_setxattr(dentry, key, NULL, 0, 0);
516 buflen = posix_acl_xattr_size(pacl->a_count);
517 buf = kmalloc(buflen, GFP_KERNEL);
522 len = posix_acl_to_xattr(pacl, buf, buflen);
528 error = vfs_setxattr(dentry, key, buf, len, 0);
535 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
536 struct nfs4_acl *acl)
540 struct dentry *dentry;
542 struct posix_acl *pacl = NULL, *dpacl = NULL;
543 unsigned int flags = 0;
546 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
550 dentry = fhp->fh_dentry;
551 inode = dentry->d_inode;
552 if (S_ISDIR(inode->i_mode))
553 flags = NFS4_ACL_DIR;
555 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
556 if (host_error == -EINVAL) {
557 return nfserr_attrnotsupp;
558 } else if (host_error < 0)
561 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
565 if (S_ISDIR(inode->i_mode))
566 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
569 posix_acl_release(pacl);
570 posix_acl_release(dpacl);
572 if (host_error == -EOPNOTSUPP)
573 return nfserr_attrnotsupp;
575 return nfserrno(host_error);
578 static struct posix_acl *
579 _get_posix_acl(struct dentry *dentry, char *key)
582 struct posix_acl *pacl = NULL;
585 buflen = nfsd_getxattr(dentry, key, &buf);
589 return ERR_PTR(buflen);
591 pacl = posix_acl_from_xattr(buf, buflen);
597 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
599 struct inode *inode = dentry->d_inode;
601 struct posix_acl *pacl = NULL, *dpacl = NULL;
602 unsigned int flags = 0;
604 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
605 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
606 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
608 error = PTR_ERR(pacl);
613 if (S_ISDIR(inode->i_mode)) {
614 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
615 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
617 else if (IS_ERR(dpacl)) {
618 error = PTR_ERR(dpacl);
622 flags = NFS4_ACL_DIR;
625 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
627 error = PTR_ERR(*acl);
631 posix_acl_release(pacl);
632 posix_acl_release(dpacl);
636 #define NFSD_XATTR_JUNCTION_PREFIX XATTR_TRUSTED_PREFIX "junction."
637 #define NFSD_XATTR_JUNCTION_TYPE NFSD_XATTR_JUNCTION_PREFIX "type"
638 int nfsd4_is_junction(struct dentry *dentry)
640 struct inode *inode = dentry->d_inode;
644 if (inode->i_mode & S_IXUGO)
646 if (!(inode->i_mode & S_ISVTX))
648 if (vfs_getxattr(dentry, NFSD_XATTR_JUNCTION_TYPE, NULL, 0) <= 0)
652 #endif /* defined(CONFIG_NFSD_V4) */
654 #ifdef CONFIG_NFSD_V3
656 * Check server access rights to a file system object
662 static struct accessmap nfs3_regaccess[] = {
663 { NFS3_ACCESS_READ, NFSD_MAY_READ },
664 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
665 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
666 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
671 static struct accessmap nfs3_diraccess[] = {
672 { NFS3_ACCESS_READ, NFSD_MAY_READ },
673 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
674 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
675 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
676 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
681 static struct accessmap nfs3_anyaccess[] = {
682 /* Some clients - Solaris 2.6 at least, make an access call
683 * to the server to check for access for things like /dev/null
684 * (which really, the server doesn't care about). So
685 * We provide simple access checking for them, looking
686 * mainly at mode bits, and we make sure to ignore read-only
689 { NFS3_ACCESS_READ, NFSD_MAY_READ },
690 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
691 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
692 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
698 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
700 struct accessmap *map;
701 struct svc_export *export;
702 struct dentry *dentry;
703 u32 query, result = 0, sresult = 0;
706 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
710 export = fhp->fh_export;
711 dentry = fhp->fh_dentry;
713 if (S_ISREG(dentry->d_inode->i_mode))
714 map = nfs3_regaccess;
715 else if (S_ISDIR(dentry->d_inode->i_mode))
716 map = nfs3_diraccess;
718 map = nfs3_anyaccess;
722 for (; map->access; map++) {
723 if (map->access & query) {
726 sresult |= map->access;
728 err2 = nfsd_permission(rqstp, export, dentry, map->how);
731 result |= map->access;
734 /* the following error codes just mean the access was not allowed,
735 * rather than an error occurred */
739 /* simply don't "or" in the access bit. */
749 *supported = sresult;
754 #endif /* CONFIG_NFSD_V3 */
756 static int nfsd_open_break_lease(struct inode *inode, int access)
760 if (access & NFSD_MAY_NOT_BREAK_LEASE)
762 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
763 return break_lease(inode, mode | O_NONBLOCK);
767 * Open an existing file or directory.
768 * The may_flags argument indicates the type of open (read/write/lock)
769 * and additional flags.
770 * N.B. After this call fhp needs an fh_put
773 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
774 int may_flags, struct file **filp)
776 struct dentry *dentry;
778 int flags = O_RDONLY|O_LARGEFILE;
782 validate_process_creds();
785 * If we get here, then the client has already done an "open",
786 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
787 * in case a chmod has now revoked permission.
789 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE);
793 dentry = fhp->fh_dentry;
794 inode = dentry->d_inode;
796 /* Disallow write access to files with the append-only bit set
797 * or any access when mandatory locking enabled
800 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
803 * We must ignore files (but only files) which might have mandatory
804 * locks on them because there is no way to know if the accesser has
807 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
813 host_err = nfsd_open_break_lease(inode, may_flags);
814 if (host_err) /* NOMEM or WOULDBLOCK */
817 if (may_flags & NFSD_MAY_WRITE) {
818 if (may_flags & NFSD_MAY_READ)
819 flags = O_RDWR|O_LARGEFILE;
821 flags = O_WRONLY|O_LARGEFILE;
823 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
824 flags, current_cred());
826 host_err = PTR_ERR(*filp);
829 host_err = ima_file_check(*filp, may_flags);
831 if (may_flags & NFSD_MAY_64BIT_COOKIE)
832 (*filp)->f_mode |= FMODE_64BITHASH;
834 (*filp)->f_mode |= FMODE_32BITHASH;
838 err = nfserrno(host_err);
840 validate_process_creds();
848 nfsd_close(struct file *filp)
854 * Obtain the readahead parameters for the file
855 * specified by (dev, ino).
858 static inline struct raparms *
859 nfsd_get_raparms(dev_t dev, ino_t ino)
861 struct raparms *ra, **rap, **frap = NULL;
864 struct raparm_hbucket *rab;
866 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
867 rab = &raparm_hash[hash];
869 spin_lock(&rab->pb_lock);
870 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
871 if (ra->p_ino == ino && ra->p_dev == dev)
874 if (ra->p_count == 0)
877 depth = nfsdstats.ra_size;
879 spin_unlock(&rab->pb_lock);
889 if (rap != &rab->pb_head) {
891 ra->p_next = rab->pb_head;
895 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
896 spin_unlock(&rab->pb_lock);
901 * Grab and keep cached pages associated with a file in the svc_rqst
902 * so that they can be passed to the network sendmsg/sendpage routines
903 * directly. They will be released after the sending has completed.
906 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
907 struct splice_desc *sd)
909 struct svc_rqst *rqstp = sd->u.data;
910 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
911 struct page *page = buf->page;
916 if (rqstp->rq_res.page_len == 0) {
921 rqstp->rq_res.page_base = buf->offset;
922 rqstp->rq_res.page_len = size;
923 } else if (page != pp[-1]) {
929 rqstp->rq_res.page_len += size;
931 rqstp->rq_res.page_len += size;
936 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
937 struct splice_desc *sd)
939 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
943 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
944 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
952 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
953 struct splice_desc sd = {
960 rqstp->rq_resused = 1;
961 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
965 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
970 nfsdstats.io_read += host_err;
973 fsnotify_access(file);
975 err = nfserrno(host_err);
979 static void kill_suid(struct dentry *dentry)
982 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
984 mutex_lock(&dentry->d_inode->i_mutex);
985 notify_change(dentry, &ia);
986 mutex_unlock(&dentry->d_inode->i_mutex);
990 * Gathered writes: If another process is currently writing to the file,
991 * there's a high chance this is another nfsd (triggered by a bulk write
992 * from a client's biod). Rather than syncing the file with each write
993 * request, we sleep for 10 msec.
995 * I don't know if this roughly approximates C. Juszak's idea of
996 * gathered writes, but it's a nice and simple solution (IMHO), and it
999 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1000 * better tool (separate unstable writes and commits) for solving this
1003 static int wait_for_concurrent_writes(struct file *file)
1005 struct inode *inode = file->f_path.dentry->d_inode;
1006 static ino_t last_ino;
1007 static dev_t last_dev;
1010 if (atomic_read(&inode->i_writecount) > 1
1011 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1012 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1014 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1017 if (inode->i_state & I_DIRTY) {
1018 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1019 err = vfs_fsync(file, 0);
1021 last_ino = inode->i_ino;
1022 last_dev = inode->i_sb->s_dev;
1027 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1028 loff_t offset, struct kvec *vec, int vlen,
1029 unsigned long *cnt, int *stablep)
1031 struct svc_export *exp;
1032 struct dentry *dentry;
1033 struct inode *inode;
1037 int stable = *stablep;
1040 dentry = file->f_path.dentry;
1041 inode = dentry->d_inode;
1042 exp = fhp->fh_export;
1045 * Request sync writes if
1046 * - the sync export option has been set, or
1047 * - the client requested O_SYNC behavior (NFSv3 feature).
1048 * - The file system doesn't support fsync().
1049 * When NFSv2 gathered writes have been configured for this volume,
1050 * flushing the data to disk is handled separately below.
1052 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1054 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1056 *stablep = 2; /* FILE_SYNC */
1059 if (!EX_ISSYNC(exp))
1061 if (stable && !use_wgather) {
1062 spin_lock(&file->f_lock);
1063 file->f_flags |= O_SYNC;
1064 spin_unlock(&file->f_lock);
1067 /* Write the data. */
1068 oldfs = get_fs(); set_fs(KERNEL_DS);
1069 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1074 nfsdstats.io_write += host_err;
1075 fsnotify_modify(file);
1077 /* clear setuid/setgid flag after write */
1078 if (inode->i_mode & (S_ISUID | S_ISGID))
1081 if (stable && use_wgather)
1082 host_err = wait_for_concurrent_writes(file);
1085 dprintk("nfsd: write complete host_err=%d\n", host_err);
1089 err = nfserrno(host_err);
1094 * Read data from a file. count must contain the requested read count
1095 * on entry. On return, *count contains the number of bytes actually read.
1096 * N.B. After this call fhp needs an fh_put
1098 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1099 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1102 struct inode *inode;
1106 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1110 inode = file->f_path.dentry->d_inode;
1112 /* Get readahead parameters */
1113 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1115 if (ra && ra->p_set)
1116 file->f_ra = ra->p_ra;
1118 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1120 /* Write back readahead params */
1122 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1123 spin_lock(&rab->pb_lock);
1124 ra->p_ra = file->f_ra;
1127 spin_unlock(&rab->pb_lock);
1134 /* As above, but use the provided file descriptor. */
1136 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1137 loff_t offset, struct kvec *vec, int vlen,
1138 unsigned long *count)
1143 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1144 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1147 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1148 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1149 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1155 * Write data to a file.
1156 * The stable flag requests synchronous writes.
1157 * N.B. After this call fhp needs an fh_put
1160 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1161 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1167 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1168 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1171 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1174 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1179 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1187 #ifdef CONFIG_NFSD_V3
1189 * Commit all pending writes to stable storage.
1191 * Note: we only guarantee that data that lies within the range specified
1192 * by the 'offset' and 'count' parameters will be synced.
1194 * Unfortunately we cannot lock the file to make sure we return full WCC
1195 * data to the client, as locking happens lower down in the filesystem.
1198 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1199 loff_t offset, unsigned long count)
1202 loff_t end = LLONG_MAX;
1203 __be32 err = nfserr_inval;
1208 end = offset + (loff_t)count - 1;
1213 err = nfsd_open(rqstp, fhp, S_IFREG,
1214 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1217 if (EX_ISSYNC(fhp->fh_export)) {
1218 int err2 = vfs_fsync_range(file, offset, end, 0);
1220 if (err2 != -EINVAL)
1221 err = nfserrno(err2);
1223 err = nfserr_notsupp;
1230 #endif /* CONFIG_NFSD_V3 */
1233 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1237 * Mode has already been set earlier in create:
1239 iap->ia_valid &= ~ATTR_MODE;
1241 * Setting uid/gid works only for root. Irix appears to
1242 * send along the gid on create when it tries to implement
1243 * setgid directories via NFS:
1245 if (current_fsuid() != 0)
1246 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1248 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1252 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1253 * setting size to 0 may fail for some specific file systems by the permission
1254 * checking which requires WRITE permission but the mode is 000.
1255 * we ignore the resizing(to 0) on the just new created file, since the size is
1256 * 0 after file created.
1258 * call this only after vfs_create() is called.
1261 nfsd_check_ignore_resizing(struct iattr *iap)
1263 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1264 iap->ia_valid &= ~ATTR_SIZE;
1268 * Create a file (regular, directory, device, fifo); UNIX sockets
1269 * not yet implemented.
1270 * If the response fh has been verified, the parent directory should
1271 * already be locked. Note that the parent directory is left locked.
1273 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1276 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1277 char *fname, int flen, struct iattr *iap,
1278 int type, dev_t rdev, struct svc_fh *resfhp)
1280 struct dentry *dentry, *dchild = NULL;
1290 if (isdotent(fname, flen))
1293 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1297 dentry = fhp->fh_dentry;
1298 dirp = dentry->d_inode;
1300 err = nfserr_notdir;
1301 if (!dirp->i_op->lookup)
1304 * Check whether the response file handle has been verified yet.
1305 * If it has, the parent directory should already be locked.
1307 if (!resfhp->fh_dentry) {
1308 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1309 fh_lock_nested(fhp, I_MUTEX_PARENT);
1310 dchild = lookup_one_len(fname, dentry, flen);
1311 host_err = PTR_ERR(dchild);
1314 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1318 /* called from nfsd_proc_create */
1319 dchild = dget(resfhp->fh_dentry);
1320 if (!fhp->fh_locked) {
1321 /* not actually possible */
1323 "nfsd_create: parent %s/%s not locked!\n",
1324 dentry->d_parent->d_name.name,
1325 dentry->d_name.name);
1331 * Make sure the child dentry is still negative ...
1334 if (dchild->d_inode) {
1335 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1336 dentry->d_name.name, dchild->d_name.name);
1340 if (!(iap->ia_valid & ATTR_MODE))
1342 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1345 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1346 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1351 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1356 * Get the dir op function pointer.
1361 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1363 nfsd_check_ignore_resizing(iap);
1366 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1372 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1376 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1380 err = nfsd_create_setattr(rqstp, resfhp, iap);
1383 * nfsd_setattr already committed the child. Transactional filesystems
1384 * had a chance to commit changes for both parent and child
1385 * simultaneously making the following commit_metadata a noop.
1387 err2 = nfserrno(commit_metadata(fhp));
1390 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1392 * Update the file handle to get the new inode info.
1395 err = fh_update(resfhp);
1397 if (dchild && !IS_ERR(dchild))
1402 err = nfserrno(host_err);
1406 #ifdef CONFIG_NFSD_V3
1408 static inline int nfsd_create_is_exclusive(int createmode)
1410 return createmode == NFS3_CREATE_EXCLUSIVE
1411 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1415 * NFSv3 and NFSv4 version of nfsd_create
1418 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1419 char *fname, int flen, struct iattr *iap,
1420 struct svc_fh *resfhp, int createmode, u32 *verifier,
1421 bool *truncp, bool *created)
1423 struct dentry *dentry, *dchild = NULL;
1427 __u32 v_mtime=0, v_atime=0;
1433 if (isdotent(fname, flen))
1435 if (!(iap->ia_valid & ATTR_MODE))
1437 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1441 dentry = fhp->fh_dentry;
1442 dirp = dentry->d_inode;
1444 /* Get all the sanity checks out of the way before
1445 * we lock the parent. */
1446 err = nfserr_notdir;
1447 if (!dirp->i_op->lookup)
1449 fh_lock_nested(fhp, I_MUTEX_PARENT);
1452 * Compose the response file handle.
1454 dchild = lookup_one_len(fname, dentry, flen);
1455 host_err = PTR_ERR(dchild);
1459 /* If file doesn't exist, check for permissions to create one */
1460 if (!dchild->d_inode) {
1461 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1466 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1470 if (nfsd_create_is_exclusive(createmode)) {
1471 /* solaris7 gets confused (bugid 4218508) if these have
1472 * the high bit set, so just clear the high bits. If this is
1473 * ever changed to use different attrs for storing the
1474 * verifier, then do_open_lookup() will also need to be fixed
1477 v_mtime = verifier[0]&0x7fffffff;
1478 v_atime = verifier[1]&0x7fffffff;
1481 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1484 if (dchild->d_inode) {
1487 switch (createmode) {
1488 case NFS3_CREATE_UNCHECKED:
1489 if (! S_ISREG(dchild->d_inode->i_mode))
1492 /* in nfsv4, we need to treat this case a little
1493 * differently. we don't want to truncate the
1494 * file now; this would be wrong if the OPEN
1495 * fails for some other reason. furthermore,
1496 * if the size is nonzero, we should ignore it
1497 * according to spec!
1499 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1502 iap->ia_valid &= ATTR_SIZE;
1506 case NFS3_CREATE_EXCLUSIVE:
1507 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1508 && dchild->d_inode->i_atime.tv_sec == v_atime
1509 && dchild->d_inode->i_size == 0 ) {
1514 case NFS4_CREATE_EXCLUSIVE4_1:
1515 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1516 && dchild->d_inode->i_atime.tv_sec == v_atime
1517 && dchild->d_inode->i_size == 0 ) {
1523 case NFS3_CREATE_GUARDED:
1526 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1530 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1532 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1538 nfsd_check_ignore_resizing(iap);
1540 if (nfsd_create_is_exclusive(createmode)) {
1541 /* Cram the verifier into atime/mtime */
1542 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1543 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1544 /* XXX someone who knows this better please fix it for nsec */
1545 iap->ia_mtime.tv_sec = v_mtime;
1546 iap->ia_atime.tv_sec = v_atime;
1547 iap->ia_mtime.tv_nsec = 0;
1548 iap->ia_atime.tv_nsec = 0;
1552 err = nfsd_create_setattr(rqstp, resfhp, iap);
1555 * nfsd_setattr already committed the child (and possibly also the parent).
1558 err = nfserrno(commit_metadata(fhp));
1560 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1562 * Update the filehandle to get the new inode info.
1565 err = fh_update(resfhp);
1569 if (dchild && !IS_ERR(dchild))
1574 err = nfserrno(host_err);
1577 #endif /* CONFIG_NFSD_V3 */
1580 * Read a symlink. On entry, *lenp must contain the maximum path length that
1581 * fits into the buffer. On return, it contains the true length.
1582 * N.B. After this call fhp needs an fh_put
1585 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1587 struct dentry *dentry;
1588 struct inode *inode;
1593 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1597 dentry = fhp->fh_dentry;
1598 inode = dentry->d_inode;
1601 if (!inode->i_op->readlink)
1604 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1605 /* N.B. Why does this call need a get_fs()??
1606 * Remove the set_fs and watch the fireworks:-) --okir
1609 oldfs = get_fs(); set_fs(KERNEL_DS);
1610 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1621 err = nfserrno(host_err);
1626 * Create a symlink and look up its inode
1627 * N.B. After this call _both_ fhp and resfhp need an fh_put
1630 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1631 char *fname, int flen,
1632 char *path, int plen,
1633 struct svc_fh *resfhp,
1636 struct dentry *dentry, *dnew;
1644 if (isdotent(fname, flen))
1647 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1651 dentry = fhp->fh_dentry;
1652 dnew = lookup_one_len(fname, dentry, flen);
1653 host_err = PTR_ERR(dnew);
1657 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1661 if (unlikely(path[plen] != 0)) {
1662 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1663 if (path_alloced == NULL)
1666 strncpy(path_alloced, path, plen);
1667 path_alloced[plen] = 0;
1668 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1669 kfree(path_alloced);
1672 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1673 err = nfserrno(host_err);
1675 err = nfserrno(commit_metadata(fhp));
1678 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1680 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1682 if (err==0) err = cerr;
1687 err = nfserrno(host_err);
1693 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1696 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1697 char *name, int len, struct svc_fh *tfhp)
1699 struct dentry *ddir, *dnew, *dold;
1704 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1707 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1711 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1717 if (isdotent(name, len))
1720 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1721 ddir = ffhp->fh_dentry;
1722 dirp = ddir->d_inode;
1724 dnew = lookup_one_len(name, ddir, len);
1725 host_err = PTR_ERR(dnew);
1729 dold = tfhp->fh_dentry;
1731 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1733 err = nfserrno(host_err);
1738 goto out_drop_write;
1739 host_err = nfsd_break_lease(dold->d_inode);
1741 err = nfserrno(host_err);
1742 goto out_drop_write;
1744 host_err = vfs_link(dold, dirp, dnew);
1746 err = nfserrno(commit_metadata(ffhp));
1748 err = nfserrno(commit_metadata(tfhp));
1750 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1753 err = nfserrno(host_err);
1756 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1765 err = nfserrno(host_err);
1771 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1774 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1775 struct svc_fh *tfhp, char *tname, int tlen)
1777 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1778 struct inode *fdir, *tdir;
1782 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1785 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1789 fdentry = ffhp->fh_dentry;
1790 fdir = fdentry->d_inode;
1792 tdentry = tfhp->fh_dentry;
1793 tdir = tdentry->d_inode;
1795 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1796 if (ffhp->fh_export != tfhp->fh_export)
1800 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1803 /* cannot use fh_lock as we need deadlock protective ordering
1804 * so do it by hand */
1805 trap = lock_rename(tdentry, fdentry);
1806 ffhp->fh_locked = tfhp->fh_locked = 1;
1810 odentry = lookup_one_len(fname, fdentry, flen);
1811 host_err = PTR_ERR(odentry);
1812 if (IS_ERR(odentry))
1816 if (!odentry->d_inode)
1819 if (odentry == trap)
1822 ndentry = lookup_one_len(tname, tdentry, tlen);
1823 host_err = PTR_ERR(ndentry);
1824 if (IS_ERR(ndentry))
1826 host_err = -ENOTEMPTY;
1827 if (ndentry == trap)
1831 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1833 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1837 host_err = nfsd_break_lease(odentry->d_inode);
1839 goto out_drop_write;
1840 if (ndentry->d_inode) {
1841 host_err = nfsd_break_lease(ndentry->d_inode);
1843 goto out_drop_write;
1845 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1847 host_err = commit_metadata(tfhp);
1849 host_err = commit_metadata(ffhp);
1852 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1858 err = nfserrno(host_err);
1860 /* we cannot reply on fh_unlock on the two filehandles,
1861 * as that would do the wrong thing if the two directories
1862 * were the same, so again we do it by hand
1864 fill_post_wcc(ffhp);
1865 fill_post_wcc(tfhp);
1866 unlock_rename(tdentry, fdentry);
1867 ffhp->fh_locked = tfhp->fh_locked = 0;
1874 * Unlink a file or directory
1875 * N.B. After this call fhp needs an fh_put
1878 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1879 char *fname, int flen)
1881 struct dentry *dentry, *rdentry;
1887 if (!flen || isdotent(fname, flen))
1889 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1893 fh_lock_nested(fhp, I_MUTEX_PARENT);
1894 dentry = fhp->fh_dentry;
1895 dirp = dentry->d_inode;
1897 rdentry = lookup_one_len(fname, dentry, flen);
1898 host_err = PTR_ERR(rdentry);
1899 if (IS_ERR(rdentry))
1902 if (!rdentry->d_inode) {
1909 type = rdentry->d_inode->i_mode & S_IFMT;
1911 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1915 host_err = nfsd_break_lease(rdentry->d_inode);
1917 goto out_drop_write;
1918 if (type != S_IFDIR)
1919 host_err = vfs_unlink(dirp, rdentry);
1921 host_err = vfs_rmdir(dirp, rdentry);
1923 host_err = commit_metadata(fhp);
1925 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1930 err = nfserrno(host_err);
1936 * We do this buffering because we must not call back into the file
1937 * system's ->lookup() method from the filldir callback. That may well
1938 * deadlock a number of file systems.
1940 * This is based heavily on the implementation of same in XFS.
1942 struct buffered_dirent {
1946 unsigned int d_type;
1950 struct readdir_data {
1956 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1957 loff_t offset, u64 ino, unsigned int d_type)
1959 struct readdir_data *buf = __buf;
1960 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1961 unsigned int reclen;
1963 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1964 if (buf->used + reclen > PAGE_SIZE) {
1969 de->namlen = namlen;
1970 de->offset = offset;
1972 de->d_type = d_type;
1973 memcpy(de->name, name, namlen);
1974 buf->used += reclen;
1979 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1980 struct readdir_cd *cdp, loff_t *offsetp)
1982 struct readdir_data buf;
1983 struct buffered_dirent *de;
1988 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1990 return nfserrno(-ENOMEM);
1995 struct inode *dir_inode = file->f_path.dentry->d_inode;
1996 unsigned int reclen;
1998 cdp->err = nfserr_eof; /* will be cleared on successful read */
2002 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2015 * Various filldir functions may end up calling back into
2016 * lookup_one_len() and the file system's ->lookup() method.
2017 * These expect i_mutex to be held, as it would within readdir.
2019 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2023 de = (struct buffered_dirent *)buf.dirent;
2025 offset = de->offset;
2027 if (func(cdp, de->name, de->namlen, de->offset,
2028 de->ino, de->d_type))
2031 if (cdp->err != nfs_ok)
2034 reclen = ALIGN(sizeof(*de) + de->namlen,
2037 de = (struct buffered_dirent *)((char *)de + reclen);
2039 mutex_unlock(&dir_inode->i_mutex);
2040 if (size > 0) /* We bailed out early */
2043 offset = vfs_llseek(file, 0, SEEK_CUR);
2046 free_page((unsigned long)(buf.dirent));
2049 return nfserrno(host_err);
2056 * Read entries from a directory.
2057 * The NFSv3/4 verifier we ignore for now.
2060 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2061 struct readdir_cd *cdp, filldir_t func)
2065 loff_t offset = *offsetp;
2066 int may_flags = NFSD_MAY_READ;
2068 /* NFSv2 only supports 32 bit cookies */
2069 if (rqstp->rq_vers > 2)
2070 may_flags |= NFSD_MAY_64BIT_COOKIE;
2072 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2076 offset = vfs_llseek(file, offset, 0);
2078 err = nfserrno((int)offset);
2082 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2084 if (err == nfserr_eof || err == nfserr_toosmall)
2085 err = nfs_ok; /* can still be found in ->err */
2093 * Get file system stats
2094 * N.B. After this call fhp needs an fh_put
2097 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2101 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2103 struct path path = {
2104 .mnt = fhp->fh_export->ex_path.mnt,
2105 .dentry = fhp->fh_dentry,
2107 if (vfs_statfs(&path, stat))
2113 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2115 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2119 * Check for a user's access permissions to this inode.
2122 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2123 struct dentry *dentry, int acc)
2125 struct inode *inode = dentry->d_inode;
2128 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2131 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2133 (acc & NFSD_MAY_READ)? " read" : "",
2134 (acc & NFSD_MAY_WRITE)? " write" : "",
2135 (acc & NFSD_MAY_EXEC)? " exec" : "",
2136 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2137 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2138 (acc & NFSD_MAY_LOCK)? " lock" : "",
2139 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2141 IS_IMMUTABLE(inode)? " immut" : "",
2142 IS_APPEND(inode)? " append" : "",
2143 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2144 dprintk(" owner %d/%d user %d/%d\n",
2145 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2148 /* Normally we reject any write/sattr etc access on a read-only file
2149 * system. But if it is IRIX doing check on write-access for a
2150 * device special file, we ignore rofs.
2152 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2153 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2154 if (exp_rdonly(rqstp, exp) ||
2155 __mnt_is_readonly(exp->ex_path.mnt))
2157 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2160 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2163 if (acc & NFSD_MAY_LOCK) {
2164 /* If we cannot rely on authentication in NLM requests,
2165 * just allow locks, otherwise require read permission, or
2168 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2171 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2174 * The file owner always gets access permission for accesses that
2175 * would normally be checked at open time. This is to make
2176 * file access work even when the client has done a fchmod(fd, 0).
2178 * However, `cp foo bar' should fail nevertheless when bar is
2179 * readonly. A sensible way to do this might be to reject all
2180 * attempts to truncate a read-only file, because a creat() call
2181 * always implies file truncation.
2182 * ... but this isn't really fair. A process may reasonably call
2183 * ftruncate on an open file descriptor on a file with perm 000.
2184 * We must trust the client to do permission checking - using "ACCESS"
2187 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2188 inode->i_uid == current_fsuid())
2191 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2192 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2194 /* Allow read access to binaries even when mode 111 */
2195 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2196 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2197 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2198 err = inode_permission(inode, MAY_EXEC);
2200 return err? nfserrno(err) : 0;
2204 nfsd_racache_shutdown(void)
2206 struct raparms *raparm, *last_raparm;
2209 dprintk("nfsd: freeing readahead buffers.\n");
2211 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2212 raparm = raparm_hash[i].pb_head;
2214 last_raparm = raparm;
2215 raparm = raparm->p_next;
2218 raparm_hash[i].pb_head = NULL;
2222 * Initialize readahead param cache
2225 nfsd_racache_init(int cache_size)
2230 struct raparms **raparm = NULL;
2233 if (raparm_hash[0].pb_head)
2235 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2238 cache_size = nperbucket * RAPARM_HASH_SIZE;
2240 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2242 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2243 spin_lock_init(&raparm_hash[i].pb_lock);
2245 raparm = &raparm_hash[i].pb_head;
2246 for (j = 0; j < nperbucket; j++) {
2247 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2250 raparm = &(*raparm)->p_next;
2255 nfsdstats.ra_size = cache_size;
2259 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2260 nfsd_racache_shutdown();
2264 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2266 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2268 struct inode *inode = fhp->fh_dentry->d_inode;
2272 struct posix_acl *acl;
2274 if (!IS_POSIXACL(inode))
2275 return ERR_PTR(-EOPNOTSUPP);
2278 case ACL_TYPE_ACCESS:
2279 name = POSIX_ACL_XATTR_ACCESS;
2281 case ACL_TYPE_DEFAULT:
2282 name = POSIX_ACL_XATTR_DEFAULT;
2285 return ERR_PTR(-EOPNOTSUPP);
2288 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2290 return ERR_PTR(size);
2292 acl = posix_acl_from_xattr(value, size);
2298 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2300 struct inode *inode = fhp->fh_dentry->d_inode;
2306 if (!IS_POSIXACL(inode) ||
2307 !inode->i_op->setxattr || !inode->i_op->removexattr)
2310 case ACL_TYPE_ACCESS:
2311 name = POSIX_ACL_XATTR_ACCESS;
2313 case ACL_TYPE_DEFAULT:
2314 name = POSIX_ACL_XATTR_DEFAULT;
2320 if (acl && acl->a_count) {
2321 size = posix_acl_xattr_size(acl->a_count);
2322 value = kmalloc(size, GFP_KERNEL);
2325 error = posix_acl_to_xattr(acl, value, size);
2332 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2336 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2338 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2341 error = vfs_removexattr(fhp->fh_dentry, name);
2342 if (error == -ENODATA)
2346 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2352 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */