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 while (d_mountpoint(path.dentry) && follow_down(&path))
93 exp2 = rqst_exp_get_by_name(rqstp, &path);
97 * We normally allow NFS clients to continue
98 * "underneath" a mountpoint that is not exported.
99 * The exception is V4ROOT, where no traversal is ever
100 * allowed without an explicit export of the new
103 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
108 if (nfsd_v4client(rqstp) ||
109 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
110 /* successfully crossed mount point */
112 * This is subtle: path.dentry is *not* on path.mnt
113 * at this point. The only reason we are safe is that
114 * original mnt is pinned down by exp, so we should
115 * put path *before* putting exp
118 path.dentry = dentry;
128 static void follow_to_parent(struct path *path)
132 while (path->dentry == path->mnt->mnt_root && follow_up(path))
134 dp = dget_parent(path->dentry);
139 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
141 struct svc_export *exp2;
142 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
143 .dentry = dget(dparent)};
145 follow_to_parent(&path);
147 exp2 = rqst_exp_parent(rqstp, &path);
148 if (PTR_ERR(exp2) == -ENOENT) {
149 *dentryp = dget(dparent);
150 } else if (IS_ERR(exp2)) {
152 return PTR_ERR(exp2);
154 *dentryp = dget(path.dentry);
163 * For nfsd purposes, we treat V4ROOT exports as though there was an
164 * export at *every* directory.
166 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
168 if (d_mountpoint(dentry))
170 if (!(exp->ex_flags & NFSEXP_V4ROOT))
172 return dentry->d_inode != NULL;
176 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
177 const char *name, unsigned int len,
178 struct svc_export **exp_ret, struct dentry **dentry_ret)
180 struct svc_export *exp;
181 struct dentry *dparent;
182 struct dentry *dentry;
186 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
188 /* Obtain dentry and export. */
189 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
193 dparent = fhp->fh_dentry;
194 exp = fhp->fh_export;
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name, len)) {
200 dentry = dget(dparent);
201 else if (dparent != exp->ex_path.dentry)
202 dentry = dget_parent(dparent);
203 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
204 dentry = dget(dparent); /* .. == . just like at / */
206 /* checking mountpoint crossing is very different when stepping up */
207 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
213 dentry = lookup_one_len(name, dparent, len);
214 host_err = PTR_ERR(dentry);
218 * check if we have crossed a mount point ...
220 if (nfsd_mountpoint(dentry, exp)) {
221 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
227 *dentry_ret = dentry;
233 return nfserrno(host_err);
237 * Look up one component of a pathname.
238 * N.B. After this call _both_ fhp and resfh need an fh_put
240 * If the lookup would cross a mountpoint, and the mounted filesystem
241 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
242 * accepted as it stands and the mounted directory is
243 * returned. Otherwise the covered directory is returned.
244 * NOTE: this mountpoint crossing is not supported properly by all
245 * clients and is explicitly disallowed for NFSv3
246 * NeilBrown <neilb@cse.unsw.edu.au>
249 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
250 unsigned int len, struct svc_fh *resfh)
252 struct svc_export *exp;
253 struct dentry *dentry;
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 /* S_IFREG */, 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);
597 #endif /* defined(CONFIG_NFSD_V4) */
599 #ifdef CONFIG_NFSD_V3
601 * Check server access rights to a file system object
607 static struct accessmap nfs3_regaccess[] = {
608 { NFS3_ACCESS_READ, NFSD_MAY_READ },
609 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
610 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
611 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
616 static struct accessmap nfs3_diraccess[] = {
617 { NFS3_ACCESS_READ, NFSD_MAY_READ },
618 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
619 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
620 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
621 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
626 static struct accessmap nfs3_anyaccess[] = {
627 /* Some clients - Solaris 2.6 at least, make an access call
628 * to the server to check for access for things like /dev/null
629 * (which really, the server doesn't care about). So
630 * We provide simple access checking for them, looking
631 * mainly at mode bits, and we make sure to ignore read-only
634 { NFS3_ACCESS_READ, NFSD_MAY_READ },
635 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
636 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
637 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
643 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
645 struct accessmap *map;
646 struct svc_export *export;
647 struct dentry *dentry;
648 u32 query, result = 0, sresult = 0;
651 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
655 export = fhp->fh_export;
656 dentry = fhp->fh_dentry;
658 if (S_ISREG(dentry->d_inode->i_mode))
659 map = nfs3_regaccess;
660 else if (S_ISDIR(dentry->d_inode->i_mode))
661 map = nfs3_diraccess;
663 map = nfs3_anyaccess;
667 for (; map->access; map++) {
668 if (map->access & query) {
671 sresult |= map->access;
673 err2 = nfsd_permission(rqstp, export, dentry, map->how);
676 result |= map->access;
679 /* the following error codes just mean the access was not allowed,
680 * rather than an error occurred */
684 /* simply don't "or" in the access bit. */
694 *supported = sresult;
699 #endif /* CONFIG_NFSD_V3 */
704 * Open an existing file or directory.
705 * The access argument indicates the type of open (read/write/lock)
706 * N.B. After this call fhp needs an fh_put
709 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
710 int access, struct file **filp)
712 struct dentry *dentry;
714 int flags = O_RDONLY|O_LARGEFILE;
718 validate_process_creds();
721 * If we get here, then the client has already done an "open",
722 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
723 * in case a chmod has now revoked permission.
725 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
729 dentry = fhp->fh_dentry;
730 inode = dentry->d_inode;
732 /* Disallow write access to files with the append-only bit set
733 * or any access when mandatory locking enabled
736 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
739 * We must ignore files (but only files) which might have mandatory
740 * locks on them because there is no way to know if the accesser has
743 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
750 * Check to see if there are any leases on this file.
751 * This may block while leases are broken.
753 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
754 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
755 if (host_err) /* NOMEM or WOULDBLOCK */
758 if (access & NFSD_MAY_WRITE) {
759 if (access & NFSD_MAY_READ)
760 flags = O_RDWR|O_LARGEFILE;
762 flags = O_WRONLY|O_LARGEFILE;
764 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
765 flags, current_cred());
767 host_err = PTR_ERR(*filp);
769 host_err = ima_file_check(*filp, access);
771 err = nfserrno(host_err);
773 validate_process_creds();
781 nfsd_close(struct file *filp)
787 * Obtain the readahead parameters for the file
788 * specified by (dev, ino).
791 static inline struct raparms *
792 nfsd_get_raparms(dev_t dev, ino_t ino)
794 struct raparms *ra, **rap, **frap = NULL;
797 struct raparm_hbucket *rab;
799 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
800 rab = &raparm_hash[hash];
802 spin_lock(&rab->pb_lock);
803 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
804 if (ra->p_ino == ino && ra->p_dev == dev)
807 if (ra->p_count == 0)
810 depth = nfsdstats.ra_size*11/10;
812 spin_unlock(&rab->pb_lock);
822 if (rap != &rab->pb_head) {
824 ra->p_next = rab->pb_head;
828 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
829 spin_unlock(&rab->pb_lock);
834 * Grab and keep cached pages associated with a file in the svc_rqst
835 * so that they can be passed to the network sendmsg/sendpage routines
836 * directly. They will be released after the sending has completed.
839 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
840 struct splice_desc *sd)
842 struct svc_rqst *rqstp = sd->u.data;
843 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
844 struct page *page = buf->page;
849 if (rqstp->rq_res.page_len == 0) {
854 rqstp->rq_res.page_base = buf->offset;
855 rqstp->rq_res.page_len = size;
856 } else if (page != pp[-1]) {
862 rqstp->rq_res.page_len += size;
864 rqstp->rq_res.page_len += size;
869 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
870 struct splice_desc *sd)
872 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
876 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
877 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
885 inode = file->f_path.dentry->d_inode;
887 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
888 struct splice_desc sd = {
895 rqstp->rq_resused = 1;
896 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
900 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
905 nfsdstats.io_read += host_err;
908 fsnotify_access(file);
910 err = nfserrno(host_err);
914 static void kill_suid(struct dentry *dentry)
917 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
919 mutex_lock(&dentry->d_inode->i_mutex);
920 notify_change(dentry, &ia);
921 mutex_unlock(&dentry->d_inode->i_mutex);
925 * Gathered writes: If another process is currently writing to the file,
926 * there's a high chance this is another nfsd (triggered by a bulk write
927 * from a client's biod). Rather than syncing the file with each write
928 * request, we sleep for 10 msec.
930 * I don't know if this roughly approximates C. Juszak's idea of
931 * gathered writes, but it's a nice and simple solution (IMHO), and it
934 * Note: we do this only in the NFSv2 case, since v3 and higher have a
935 * better tool (separate unstable writes and commits) for solving this
938 static int wait_for_concurrent_writes(struct file *file)
940 struct inode *inode = file->f_path.dentry->d_inode;
941 static ino_t last_ino;
942 static dev_t last_dev;
945 if (atomic_read(&inode->i_writecount) > 1
946 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
947 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
949 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
952 if (inode->i_state & I_DIRTY) {
953 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
954 err = vfs_fsync(file, 0);
956 last_ino = inode->i_ino;
957 last_dev = inode->i_sb->s_dev;
962 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
963 loff_t offset, struct kvec *vec, int vlen,
964 unsigned long *cnt, int *stablep)
966 struct svc_export *exp;
967 struct dentry *dentry;
972 int stable = *stablep;
975 dentry = file->f_path.dentry;
976 inode = dentry->d_inode;
977 exp = fhp->fh_export;
980 * Request sync writes if
981 * - the sync export option has been set, or
982 * - the client requested O_SYNC behavior (NFSv3 feature).
983 * - The file system doesn't support fsync().
984 * When NFSv2 gathered writes have been configured for this volume,
985 * flushing the data to disk is handled separately below.
987 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
989 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
991 *stablep = 2; /* FILE_SYNC */
996 if (stable && !use_wgather) {
997 spin_lock(&file->f_lock);
998 file->f_flags |= O_SYNC;
999 spin_unlock(&file->f_lock);
1002 /* Write the data. */
1003 oldfs = get_fs(); set_fs(KERNEL_DS);
1004 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1009 nfsdstats.io_write += host_err;
1010 fsnotify_modify(file);
1012 /* clear setuid/setgid flag after write */
1013 if (inode->i_mode & (S_ISUID | S_ISGID))
1016 if (stable && use_wgather)
1017 host_err = wait_for_concurrent_writes(file);
1020 dprintk("nfsd: write complete host_err=%d\n", host_err);
1024 err = nfserrno(host_err);
1029 * Read data from a file. count must contain the requested read count
1030 * on entry. On return, *count contains the number of bytes actually read.
1031 * N.B. After this call fhp needs an fh_put
1033 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1034 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1037 struct inode *inode;
1041 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1045 inode = file->f_path.dentry->d_inode;
1047 /* Get readahead parameters */
1048 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1050 if (ra && ra->p_set)
1051 file->f_ra = ra->p_ra;
1053 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1055 /* Write back readahead params */
1057 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1058 spin_lock(&rab->pb_lock);
1059 ra->p_ra = file->f_ra;
1062 spin_unlock(&rab->pb_lock);
1069 /* As above, but use the provided file descriptor. */
1071 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1072 loff_t offset, struct kvec *vec, int vlen,
1073 unsigned long *count)
1078 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1079 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1082 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1083 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1084 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1090 * Write data to a file.
1091 * The stable flag requests synchronous writes.
1092 * N.B. After this call fhp needs an fh_put
1095 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1096 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1102 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1103 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1106 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1109 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1114 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1122 #ifdef CONFIG_NFSD_V3
1124 * Commit all pending writes to stable storage.
1126 * Note: we only guarantee that data that lies within the range specified
1127 * by the 'offset' and 'count' parameters will be synced.
1129 * Unfortunately we cannot lock the file to make sure we return full WCC
1130 * data to the client, as locking happens lower down in the filesystem.
1133 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1134 loff_t offset, unsigned long count)
1137 loff_t end = LLONG_MAX;
1138 __be32 err = nfserr_inval;
1143 end = offset + (loff_t)count - 1;
1148 err = nfsd_open(rqstp, fhp, S_IFREG,
1149 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1152 if (EX_ISSYNC(fhp->fh_export)) {
1153 int err2 = vfs_fsync_range(file, offset, end, 0);
1155 if (err2 != -EINVAL)
1156 err = nfserrno(err2);
1158 err = nfserr_notsupp;
1165 #endif /* CONFIG_NFSD_V3 */
1168 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1172 * Mode has already been set earlier in create:
1174 iap->ia_valid &= ~ATTR_MODE;
1176 * Setting uid/gid works only for root. Irix appears to
1177 * send along the gid on create when it tries to implement
1178 * setgid directories via NFS:
1180 if (current_fsuid() != 0)
1181 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1183 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1187 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1188 * setting size to 0 may fail for some specific file systems by the permission
1189 * checking which requires WRITE permission but the mode is 000.
1190 * we ignore the resizing(to 0) on the just new created file, since the size is
1191 * 0 after file created.
1193 * call this only after vfs_create() is called.
1196 nfsd_check_ignore_resizing(struct iattr *iap)
1198 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1199 iap->ia_valid &= ~ATTR_SIZE;
1203 * Create a file (regular, directory, device, fifo); UNIX sockets
1204 * not yet implemented.
1205 * If the response fh has been verified, the parent directory should
1206 * already be locked. Note that the parent directory is left locked.
1208 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1211 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1212 char *fname, int flen, struct iattr *iap,
1213 int type, dev_t rdev, struct svc_fh *resfhp)
1215 struct dentry *dentry, *dchild = NULL;
1225 if (isdotent(fname, flen))
1228 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1232 dentry = fhp->fh_dentry;
1233 dirp = dentry->d_inode;
1235 err = nfserr_notdir;
1236 if (!dirp->i_op->lookup)
1239 * Check whether the response file handle has been verified yet.
1240 * If it has, the parent directory should already be locked.
1242 if (!resfhp->fh_dentry) {
1243 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1244 fh_lock_nested(fhp, I_MUTEX_PARENT);
1245 dchild = lookup_one_len(fname, dentry, flen);
1246 host_err = PTR_ERR(dchild);
1249 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1253 /* called from nfsd_proc_create */
1254 dchild = dget(resfhp->fh_dentry);
1255 if (!fhp->fh_locked) {
1256 /* not actually possible */
1258 "nfsd_create: parent %s/%s not locked!\n",
1259 dentry->d_parent->d_name.name,
1260 dentry->d_name.name);
1266 * Make sure the child dentry is still negative ...
1269 if (dchild->d_inode) {
1270 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1271 dentry->d_name.name, dchild->d_name.name);
1275 if (!(iap->ia_valid & ATTR_MODE))
1277 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1280 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1281 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1286 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1291 * Get the dir op function pointer.
1296 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1298 nfsd_check_ignore_resizing(iap);
1301 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1307 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1311 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1315 err = nfsd_create_setattr(rqstp, resfhp, iap);
1318 * nfsd_setattr already committed the child. Transactional filesystems
1319 * had a chance to commit changes for both parent and child
1320 * simultaneously making the following commit_metadata a noop.
1322 err2 = nfserrno(commit_metadata(fhp));
1325 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1327 * Update the file handle to get the new inode info.
1330 err = fh_update(resfhp);
1332 if (dchild && !IS_ERR(dchild))
1337 err = nfserrno(host_err);
1341 #ifdef CONFIG_NFSD_V3
1343 * NFSv3 version of nfsd_create
1346 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1347 char *fname, int flen, struct iattr *iap,
1348 struct svc_fh *resfhp, int createmode, u32 *verifier,
1349 int *truncp, int *created)
1351 struct dentry *dentry, *dchild = NULL;
1355 __u32 v_mtime=0, v_atime=0;
1361 if (isdotent(fname, flen))
1363 if (!(iap->ia_valid & ATTR_MODE))
1365 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1369 dentry = fhp->fh_dentry;
1370 dirp = dentry->d_inode;
1372 /* Get all the sanity checks out of the way before
1373 * we lock the parent. */
1374 err = nfserr_notdir;
1375 if (!dirp->i_op->lookup)
1377 fh_lock_nested(fhp, I_MUTEX_PARENT);
1380 * Compose the response file handle.
1382 dchild = lookup_one_len(fname, dentry, flen);
1383 host_err = PTR_ERR(dchild);
1387 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1391 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1392 /* solaris7 gets confused (bugid 4218508) if these have
1393 * the high bit set, so just clear the high bits. If this is
1394 * ever changed to use different attrs for storing the
1395 * verifier, then do_open_lookup() will also need to be fixed
1398 v_mtime = verifier[0]&0x7fffffff;
1399 v_atime = verifier[1]&0x7fffffff;
1402 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1405 if (dchild->d_inode) {
1408 switch (createmode) {
1409 case NFS3_CREATE_UNCHECKED:
1410 if (! S_ISREG(dchild->d_inode->i_mode))
1413 /* in nfsv4, we need to treat this case a little
1414 * differently. we don't want to truncate the
1415 * file now; this would be wrong if the OPEN
1416 * fails for some other reason. furthermore,
1417 * if the size is nonzero, we should ignore it
1418 * according to spec!
1420 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1423 iap->ia_valid &= ATTR_SIZE;
1427 case NFS3_CREATE_EXCLUSIVE:
1428 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1429 && dchild->d_inode->i_atime.tv_sec == v_atime
1430 && dchild->d_inode->i_size == 0 )
1433 case NFS3_CREATE_GUARDED:
1436 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1440 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1442 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1448 nfsd_check_ignore_resizing(iap);
1450 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1451 /* Cram the verifier into atime/mtime */
1452 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1453 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1454 /* XXX someone who knows this better please fix it for nsec */
1455 iap->ia_mtime.tv_sec = v_mtime;
1456 iap->ia_atime.tv_sec = v_atime;
1457 iap->ia_mtime.tv_nsec = 0;
1458 iap->ia_atime.tv_nsec = 0;
1462 err = nfsd_create_setattr(rqstp, resfhp, iap);
1465 * nfsd_setattr already committed the child (and possibly also the parent).
1468 err = nfserrno(commit_metadata(fhp));
1470 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1472 * Update the filehandle to get the new inode info.
1475 err = fh_update(resfhp);
1479 if (dchild && !IS_ERR(dchild))
1484 err = nfserrno(host_err);
1487 #endif /* CONFIG_NFSD_V3 */
1490 * Read a symlink. On entry, *lenp must contain the maximum path length that
1491 * fits into the buffer. On return, it contains the true length.
1492 * N.B. After this call fhp needs an fh_put
1495 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1497 struct dentry *dentry;
1498 struct inode *inode;
1503 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1507 dentry = fhp->fh_dentry;
1508 inode = dentry->d_inode;
1511 if (!inode->i_op->readlink)
1514 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1515 /* N.B. Why does this call need a get_fs()??
1516 * Remove the set_fs and watch the fireworks:-) --okir
1519 oldfs = get_fs(); set_fs(KERNEL_DS);
1520 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1531 err = nfserrno(host_err);
1536 * Create a symlink and look up its inode
1537 * N.B. After this call _both_ fhp and resfhp need an fh_put
1540 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1541 char *fname, int flen,
1542 char *path, int plen,
1543 struct svc_fh *resfhp,
1546 struct dentry *dentry, *dnew;
1554 if (isdotent(fname, flen))
1557 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1561 dentry = fhp->fh_dentry;
1562 dnew = lookup_one_len(fname, dentry, flen);
1563 host_err = PTR_ERR(dnew);
1567 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1571 if (unlikely(path[plen] != 0)) {
1572 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1573 if (path_alloced == NULL)
1576 strncpy(path_alloced, path, plen);
1577 path_alloced[plen] = 0;
1578 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1579 kfree(path_alloced);
1582 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1583 err = nfserrno(host_err);
1585 err = nfserrno(commit_metadata(fhp));
1588 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1590 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1592 if (err==0) err = cerr;
1597 err = nfserrno(host_err);
1603 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1606 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1607 char *name, int len, struct svc_fh *tfhp)
1609 struct dentry *ddir, *dnew, *dold;
1614 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1617 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1625 if (isdotent(name, len))
1628 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1629 ddir = ffhp->fh_dentry;
1630 dirp = ddir->d_inode;
1632 dnew = lookup_one_len(name, ddir, len);
1633 host_err = PTR_ERR(dnew);
1637 dold = tfhp->fh_dentry;
1639 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1641 err = nfserrno(host_err);
1646 goto out_drop_write;
1647 host_err = nfsd_break_lease(dold->d_inode);
1649 goto out_drop_write;
1650 host_err = vfs_link(dold, dirp, dnew);
1652 err = nfserrno(commit_metadata(ffhp));
1654 err = nfserrno(commit_metadata(tfhp));
1656 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1659 err = nfserrno(host_err);
1662 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1671 err = nfserrno(host_err);
1677 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1680 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1681 struct svc_fh *tfhp, char *tname, int tlen)
1683 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1684 struct inode *fdir, *tdir;
1688 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1691 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1695 fdentry = ffhp->fh_dentry;
1696 fdir = fdentry->d_inode;
1698 tdentry = tfhp->fh_dentry;
1699 tdir = tdentry->d_inode;
1701 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1702 if (ffhp->fh_export != tfhp->fh_export)
1706 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1709 /* cannot use fh_lock as we need deadlock protective ordering
1710 * so do it by hand */
1711 trap = lock_rename(tdentry, fdentry);
1712 ffhp->fh_locked = tfhp->fh_locked = 1;
1716 odentry = lookup_one_len(fname, fdentry, flen);
1717 host_err = PTR_ERR(odentry);
1718 if (IS_ERR(odentry))
1722 if (!odentry->d_inode)
1725 if (odentry == trap)
1728 ndentry = lookup_one_len(tname, tdentry, tlen);
1729 host_err = PTR_ERR(ndentry);
1730 if (IS_ERR(ndentry))
1732 host_err = -ENOTEMPTY;
1733 if (ndentry == trap)
1737 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1739 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1743 host_err = nfsd_break_lease(odentry->d_inode);
1745 goto out_drop_write;
1746 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1748 host_err = commit_metadata(tfhp);
1750 host_err = commit_metadata(ffhp);
1753 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1759 err = nfserrno(host_err);
1761 /* we cannot reply on fh_unlock on the two filehandles,
1762 * as that would do the wrong thing if the two directories
1763 * were the same, so again we do it by hand
1765 fill_post_wcc(ffhp);
1766 fill_post_wcc(tfhp);
1767 unlock_rename(tdentry, fdentry);
1768 ffhp->fh_locked = tfhp->fh_locked = 0;
1775 * Unlink a file or directory
1776 * N.B. After this call fhp needs an fh_put
1779 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1780 char *fname, int flen)
1782 struct dentry *dentry, *rdentry;
1788 if (!flen || isdotent(fname, flen))
1790 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1794 fh_lock_nested(fhp, I_MUTEX_PARENT);
1795 dentry = fhp->fh_dentry;
1796 dirp = dentry->d_inode;
1798 rdentry = lookup_one_len(fname, dentry, flen);
1799 host_err = PTR_ERR(rdentry);
1800 if (IS_ERR(rdentry))
1803 if (!rdentry->d_inode) {
1810 type = rdentry->d_inode->i_mode & S_IFMT;
1812 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1816 host_err = nfsd_break_lease(rdentry->d_inode);
1819 if (type != S_IFDIR)
1820 host_err = vfs_unlink(dirp, rdentry);
1822 host_err = vfs_rmdir(dirp, rdentry);
1827 host_err = commit_metadata(fhp);
1829 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1831 err = nfserrno(host_err);
1837 * We do this buffering because we must not call back into the file
1838 * system's ->lookup() method from the filldir callback. That may well
1839 * deadlock a number of file systems.
1841 * This is based heavily on the implementation of same in XFS.
1843 struct buffered_dirent {
1847 unsigned int d_type;
1851 struct readdir_data {
1857 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1858 loff_t offset, u64 ino, unsigned int d_type)
1860 struct readdir_data *buf = __buf;
1861 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1862 unsigned int reclen;
1864 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1865 if (buf->used + reclen > PAGE_SIZE) {
1870 de->namlen = namlen;
1871 de->offset = offset;
1873 de->d_type = d_type;
1874 memcpy(de->name, name, namlen);
1875 buf->used += reclen;
1880 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1881 struct readdir_cd *cdp, loff_t *offsetp)
1883 struct readdir_data buf;
1884 struct buffered_dirent *de;
1889 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1891 return nfserrno(-ENOMEM);
1896 struct inode *dir_inode = file->f_path.dentry->d_inode;
1897 unsigned int reclen;
1899 cdp->err = nfserr_eof; /* will be cleared on successful read */
1903 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1916 * Various filldir functions may end up calling back into
1917 * lookup_one_len() and the file system's ->lookup() method.
1918 * These expect i_mutex to be held, as it would within readdir.
1920 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1924 de = (struct buffered_dirent *)buf.dirent;
1926 offset = de->offset;
1928 if (func(cdp, de->name, de->namlen, de->offset,
1929 de->ino, de->d_type))
1932 if (cdp->err != nfs_ok)
1935 reclen = ALIGN(sizeof(*de) + de->namlen,
1938 de = (struct buffered_dirent *)((char *)de + reclen);
1940 mutex_unlock(&dir_inode->i_mutex);
1941 if (size > 0) /* We bailed out early */
1944 offset = vfs_llseek(file, 0, SEEK_CUR);
1947 free_page((unsigned long)(buf.dirent));
1950 return nfserrno(host_err);
1957 * Read entries from a directory.
1958 * The NFSv3/4 verifier we ignore for now.
1961 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1962 struct readdir_cd *cdp, filldir_t func)
1966 loff_t offset = *offsetp;
1968 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1972 offset = vfs_llseek(file, offset, 0);
1974 err = nfserrno((int)offset);
1978 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1980 if (err == nfserr_eof || err == nfserr_toosmall)
1981 err = nfs_ok; /* can still be found in ->err */
1989 * Get file system stats
1990 * N.B. After this call fhp needs an fh_put
1993 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1997 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1999 struct path path = {
2000 .mnt = fhp->fh_export->ex_path.mnt,
2001 .dentry = fhp->fh_dentry,
2003 if (vfs_statfs(&path, stat))
2009 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2011 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2015 * Check for a user's access permissions to this inode.
2018 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2019 struct dentry *dentry, int acc)
2021 struct inode *inode = dentry->d_inode;
2024 if (acc == NFSD_MAY_NOP)
2027 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2029 (acc & NFSD_MAY_READ)? " read" : "",
2030 (acc & NFSD_MAY_WRITE)? " write" : "",
2031 (acc & NFSD_MAY_EXEC)? " exec" : "",
2032 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2033 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2034 (acc & NFSD_MAY_LOCK)? " lock" : "",
2035 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2037 IS_IMMUTABLE(inode)? " immut" : "",
2038 IS_APPEND(inode)? " append" : "",
2039 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2040 dprintk(" owner %d/%d user %d/%d\n",
2041 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2044 /* Normally we reject any write/sattr etc access on a read-only file
2045 * system. But if it is IRIX doing check on write-access for a
2046 * device special file, we ignore rofs.
2048 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2049 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2050 if (exp_rdonly(rqstp, exp) ||
2051 __mnt_is_readonly(exp->ex_path.mnt))
2053 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2056 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2059 if (acc & NFSD_MAY_LOCK) {
2060 /* If we cannot rely on authentication in NLM requests,
2061 * just allow locks, otherwise require read permission, or
2064 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2067 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2070 * The file owner always gets access permission for accesses that
2071 * would normally be checked at open time. This is to make
2072 * file access work even when the client has done a fchmod(fd, 0).
2074 * However, `cp foo bar' should fail nevertheless when bar is
2075 * readonly. A sensible way to do this might be to reject all
2076 * attempts to truncate a read-only file, because a creat() call
2077 * always implies file truncation.
2078 * ... but this isn't really fair. A process may reasonably call
2079 * ftruncate on an open file descriptor on a file with perm 000.
2080 * We must trust the client to do permission checking - using "ACCESS"
2083 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2084 inode->i_uid == current_fsuid())
2087 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2088 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2090 /* Allow read access to binaries even when mode 111 */
2091 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2092 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2093 err = inode_permission(inode, MAY_EXEC);
2095 return err? nfserrno(err) : 0;
2099 nfsd_racache_shutdown(void)
2101 struct raparms *raparm, *last_raparm;
2104 dprintk("nfsd: freeing readahead buffers.\n");
2106 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2107 raparm = raparm_hash[i].pb_head;
2109 last_raparm = raparm;
2110 raparm = raparm->p_next;
2113 raparm_hash[i].pb_head = NULL;
2117 * Initialize readahead param cache
2120 nfsd_racache_init(int cache_size)
2125 struct raparms **raparm = NULL;
2128 if (raparm_hash[0].pb_head)
2130 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2133 cache_size = nperbucket * RAPARM_HASH_SIZE;
2135 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2137 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2138 spin_lock_init(&raparm_hash[i].pb_lock);
2140 raparm = &raparm_hash[i].pb_head;
2141 for (j = 0; j < nperbucket; j++) {
2142 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2145 raparm = &(*raparm)->p_next;
2150 nfsdstats.ra_size = cache_size;
2154 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2155 nfsd_racache_shutdown();
2159 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2161 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2163 struct inode *inode = fhp->fh_dentry->d_inode;
2167 struct posix_acl *acl;
2169 if (!IS_POSIXACL(inode))
2170 return ERR_PTR(-EOPNOTSUPP);
2173 case ACL_TYPE_ACCESS:
2174 name = POSIX_ACL_XATTR_ACCESS;
2176 case ACL_TYPE_DEFAULT:
2177 name = POSIX_ACL_XATTR_DEFAULT;
2180 return ERR_PTR(-EOPNOTSUPP);
2183 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2185 return ERR_PTR(size);
2187 acl = posix_acl_from_xattr(value, size);
2193 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2195 struct inode *inode = fhp->fh_dentry->d_inode;
2201 if (!IS_POSIXACL(inode) ||
2202 !inode->i_op->setxattr || !inode->i_op->removexattr)
2205 case ACL_TYPE_ACCESS:
2206 name = POSIX_ACL_XATTR_ACCESS;
2208 case ACL_TYPE_DEFAULT:
2209 name = POSIX_ACL_XATTR_DEFAULT;
2215 if (acl && acl->a_count) {
2216 size = posix_acl_xattr_size(acl->a_count);
2217 value = kmalloc(size, GFP_KERNEL);
2220 error = posix_acl_to_xattr(acl, value, size);
2227 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2231 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2233 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2236 error = vfs_removexattr(fhp->fh_dentry, name);
2237 if (error == -ENODATA)
2241 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2247 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */