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 * Set various file attributes.
301 * N.B. After this call fhp needs an fh_put
304 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
305 int check_guard, time_t guardtime)
307 struct dentry *dentry;
309 int accmode = NFSD_MAY_SATTR;
315 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
316 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
317 if (iap->ia_valid & ATTR_SIZE)
321 err = fh_verify(rqstp, fhp, ftype, accmode);
325 dentry = fhp->fh_dentry;
326 inode = dentry->d_inode;
328 /* Ignore any mode updates on symlinks */
329 if (S_ISLNK(inode->i_mode))
330 iap->ia_valid &= ~ATTR_MODE;
336 * NFSv2 does not differentiate between "set-[ac]time-to-now"
337 * which only requires access, and "set-[ac]time-to-X" which
338 * requires ownership.
339 * So if it looks like it might be "set both to the same time which
340 * is close to now", and if inode_change_ok fails, then we
341 * convert to "set to now" instead of "set to explicit time"
343 * We only call inode_change_ok as the last test as technically
344 * it is not an interface that we should be using. It is only
345 * valid if the filesystem does not define it's own i_op->setattr.
347 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
348 #define MAX_TOUCH_TIME_ERROR (30*60)
349 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
350 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
354 * Now just make sure time is in the right ballpark.
355 * Solaris, at least, doesn't seem to care what the time
356 * request is. We require it be within 30 minutes of now.
358 time_t delta = iap->ia_atime.tv_sec - get_seconds();
361 if (delta < MAX_TOUCH_TIME_ERROR &&
362 inode_change_ok(inode, iap) != 0) {
364 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
365 * This will cause notify_change to set these times
368 iap->ia_valid &= ~BOTH_TIME_SET;
373 * The size case is special.
374 * It changes the file as well as the attributes.
376 if (iap->ia_valid & ATTR_SIZE) {
377 if (iap->ia_size < inode->i_size) {
378 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
379 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
384 host_err = get_write_access(inode);
389 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
391 put_write_access(inode);
396 /* sanitize the mode change */
397 if (iap->ia_valid & ATTR_MODE) {
398 iap->ia_mode &= S_IALLUGO;
399 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
402 /* Revoke setuid/setgid on chown */
403 if (!S_ISDIR(inode->i_mode) &&
404 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
405 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
406 iap->ia_valid |= ATTR_KILL_PRIV;
407 if (iap->ia_valid & ATTR_MODE) {
408 /* we're setting mode too, just clear the s*id bits */
409 iap->ia_mode &= ~S_ISUID;
410 if (iap->ia_mode & S_IXGRP)
411 iap->ia_mode &= ~S_ISGID;
413 /* set ATTR_KILL_* bits and let VFS handle it */
414 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
418 /* Change the attributes. */
420 iap->ia_valid |= ATTR_CTIME;
422 err = nfserr_notsync;
423 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
424 host_err = nfsd_break_lease(inode);
429 host_err = notify_change(dentry, iap);
430 err = nfserrno(host_err);
434 put_write_access(inode);
436 commit_metadata(fhp);
441 err = nfserrno(host_err);
445 #if defined(CONFIG_NFSD_V2_ACL) || \
446 defined(CONFIG_NFSD_V3_ACL) || \
447 defined(CONFIG_NFSD_V4)
448 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
453 buflen = vfs_getxattr(dentry, key, NULL, 0);
457 *buf = kmalloc(buflen, GFP_KERNEL);
461 ret = vfs_getxattr(dentry, key, *buf, buflen);
468 #if defined(CONFIG_NFSD_V4)
470 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
477 buflen = posix_acl_xattr_size(pacl->a_count);
478 buf = kmalloc(buflen, GFP_KERNEL);
483 len = posix_acl_to_xattr(pacl, buf, buflen);
489 error = vfs_setxattr(dentry, key, buf, len, 0);
496 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
497 struct nfs4_acl *acl)
501 struct dentry *dentry;
503 struct posix_acl *pacl = NULL, *dpacl = NULL;
504 unsigned int flags = 0;
507 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
511 dentry = fhp->fh_dentry;
512 inode = dentry->d_inode;
513 if (S_ISDIR(inode->i_mode))
514 flags = NFS4_ACL_DIR;
516 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
517 if (host_error == -EINVAL) {
518 return nfserr_attrnotsupp;
519 } else if (host_error < 0)
522 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
526 if (S_ISDIR(inode->i_mode))
527 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
530 posix_acl_release(pacl);
531 posix_acl_release(dpacl);
533 if (host_error == -EOPNOTSUPP)
534 return nfserr_attrnotsupp;
536 return nfserrno(host_error);
539 static struct posix_acl *
540 _get_posix_acl(struct dentry *dentry, char *key)
543 struct posix_acl *pacl = NULL;
546 buflen = nfsd_getxattr(dentry, key, &buf);
550 return ERR_PTR(buflen);
552 pacl = posix_acl_from_xattr(buf, buflen);
558 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
560 struct inode *inode = dentry->d_inode;
562 struct posix_acl *pacl = NULL, *dpacl = NULL;
563 unsigned int flags = 0;
565 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
566 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
567 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
569 error = PTR_ERR(pacl);
574 if (S_ISDIR(inode->i_mode)) {
575 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
576 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
578 else if (IS_ERR(dpacl)) {
579 error = PTR_ERR(dpacl);
583 flags = NFS4_ACL_DIR;
586 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
588 error = PTR_ERR(*acl);
592 posix_acl_release(pacl);
593 posix_acl_release(dpacl);
598 * NFS junction information is stored in an extended attribute.
600 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
603 * nfsd4_is_junction - Test if an object could be an NFS junction
605 * @dentry: object to test
607 * Returns 1 if "dentry" appears to contain NFS junction information.
608 * Otherwise 0 is returned.
610 int nfsd4_is_junction(struct dentry *dentry)
612 struct inode *inode = dentry->d_inode;
616 if (inode->i_mode & S_IXUGO)
618 if (!(inode->i_mode & S_ISVTX))
620 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
624 #endif /* defined(CONFIG_NFSD_V4) */
626 #ifdef CONFIG_NFSD_V3
628 * Check server access rights to a file system object
634 static struct accessmap nfs3_regaccess[] = {
635 { NFS3_ACCESS_READ, NFSD_MAY_READ },
636 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
637 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
638 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
643 static struct accessmap nfs3_diraccess[] = {
644 { NFS3_ACCESS_READ, NFSD_MAY_READ },
645 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
646 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
647 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
648 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
653 static struct accessmap nfs3_anyaccess[] = {
654 /* Some clients - Solaris 2.6 at least, make an access call
655 * to the server to check for access for things like /dev/null
656 * (which really, the server doesn't care about). So
657 * We provide simple access checking for them, looking
658 * mainly at mode bits, and we make sure to ignore read-only
661 { NFS3_ACCESS_READ, NFSD_MAY_READ },
662 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
663 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
664 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
670 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
672 struct accessmap *map;
673 struct svc_export *export;
674 struct dentry *dentry;
675 u32 query, result = 0, sresult = 0;
678 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
682 export = fhp->fh_export;
683 dentry = fhp->fh_dentry;
685 if (S_ISREG(dentry->d_inode->i_mode))
686 map = nfs3_regaccess;
687 else if (S_ISDIR(dentry->d_inode->i_mode))
688 map = nfs3_diraccess;
690 map = nfs3_anyaccess;
694 for (; map->access; map++) {
695 if (map->access & query) {
698 sresult |= map->access;
700 err2 = nfsd_permission(rqstp, export, dentry, map->how);
703 result |= map->access;
706 /* the following error codes just mean the access was not allowed,
707 * rather than an error occurred */
711 /* simply don't "or" in the access bit. */
721 *supported = sresult;
726 #endif /* CONFIG_NFSD_V3 */
728 static int nfsd_open_break_lease(struct inode *inode, int access)
732 if (access & NFSD_MAY_NOT_BREAK_LEASE)
734 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
735 return break_lease(inode, mode | O_NONBLOCK);
739 * Open an existing file or directory.
740 * The may_flags argument indicates the type of open (read/write/lock)
741 * and additional flags.
742 * N.B. After this call fhp needs an fh_put
745 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
746 int may_flags, struct file **filp)
750 int flags = O_RDONLY|O_LARGEFILE;
754 validate_process_creds();
757 * If we get here, then the client has already done an "open",
758 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
759 * in case a chmod has now revoked permission.
761 * Arguably we should also allow the owner override for
762 * directories, but we never have and it doesn't seem to have
763 * caused anyone a problem. If we were to change this, note
764 * also that our filldir callbacks would need a variant of
765 * lookup_one_len that doesn't check permissions.
768 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
769 err = fh_verify(rqstp, fhp, type, may_flags);
773 path.mnt = fhp->fh_export->ex_path.mnt;
774 path.dentry = fhp->fh_dentry;
775 inode = path.dentry->d_inode;
777 /* Disallow write access to files with the append-only bit set
778 * or any access when mandatory locking enabled
781 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
784 * We must ignore files (but only files) which might have mandatory
785 * locks on them because there is no way to know if the accesser has
788 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
794 host_err = nfsd_open_break_lease(inode, may_flags);
795 if (host_err) /* NOMEM or WOULDBLOCK */
798 if (may_flags & NFSD_MAY_WRITE) {
799 if (may_flags & NFSD_MAY_READ)
800 flags = O_RDWR|O_LARGEFILE;
802 flags = O_WRONLY|O_LARGEFILE;
804 *filp = dentry_open(&path, flags, current_cred());
806 host_err = PTR_ERR(*filp);
808 host_err = ima_file_check(*filp, may_flags);
810 if (may_flags & NFSD_MAY_64BIT_COOKIE)
811 (*filp)->f_mode |= FMODE_64BITHASH;
813 (*filp)->f_mode |= FMODE_32BITHASH;
817 err = nfserrno(host_err);
819 validate_process_creds();
827 nfsd_close(struct file *filp)
833 * Obtain the readahead parameters for the file
834 * specified by (dev, ino).
837 static inline struct raparms *
838 nfsd_get_raparms(dev_t dev, ino_t ino)
840 struct raparms *ra, **rap, **frap = NULL;
843 struct raparm_hbucket *rab;
845 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
846 rab = &raparm_hash[hash];
848 spin_lock(&rab->pb_lock);
849 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
850 if (ra->p_ino == ino && ra->p_dev == dev)
853 if (ra->p_count == 0)
856 depth = nfsdstats.ra_size;
858 spin_unlock(&rab->pb_lock);
868 if (rap != &rab->pb_head) {
870 ra->p_next = rab->pb_head;
874 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
875 spin_unlock(&rab->pb_lock);
880 * Grab and keep cached pages associated with a file in the svc_rqst
881 * so that they can be passed to the network sendmsg/sendpage routines
882 * directly. They will be released after the sending has completed.
885 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
886 struct splice_desc *sd)
888 struct svc_rqst *rqstp = sd->u.data;
889 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
890 struct page *page = buf->page;
895 if (rqstp->rq_res.page_len == 0) {
900 rqstp->rq_res.page_base = buf->offset;
901 rqstp->rq_res.page_len = size;
902 } else if (page != pp[-1]) {
908 rqstp->rq_res.page_len += size;
910 rqstp->rq_res.page_len += size;
915 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
916 struct splice_desc *sd)
918 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
922 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
923 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
931 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
932 struct splice_desc sd = {
939 rqstp->rq_resused = 1;
940 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
944 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
949 nfsdstats.io_read += host_err;
952 fsnotify_access(file);
954 err = nfserrno(host_err);
958 static void kill_suid(struct dentry *dentry)
961 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
963 mutex_lock(&dentry->d_inode->i_mutex);
964 notify_change(dentry, &ia);
965 mutex_unlock(&dentry->d_inode->i_mutex);
969 * Gathered writes: If another process is currently writing to the file,
970 * there's a high chance this is another nfsd (triggered by a bulk write
971 * from a client's biod). Rather than syncing the file with each write
972 * request, we sleep for 10 msec.
974 * I don't know if this roughly approximates C. Juszak's idea of
975 * gathered writes, but it's a nice and simple solution (IMHO), and it
978 * Note: we do this only in the NFSv2 case, since v3 and higher have a
979 * better tool (separate unstable writes and commits) for solving this
982 static int wait_for_concurrent_writes(struct file *file)
984 struct inode *inode = file->f_path.dentry->d_inode;
985 static ino_t last_ino;
986 static dev_t last_dev;
989 if (atomic_read(&inode->i_writecount) > 1
990 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
991 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
993 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
996 if (inode->i_state & I_DIRTY) {
997 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
998 err = vfs_fsync(file, 0);
1000 last_ino = inode->i_ino;
1001 last_dev = inode->i_sb->s_dev;
1006 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1007 loff_t offset, struct kvec *vec, int vlen,
1008 unsigned long *cnt, int *stablep)
1010 struct svc_export *exp;
1011 struct dentry *dentry;
1012 struct inode *inode;
1016 int stable = *stablep;
1019 dentry = file->f_path.dentry;
1020 inode = dentry->d_inode;
1021 exp = fhp->fh_export;
1024 * Request sync writes if
1025 * - the sync export option has been set, or
1026 * - the client requested O_SYNC behavior (NFSv3 feature).
1027 * - The file system doesn't support fsync().
1028 * When NFSv2 gathered writes have been configured for this volume,
1029 * flushing the data to disk is handled separately below.
1031 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1033 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1035 *stablep = 2; /* FILE_SYNC */
1038 if (!EX_ISSYNC(exp))
1040 if (stable && !use_wgather) {
1041 spin_lock(&file->f_lock);
1042 file->f_flags |= O_SYNC;
1043 spin_unlock(&file->f_lock);
1046 /* Write the data. */
1047 oldfs = get_fs(); set_fs(KERNEL_DS);
1048 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1053 nfsdstats.io_write += host_err;
1054 fsnotify_modify(file);
1056 /* clear setuid/setgid flag after write */
1057 if (inode->i_mode & (S_ISUID | S_ISGID))
1060 if (stable && use_wgather)
1061 host_err = wait_for_concurrent_writes(file);
1064 dprintk("nfsd: write complete host_err=%d\n", host_err);
1068 err = nfserrno(host_err);
1073 * Read data from a file. count must contain the requested read count
1074 * on entry. On return, *count contains the number of bytes actually read.
1075 * N.B. After this call fhp needs an fh_put
1077 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1078 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1081 struct inode *inode;
1085 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1089 inode = file->f_path.dentry->d_inode;
1091 /* Get readahead parameters */
1092 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1094 if (ra && ra->p_set)
1095 file->f_ra = ra->p_ra;
1097 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1099 /* Write back readahead params */
1101 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1102 spin_lock(&rab->pb_lock);
1103 ra->p_ra = file->f_ra;
1106 spin_unlock(&rab->pb_lock);
1113 /* As above, but use the provided file descriptor. */
1115 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1116 loff_t offset, struct kvec *vec, int vlen,
1117 unsigned long *count)
1122 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1123 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1126 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1127 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1128 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1134 * Write data to a file.
1135 * The stable flag requests synchronous writes.
1136 * N.B. After this call fhp needs an fh_put
1139 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1140 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1146 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1147 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1150 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1153 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1158 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1166 #ifdef CONFIG_NFSD_V3
1168 * Commit all pending writes to stable storage.
1170 * Note: we only guarantee that data that lies within the range specified
1171 * by the 'offset' and 'count' parameters will be synced.
1173 * Unfortunately we cannot lock the file to make sure we return full WCC
1174 * data to the client, as locking happens lower down in the filesystem.
1177 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1178 loff_t offset, unsigned long count)
1181 loff_t end = LLONG_MAX;
1182 __be32 err = nfserr_inval;
1187 end = offset + (loff_t)count - 1;
1192 err = nfsd_open(rqstp, fhp, S_IFREG,
1193 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1196 if (EX_ISSYNC(fhp->fh_export)) {
1197 int err2 = vfs_fsync_range(file, offset, end, 0);
1199 if (err2 != -EINVAL)
1200 err = nfserrno(err2);
1202 err = nfserr_notsupp;
1209 #endif /* CONFIG_NFSD_V3 */
1212 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1216 * Mode has already been set earlier in create:
1218 iap->ia_valid &= ~ATTR_MODE;
1220 * Setting uid/gid works only for root. Irix appears to
1221 * send along the gid on create when it tries to implement
1222 * setgid directories via NFS:
1224 if (current_fsuid() != 0)
1225 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1227 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1231 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1232 * setting size to 0 may fail for some specific file systems by the permission
1233 * checking which requires WRITE permission but the mode is 000.
1234 * we ignore the resizing(to 0) on the just new created file, since the size is
1235 * 0 after file created.
1237 * call this only after vfs_create() is called.
1240 nfsd_check_ignore_resizing(struct iattr *iap)
1242 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1243 iap->ia_valid &= ~ATTR_SIZE;
1247 * Create a file (regular, directory, device, fifo); UNIX sockets
1248 * not yet implemented.
1249 * If the response fh has been verified, the parent directory should
1250 * already be locked. Note that the parent directory is left locked.
1252 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1255 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1256 char *fname, int flen, struct iattr *iap,
1257 int type, dev_t rdev, struct svc_fh *resfhp)
1259 struct dentry *dentry, *dchild = NULL;
1269 if (isdotent(fname, flen))
1272 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1276 dentry = fhp->fh_dentry;
1277 dirp = dentry->d_inode;
1279 err = nfserr_notdir;
1280 if (!dirp->i_op->lookup)
1283 * Check whether the response file handle has been verified yet.
1284 * If it has, the parent directory should already be locked.
1286 if (!resfhp->fh_dentry) {
1287 host_err = fh_want_write(fhp);
1291 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1292 fh_lock_nested(fhp, I_MUTEX_PARENT);
1293 dchild = lookup_one_len(fname, dentry, flen);
1294 host_err = PTR_ERR(dchild);
1297 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1301 /* called from nfsd_proc_create */
1302 dchild = dget(resfhp->fh_dentry);
1303 if (!fhp->fh_locked) {
1304 /* not actually possible */
1306 "nfsd_create: parent %s/%s not locked!\n",
1307 dentry->d_parent->d_name.name,
1308 dentry->d_name.name);
1314 * Make sure the child dentry is still negative ...
1317 if (dchild->d_inode) {
1318 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1319 dentry->d_name.name, dchild->d_name.name);
1323 if (!(iap->ia_valid & ATTR_MODE))
1325 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1328 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1329 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1335 * Get the dir op function pointer.
1341 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1343 nfsd_check_ignore_resizing(iap);
1346 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1352 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1358 err = nfsd_create_setattr(rqstp, resfhp, iap);
1361 * nfsd_setattr already committed the child. Transactional filesystems
1362 * had a chance to commit changes for both parent and child
1363 * simultaneously making the following commit_metadata a noop.
1365 err2 = nfserrno(commit_metadata(fhp));
1369 * Update the file handle to get the new inode info.
1372 err = fh_update(resfhp);
1374 if (dchild && !IS_ERR(dchild))
1379 err = nfserrno(host_err);
1383 #ifdef CONFIG_NFSD_V3
1385 static inline int nfsd_create_is_exclusive(int createmode)
1387 return createmode == NFS3_CREATE_EXCLUSIVE
1388 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1392 * NFSv3 and NFSv4 version of nfsd_create
1395 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1396 char *fname, int flen, struct iattr *iap,
1397 struct svc_fh *resfhp, int createmode, u32 *verifier,
1398 bool *truncp, bool *created)
1400 struct dentry *dentry, *dchild = NULL;
1404 __u32 v_mtime=0, v_atime=0;
1410 if (isdotent(fname, flen))
1412 if (!(iap->ia_valid & ATTR_MODE))
1414 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1418 dentry = fhp->fh_dentry;
1419 dirp = dentry->d_inode;
1421 /* Get all the sanity checks out of the way before
1422 * we lock the parent. */
1423 err = nfserr_notdir;
1424 if (!dirp->i_op->lookup)
1427 host_err = fh_want_write(fhp);
1431 fh_lock_nested(fhp, I_MUTEX_PARENT);
1434 * Compose the response file handle.
1436 dchild = lookup_one_len(fname, dentry, flen);
1437 host_err = PTR_ERR(dchild);
1441 /* If file doesn't exist, check for permissions to create one */
1442 if (!dchild->d_inode) {
1443 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1448 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1452 if (nfsd_create_is_exclusive(createmode)) {
1453 /* solaris7 gets confused (bugid 4218508) if these have
1454 * the high bit set, so just clear the high bits. If this is
1455 * ever changed to use different attrs for storing the
1456 * verifier, then do_open_lookup() will also need to be fixed
1459 v_mtime = verifier[0]&0x7fffffff;
1460 v_atime = verifier[1]&0x7fffffff;
1463 if (dchild->d_inode) {
1466 switch (createmode) {
1467 case NFS3_CREATE_UNCHECKED:
1468 if (! S_ISREG(dchild->d_inode->i_mode))
1471 /* in nfsv4, we need to treat this case a little
1472 * differently. we don't want to truncate the
1473 * file now; this would be wrong if the OPEN
1474 * fails for some other reason. furthermore,
1475 * if the size is nonzero, we should ignore it
1476 * according to spec!
1478 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1481 iap->ia_valid &= ATTR_SIZE;
1485 case NFS3_CREATE_EXCLUSIVE:
1486 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1487 && dchild->d_inode->i_atime.tv_sec == v_atime
1488 && dchild->d_inode->i_size == 0 )
1490 case NFS4_CREATE_EXCLUSIVE4_1:
1491 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1492 && dchild->d_inode->i_atime.tv_sec == v_atime
1493 && dchild->d_inode->i_size == 0 )
1496 case NFS3_CREATE_GUARDED:
1503 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1511 nfsd_check_ignore_resizing(iap);
1513 if (nfsd_create_is_exclusive(createmode)) {
1514 /* Cram the verifier into atime/mtime */
1515 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1516 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1517 /* XXX someone who knows this better please fix it for nsec */
1518 iap->ia_mtime.tv_sec = v_mtime;
1519 iap->ia_atime.tv_sec = v_atime;
1520 iap->ia_mtime.tv_nsec = 0;
1521 iap->ia_atime.tv_nsec = 0;
1525 err = nfsd_create_setattr(rqstp, resfhp, iap);
1528 * nfsd_setattr already committed the child (and possibly also the parent).
1531 err = nfserrno(commit_metadata(fhp));
1534 * Update the filehandle to get the new inode info.
1537 err = fh_update(resfhp);
1541 if (dchild && !IS_ERR(dchild))
1547 err = nfserrno(host_err);
1550 #endif /* CONFIG_NFSD_V3 */
1553 * Read a symlink. On entry, *lenp must contain the maximum path length that
1554 * fits into the buffer. On return, it contains the true length.
1555 * N.B. After this call fhp needs an fh_put
1558 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1560 struct inode *inode;
1566 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1570 path.mnt = fhp->fh_export->ex_path.mnt;
1571 path.dentry = fhp->fh_dentry;
1572 inode = path.dentry->d_inode;
1575 if (!inode->i_op->readlink)
1579 /* N.B. Why does this call need a get_fs()??
1580 * Remove the set_fs and watch the fireworks:-) --okir
1583 oldfs = get_fs(); set_fs(KERNEL_DS);
1584 host_err = inode->i_op->readlink(path.dentry, buf, *lenp);
1595 err = nfserrno(host_err);
1600 * Create a symlink and look up its inode
1601 * N.B. After this call _both_ fhp and resfhp need an fh_put
1604 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1605 char *fname, int flen,
1606 char *path, int plen,
1607 struct svc_fh *resfhp,
1610 struct dentry *dentry, *dnew;
1618 if (isdotent(fname, flen))
1621 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1625 host_err = fh_want_write(fhp);
1630 dentry = fhp->fh_dentry;
1631 dnew = lookup_one_len(fname, dentry, flen);
1632 host_err = PTR_ERR(dnew);
1636 if (unlikely(path[plen] != 0)) {
1637 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1638 if (path_alloced == NULL)
1641 strncpy(path_alloced, path, plen);
1642 path_alloced[plen] = 0;
1643 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1644 kfree(path_alloced);
1647 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1648 err = nfserrno(host_err);
1650 err = nfserrno(commit_metadata(fhp));
1655 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1657 if (err==0) err = cerr;
1662 err = nfserrno(host_err);
1668 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1671 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1672 char *name, int len, struct svc_fh *tfhp)
1674 struct dentry *ddir, *dnew, *dold;
1679 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1682 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1686 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1692 if (isdotent(name, len))
1695 host_err = fh_want_write(tfhp);
1697 err = nfserrno(host_err);
1701 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1702 ddir = ffhp->fh_dentry;
1703 dirp = ddir->d_inode;
1705 dnew = lookup_one_len(name, ddir, len);
1706 host_err = PTR_ERR(dnew);
1710 dold = tfhp->fh_dentry;
1715 host_err = nfsd_break_lease(dold->d_inode);
1717 err = nfserrno(host_err);
1720 host_err = vfs_link(dold, dirp, dnew);
1722 err = nfserrno(commit_metadata(ffhp));
1724 err = nfserrno(commit_metadata(tfhp));
1726 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1729 err = nfserrno(host_err);
1735 fh_drop_write(tfhp);
1740 err = nfserrno(host_err);
1746 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1749 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1750 struct svc_fh *tfhp, char *tname, int tlen)
1752 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1753 struct inode *fdir, *tdir;
1757 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1760 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1764 fdentry = ffhp->fh_dentry;
1765 fdir = fdentry->d_inode;
1767 tdentry = tfhp->fh_dentry;
1768 tdir = tdentry->d_inode;
1770 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1771 if (ffhp->fh_export != tfhp->fh_export)
1775 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1778 host_err = fh_want_write(ffhp);
1780 err = nfserrno(host_err);
1784 /* cannot use fh_lock as we need deadlock protective ordering
1785 * so do it by hand */
1786 trap = lock_rename(tdentry, fdentry);
1787 ffhp->fh_locked = tfhp->fh_locked = 1;
1791 odentry = lookup_one_len(fname, fdentry, flen);
1792 host_err = PTR_ERR(odentry);
1793 if (IS_ERR(odentry))
1797 if (!odentry->d_inode)
1800 if (odentry == trap)
1803 ndentry = lookup_one_len(tname, tdentry, tlen);
1804 host_err = PTR_ERR(ndentry);
1805 if (IS_ERR(ndentry))
1807 host_err = -ENOTEMPTY;
1808 if (ndentry == trap)
1812 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1815 host_err = nfsd_break_lease(odentry->d_inode);
1818 if (ndentry->d_inode) {
1819 host_err = nfsd_break_lease(ndentry->d_inode);
1823 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1825 host_err = commit_metadata(tfhp);
1827 host_err = commit_metadata(ffhp);
1834 err = nfserrno(host_err);
1836 /* we cannot reply on fh_unlock on the two filehandles,
1837 * as that would do the wrong thing if the two directories
1838 * were the same, so again we do it by hand
1840 fill_post_wcc(ffhp);
1841 fill_post_wcc(tfhp);
1842 unlock_rename(tdentry, fdentry);
1843 ffhp->fh_locked = tfhp->fh_locked = 0;
1844 fh_drop_write(ffhp);
1851 * Unlink a file or directory
1852 * N.B. After this call fhp needs an fh_put
1855 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1856 char *fname, int flen)
1858 struct dentry *dentry, *rdentry;
1864 if (!flen || isdotent(fname, flen))
1866 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1870 host_err = fh_want_write(fhp);
1874 fh_lock_nested(fhp, I_MUTEX_PARENT);
1875 dentry = fhp->fh_dentry;
1876 dirp = dentry->d_inode;
1878 rdentry = lookup_one_len(fname, dentry, flen);
1879 host_err = PTR_ERR(rdentry);
1880 if (IS_ERR(rdentry))
1883 if (!rdentry->d_inode) {
1890 type = rdentry->d_inode->i_mode & S_IFMT;
1892 host_err = nfsd_break_lease(rdentry->d_inode);
1895 if (type != S_IFDIR)
1896 host_err = vfs_unlink(dirp, rdentry);
1898 host_err = vfs_rmdir(dirp, rdentry);
1900 host_err = commit_metadata(fhp);
1905 err = nfserrno(host_err);
1911 * We do this buffering because we must not call back into the file
1912 * system's ->lookup() method from the filldir callback. That may well
1913 * deadlock a number of file systems.
1915 * This is based heavily on the implementation of same in XFS.
1917 struct buffered_dirent {
1921 unsigned int d_type;
1925 struct readdir_data {
1931 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1932 loff_t offset, u64 ino, unsigned int d_type)
1934 struct readdir_data *buf = __buf;
1935 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1936 unsigned int reclen;
1938 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1939 if (buf->used + reclen > PAGE_SIZE) {
1944 de->namlen = namlen;
1945 de->offset = offset;
1947 de->d_type = d_type;
1948 memcpy(de->name, name, namlen);
1949 buf->used += reclen;
1954 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1955 struct readdir_cd *cdp, loff_t *offsetp)
1957 struct readdir_data buf;
1958 struct buffered_dirent *de;
1963 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1965 return nfserrno(-ENOMEM);
1970 struct inode *dir_inode = file->f_path.dentry->d_inode;
1971 unsigned int reclen;
1973 cdp->err = nfserr_eof; /* will be cleared on successful read */
1977 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1990 * Various filldir functions may end up calling back into
1991 * lookup_one_len() and the file system's ->lookup() method.
1992 * These expect i_mutex to be held, as it would within readdir.
1994 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1998 de = (struct buffered_dirent *)buf.dirent;
2000 offset = de->offset;
2002 if (func(cdp, de->name, de->namlen, de->offset,
2003 de->ino, de->d_type))
2006 if (cdp->err != nfs_ok)
2009 reclen = ALIGN(sizeof(*de) + de->namlen,
2012 de = (struct buffered_dirent *)((char *)de + reclen);
2014 mutex_unlock(&dir_inode->i_mutex);
2015 if (size > 0) /* We bailed out early */
2018 offset = vfs_llseek(file, 0, SEEK_CUR);
2021 free_page((unsigned long)(buf.dirent));
2024 return nfserrno(host_err);
2031 * Read entries from a directory.
2032 * The NFSv3/4 verifier we ignore for now.
2035 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2036 struct readdir_cd *cdp, filldir_t func)
2040 loff_t offset = *offsetp;
2041 int may_flags = NFSD_MAY_READ;
2043 /* NFSv2 only supports 32 bit cookies */
2044 if (rqstp->rq_vers > 2)
2045 may_flags |= NFSD_MAY_64BIT_COOKIE;
2047 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2051 offset = vfs_llseek(file, offset, SEEK_SET);
2053 err = nfserrno((int)offset);
2057 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2059 if (err == nfserr_eof || err == nfserr_toosmall)
2060 err = nfs_ok; /* can still be found in ->err */
2068 * Get file system stats
2069 * N.B. After this call fhp needs an fh_put
2072 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2076 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2078 struct path path = {
2079 .mnt = fhp->fh_export->ex_path.mnt,
2080 .dentry = fhp->fh_dentry,
2082 if (vfs_statfs(&path, stat))
2088 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2090 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2094 * Check for a user's access permissions to this inode.
2097 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2098 struct dentry *dentry, int acc)
2100 struct inode *inode = dentry->d_inode;
2103 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2106 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2108 (acc & NFSD_MAY_READ)? " read" : "",
2109 (acc & NFSD_MAY_WRITE)? " write" : "",
2110 (acc & NFSD_MAY_EXEC)? " exec" : "",
2111 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2112 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2113 (acc & NFSD_MAY_LOCK)? " lock" : "",
2114 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2116 IS_IMMUTABLE(inode)? " immut" : "",
2117 IS_APPEND(inode)? " append" : "",
2118 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2119 dprintk(" owner %d/%d user %d/%d\n",
2120 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2123 /* Normally we reject any write/sattr etc access on a read-only file
2124 * system. But if it is IRIX doing check on write-access for a
2125 * device special file, we ignore rofs.
2127 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2128 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2129 if (exp_rdonly(rqstp, exp) ||
2130 __mnt_is_readonly(exp->ex_path.mnt))
2132 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2135 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2138 if (acc & NFSD_MAY_LOCK) {
2139 /* If we cannot rely on authentication in NLM requests,
2140 * just allow locks, otherwise require read permission, or
2143 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2146 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2149 * The file owner always gets access permission for accesses that
2150 * would normally be checked at open time. This is to make
2151 * file access work even when the client has done a fchmod(fd, 0).
2153 * However, `cp foo bar' should fail nevertheless when bar is
2154 * readonly. A sensible way to do this might be to reject all
2155 * attempts to truncate a read-only file, because a creat() call
2156 * always implies file truncation.
2157 * ... but this isn't really fair. A process may reasonably call
2158 * ftruncate on an open file descriptor on a file with perm 000.
2159 * We must trust the client to do permission checking - using "ACCESS"
2162 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2163 inode->i_uid == current_fsuid())
2166 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2167 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2169 /* Allow read access to binaries even when mode 111 */
2170 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2171 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2172 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2173 err = inode_permission(inode, MAY_EXEC);
2175 return err? nfserrno(err) : 0;
2179 nfsd_racache_shutdown(void)
2181 struct raparms *raparm, *last_raparm;
2184 dprintk("nfsd: freeing readahead buffers.\n");
2186 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2187 raparm = raparm_hash[i].pb_head;
2189 last_raparm = raparm;
2190 raparm = raparm->p_next;
2193 raparm_hash[i].pb_head = NULL;
2197 * Initialize readahead param cache
2200 nfsd_racache_init(int cache_size)
2205 struct raparms **raparm = NULL;
2208 if (raparm_hash[0].pb_head)
2210 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2213 cache_size = nperbucket * RAPARM_HASH_SIZE;
2215 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2217 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2218 spin_lock_init(&raparm_hash[i].pb_lock);
2220 raparm = &raparm_hash[i].pb_head;
2221 for (j = 0; j < nperbucket; j++) {
2222 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2225 raparm = &(*raparm)->p_next;
2230 nfsdstats.ra_size = cache_size;
2234 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2235 nfsd_racache_shutdown();
2239 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2241 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2243 struct inode *inode = fhp->fh_dentry->d_inode;
2247 struct posix_acl *acl;
2249 if (!IS_POSIXACL(inode))
2250 return ERR_PTR(-EOPNOTSUPP);
2253 case ACL_TYPE_ACCESS:
2254 name = POSIX_ACL_XATTR_ACCESS;
2256 case ACL_TYPE_DEFAULT:
2257 name = POSIX_ACL_XATTR_DEFAULT;
2260 return ERR_PTR(-EOPNOTSUPP);
2263 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2265 return ERR_PTR(size);
2267 acl = posix_acl_from_xattr(value, size);
2273 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2275 struct inode *inode = fhp->fh_dentry->d_inode;
2281 if (!IS_POSIXACL(inode) ||
2282 !inode->i_op->setxattr || !inode->i_op->removexattr)
2285 case ACL_TYPE_ACCESS:
2286 name = POSIX_ACL_XATTR_ACCESS;
2288 case ACL_TYPE_DEFAULT:
2289 name = POSIX_ACL_XATTR_DEFAULT;
2295 if (acl && acl->a_count) {
2296 size = posix_acl_xattr_size(acl->a_count);
2297 value = kmalloc(size, GFP_KERNEL);
2300 error = posix_acl_to_xattr(acl, value, size);
2307 error = fh_want_write(fhp);
2311 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2313 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2316 error = vfs_removexattr(fhp->fh_dentry, name);
2317 if (error == -ENODATA)
2327 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */