2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
39 #endif /* CONFIG_NFSD_V4 */
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
55 struct raparms *p_next;
60 struct file_ra_state p_ra;
61 unsigned int p_hindex;
64 struct raparm_hbucket {
65 struct raparms *pb_head;
67 } ____cacheline_aligned_in_smp;
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
90 err = follow_down(&path);
94 exp2 = rqst_exp_get_by_name(rqstp, &path);
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
104 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
109 if (nfsd_v4client(rqstp) ||
110 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111 /* successfully crossed mount point */
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
119 path.dentry = dentry;
129 static void follow_to_parent(struct path *path)
133 while (path->dentry == path->mnt->mnt_root && follow_up(path))
135 dp = dget_parent(path->dentry);
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
142 struct svc_export *exp2;
143 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144 .dentry = dget(dparent)};
146 follow_to_parent(&path);
148 exp2 = rqst_exp_parent(rqstp, &path);
149 if (PTR_ERR(exp2) == -ENOENT) {
150 *dentryp = dget(dparent);
151 } else if (IS_ERR(exp2)) {
153 return PTR_ERR(exp2);
155 *dentryp = dget(path.dentry);
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
169 if (d_mountpoint(dentry))
171 if (nfsd4_is_junction(dentry))
173 if (!(exp->ex_flags & NFSEXP_V4ROOT))
175 return dentry->d_inode != NULL;
179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
180 const char *name, unsigned int len,
181 struct svc_export **exp_ret, struct dentry **dentry_ret)
183 struct svc_export *exp;
184 struct dentry *dparent;
185 struct dentry *dentry;
188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
190 dparent = fhp->fh_dentry;
191 exp = fhp->fh_export;
194 /* Lookup the name, but don't follow links */
195 if (isdotent(name, len)) {
197 dentry = dget(dparent);
198 else if (dparent != exp->ex_path.dentry)
199 dentry = dget_parent(dparent);
200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
201 dentry = dget(dparent); /* .. == . just like at / */
203 /* checking mountpoint crossing is very different when stepping up */
204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
210 dentry = lookup_one_len(name, dparent, len);
211 host_err = PTR_ERR(dentry);
215 * check if we have crossed a mount point ...
217 if (nfsd_mountpoint(dentry, exp)) {
218 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
224 *dentry_ret = dentry;
230 return nfserrno(host_err);
234 * Look up one component of a pathname.
235 * N.B. After this call _both_ fhp and resfh need an fh_put
237 * If the lookup would cross a mountpoint, and the mounted filesystem
238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
239 * accepted as it stands and the mounted directory is
240 * returned. Otherwise the covered directory is returned.
241 * NOTE: this mountpoint crossing is not supported properly by all
242 * clients and is explicitly disallowed for NFSv3
243 * NeilBrown <neilb@cse.unsw.edu.au>
246 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
247 unsigned int len, struct svc_fh *resfh)
249 struct svc_export *exp;
250 struct dentry *dentry;
253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
259 err = check_nfsd_access(exp, rqstp);
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
275 static int nfsd_break_lease(struct inode *inode)
277 if (!S_ISREG(inode->i_mode))
279 return break_lease(inode, O_WRONLY | O_NONBLOCK);
283 * Commit metadata changes to stable storage.
286 commit_metadata(struct svc_fh *fhp)
288 struct inode *inode = fhp->fh_dentry->d_inode;
289 const struct export_operations *export_ops = inode->i_sb->s_export_op;
291 if (!EX_ISSYNC(fhp->fh_export))
294 if (export_ops->commit_metadata)
295 return export_ops->commit_metadata(inode);
296 return sync_inode_metadata(inode, 1);
300 * Set various file attributes.
301 * N.B. After this call fhp needs an fh_put
304 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
305 int check_guard, time_t guardtime)
307 struct dentry *dentry;
309 int accmode = NFSD_MAY_SATTR;
315 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
316 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
317 if (iap->ia_valid & ATTR_SIZE)
321 err = fh_verify(rqstp, fhp, ftype, accmode);
325 dentry = fhp->fh_dentry;
326 inode = dentry->d_inode;
328 /* Ignore any mode updates on symlinks */
329 if (S_ISLNK(inode->i_mode))
330 iap->ia_valid &= ~ATTR_MODE;
336 * NFSv2 does not differentiate between "set-[ac]time-to-now"
337 * which only requires access, and "set-[ac]time-to-X" which
338 * requires ownership.
339 * So if it looks like it might be "set both to the same time which
340 * is close to now", and if inode_change_ok fails, then we
341 * convert to "set to now" instead of "set to explicit time"
343 * We only call inode_change_ok as the last test as technically
344 * it is not an interface that we should be using. It is only
345 * valid if the filesystem does not define it's own i_op->setattr.
347 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
348 #define MAX_TOUCH_TIME_ERROR (30*60)
349 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
350 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
354 * Now just make sure time is in the right ballpark.
355 * Solaris, at least, doesn't seem to care what the time
356 * request is. We require it be within 30 minutes of now.
358 time_t delta = iap->ia_atime.tv_sec - get_seconds();
361 if (delta < MAX_TOUCH_TIME_ERROR &&
362 inode_change_ok(inode, iap) != 0) {
364 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
365 * This will cause notify_change to set these times
368 iap->ia_valid &= ~BOTH_TIME_SET;
373 * The size case is special.
374 * It changes the file as well as the attributes.
376 if (iap->ia_valid & ATTR_SIZE) {
377 if (iap->ia_size < inode->i_size) {
378 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
379 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
384 host_err = get_write_access(inode);
389 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
391 put_write_access(inode);
396 /* sanitize the mode change */
397 if (iap->ia_valid & ATTR_MODE) {
398 iap->ia_mode &= S_IALLUGO;
399 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
402 /* Revoke setuid/setgid on chown */
403 if (!S_ISDIR(inode->i_mode) &&
404 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
405 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
406 iap->ia_valid |= ATTR_KILL_PRIV;
407 if (iap->ia_valid & ATTR_MODE) {
408 /* we're setting mode too, just clear the s*id bits */
409 iap->ia_mode &= ~S_ISUID;
410 if (iap->ia_mode & S_IXGRP)
411 iap->ia_mode &= ~S_ISGID;
413 /* set ATTR_KILL_* bits and let VFS handle it */
414 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
418 /* Change the attributes. */
420 iap->ia_valid |= ATTR_CTIME;
422 err = nfserr_notsync;
423 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
424 host_err = nfsd_break_lease(inode);
429 host_err = notify_change(dentry, iap);
430 err = nfserrno(host_err);
434 put_write_access(inode);
436 commit_metadata(fhp);
441 err = nfserrno(host_err);
445 #if defined(CONFIG_NFSD_V2_ACL) || \
446 defined(CONFIG_NFSD_V3_ACL) || \
447 defined(CONFIG_NFSD_V4)
448 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
453 buflen = vfs_getxattr(dentry, key, NULL, 0);
457 *buf = kmalloc(buflen, GFP_KERNEL);
461 ret = vfs_getxattr(dentry, key, *buf, buflen);
468 #if defined(CONFIG_NFSD_V4)
470 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
477 buflen = posix_acl_xattr_size(pacl->a_count);
478 buf = kmalloc(buflen, GFP_KERNEL);
483 len = posix_acl_to_xattr(pacl, buf, buflen);
489 error = vfs_setxattr(dentry, key, buf, len, 0);
496 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
497 struct nfs4_acl *acl)
501 struct dentry *dentry;
503 struct posix_acl *pacl = NULL, *dpacl = NULL;
504 unsigned int flags = 0;
507 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
511 dentry = fhp->fh_dentry;
512 inode = dentry->d_inode;
513 if (S_ISDIR(inode->i_mode))
514 flags = NFS4_ACL_DIR;
516 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
517 if (host_error == -EINVAL) {
518 return nfserr_attrnotsupp;
519 } else if (host_error < 0)
522 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
526 if (S_ISDIR(inode->i_mode))
527 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
530 posix_acl_release(pacl);
531 posix_acl_release(dpacl);
533 if (host_error == -EOPNOTSUPP)
534 return nfserr_attrnotsupp;
536 return nfserrno(host_error);
539 static struct posix_acl *
540 _get_posix_acl(struct dentry *dentry, char *key)
543 struct posix_acl *pacl = NULL;
546 buflen = nfsd_getxattr(dentry, key, &buf);
550 return ERR_PTR(buflen);
552 pacl = posix_acl_from_xattr(buf, buflen);
558 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
560 struct inode *inode = dentry->d_inode;
562 struct posix_acl *pacl = NULL, *dpacl = NULL;
563 unsigned int flags = 0;
565 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
566 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
567 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
569 error = PTR_ERR(pacl);
574 if (S_ISDIR(inode->i_mode)) {
575 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
576 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
578 else if (IS_ERR(dpacl)) {
579 error = PTR_ERR(dpacl);
583 flags = NFS4_ACL_DIR;
586 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
588 error = PTR_ERR(*acl);
592 posix_acl_release(pacl);
593 posix_acl_release(dpacl);
598 * NFS junction information is stored in an extended attribute.
600 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
603 * nfsd4_is_junction - Test if an object could be an NFS junction
605 * @dentry: object to test
607 * Returns 1 if "dentry" appears to contain NFS junction information.
608 * Otherwise 0 is returned.
610 int nfsd4_is_junction(struct dentry *dentry)
612 struct inode *inode = dentry->d_inode;
616 if (inode->i_mode & S_IXUGO)
618 if (!(inode->i_mode & S_ISVTX))
620 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
624 #endif /* defined(CONFIG_NFSD_V4) */
626 #ifdef CONFIG_NFSD_V3
628 * Check server access rights to a file system object
634 static struct accessmap nfs3_regaccess[] = {
635 { NFS3_ACCESS_READ, NFSD_MAY_READ },
636 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
637 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
638 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
643 static struct accessmap nfs3_diraccess[] = {
644 { NFS3_ACCESS_READ, NFSD_MAY_READ },
645 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
646 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
647 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
648 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
653 static struct accessmap nfs3_anyaccess[] = {
654 /* Some clients - Solaris 2.6 at least, make an access call
655 * to the server to check for access for things like /dev/null
656 * (which really, the server doesn't care about). So
657 * We provide simple access checking for them, looking
658 * mainly at mode bits, and we make sure to ignore read-only
661 { NFS3_ACCESS_READ, NFSD_MAY_READ },
662 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
663 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
664 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
670 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
672 struct accessmap *map;
673 struct svc_export *export;
674 struct dentry *dentry;
675 u32 query, result = 0, sresult = 0;
678 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
682 export = fhp->fh_export;
683 dentry = fhp->fh_dentry;
685 if (S_ISREG(dentry->d_inode->i_mode))
686 map = nfs3_regaccess;
687 else if (S_ISDIR(dentry->d_inode->i_mode))
688 map = nfs3_diraccess;
690 map = nfs3_anyaccess;
694 for (; map->access; map++) {
695 if (map->access & query) {
698 sresult |= map->access;
700 err2 = nfsd_permission(rqstp, export, dentry, map->how);
703 result |= map->access;
706 /* the following error codes just mean the access was not allowed,
707 * rather than an error occurred */
711 /* simply don't "or" in the access bit. */
721 *supported = sresult;
726 #endif /* CONFIG_NFSD_V3 */
728 static int nfsd_open_break_lease(struct inode *inode, int access)
732 if (access & NFSD_MAY_NOT_BREAK_LEASE)
734 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
735 return break_lease(inode, mode | O_NONBLOCK);
739 * Open an existing file or directory.
740 * The may_flags argument indicates the type of open (read/write/lock)
741 * and additional flags.
742 * N.B. After this call fhp needs an fh_put
745 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
746 int may_flags, struct file **filp)
748 struct dentry *dentry;
750 int flags = O_RDONLY|O_LARGEFILE;
754 validate_process_creds();
757 * If we get here, then the client has already done an "open",
758 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
759 * in case a chmod has now revoked permission.
761 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE);
765 dentry = fhp->fh_dentry;
766 inode = dentry->d_inode;
768 /* Disallow write access to files with the append-only bit set
769 * or any access when mandatory locking enabled
772 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
775 * We must ignore files (but only files) which might have mandatory
776 * locks on them because there is no way to know if the accesser has
779 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
785 host_err = nfsd_open_break_lease(inode, may_flags);
786 if (host_err) /* NOMEM or WOULDBLOCK */
789 if (may_flags & NFSD_MAY_WRITE) {
790 if (may_flags & NFSD_MAY_READ)
791 flags = O_RDWR|O_LARGEFILE;
793 flags = O_WRONLY|O_LARGEFILE;
795 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
796 flags, current_cred());
798 host_err = PTR_ERR(*filp);
800 host_err = ima_file_check(*filp, may_flags);
802 if (may_flags & NFSD_MAY_64BIT_COOKIE)
803 (*filp)->f_mode |= FMODE_64BITHASH;
805 (*filp)->f_mode |= FMODE_32BITHASH;
809 err = nfserrno(host_err);
811 validate_process_creds();
819 nfsd_close(struct file *filp)
825 * Obtain the readahead parameters for the file
826 * specified by (dev, ino).
829 static inline struct raparms *
830 nfsd_get_raparms(dev_t dev, ino_t ino)
832 struct raparms *ra, **rap, **frap = NULL;
835 struct raparm_hbucket *rab;
837 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
838 rab = &raparm_hash[hash];
840 spin_lock(&rab->pb_lock);
841 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
842 if (ra->p_ino == ino && ra->p_dev == dev)
845 if (ra->p_count == 0)
848 depth = nfsdstats.ra_size;
850 spin_unlock(&rab->pb_lock);
860 if (rap != &rab->pb_head) {
862 ra->p_next = rab->pb_head;
866 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
867 spin_unlock(&rab->pb_lock);
872 * Grab and keep cached pages associated with a file in the svc_rqst
873 * so that they can be passed to the network sendmsg/sendpage routines
874 * directly. They will be released after the sending has completed.
877 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
878 struct splice_desc *sd)
880 struct svc_rqst *rqstp = sd->u.data;
881 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
882 struct page *page = buf->page;
887 if (rqstp->rq_res.page_len == 0) {
892 rqstp->rq_res.page_base = buf->offset;
893 rqstp->rq_res.page_len = size;
894 } else if (page != pp[-1]) {
900 rqstp->rq_res.page_len += size;
902 rqstp->rq_res.page_len += size;
907 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
908 struct splice_desc *sd)
910 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
914 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
915 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
923 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
924 struct splice_desc sd = {
931 rqstp->rq_resused = 1;
932 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
936 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
941 nfsdstats.io_read += host_err;
944 fsnotify_access(file);
946 err = nfserrno(host_err);
950 static void kill_suid(struct dentry *dentry)
953 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
955 mutex_lock(&dentry->d_inode->i_mutex);
956 notify_change(dentry, &ia);
957 mutex_unlock(&dentry->d_inode->i_mutex);
961 * Gathered writes: If another process is currently writing to the file,
962 * there's a high chance this is another nfsd (triggered by a bulk write
963 * from a client's biod). Rather than syncing the file with each write
964 * request, we sleep for 10 msec.
966 * I don't know if this roughly approximates C. Juszak's idea of
967 * gathered writes, but it's a nice and simple solution (IMHO), and it
970 * Note: we do this only in the NFSv2 case, since v3 and higher have a
971 * better tool (separate unstable writes and commits) for solving this
974 static int wait_for_concurrent_writes(struct file *file)
976 struct inode *inode = file->f_path.dentry->d_inode;
977 static ino_t last_ino;
978 static dev_t last_dev;
981 if (atomic_read(&inode->i_writecount) > 1
982 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
983 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
985 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
988 if (inode->i_state & I_DIRTY) {
989 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
990 err = vfs_fsync(file, 0);
992 last_ino = inode->i_ino;
993 last_dev = inode->i_sb->s_dev;
998 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
999 loff_t offset, struct kvec *vec, int vlen,
1000 unsigned long *cnt, int *stablep)
1002 struct svc_export *exp;
1003 struct dentry *dentry;
1004 struct inode *inode;
1008 int stable = *stablep;
1011 dentry = file->f_path.dentry;
1012 inode = dentry->d_inode;
1013 exp = fhp->fh_export;
1016 * Request sync writes if
1017 * - the sync export option has been set, or
1018 * - the client requested O_SYNC behavior (NFSv3 feature).
1019 * - The file system doesn't support fsync().
1020 * When NFSv2 gathered writes have been configured for this volume,
1021 * flushing the data to disk is handled separately below.
1023 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1025 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1027 *stablep = 2; /* FILE_SYNC */
1030 if (!EX_ISSYNC(exp))
1032 if (stable && !use_wgather) {
1033 spin_lock(&file->f_lock);
1034 file->f_flags |= O_SYNC;
1035 spin_unlock(&file->f_lock);
1038 /* Write the data. */
1039 oldfs = get_fs(); set_fs(KERNEL_DS);
1040 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1045 nfsdstats.io_write += host_err;
1046 fsnotify_modify(file);
1048 /* clear setuid/setgid flag after write */
1049 if (inode->i_mode & (S_ISUID | S_ISGID))
1052 if (stable && use_wgather)
1053 host_err = wait_for_concurrent_writes(file);
1056 dprintk("nfsd: write complete host_err=%d\n", host_err);
1060 err = nfserrno(host_err);
1065 * Read data from a file. count must contain the requested read count
1066 * on entry. On return, *count contains the number of bytes actually read.
1067 * N.B. After this call fhp needs an fh_put
1069 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1070 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1073 struct inode *inode;
1077 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1081 inode = file->f_path.dentry->d_inode;
1083 /* Get readahead parameters */
1084 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1086 if (ra && ra->p_set)
1087 file->f_ra = ra->p_ra;
1089 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1091 /* Write back readahead params */
1093 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1094 spin_lock(&rab->pb_lock);
1095 ra->p_ra = file->f_ra;
1098 spin_unlock(&rab->pb_lock);
1105 /* As above, but use the provided file descriptor. */
1107 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1108 loff_t offset, struct kvec *vec, int vlen,
1109 unsigned long *count)
1114 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1115 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1118 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1119 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1120 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1126 * Write data to a file.
1127 * The stable flag requests synchronous writes.
1128 * N.B. After this call fhp needs an fh_put
1131 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1132 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1138 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1139 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1142 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1145 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1150 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1158 #ifdef CONFIG_NFSD_V3
1160 * Commit all pending writes to stable storage.
1162 * Note: we only guarantee that data that lies within the range specified
1163 * by the 'offset' and 'count' parameters will be synced.
1165 * Unfortunately we cannot lock the file to make sure we return full WCC
1166 * data to the client, as locking happens lower down in the filesystem.
1169 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1170 loff_t offset, unsigned long count)
1173 loff_t end = LLONG_MAX;
1174 __be32 err = nfserr_inval;
1179 end = offset + (loff_t)count - 1;
1184 err = nfsd_open(rqstp, fhp, S_IFREG,
1185 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1188 if (EX_ISSYNC(fhp->fh_export)) {
1189 int err2 = vfs_fsync_range(file, offset, end, 0);
1191 if (err2 != -EINVAL)
1192 err = nfserrno(err2);
1194 err = nfserr_notsupp;
1201 #endif /* CONFIG_NFSD_V3 */
1204 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1208 * Mode has already been set earlier in create:
1210 iap->ia_valid &= ~ATTR_MODE;
1212 * Setting uid/gid works only for root. Irix appears to
1213 * send along the gid on create when it tries to implement
1214 * setgid directories via NFS:
1216 if (current_fsuid() != 0)
1217 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1219 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1223 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1224 * setting size to 0 may fail for some specific file systems by the permission
1225 * checking which requires WRITE permission but the mode is 000.
1226 * we ignore the resizing(to 0) on the just new created file, since the size is
1227 * 0 after file created.
1229 * call this only after vfs_create() is called.
1232 nfsd_check_ignore_resizing(struct iattr *iap)
1234 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1235 iap->ia_valid &= ~ATTR_SIZE;
1239 * Create a file (regular, directory, device, fifo); UNIX sockets
1240 * not yet implemented.
1241 * If the response fh has been verified, the parent directory should
1242 * already be locked. Note that the parent directory is left locked.
1244 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1247 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1248 char *fname, int flen, struct iattr *iap,
1249 int type, dev_t rdev, struct svc_fh *resfhp)
1251 struct dentry *dentry, *dchild = NULL;
1261 if (isdotent(fname, flen))
1264 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1268 dentry = fhp->fh_dentry;
1269 dirp = dentry->d_inode;
1271 err = nfserr_notdir;
1272 if (!dirp->i_op->lookup)
1275 * Check whether the response file handle has been verified yet.
1276 * If it has, the parent directory should already be locked.
1278 if (!resfhp->fh_dentry) {
1279 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1280 fh_lock_nested(fhp, I_MUTEX_PARENT);
1281 dchild = lookup_one_len(fname, dentry, flen);
1282 host_err = PTR_ERR(dchild);
1285 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1289 /* called from nfsd_proc_create */
1290 dchild = dget(resfhp->fh_dentry);
1291 if (!fhp->fh_locked) {
1292 /* not actually possible */
1294 "nfsd_create: parent %s/%s not locked!\n",
1295 dentry->d_parent->d_name.name,
1296 dentry->d_name.name);
1302 * Make sure the child dentry is still negative ...
1305 if (dchild->d_inode) {
1306 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1307 dentry->d_name.name, dchild->d_name.name);
1311 if (!(iap->ia_valid & ATTR_MODE))
1313 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1316 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1317 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1322 host_err = fh_want_write(fhp);
1327 * Get the dir op function pointer.
1332 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1334 nfsd_check_ignore_resizing(iap);
1337 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1343 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1351 err = nfsd_create_setattr(rqstp, resfhp, iap);
1354 * nfsd_setattr already committed the child. Transactional filesystems
1355 * had a chance to commit changes for both parent and child
1356 * simultaneously making the following commit_metadata a noop.
1358 err2 = nfserrno(commit_metadata(fhp));
1363 * Update the file handle to get the new inode info.
1366 err = fh_update(resfhp);
1368 if (dchild && !IS_ERR(dchild))
1373 err = nfserrno(host_err);
1377 #ifdef CONFIG_NFSD_V3
1379 static inline int nfsd_create_is_exclusive(int createmode)
1381 return createmode == NFS3_CREATE_EXCLUSIVE
1382 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1386 * NFSv3 and NFSv4 version of nfsd_create
1389 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1390 char *fname, int flen, struct iattr *iap,
1391 struct svc_fh *resfhp, int createmode, u32 *verifier,
1392 bool *truncp, bool *created)
1394 struct dentry *dentry, *dchild = NULL;
1398 __u32 v_mtime=0, v_atime=0;
1404 if (isdotent(fname, flen))
1406 if (!(iap->ia_valid & ATTR_MODE))
1408 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1412 dentry = fhp->fh_dentry;
1413 dirp = dentry->d_inode;
1415 /* Get all the sanity checks out of the way before
1416 * we lock the parent. */
1417 err = nfserr_notdir;
1418 if (!dirp->i_op->lookup)
1420 fh_lock_nested(fhp, I_MUTEX_PARENT);
1423 * Compose the response file handle.
1425 dchild = lookup_one_len(fname, dentry, flen);
1426 host_err = PTR_ERR(dchild);
1430 /* If file doesn't exist, check for permissions to create one */
1431 if (!dchild->d_inode) {
1432 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1437 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1441 if (nfsd_create_is_exclusive(createmode)) {
1442 /* solaris7 gets confused (bugid 4218508) if these have
1443 * the high bit set, so just clear the high bits. If this is
1444 * ever changed to use different attrs for storing the
1445 * verifier, then do_open_lookup() will also need to be fixed
1448 v_mtime = verifier[0]&0x7fffffff;
1449 v_atime = verifier[1]&0x7fffffff;
1452 host_err = fh_want_write(fhp);
1455 if (dchild->d_inode) {
1458 switch (createmode) {
1459 case NFS3_CREATE_UNCHECKED:
1460 if (! S_ISREG(dchild->d_inode->i_mode))
1463 /* in nfsv4, we need to treat this case a little
1464 * differently. we don't want to truncate the
1465 * file now; this would be wrong if the OPEN
1466 * fails for some other reason. furthermore,
1467 * if the size is nonzero, we should ignore it
1468 * according to spec!
1470 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1473 iap->ia_valid &= ATTR_SIZE;
1477 case NFS3_CREATE_EXCLUSIVE:
1478 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1479 && dchild->d_inode->i_atime.tv_sec == v_atime
1480 && dchild->d_inode->i_size == 0 )
1482 case NFS4_CREATE_EXCLUSIVE4_1:
1483 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1484 && dchild->d_inode->i_atime.tv_sec == v_atime
1485 && dchild->d_inode->i_size == 0 )
1488 case NFS3_CREATE_GUARDED:
1495 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1503 nfsd_check_ignore_resizing(iap);
1505 if (nfsd_create_is_exclusive(createmode)) {
1506 /* Cram the verifier into atime/mtime */
1507 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1508 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1509 /* XXX someone who knows this better please fix it for nsec */
1510 iap->ia_mtime.tv_sec = v_mtime;
1511 iap->ia_atime.tv_sec = v_atime;
1512 iap->ia_mtime.tv_nsec = 0;
1513 iap->ia_atime.tv_nsec = 0;
1517 err = nfsd_create_setattr(rqstp, resfhp, iap);
1520 * nfsd_setattr already committed the child (and possibly also the parent).
1523 err = nfserrno(commit_metadata(fhp));
1527 * Update the filehandle to get the new inode info.
1530 err = fh_update(resfhp);
1534 if (dchild && !IS_ERR(dchild))
1539 err = nfserrno(host_err);
1542 #endif /* CONFIG_NFSD_V3 */
1545 * Read a symlink. On entry, *lenp must contain the maximum path length that
1546 * fits into the buffer. On return, it contains the true length.
1547 * N.B. After this call fhp needs an fh_put
1550 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1552 struct inode *inode;
1558 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1562 path.mnt = fhp->fh_export->ex_path.mnt;
1563 path.dentry = fhp->fh_dentry;
1564 inode = path.dentry->d_inode;
1567 if (!inode->i_op->readlink)
1571 /* N.B. Why does this call need a get_fs()??
1572 * Remove the set_fs and watch the fireworks:-) --okir
1575 oldfs = get_fs(); set_fs(KERNEL_DS);
1576 host_err = inode->i_op->readlink(path.dentry, buf, *lenp);
1587 err = nfserrno(host_err);
1592 * Create a symlink and look up its inode
1593 * N.B. After this call _both_ fhp and resfhp need an fh_put
1596 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1597 char *fname, int flen,
1598 char *path, int plen,
1599 struct svc_fh *resfhp,
1602 struct dentry *dentry, *dnew;
1610 if (isdotent(fname, flen))
1613 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1617 dentry = fhp->fh_dentry;
1618 dnew = lookup_one_len(fname, dentry, flen);
1619 host_err = PTR_ERR(dnew);
1623 host_err = fh_want_write(fhp);
1627 if (unlikely(path[plen] != 0)) {
1628 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1629 if (path_alloced == NULL)
1632 strncpy(path_alloced, path, plen);
1633 path_alloced[plen] = 0;
1634 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1635 kfree(path_alloced);
1638 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1639 err = nfserrno(host_err);
1641 err = nfserrno(commit_metadata(fhp));
1646 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1648 if (err==0) err = cerr;
1653 err = nfserrno(host_err);
1659 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1662 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1663 char *name, int len, struct svc_fh *tfhp)
1665 struct dentry *ddir, *dnew, *dold;
1670 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1673 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1677 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1683 if (isdotent(name, len))
1686 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1687 ddir = ffhp->fh_dentry;
1688 dirp = ddir->d_inode;
1690 dnew = lookup_one_len(name, ddir, len);
1691 host_err = PTR_ERR(dnew);
1695 dold = tfhp->fh_dentry;
1697 host_err = fh_want_write(tfhp);
1699 err = nfserrno(host_err);
1704 goto out_drop_write;
1705 host_err = nfsd_break_lease(dold->d_inode);
1707 err = nfserrno(host_err);
1708 goto out_drop_write;
1710 host_err = vfs_link(dold, dirp, dnew);
1712 err = nfserrno(commit_metadata(ffhp));
1714 err = nfserrno(commit_metadata(tfhp));
1716 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1719 err = nfserrno(host_err);
1722 fh_drop_write(tfhp);
1731 err = nfserrno(host_err);
1737 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1740 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1741 struct svc_fh *tfhp, char *tname, int tlen)
1743 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1744 struct inode *fdir, *tdir;
1748 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1751 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1755 fdentry = ffhp->fh_dentry;
1756 fdir = fdentry->d_inode;
1758 tdentry = tfhp->fh_dentry;
1759 tdir = tdentry->d_inode;
1761 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1762 if (ffhp->fh_export != tfhp->fh_export)
1766 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1769 /* cannot use fh_lock as we need deadlock protective ordering
1770 * so do it by hand */
1771 trap = lock_rename(tdentry, fdentry);
1772 ffhp->fh_locked = tfhp->fh_locked = 1;
1776 odentry = lookup_one_len(fname, fdentry, flen);
1777 host_err = PTR_ERR(odentry);
1778 if (IS_ERR(odentry))
1782 if (!odentry->d_inode)
1785 if (odentry == trap)
1788 ndentry = lookup_one_len(tname, tdentry, tlen);
1789 host_err = PTR_ERR(ndentry);
1790 if (IS_ERR(ndentry))
1792 host_err = -ENOTEMPTY;
1793 if (ndentry == trap)
1797 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1799 host_err = fh_want_write(ffhp);
1803 host_err = nfsd_break_lease(odentry->d_inode);
1805 goto out_drop_write;
1806 if (ndentry->d_inode) {
1807 host_err = nfsd_break_lease(ndentry->d_inode);
1809 goto out_drop_write;
1811 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1813 host_err = commit_metadata(tfhp);
1815 host_err = commit_metadata(ffhp);
1818 fh_drop_write(ffhp);
1824 err = nfserrno(host_err);
1826 /* we cannot reply on fh_unlock on the two filehandles,
1827 * as that would do the wrong thing if the two directories
1828 * were the same, so again we do it by hand
1830 fill_post_wcc(ffhp);
1831 fill_post_wcc(tfhp);
1832 unlock_rename(tdentry, fdentry);
1833 ffhp->fh_locked = tfhp->fh_locked = 0;
1840 * Unlink a file or directory
1841 * N.B. After this call fhp needs an fh_put
1844 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1845 char *fname, int flen)
1847 struct dentry *dentry, *rdentry;
1853 if (!flen || isdotent(fname, flen))
1855 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1859 fh_lock_nested(fhp, I_MUTEX_PARENT);
1860 dentry = fhp->fh_dentry;
1861 dirp = dentry->d_inode;
1863 rdentry = lookup_one_len(fname, dentry, flen);
1864 host_err = PTR_ERR(rdentry);
1865 if (IS_ERR(rdentry))
1868 if (!rdentry->d_inode) {
1875 type = rdentry->d_inode->i_mode & S_IFMT;
1877 host_err = fh_want_write(fhp);
1881 host_err = nfsd_break_lease(rdentry->d_inode);
1883 goto out_drop_write;
1884 if (type != S_IFDIR)
1885 host_err = vfs_unlink(dirp, rdentry);
1887 host_err = vfs_rmdir(dirp, rdentry);
1889 host_err = commit_metadata(fhp);
1896 err = nfserrno(host_err);
1902 * We do this buffering because we must not call back into the file
1903 * system's ->lookup() method from the filldir callback. That may well
1904 * deadlock a number of file systems.
1906 * This is based heavily on the implementation of same in XFS.
1908 struct buffered_dirent {
1912 unsigned int d_type;
1916 struct readdir_data {
1922 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1923 loff_t offset, u64 ino, unsigned int d_type)
1925 struct readdir_data *buf = __buf;
1926 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1927 unsigned int reclen;
1929 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1930 if (buf->used + reclen > PAGE_SIZE) {
1935 de->namlen = namlen;
1936 de->offset = offset;
1938 de->d_type = d_type;
1939 memcpy(de->name, name, namlen);
1940 buf->used += reclen;
1945 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1946 struct readdir_cd *cdp, loff_t *offsetp)
1948 struct readdir_data buf;
1949 struct buffered_dirent *de;
1954 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1956 return nfserrno(-ENOMEM);
1961 struct inode *dir_inode = file->f_path.dentry->d_inode;
1962 unsigned int reclen;
1964 cdp->err = nfserr_eof; /* will be cleared on successful read */
1968 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1981 * Various filldir functions may end up calling back into
1982 * lookup_one_len() and the file system's ->lookup() method.
1983 * These expect i_mutex to be held, as it would within readdir.
1985 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1989 de = (struct buffered_dirent *)buf.dirent;
1991 offset = de->offset;
1993 if (func(cdp, de->name, de->namlen, de->offset,
1994 de->ino, de->d_type))
1997 if (cdp->err != nfs_ok)
2000 reclen = ALIGN(sizeof(*de) + de->namlen,
2003 de = (struct buffered_dirent *)((char *)de + reclen);
2005 mutex_unlock(&dir_inode->i_mutex);
2006 if (size > 0) /* We bailed out early */
2009 offset = vfs_llseek(file, 0, SEEK_CUR);
2012 free_page((unsigned long)(buf.dirent));
2015 return nfserrno(host_err);
2022 * Read entries from a directory.
2023 * The NFSv3/4 verifier we ignore for now.
2026 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2027 struct readdir_cd *cdp, filldir_t func)
2031 loff_t offset = *offsetp;
2032 int may_flags = NFSD_MAY_READ;
2034 /* NFSv2 only supports 32 bit cookies */
2035 if (rqstp->rq_vers > 2)
2036 may_flags |= NFSD_MAY_64BIT_COOKIE;
2038 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2042 offset = vfs_llseek(file, offset, SEEK_SET);
2044 err = nfserrno((int)offset);
2048 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2050 if (err == nfserr_eof || err == nfserr_toosmall)
2051 err = nfs_ok; /* can still be found in ->err */
2059 * Get file system stats
2060 * N.B. After this call fhp needs an fh_put
2063 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2067 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2069 struct path path = {
2070 .mnt = fhp->fh_export->ex_path.mnt,
2071 .dentry = fhp->fh_dentry,
2073 if (vfs_statfs(&path, stat))
2079 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2081 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2085 * Check for a user's access permissions to this inode.
2088 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2089 struct dentry *dentry, int acc)
2091 struct inode *inode = dentry->d_inode;
2094 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2097 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2099 (acc & NFSD_MAY_READ)? " read" : "",
2100 (acc & NFSD_MAY_WRITE)? " write" : "",
2101 (acc & NFSD_MAY_EXEC)? " exec" : "",
2102 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2103 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2104 (acc & NFSD_MAY_LOCK)? " lock" : "",
2105 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2107 IS_IMMUTABLE(inode)? " immut" : "",
2108 IS_APPEND(inode)? " append" : "",
2109 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2110 dprintk(" owner %d/%d user %d/%d\n",
2111 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2114 /* Normally we reject any write/sattr etc access on a read-only file
2115 * system. But if it is IRIX doing check on write-access for a
2116 * device special file, we ignore rofs.
2118 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2119 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2120 if (exp_rdonly(rqstp, exp) ||
2121 __mnt_is_readonly(exp->ex_path.mnt))
2123 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2126 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2129 if (acc & NFSD_MAY_LOCK) {
2130 /* If we cannot rely on authentication in NLM requests,
2131 * just allow locks, otherwise require read permission, or
2134 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2137 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2140 * The file owner always gets access permission for accesses that
2141 * would normally be checked at open time. This is to make
2142 * file access work even when the client has done a fchmod(fd, 0).
2144 * However, `cp foo bar' should fail nevertheless when bar is
2145 * readonly. A sensible way to do this might be to reject all
2146 * attempts to truncate a read-only file, because a creat() call
2147 * always implies file truncation.
2148 * ... but this isn't really fair. A process may reasonably call
2149 * ftruncate on an open file descriptor on a file with perm 000.
2150 * We must trust the client to do permission checking - using "ACCESS"
2153 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2154 inode->i_uid == current_fsuid())
2157 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2158 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2160 /* Allow read access to binaries even when mode 111 */
2161 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2162 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2163 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2164 err = inode_permission(inode, MAY_EXEC);
2166 return err? nfserrno(err) : 0;
2170 nfsd_racache_shutdown(void)
2172 struct raparms *raparm, *last_raparm;
2175 dprintk("nfsd: freeing readahead buffers.\n");
2177 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2178 raparm = raparm_hash[i].pb_head;
2180 last_raparm = raparm;
2181 raparm = raparm->p_next;
2184 raparm_hash[i].pb_head = NULL;
2188 * Initialize readahead param cache
2191 nfsd_racache_init(int cache_size)
2196 struct raparms **raparm = NULL;
2199 if (raparm_hash[0].pb_head)
2201 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2204 cache_size = nperbucket * RAPARM_HASH_SIZE;
2206 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2208 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2209 spin_lock_init(&raparm_hash[i].pb_lock);
2211 raparm = &raparm_hash[i].pb_head;
2212 for (j = 0; j < nperbucket; j++) {
2213 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2216 raparm = &(*raparm)->p_next;
2221 nfsdstats.ra_size = cache_size;
2225 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2226 nfsd_racache_shutdown();
2230 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2232 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2234 struct inode *inode = fhp->fh_dentry->d_inode;
2238 struct posix_acl *acl;
2240 if (!IS_POSIXACL(inode))
2241 return ERR_PTR(-EOPNOTSUPP);
2244 case ACL_TYPE_ACCESS:
2245 name = POSIX_ACL_XATTR_ACCESS;
2247 case ACL_TYPE_DEFAULT:
2248 name = POSIX_ACL_XATTR_DEFAULT;
2251 return ERR_PTR(-EOPNOTSUPP);
2254 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2256 return ERR_PTR(size);
2258 acl = posix_acl_from_xattr(value, size);
2264 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2266 struct inode *inode = fhp->fh_dentry->d_inode;
2272 if (!IS_POSIXACL(inode) ||
2273 !inode->i_op->setxattr || !inode->i_op->removexattr)
2276 case ACL_TYPE_ACCESS:
2277 name = POSIX_ACL_XATTR_ACCESS;
2279 case ACL_TYPE_DEFAULT:
2280 name = POSIX_ACL_XATTR_DEFAULT;
2286 if (acl && acl->a_count) {
2287 size = posix_acl_xattr_size(acl->a_count);
2288 value = kmalloc(size, GFP_KERNEL);
2291 error = posix_acl_to_xattr(acl, value, size);
2298 error = fh_want_write(fhp);
2302 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2304 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2307 error = vfs_removexattr(fhp->fh_dentry, name);
2308 if (error == -ENODATA)
2318 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */