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 * RFC5661, Section 18.30.4:
456 * Changing the size of a file with SETATTR indirectly
457 * changes the time_modify and change attributes.
459 * (and similar for the older RFCs)
461 if (iap->ia_size != i_size_read(inode))
462 iap->ia_valid |= ATTR_MTIME;
465 iap->ia_valid |= ATTR_CTIME;
467 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
468 err = nfserr_notsync;
469 goto out_put_write_access;
472 host_err = nfsd_break_lease(inode);
474 goto out_put_write_access_nfserror;
477 host_err = notify_change(dentry, iap);
480 out_put_write_access_nfserror:
481 err = nfserrno(host_err);
482 out_put_write_access:
484 put_write_access(inode);
486 commit_metadata(fhp);
491 #if defined(CONFIG_NFSD_V2_ACL) || \
492 defined(CONFIG_NFSD_V3_ACL) || \
493 defined(CONFIG_NFSD_V4)
494 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
499 buflen = vfs_getxattr(dentry, key, NULL, 0);
503 *buf = kmalloc(buflen, GFP_KERNEL);
507 ret = vfs_getxattr(dentry, key, *buf, buflen);
514 #if defined(CONFIG_NFSD_V4)
516 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
524 return vfs_setxattr(dentry, key, NULL, 0, 0);
526 buflen = posix_acl_xattr_size(pacl->a_count);
527 buf = kmalloc(buflen, GFP_KERNEL);
532 len = posix_acl_to_xattr(pacl, buf, buflen);
538 error = vfs_setxattr(dentry, key, buf, len, 0);
545 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
546 struct nfs4_acl *acl)
550 struct dentry *dentry;
552 struct posix_acl *pacl = NULL, *dpacl = NULL;
553 unsigned int flags = 0;
556 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
560 dentry = fhp->fh_dentry;
561 inode = dentry->d_inode;
562 if (S_ISDIR(inode->i_mode))
563 flags = NFS4_ACL_DIR;
565 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
566 if (host_error == -EINVAL) {
567 return nfserr_attrnotsupp;
568 } else if (host_error < 0)
571 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
575 if (S_ISDIR(inode->i_mode))
576 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
579 posix_acl_release(pacl);
580 posix_acl_release(dpacl);
582 if (host_error == -EOPNOTSUPP)
583 return nfserr_attrnotsupp;
585 return nfserrno(host_error);
588 static struct posix_acl *
589 _get_posix_acl(struct dentry *dentry, char *key)
592 struct posix_acl *pacl = NULL;
595 buflen = nfsd_getxattr(dentry, key, &buf);
599 return ERR_PTR(buflen);
601 pacl = posix_acl_from_xattr(buf, buflen);
607 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
609 struct inode *inode = dentry->d_inode;
611 struct posix_acl *pacl = NULL, *dpacl = NULL;
612 unsigned int flags = 0;
614 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
615 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
616 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
618 error = PTR_ERR(pacl);
623 if (S_ISDIR(inode->i_mode)) {
624 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
625 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
627 else if (IS_ERR(dpacl)) {
628 error = PTR_ERR(dpacl);
632 flags = NFS4_ACL_DIR;
635 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
637 error = PTR_ERR(*acl);
641 posix_acl_release(pacl);
642 posix_acl_release(dpacl);
646 #define NFSD_XATTR_JUNCTION_PREFIX XATTR_TRUSTED_PREFIX "junction."
647 #define NFSD_XATTR_JUNCTION_TYPE NFSD_XATTR_JUNCTION_PREFIX "type"
648 int nfsd4_is_junction(struct dentry *dentry)
650 struct inode *inode = dentry->d_inode;
654 if (inode->i_mode & S_IXUGO)
656 if (!(inode->i_mode & S_ISVTX))
658 if (vfs_getxattr(dentry, NFSD_XATTR_JUNCTION_TYPE, NULL, 0) <= 0)
662 #endif /* defined(CONFIG_NFSD_V4) */
664 #ifdef CONFIG_NFSD_V3
666 * Check server access rights to a file system object
672 static struct accessmap nfs3_regaccess[] = {
673 { NFS3_ACCESS_READ, NFSD_MAY_READ },
674 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
675 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
676 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
681 static struct accessmap nfs3_diraccess[] = {
682 { NFS3_ACCESS_READ, NFSD_MAY_READ },
683 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
684 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
685 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
686 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
691 static struct accessmap nfs3_anyaccess[] = {
692 /* Some clients - Solaris 2.6 at least, make an access call
693 * to the server to check for access for things like /dev/null
694 * (which really, the server doesn't care about). So
695 * We provide simple access checking for them, looking
696 * mainly at mode bits, and we make sure to ignore read-only
699 { NFS3_ACCESS_READ, NFSD_MAY_READ },
700 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
701 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
702 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
708 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
710 struct accessmap *map;
711 struct svc_export *export;
712 struct dentry *dentry;
713 u32 query, result = 0, sresult = 0;
716 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
720 export = fhp->fh_export;
721 dentry = fhp->fh_dentry;
723 if (S_ISREG(dentry->d_inode->i_mode))
724 map = nfs3_regaccess;
725 else if (S_ISDIR(dentry->d_inode->i_mode))
726 map = nfs3_diraccess;
728 map = nfs3_anyaccess;
732 for (; map->access; map++) {
733 if (map->access & query) {
736 sresult |= map->access;
738 err2 = nfsd_permission(rqstp, export, dentry, map->how);
741 result |= map->access;
744 /* the following error codes just mean the access was not allowed,
745 * rather than an error occurred */
749 /* simply don't "or" in the access bit. */
759 *supported = sresult;
764 #endif /* CONFIG_NFSD_V3 */
766 static int nfsd_open_break_lease(struct inode *inode, int access)
770 if (access & NFSD_MAY_NOT_BREAK_LEASE)
772 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
773 return break_lease(inode, mode | O_NONBLOCK);
777 * Open an existing file or directory.
778 * The may_flags argument indicates the type of open (read/write/lock)
779 * and additional flags.
780 * N.B. After this call fhp needs an fh_put
783 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
784 int may_flags, struct file **filp)
786 struct dentry *dentry;
788 int flags = O_RDONLY|O_LARGEFILE;
792 validate_process_creds();
795 * If we get here, then the client has already done an "open",
796 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
797 * in case a chmod has now revoked permission.
799 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE);
803 dentry = fhp->fh_dentry;
804 inode = dentry->d_inode;
806 /* Disallow write access to files with the append-only bit set
807 * or any access when mandatory locking enabled
810 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
813 * We must ignore files (but only files) which might have mandatory
814 * locks on them because there is no way to know if the accesser has
817 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
823 host_err = nfsd_open_break_lease(inode, may_flags);
824 if (host_err) /* NOMEM or WOULDBLOCK */
827 if (may_flags & NFSD_MAY_WRITE) {
828 if (may_flags & NFSD_MAY_READ)
829 flags = O_RDWR|O_LARGEFILE;
831 flags = O_WRONLY|O_LARGEFILE;
833 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
834 flags, current_cred());
836 host_err = PTR_ERR(*filp);
839 host_err = ima_file_check(*filp, may_flags);
841 if (may_flags & NFSD_MAY_64BIT_COOKIE)
842 (*filp)->f_mode |= FMODE_64BITHASH;
844 (*filp)->f_mode |= FMODE_32BITHASH;
848 err = nfserrno(host_err);
850 validate_process_creds();
858 nfsd_close(struct file *filp)
864 * Obtain the readahead parameters for the file
865 * specified by (dev, ino).
868 static inline struct raparms *
869 nfsd_get_raparms(dev_t dev, ino_t ino)
871 struct raparms *ra, **rap, **frap = NULL;
874 struct raparm_hbucket *rab;
876 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
877 rab = &raparm_hash[hash];
879 spin_lock(&rab->pb_lock);
880 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
881 if (ra->p_ino == ino && ra->p_dev == dev)
884 if (ra->p_count == 0)
887 depth = nfsdstats.ra_size;
889 spin_unlock(&rab->pb_lock);
899 if (rap != &rab->pb_head) {
901 ra->p_next = rab->pb_head;
905 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
906 spin_unlock(&rab->pb_lock);
911 * Grab and keep cached pages associated with a file in the svc_rqst
912 * so that they can be passed to the network sendmsg/sendpage routines
913 * directly. They will be released after the sending has completed.
916 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
917 struct splice_desc *sd)
919 struct svc_rqst *rqstp = sd->u.data;
920 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
921 struct page *page = buf->page;
926 if (rqstp->rq_res.page_len == 0) {
931 rqstp->rq_res.page_base = buf->offset;
932 rqstp->rq_res.page_len = size;
933 } else if (page != pp[-1]) {
939 rqstp->rq_res.page_len += size;
941 rqstp->rq_res.page_len += size;
946 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
947 struct splice_desc *sd)
949 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
953 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
954 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
962 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
963 struct splice_desc sd = {
970 rqstp->rq_resused = 1;
971 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
975 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
980 nfsdstats.io_read += host_err;
983 fsnotify_access(file);
985 err = nfserrno(host_err);
989 static void kill_suid(struct dentry *dentry)
992 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
994 mutex_lock(&dentry->d_inode->i_mutex);
995 notify_change(dentry, &ia);
996 mutex_unlock(&dentry->d_inode->i_mutex);
1000 * Gathered writes: If another process is currently writing to the file,
1001 * there's a high chance this is another nfsd (triggered by a bulk write
1002 * from a client's biod). Rather than syncing the file with each write
1003 * request, we sleep for 10 msec.
1005 * I don't know if this roughly approximates C. Juszak's idea of
1006 * gathered writes, but it's a nice and simple solution (IMHO), and it
1009 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1010 * better tool (separate unstable writes and commits) for solving this
1013 static int wait_for_concurrent_writes(struct file *file)
1015 struct inode *inode = file->f_path.dentry->d_inode;
1016 static ino_t last_ino;
1017 static dev_t last_dev;
1020 if (atomic_read(&inode->i_writecount) > 1
1021 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1022 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1024 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1027 if (inode->i_state & I_DIRTY) {
1028 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1029 err = vfs_fsync(file, 0);
1031 last_ino = inode->i_ino;
1032 last_dev = inode->i_sb->s_dev;
1037 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1038 loff_t offset, struct kvec *vec, int vlen,
1039 unsigned long *cnt, int *stablep)
1041 struct svc_export *exp;
1042 struct dentry *dentry;
1043 struct inode *inode;
1047 int stable = *stablep;
1050 dentry = file->f_path.dentry;
1051 inode = dentry->d_inode;
1052 exp = fhp->fh_export;
1055 * Request sync writes if
1056 * - the sync export option has been set, or
1057 * - the client requested O_SYNC behavior (NFSv3 feature).
1058 * - The file system doesn't support fsync().
1059 * When NFSv2 gathered writes have been configured for this volume,
1060 * flushing the data to disk is handled separately below.
1062 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1064 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1066 *stablep = 2; /* FILE_SYNC */
1069 if (!EX_ISSYNC(exp))
1071 if (stable && !use_wgather) {
1072 spin_lock(&file->f_lock);
1073 file->f_flags |= O_SYNC;
1074 spin_unlock(&file->f_lock);
1077 /* Write the data. */
1078 oldfs = get_fs(); set_fs(KERNEL_DS);
1079 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1084 nfsdstats.io_write += host_err;
1085 fsnotify_modify(file);
1087 /* clear setuid/setgid flag after write */
1088 if (inode->i_mode & (S_ISUID | S_ISGID))
1091 if (stable && use_wgather)
1092 host_err = wait_for_concurrent_writes(file);
1095 dprintk("nfsd: write complete host_err=%d\n", host_err);
1099 err = nfserrno(host_err);
1104 * Read data from a file. count must contain the requested read count
1105 * on entry. On return, *count contains the number of bytes actually read.
1106 * N.B. After this call fhp needs an fh_put
1108 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1109 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1112 struct inode *inode;
1116 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1120 inode = file->f_path.dentry->d_inode;
1122 /* Get readahead parameters */
1123 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1125 if (ra && ra->p_set)
1126 file->f_ra = ra->p_ra;
1128 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1130 /* Write back readahead params */
1132 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1133 spin_lock(&rab->pb_lock);
1134 ra->p_ra = file->f_ra;
1137 spin_unlock(&rab->pb_lock);
1144 /* As above, but use the provided file descriptor. */
1146 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1147 loff_t offset, struct kvec *vec, int vlen,
1148 unsigned long *count)
1153 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1154 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1157 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1158 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1159 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1165 * Write data to a file.
1166 * The stable flag requests synchronous writes.
1167 * N.B. After this call fhp needs an fh_put
1170 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1171 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1177 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1178 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1181 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1184 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1189 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1197 #ifdef CONFIG_NFSD_V3
1199 * Commit all pending writes to stable storage.
1201 * Note: we only guarantee that data that lies within the range specified
1202 * by the 'offset' and 'count' parameters will be synced.
1204 * Unfortunately we cannot lock the file to make sure we return full WCC
1205 * data to the client, as locking happens lower down in the filesystem.
1208 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1209 loff_t offset, unsigned long count)
1212 loff_t end = LLONG_MAX;
1213 __be32 err = nfserr_inval;
1218 end = offset + (loff_t)count - 1;
1223 err = nfsd_open(rqstp, fhp, S_IFREG,
1224 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1227 if (EX_ISSYNC(fhp->fh_export)) {
1228 int err2 = vfs_fsync_range(file, offset, end, 0);
1230 if (err2 != -EINVAL)
1231 err = nfserrno(err2);
1233 err = nfserr_notsupp;
1240 #endif /* CONFIG_NFSD_V3 */
1243 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1247 * Mode has already been set earlier in create:
1249 iap->ia_valid &= ~ATTR_MODE;
1251 * Setting uid/gid works only for root. Irix appears to
1252 * send along the gid on create when it tries to implement
1253 * setgid directories via NFS:
1255 if (current_fsuid() != 0)
1256 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1258 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1262 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1263 * setting size to 0 may fail for some specific file systems by the permission
1264 * checking which requires WRITE permission but the mode is 000.
1265 * we ignore the resizing(to 0) on the just new created file, since the size is
1266 * 0 after file created.
1268 * call this only after vfs_create() is called.
1271 nfsd_check_ignore_resizing(struct iattr *iap)
1273 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1274 iap->ia_valid &= ~ATTR_SIZE;
1278 * Create a file (regular, directory, device, fifo); UNIX sockets
1279 * not yet implemented.
1280 * If the response fh has been verified, the parent directory should
1281 * already be locked. Note that the parent directory is left locked.
1283 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1286 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1287 char *fname, int flen, struct iattr *iap,
1288 int type, dev_t rdev, struct svc_fh *resfhp)
1290 struct dentry *dentry, *dchild = NULL;
1300 if (isdotent(fname, flen))
1303 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1307 dentry = fhp->fh_dentry;
1308 dirp = dentry->d_inode;
1310 err = nfserr_notdir;
1311 if (!dirp->i_op->lookup)
1314 * Check whether the response file handle has been verified yet.
1315 * If it has, the parent directory should already be locked.
1317 if (!resfhp->fh_dentry) {
1318 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1319 fh_lock_nested(fhp, I_MUTEX_PARENT);
1320 dchild = lookup_one_len(fname, dentry, flen);
1321 host_err = PTR_ERR(dchild);
1324 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1328 /* called from nfsd_proc_create */
1329 dchild = dget(resfhp->fh_dentry);
1330 if (!fhp->fh_locked) {
1331 /* not actually possible */
1333 "nfsd_create: parent %s/%s not locked!\n",
1334 dentry->d_parent->d_name.name,
1335 dentry->d_name.name);
1341 * Make sure the child dentry is still negative ...
1344 if (dchild->d_inode) {
1345 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1346 dentry->d_name.name, dchild->d_name.name);
1350 if (!(iap->ia_valid & ATTR_MODE))
1352 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1355 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1356 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1361 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1366 * Get the dir op function pointer.
1371 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1373 nfsd_check_ignore_resizing(iap);
1376 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1382 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1386 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1390 err = nfsd_create_setattr(rqstp, resfhp, iap);
1393 * nfsd_setattr already committed the child. Transactional filesystems
1394 * had a chance to commit changes for both parent and child
1395 * simultaneously making the following commit_metadata a noop.
1397 err2 = nfserrno(commit_metadata(fhp));
1400 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1402 * Update the file handle to get the new inode info.
1405 err = fh_update(resfhp);
1407 if (dchild && !IS_ERR(dchild))
1412 err = nfserrno(host_err);
1416 #ifdef CONFIG_NFSD_V3
1418 static inline int nfsd_create_is_exclusive(int createmode)
1420 return createmode == NFS3_CREATE_EXCLUSIVE
1421 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1425 * NFSv3 and NFSv4 version of nfsd_create
1428 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1429 char *fname, int flen, struct iattr *iap,
1430 struct svc_fh *resfhp, int createmode, u32 *verifier,
1431 bool *truncp, bool *created)
1433 struct dentry *dentry, *dchild = NULL;
1437 __u32 v_mtime=0, v_atime=0;
1443 if (isdotent(fname, flen))
1445 if (!(iap->ia_valid & ATTR_MODE))
1447 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1451 dentry = fhp->fh_dentry;
1452 dirp = dentry->d_inode;
1454 /* Get all the sanity checks out of the way before
1455 * we lock the parent. */
1456 err = nfserr_notdir;
1457 if (!dirp->i_op->lookup)
1459 fh_lock_nested(fhp, I_MUTEX_PARENT);
1462 * Compose the response file handle.
1464 dchild = lookup_one_len(fname, dentry, flen);
1465 host_err = PTR_ERR(dchild);
1469 /* If file doesn't exist, check for permissions to create one */
1470 if (!dchild->d_inode) {
1471 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1476 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1480 if (nfsd_create_is_exclusive(createmode)) {
1481 /* solaris7 gets confused (bugid 4218508) if these have
1482 * the high bit set, so just clear the high bits. If this is
1483 * ever changed to use different attrs for storing the
1484 * verifier, then do_open_lookup() will also need to be fixed
1487 v_mtime = verifier[0]&0x7fffffff;
1488 v_atime = verifier[1]&0x7fffffff;
1491 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1494 if (dchild->d_inode) {
1497 switch (createmode) {
1498 case NFS3_CREATE_UNCHECKED:
1499 if (! S_ISREG(dchild->d_inode->i_mode))
1502 /* in nfsv4, we need to treat this case a little
1503 * differently. we don't want to truncate the
1504 * file now; this would be wrong if the OPEN
1505 * fails for some other reason. furthermore,
1506 * if the size is nonzero, we should ignore it
1507 * according to spec!
1509 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1512 iap->ia_valid &= ATTR_SIZE;
1516 case NFS3_CREATE_EXCLUSIVE:
1517 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1518 && dchild->d_inode->i_atime.tv_sec == v_atime
1519 && dchild->d_inode->i_size == 0 ) {
1524 case NFS4_CREATE_EXCLUSIVE4_1:
1525 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1526 && dchild->d_inode->i_atime.tv_sec == v_atime
1527 && dchild->d_inode->i_size == 0 ) {
1533 case NFS3_CREATE_GUARDED:
1536 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1540 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1542 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1548 nfsd_check_ignore_resizing(iap);
1550 if (nfsd_create_is_exclusive(createmode)) {
1551 /* Cram the verifier into atime/mtime */
1552 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1553 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1554 /* XXX someone who knows this better please fix it for nsec */
1555 iap->ia_mtime.tv_sec = v_mtime;
1556 iap->ia_atime.tv_sec = v_atime;
1557 iap->ia_mtime.tv_nsec = 0;
1558 iap->ia_atime.tv_nsec = 0;
1562 err = nfsd_create_setattr(rqstp, resfhp, iap);
1565 * nfsd_setattr already committed the child (and possibly also the parent).
1568 err = nfserrno(commit_metadata(fhp));
1570 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1572 * Update the filehandle to get the new inode info.
1575 err = fh_update(resfhp);
1579 if (dchild && !IS_ERR(dchild))
1584 err = nfserrno(host_err);
1587 #endif /* CONFIG_NFSD_V3 */
1590 * Read a symlink. On entry, *lenp must contain the maximum path length that
1591 * fits into the buffer. On return, it contains the true length.
1592 * N.B. After this call fhp needs an fh_put
1595 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1597 struct dentry *dentry;
1598 struct inode *inode;
1603 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1607 dentry = fhp->fh_dentry;
1608 inode = dentry->d_inode;
1611 if (!inode->i_op->readlink)
1614 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1615 /* N.B. Why does this call need a get_fs()??
1616 * Remove the set_fs and watch the fireworks:-) --okir
1619 oldfs = get_fs(); set_fs(KERNEL_DS);
1620 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1631 err = nfserrno(host_err);
1636 * Create a symlink and look up its inode
1637 * N.B. After this call _both_ fhp and resfhp need an fh_put
1640 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1641 char *fname, int flen,
1642 char *path, int plen,
1643 struct svc_fh *resfhp,
1646 struct dentry *dentry, *dnew;
1654 if (isdotent(fname, flen))
1657 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1661 dentry = fhp->fh_dentry;
1662 dnew = lookup_one_len(fname, dentry, flen);
1663 host_err = PTR_ERR(dnew);
1667 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1671 if (unlikely(path[plen] != 0)) {
1672 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1673 if (path_alloced == NULL)
1676 strncpy(path_alloced, path, plen);
1677 path_alloced[plen] = 0;
1678 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1679 kfree(path_alloced);
1682 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1683 err = nfserrno(host_err);
1685 err = nfserrno(commit_metadata(fhp));
1688 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1690 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1692 if (err==0) err = cerr;
1697 err = nfserrno(host_err);
1703 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1706 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1707 char *name, int len, struct svc_fh *tfhp)
1709 struct dentry *ddir, *dnew, *dold;
1714 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1717 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1721 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1727 if (isdotent(name, len))
1730 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1731 ddir = ffhp->fh_dentry;
1732 dirp = ddir->d_inode;
1734 dnew = lookup_one_len(name, ddir, len);
1735 host_err = PTR_ERR(dnew);
1739 dold = tfhp->fh_dentry;
1741 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1743 err = nfserrno(host_err);
1748 goto out_drop_write;
1749 host_err = nfsd_break_lease(dold->d_inode);
1751 err = nfserrno(host_err);
1752 goto out_drop_write;
1754 host_err = vfs_link(dold, dirp, dnew);
1756 err = nfserrno(commit_metadata(ffhp));
1758 err = nfserrno(commit_metadata(tfhp));
1760 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1763 err = nfserrno(host_err);
1766 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1775 err = nfserrno(host_err);
1781 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1784 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1785 struct svc_fh *tfhp, char *tname, int tlen)
1787 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1788 struct inode *fdir, *tdir;
1792 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1795 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1799 fdentry = ffhp->fh_dentry;
1800 fdir = fdentry->d_inode;
1802 tdentry = tfhp->fh_dentry;
1803 tdir = tdentry->d_inode;
1805 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1806 if (ffhp->fh_export != tfhp->fh_export)
1810 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1813 /* cannot use fh_lock as we need deadlock protective ordering
1814 * so do it by hand */
1815 trap = lock_rename(tdentry, fdentry);
1816 ffhp->fh_locked = tfhp->fh_locked = 1;
1820 odentry = lookup_one_len(fname, fdentry, flen);
1821 host_err = PTR_ERR(odentry);
1822 if (IS_ERR(odentry))
1826 if (!odentry->d_inode)
1829 if (odentry == trap)
1832 ndentry = lookup_one_len(tname, tdentry, tlen);
1833 host_err = PTR_ERR(ndentry);
1834 if (IS_ERR(ndentry))
1836 host_err = -ENOTEMPTY;
1837 if (ndentry == trap)
1841 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1843 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1847 host_err = nfsd_break_lease(odentry->d_inode);
1849 goto out_drop_write;
1850 if (ndentry->d_inode) {
1851 host_err = nfsd_break_lease(ndentry->d_inode);
1853 goto out_drop_write;
1855 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1857 host_err = commit_metadata(tfhp);
1859 host_err = commit_metadata(ffhp);
1862 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1868 err = nfserrno(host_err);
1870 /* we cannot reply on fh_unlock on the two filehandles,
1871 * as that would do the wrong thing if the two directories
1872 * were the same, so again we do it by hand
1874 fill_post_wcc(ffhp);
1875 fill_post_wcc(tfhp);
1876 unlock_rename(tdentry, fdentry);
1877 ffhp->fh_locked = tfhp->fh_locked = 0;
1884 * Unlink a file or directory
1885 * N.B. After this call fhp needs an fh_put
1888 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1889 char *fname, int flen)
1891 struct dentry *dentry, *rdentry;
1897 if (!flen || isdotent(fname, flen))
1899 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1903 fh_lock_nested(fhp, I_MUTEX_PARENT);
1904 dentry = fhp->fh_dentry;
1905 dirp = dentry->d_inode;
1907 rdentry = lookup_one_len(fname, dentry, flen);
1908 host_err = PTR_ERR(rdentry);
1909 if (IS_ERR(rdentry))
1912 if (!rdentry->d_inode) {
1919 type = rdentry->d_inode->i_mode & S_IFMT;
1921 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1925 host_err = nfsd_break_lease(rdentry->d_inode);
1927 goto out_drop_write;
1928 if (type != S_IFDIR)
1929 host_err = vfs_unlink(dirp, rdentry);
1931 host_err = vfs_rmdir(dirp, rdentry);
1933 host_err = commit_metadata(fhp);
1935 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1940 err = nfserrno(host_err);
1946 * We do this buffering because we must not call back into the file
1947 * system's ->lookup() method from the filldir callback. That may well
1948 * deadlock a number of file systems.
1950 * This is based heavily on the implementation of same in XFS.
1952 struct buffered_dirent {
1956 unsigned int d_type;
1960 struct readdir_data {
1966 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1967 loff_t offset, u64 ino, unsigned int d_type)
1969 struct readdir_data *buf = __buf;
1970 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1971 unsigned int reclen;
1973 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1974 if (buf->used + reclen > PAGE_SIZE) {
1979 de->namlen = namlen;
1980 de->offset = offset;
1982 de->d_type = d_type;
1983 memcpy(de->name, name, namlen);
1984 buf->used += reclen;
1989 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1990 struct readdir_cd *cdp, loff_t *offsetp)
1992 struct readdir_data buf;
1993 struct buffered_dirent *de;
1998 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2000 return nfserrno(-ENOMEM);
2005 struct inode *dir_inode = file->f_path.dentry->d_inode;
2006 unsigned int reclen;
2008 cdp->err = nfserr_eof; /* will be cleared on successful read */
2012 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2025 * Various filldir functions may end up calling back into
2026 * lookup_one_len() and the file system's ->lookup() method.
2027 * These expect i_mutex to be held, as it would within readdir.
2029 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2033 de = (struct buffered_dirent *)buf.dirent;
2035 offset = de->offset;
2037 if (func(cdp, de->name, de->namlen, de->offset,
2038 de->ino, de->d_type))
2041 if (cdp->err != nfs_ok)
2044 reclen = ALIGN(sizeof(*de) + de->namlen,
2047 de = (struct buffered_dirent *)((char *)de + reclen);
2049 mutex_unlock(&dir_inode->i_mutex);
2050 if (size > 0) /* We bailed out early */
2053 offset = vfs_llseek(file, 0, SEEK_CUR);
2056 free_page((unsigned long)(buf.dirent));
2059 return nfserrno(host_err);
2066 * Read entries from a directory.
2067 * The NFSv3/4 verifier we ignore for now.
2070 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2071 struct readdir_cd *cdp, filldir_t func)
2075 loff_t offset = *offsetp;
2076 int may_flags = NFSD_MAY_READ;
2078 /* NFSv2 only supports 32 bit cookies */
2079 if (rqstp->rq_vers > 2)
2080 may_flags |= NFSD_MAY_64BIT_COOKIE;
2082 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2086 offset = vfs_llseek(file, offset, 0);
2088 err = nfserrno((int)offset);
2092 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2094 if (err == nfserr_eof || err == nfserr_toosmall)
2095 err = nfs_ok; /* can still be found in ->err */
2103 * Get file system stats
2104 * N.B. After this call fhp needs an fh_put
2107 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2111 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2113 struct path path = {
2114 .mnt = fhp->fh_export->ex_path.mnt,
2115 .dentry = fhp->fh_dentry,
2117 if (vfs_statfs(&path, stat))
2123 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2125 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2129 * Check for a user's access permissions to this inode.
2132 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2133 struct dentry *dentry, int acc)
2135 struct inode *inode = dentry->d_inode;
2138 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2141 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2143 (acc & NFSD_MAY_READ)? " read" : "",
2144 (acc & NFSD_MAY_WRITE)? " write" : "",
2145 (acc & NFSD_MAY_EXEC)? " exec" : "",
2146 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2147 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2148 (acc & NFSD_MAY_LOCK)? " lock" : "",
2149 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2151 IS_IMMUTABLE(inode)? " immut" : "",
2152 IS_APPEND(inode)? " append" : "",
2153 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2154 dprintk(" owner %d/%d user %d/%d\n",
2155 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2158 /* Normally we reject any write/sattr etc access on a read-only file
2159 * system. But if it is IRIX doing check on write-access for a
2160 * device special file, we ignore rofs.
2162 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2163 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2164 if (exp_rdonly(rqstp, exp) ||
2165 __mnt_is_readonly(exp->ex_path.mnt))
2167 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2170 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2173 if (acc & NFSD_MAY_LOCK) {
2174 /* If we cannot rely on authentication in NLM requests,
2175 * just allow locks, otherwise require read permission, or
2178 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2181 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2184 * The file owner always gets access permission for accesses that
2185 * would normally be checked at open time. This is to make
2186 * file access work even when the client has done a fchmod(fd, 0).
2188 * However, `cp foo bar' should fail nevertheless when bar is
2189 * readonly. A sensible way to do this might be to reject all
2190 * attempts to truncate a read-only file, because a creat() call
2191 * always implies file truncation.
2192 * ... but this isn't really fair. A process may reasonably call
2193 * ftruncate on an open file descriptor on a file with perm 000.
2194 * We must trust the client to do permission checking - using "ACCESS"
2197 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2198 inode->i_uid == current_fsuid())
2201 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2202 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2204 /* Allow read access to binaries even when mode 111 */
2205 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2206 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2207 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2208 err = inode_permission(inode, MAY_EXEC);
2210 return err? nfserrno(err) : 0;
2214 nfsd_racache_shutdown(void)
2216 struct raparms *raparm, *last_raparm;
2219 dprintk("nfsd: freeing readahead buffers.\n");
2221 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2222 raparm = raparm_hash[i].pb_head;
2224 last_raparm = raparm;
2225 raparm = raparm->p_next;
2228 raparm_hash[i].pb_head = NULL;
2232 * Initialize readahead param cache
2235 nfsd_racache_init(int cache_size)
2240 struct raparms **raparm = NULL;
2243 if (raparm_hash[0].pb_head)
2245 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2248 cache_size = nperbucket * RAPARM_HASH_SIZE;
2250 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2252 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2253 spin_lock_init(&raparm_hash[i].pb_lock);
2255 raparm = &raparm_hash[i].pb_head;
2256 for (j = 0; j < nperbucket; j++) {
2257 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2260 raparm = &(*raparm)->p_next;
2265 nfsdstats.ra_size = cache_size;
2269 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2270 nfsd_racache_shutdown();
2274 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2276 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2278 struct inode *inode = fhp->fh_dentry->d_inode;
2282 struct posix_acl *acl;
2284 if (!IS_POSIXACL(inode))
2285 return ERR_PTR(-EOPNOTSUPP);
2288 case ACL_TYPE_ACCESS:
2289 name = POSIX_ACL_XATTR_ACCESS;
2291 case ACL_TYPE_DEFAULT:
2292 name = POSIX_ACL_XATTR_DEFAULT;
2295 return ERR_PTR(-EOPNOTSUPP);
2298 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2300 return ERR_PTR(size);
2302 acl = posix_acl_from_xattr(value, size);
2308 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2310 struct inode *inode = fhp->fh_dentry->d_inode;
2316 if (!IS_POSIXACL(inode) ||
2317 !inode->i_op->setxattr || !inode->i_op->removexattr)
2320 case ACL_TYPE_ACCESS:
2321 name = POSIX_ACL_XATTR_ACCESS;
2323 case ACL_TYPE_DEFAULT:
2324 name = POSIX_ACL_XATTR_DEFAULT;
2330 if (acl && acl->a_count) {
2331 size = posix_acl_xattr_size(acl->a_count);
2332 value = kmalloc(size, GFP_KERNEL);
2335 error = posix_acl_to_xattr(acl, value, size);
2342 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2346 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2348 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2351 error = vfs_removexattr(fhp->fh_dentry, name);
2352 if (error == -ENODATA)
2356 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2362 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */