1 #define MSNFS /* HACK HACK */
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
20 #include <linux/file.h>
21 #include <linux/splice.h>
22 #include <linux/fcntl.h>
23 #include <linux/namei.h>
24 #include <linux/delay.h>
25 #include <linux/quotaops.h>
26 #include <linux/fsnotify.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/xattr.h>
29 #include <linux/jhash.h>
30 #include <linux/ima.h>
31 #include <asm/uaccess.h>
35 #endif /* CONFIG_NFSD_V3 */
38 #include <linux/nfs4_acl.h>
39 #include <linux/nfsd_idmap.h>
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms *p_next;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
68 } ____cacheline_aligned_in_smp;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
76 nfsd_v4client(struct svc_rqst *rq)
78 return rq->rq_prog == NFS_PROGRAM && rq->rq_vers == 4;
82 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
84 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
85 * or nfs_ok having possibly changed *dpp and *expp
88 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
89 struct svc_export **expp)
91 struct svc_export *exp = *expp, *exp2 = NULL;
92 struct dentry *dentry = *dpp;
93 struct path path = {.mnt = mntget(exp->ex_path.mnt),
94 .dentry = dget(dentry)};
97 while (d_mountpoint(path.dentry) && follow_down(&path))
100 exp2 = rqst_exp_get_by_name(rqstp, &path);
104 * We normally allow NFS clients to continue
105 * "underneath" a mountpoint that is not exported.
106 * The exception is V4ROOT, where no traversal is ever
107 * allowed without an explicit export of the new
110 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
115 if (nfsd_v4client(rqstp) ||
116 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
117 /* successfully crossed mount point */
119 * This is subtle: path.dentry is *not* on path.mnt
120 * at this point. The only reason we are safe is that
121 * original mnt is pinned down by exp, so we should
122 * put path *before* putting exp
125 path.dentry = dentry;
135 static void follow_to_parent(struct path *path)
139 while (path->dentry == path->mnt->mnt_root && follow_up(path))
141 dp = dget_parent(path->dentry);
146 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
148 struct svc_export *exp2;
149 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
150 .dentry = dget(dparent)};
152 follow_to_parent(&path);
154 exp2 = rqst_exp_parent(rqstp, &path);
155 if (PTR_ERR(exp2) == -ENOENT) {
156 *dentryp = dget(dparent);
157 } else if (IS_ERR(exp2)) {
159 return PTR_ERR(exp2);
161 *dentryp = dget(path.dentry);
170 * For nfsd purposes, we treat V4ROOT exports as though there was an
171 * export at *every* directory.
173 static int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
175 if (d_mountpoint(dentry))
177 if (!(exp->ex_flags & NFSEXP_V4ROOT))
179 return dentry->d_inode != NULL;
183 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
184 const char *name, unsigned int len,
185 struct svc_export **exp_ret, struct dentry **dentry_ret)
187 struct svc_export *exp;
188 struct dentry *dparent;
189 struct dentry *dentry;
193 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
195 /* Obtain dentry and export. */
196 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
200 dparent = fhp->fh_dentry;
201 exp = fhp->fh_export;
204 /* Lookup the name, but don't follow links */
205 if (isdotent(name, len)) {
207 dentry = dget(dparent);
208 else if (dparent != exp->ex_path.dentry)
209 dentry = dget_parent(dparent);
210 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
211 dentry = dget(dparent); /* .. == . just like at / */
213 /* checking mountpoint crossing is very different when stepping up */
214 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
220 dentry = lookup_one_len(name, dparent, len);
221 host_err = PTR_ERR(dentry);
225 * check if we have crossed a mount point ...
227 if (nfsd_mountpoint(dentry, exp)) {
228 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
234 *dentry_ret = dentry;
240 return nfserrno(host_err);
244 * Look up one component of a pathname.
245 * N.B. After this call _both_ fhp and resfh need an fh_put
247 * If the lookup would cross a mountpoint, and the mounted filesystem
248 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
249 * accepted as it stands and the mounted directory is
250 * returned. Otherwise the covered directory is returned.
251 * NOTE: this mountpoint crossing is not supported properly by all
252 * clients and is explicitly disallowed for NFSv3
253 * NeilBrown <neilb@cse.unsw.edu.au>
256 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
257 unsigned int len, struct svc_fh *resfh)
259 struct svc_export *exp;
260 struct dentry *dentry;
263 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
266 err = check_nfsd_access(exp, rqstp);
270 * Note: we compose the file handle now, but as the
271 * dentry may be negative, it may need to be updated.
273 err = fh_compose(resfh, exp, dentry, fhp);
274 if (!err && !dentry->d_inode)
284 * Set various file attributes.
285 * N.B. After this call fhp needs an fh_put
288 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
289 int check_guard, time_t guardtime)
291 struct dentry *dentry;
293 int accmode = NFSD_MAY_SATTR;
299 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
300 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
301 if (iap->ia_valid & ATTR_SIZE)
305 err = fh_verify(rqstp, fhp, ftype, accmode);
309 dentry = fhp->fh_dentry;
310 inode = dentry->d_inode;
312 /* Ignore any mode updates on symlinks */
313 if (S_ISLNK(inode->i_mode))
314 iap->ia_valid &= ~ATTR_MODE;
320 * NFSv2 does not differentiate between "set-[ac]time-to-now"
321 * which only requires access, and "set-[ac]time-to-X" which
322 * requires ownership.
323 * So if it looks like it might be "set both to the same time which
324 * is close to now", and if inode_change_ok fails, then we
325 * convert to "set to now" instead of "set to explicit time"
327 * We only call inode_change_ok as the last test as technically
328 * it is not an interface that we should be using. It is only
329 * valid if the filesystem does not define it's own i_op->setattr.
331 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
332 #define MAX_TOUCH_TIME_ERROR (30*60)
333 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
334 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
338 * Now just make sure time is in the right ballpark.
339 * Solaris, at least, doesn't seem to care what the time
340 * request is. We require it be within 30 minutes of now.
342 time_t delta = iap->ia_atime.tv_sec - get_seconds();
345 if (delta < MAX_TOUCH_TIME_ERROR &&
346 inode_change_ok(inode, iap) != 0) {
348 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
349 * This will cause notify_change to set these times
352 iap->ia_valid &= ~BOTH_TIME_SET;
357 * The size case is special.
358 * It changes the file as well as the attributes.
360 if (iap->ia_valid & ATTR_SIZE) {
361 if (iap->ia_size < inode->i_size) {
362 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
363 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
369 * If we are changing the size of the file, then
370 * we need to break all leases.
372 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
373 if (host_err == -EWOULDBLOCK)
374 host_err = -ETIMEDOUT;
375 if (host_err) /* ENOMEM or EWOULDBLOCK */
378 host_err = get_write_access(inode);
383 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
385 put_write_access(inode);
391 /* sanitize the mode change */
392 if (iap->ia_valid & ATTR_MODE) {
393 iap->ia_mode &= S_IALLUGO;
394 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
397 /* Revoke setuid/setgid on chown */
398 if (!S_ISDIR(inode->i_mode) &&
399 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
400 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
401 iap->ia_valid |= ATTR_KILL_PRIV;
402 if (iap->ia_valid & ATTR_MODE) {
403 /* we're setting mode too, just clear the s*id bits */
404 iap->ia_mode &= ~S_ISUID;
405 if (iap->ia_mode & S_IXGRP)
406 iap->ia_mode &= ~S_ISGID;
408 /* set ATTR_KILL_* bits and let VFS handle it */
409 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
413 /* Change the attributes. */
415 iap->ia_valid |= ATTR_CTIME;
417 err = nfserr_notsync;
418 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
420 host_err = notify_change(dentry, iap);
421 err = nfserrno(host_err);
425 put_write_access(inode);
427 if (EX_ISSYNC(fhp->fh_export))
428 write_inode_now(inode, 1);
433 err = nfserrno(host_err);
437 #if defined(CONFIG_NFSD_V2_ACL) || \
438 defined(CONFIG_NFSD_V3_ACL) || \
439 defined(CONFIG_NFSD_V4)
440 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
445 buflen = vfs_getxattr(dentry, key, NULL, 0);
449 *buf = kmalloc(buflen, GFP_KERNEL);
453 ret = vfs_getxattr(dentry, key, *buf, buflen);
460 #if defined(CONFIG_NFSD_V4)
462 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
469 buflen = posix_acl_xattr_size(pacl->a_count);
470 buf = kmalloc(buflen, GFP_KERNEL);
475 len = posix_acl_to_xattr(pacl, buf, buflen);
481 error = vfs_setxattr(dentry, key, buf, len, 0);
488 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
489 struct nfs4_acl *acl)
493 struct dentry *dentry;
495 struct posix_acl *pacl = NULL, *dpacl = NULL;
496 unsigned int flags = 0;
499 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
503 dentry = fhp->fh_dentry;
504 inode = dentry->d_inode;
505 if (S_ISDIR(inode->i_mode))
506 flags = NFS4_ACL_DIR;
508 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
509 if (host_error == -EINVAL) {
510 return nfserr_attrnotsupp;
511 } else if (host_error < 0)
514 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
518 if (S_ISDIR(inode->i_mode))
519 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
522 posix_acl_release(pacl);
523 posix_acl_release(dpacl);
525 if (host_error == -EOPNOTSUPP)
526 return nfserr_attrnotsupp;
528 return nfserrno(host_error);
531 static struct posix_acl *
532 _get_posix_acl(struct dentry *dentry, char *key)
535 struct posix_acl *pacl = NULL;
538 buflen = nfsd_getxattr(dentry, key, &buf);
542 return ERR_PTR(buflen);
544 pacl = posix_acl_from_xattr(buf, buflen);
550 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
552 struct inode *inode = dentry->d_inode;
554 struct posix_acl *pacl = NULL, *dpacl = NULL;
555 unsigned int flags = 0;
557 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
558 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
559 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
561 error = PTR_ERR(pacl);
566 if (S_ISDIR(inode->i_mode)) {
567 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
568 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
570 else if (IS_ERR(dpacl)) {
571 error = PTR_ERR(dpacl);
575 flags = NFS4_ACL_DIR;
578 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
580 error = PTR_ERR(*acl);
584 posix_acl_release(pacl);
585 posix_acl_release(dpacl);
589 #endif /* defined(CONFIG_NFS_V4) */
591 #ifdef CONFIG_NFSD_V3
593 * Check server access rights to a file system object
599 static struct accessmap nfs3_regaccess[] = {
600 { NFS3_ACCESS_READ, NFSD_MAY_READ },
601 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
602 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
603 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
608 static struct accessmap nfs3_diraccess[] = {
609 { NFS3_ACCESS_READ, NFSD_MAY_READ },
610 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
611 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
612 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
613 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
618 static struct accessmap nfs3_anyaccess[] = {
619 /* Some clients - Solaris 2.6 at least, make an access call
620 * to the server to check for access for things like /dev/null
621 * (which really, the server doesn't care about). So
622 * We provide simple access checking for them, looking
623 * mainly at mode bits, and we make sure to ignore read-only
626 { NFS3_ACCESS_READ, NFSD_MAY_READ },
627 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
628 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
629 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
635 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
637 struct accessmap *map;
638 struct svc_export *export;
639 struct dentry *dentry;
640 u32 query, result = 0, sresult = 0;
643 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
647 export = fhp->fh_export;
648 dentry = fhp->fh_dentry;
650 if (S_ISREG(dentry->d_inode->i_mode))
651 map = nfs3_regaccess;
652 else if (S_ISDIR(dentry->d_inode->i_mode))
653 map = nfs3_diraccess;
655 map = nfs3_anyaccess;
659 for (; map->access; map++) {
660 if (map->access & query) {
663 sresult |= map->access;
665 err2 = nfsd_permission(rqstp, export, dentry, map->how);
668 result |= map->access;
671 /* the following error codes just mean the access was not allowed,
672 * rather than an error occurred */
676 /* simply don't "or" in the access bit. */
686 *supported = sresult;
691 #endif /* CONFIG_NFSD_V3 */
696 * Open an existing file or directory.
697 * The access argument indicates the type of open (read/write/lock)
698 * N.B. After this call fhp needs an fh_put
701 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
702 int access, struct file **filp)
704 struct dentry *dentry;
706 int flags = O_RDONLY|O_LARGEFILE;
710 validate_process_creds();
713 * If we get here, then the client has already done an "open",
714 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
715 * in case a chmod has now revoked permission.
717 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
721 dentry = fhp->fh_dentry;
722 inode = dentry->d_inode;
724 /* Disallow write access to files with the append-only bit set
725 * or any access when mandatory locking enabled
728 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
731 * We must ignore files (but only files) which might have mandatory
732 * locks on them because there is no way to know if the accesser has
735 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
742 * Check to see if there are any leases on this file.
743 * This may block while leases are broken.
745 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
746 if (host_err == -EWOULDBLOCK)
747 host_err = -ETIMEDOUT;
748 if (host_err) /* NOMEM or WOULDBLOCK */
751 if (access & NFSD_MAY_WRITE) {
752 if (access & NFSD_MAY_READ)
753 flags = O_RDWR|O_LARGEFILE;
755 flags = O_WRONLY|O_LARGEFILE;
759 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
760 flags, current_cred());
762 host_err = PTR_ERR(*filp);
764 ima_counts_get(*filp);
766 err = nfserrno(host_err);
768 validate_process_creds();
776 nfsd_close(struct file *filp)
783 * As this calls fsync (not fdatasync) there is no need for a write_inode
786 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
787 const struct file_operations *fop)
789 struct inode *inode = dp->d_inode;
790 int (*fsync) (struct file *, struct dentry *, int);
793 err = filemap_fdatawrite(inode->i_mapping);
794 if (err == 0 && fop && (fsync = fop->fsync))
795 err = fsync(filp, dp, 0);
797 err = filemap_fdatawait(inode->i_mapping);
803 nfsd_sync(struct file *filp)
806 struct inode *inode = filp->f_path.dentry->d_inode;
807 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
808 mutex_lock(&inode->i_mutex);
809 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
810 mutex_unlock(&inode->i_mutex);
816 nfsd_sync_dir(struct dentry *dp)
818 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
822 * Obtain the readahead parameters for the file
823 * specified by (dev, ino).
826 static inline struct raparms *
827 nfsd_get_raparms(dev_t dev, ino_t ino)
829 struct raparms *ra, **rap, **frap = NULL;
832 struct raparm_hbucket *rab;
834 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
835 rab = &raparm_hash[hash];
837 spin_lock(&rab->pb_lock);
838 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
839 if (ra->p_ino == ino && ra->p_dev == dev)
842 if (ra->p_count == 0)
845 depth = nfsdstats.ra_size*11/10;
847 spin_unlock(&rab->pb_lock);
857 if (rap != &rab->pb_head) {
859 ra->p_next = rab->pb_head;
863 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
864 spin_unlock(&rab->pb_lock);
869 * Grab and keep cached pages associated with a file in the svc_rqst
870 * so that they can be passed to the network sendmsg/sendpage routines
871 * directly. They will be released after the sending has completed.
874 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
875 struct splice_desc *sd)
877 struct svc_rqst *rqstp = sd->u.data;
878 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
879 struct page *page = buf->page;
883 ret = buf->ops->confirm(pipe, buf);
889 if (rqstp->rq_res.page_len == 0) {
894 rqstp->rq_res.page_base = buf->offset;
895 rqstp->rq_res.page_len = size;
896 } else if (page != pp[-1]) {
902 rqstp->rq_res.page_len += size;
904 rqstp->rq_res.page_len += size;
909 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
910 struct splice_desc *sd)
912 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
915 static inline int svc_msnfs(struct svc_fh *ffhp)
918 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
925 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
926 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
935 inode = file->f_path.dentry->d_inode;
937 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
940 /* Get readahead parameters */
941 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
944 file->f_ra = ra->p_ra;
946 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
947 struct splice_desc sd = {
954 rqstp->rq_resused = 1;
955 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
959 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
963 /* Write back readahead params */
965 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
966 spin_lock(&rab->pb_lock);
967 ra->p_ra = file->f_ra;
970 spin_unlock(&rab->pb_lock);
974 nfsdstats.io_read += host_err;
977 fsnotify_access(file->f_path.dentry);
979 err = nfserrno(host_err);
984 static void kill_suid(struct dentry *dentry)
987 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
989 mutex_lock(&dentry->d_inode->i_mutex);
990 notify_change(dentry, &ia);
991 mutex_unlock(&dentry->d_inode->i_mutex);
995 * Gathered writes: If another process is currently writing to the file,
996 * there's a high chance this is another nfsd (triggered by a bulk write
997 * from a client's biod). Rather than syncing the file with each write
998 * request, we sleep for 10 msec.
1000 * I don't know if this roughly approximates C. Juszak's idea of
1001 * gathered writes, but it's a nice and simple solution (IMHO), and it
1004 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1005 * better tool (separate unstable writes and commits) for solving this
1008 static int wait_for_concurrent_writes(struct file *file)
1010 struct inode *inode = file->f_path.dentry->d_inode;
1011 static ino_t last_ino;
1012 static dev_t last_dev;
1015 if (atomic_read(&inode->i_writecount) > 1
1016 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1017 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1019 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1022 if (inode->i_state & I_DIRTY) {
1023 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1024 err = nfsd_sync(file);
1026 last_ino = inode->i_ino;
1027 last_dev = inode->i_sb->s_dev;
1032 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1033 loff_t offset, struct kvec *vec, int vlen,
1034 unsigned long *cnt, int *stablep)
1036 struct svc_export *exp;
1037 struct dentry *dentry;
1038 struct inode *inode;
1042 int stable = *stablep;
1048 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1049 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1053 dentry = file->f_path.dentry;
1054 inode = dentry->d_inode;
1055 exp = fhp->fh_export;
1058 * Request sync writes if
1059 * - the sync export option has been set, or
1060 * - the client requested O_SYNC behavior (NFSv3 feature).
1061 * - The file system doesn't support fsync().
1062 * When NFSv2 gathered writes have been configured for this volume,
1063 * flushing the data to disk is handled separately below.
1065 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1067 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1069 *stablep = 2; /* FILE_SYNC */
1072 if (!EX_ISSYNC(exp))
1074 if (stable && !use_wgather) {
1075 spin_lock(&file->f_lock);
1076 file->f_flags |= O_SYNC;
1077 spin_unlock(&file->f_lock);
1080 /* Write the data. */
1081 oldfs = get_fs(); set_fs(KERNEL_DS);
1082 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1087 nfsdstats.io_write += host_err;
1088 fsnotify_modify(file->f_path.dentry);
1090 /* clear setuid/setgid flag after write */
1091 if (inode->i_mode & (S_ISUID | S_ISGID))
1094 if (stable && use_wgather)
1095 host_err = wait_for_concurrent_writes(file);
1098 dprintk("nfsd: write complete host_err=%d\n", host_err);
1102 err = nfserrno(host_err);
1108 * Read data from a file. count must contain the requested read count
1109 * on entry. On return, *count contains the number of bytes actually read.
1110 * N.B. After this call fhp needs an fh_put
1113 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1114 loff_t offset, struct kvec *vec, int vlen,
1115 unsigned long *count)
1120 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1121 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1124 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1126 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1129 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1137 * Write data to a file.
1138 * The stable flag requests synchronous writes.
1139 * N.B. After this call fhp needs an fh_put
1142 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1143 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1149 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1150 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1153 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1156 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1161 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1169 #ifdef CONFIG_NFSD_V3
1171 * Commit all pending writes to stable storage.
1172 * Strictly speaking, we could sync just the indicated file region here,
1173 * but there's currently no way we can ask the VFS to do so.
1175 * Unfortunately we cannot lock the file to make sure we return full WCC
1176 * data to the client, as locking happens lower down in the filesystem.
1179 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1180 loff_t offset, unsigned long count)
1185 if ((u64)count > ~(u64)offset)
1186 return nfserr_inval;
1188 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1191 if (EX_ISSYNC(fhp->fh_export)) {
1192 if (file->f_op && file->f_op->fsync) {
1193 err = nfserrno(nfsd_sync(file));
1195 err = nfserr_notsupp;
1202 #endif /* CONFIG_NFSD_V3 */
1205 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1209 * Mode has already been set earlier in create:
1211 iap->ia_valid &= ~ATTR_MODE;
1213 * Setting uid/gid works only for root. Irix appears to
1214 * send along the gid on create when it tries to implement
1215 * setgid directories via NFS:
1217 if (current_fsuid() != 0)
1218 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1220 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1224 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1225 * setting size to 0 may fail for some specific file systems by the permission
1226 * checking which requires WRITE permission but the mode is 000.
1227 * we ignore the resizing(to 0) on the just new created file, since the size is
1228 * 0 after file created.
1230 * call this only after vfs_create() is called.
1233 nfsd_check_ignore_resizing(struct iattr *iap)
1235 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1236 iap->ia_valid &= ~ATTR_SIZE;
1240 * Create a file (regular, directory, device, fifo); UNIX sockets
1241 * not yet implemented.
1242 * If the response fh has been verified, the parent directory should
1243 * already be locked. Note that the parent directory is left locked.
1245 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1248 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1249 char *fname, int flen, struct iattr *iap,
1250 int type, dev_t rdev, struct svc_fh *resfhp)
1252 struct dentry *dentry, *dchild = NULL;
1262 if (isdotent(fname, flen))
1265 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1269 dentry = fhp->fh_dentry;
1270 dirp = dentry->d_inode;
1272 err = nfserr_notdir;
1273 if (!dirp->i_op->lookup)
1276 * Check whether the response file handle has been verified yet.
1277 * If it has, the parent directory should already be locked.
1279 if (!resfhp->fh_dentry) {
1280 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1281 fh_lock_nested(fhp, I_MUTEX_PARENT);
1282 dchild = lookup_one_len(fname, dentry, flen);
1283 host_err = PTR_ERR(dchild);
1286 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1290 /* called from nfsd_proc_create */
1291 dchild = dget(resfhp->fh_dentry);
1292 if (!fhp->fh_locked) {
1293 /* not actually possible */
1295 "nfsd_create: parent %s/%s not locked!\n",
1296 dentry->d_parent->d_name.name,
1297 dentry->d_name.name);
1303 * Make sure the child dentry is still negative ...
1306 if (dchild->d_inode) {
1307 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1308 dentry->d_name.name, dchild->d_name.name);
1312 if (!(iap->ia_valid & ATTR_MODE))
1314 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1317 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1318 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1323 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1328 * Get the dir op function pointer.
1333 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1335 nfsd_check_ignore_resizing(iap);
1338 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1344 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1348 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1352 if (EX_ISSYNC(fhp->fh_export)) {
1353 err = nfserrno(nfsd_sync_dir(dentry));
1354 write_inode_now(dchild->d_inode, 1);
1357 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1360 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1362 * Update the file handle to get the new inode info.
1365 err = fh_update(resfhp);
1367 if (dchild && !IS_ERR(dchild))
1372 err = nfserrno(host_err);
1376 #ifdef CONFIG_NFSD_V3
1378 * NFSv3 version of nfsd_create
1381 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1382 char *fname, int flen, struct iattr *iap,
1383 struct svc_fh *resfhp, int createmode, u32 *verifier,
1384 int *truncp, int *created)
1386 struct dentry *dentry, *dchild = NULL;
1391 __u32 v_mtime=0, v_atime=0;
1397 if (isdotent(fname, flen))
1399 if (!(iap->ia_valid & ATTR_MODE))
1401 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1405 dentry = fhp->fh_dentry;
1406 dirp = dentry->d_inode;
1408 /* Get all the sanity checks out of the way before
1409 * we lock the parent. */
1410 err = nfserr_notdir;
1411 if (!dirp->i_op->lookup)
1413 fh_lock_nested(fhp, I_MUTEX_PARENT);
1416 * Compose the response file handle.
1418 dchild = lookup_one_len(fname, dentry, flen);
1419 host_err = PTR_ERR(dchild);
1423 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1427 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1428 /* solaris7 gets confused (bugid 4218508) if these have
1429 * the high bit set, so just clear the high bits. If this is
1430 * ever changed to use different attrs for storing the
1431 * verifier, then do_open_lookup() will also need to be fixed
1434 v_mtime = verifier[0]&0x7fffffff;
1435 v_atime = verifier[1]&0x7fffffff;
1438 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1441 if (dchild->d_inode) {
1444 switch (createmode) {
1445 case NFS3_CREATE_UNCHECKED:
1446 if (! S_ISREG(dchild->d_inode->i_mode))
1449 /* in nfsv4, we need to treat this case a little
1450 * differently. we don't want to truncate the
1451 * file now; this would be wrong if the OPEN
1452 * fails for some other reason. furthermore,
1453 * if the size is nonzero, we should ignore it
1454 * according to spec!
1456 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1459 iap->ia_valid &= ATTR_SIZE;
1463 case NFS3_CREATE_EXCLUSIVE:
1464 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1465 && dchild->d_inode->i_atime.tv_sec == v_atime
1466 && dchild->d_inode->i_size == 0 )
1469 case NFS3_CREATE_GUARDED:
1472 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1476 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1478 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1484 if (EX_ISSYNC(fhp->fh_export)) {
1485 err = nfserrno(nfsd_sync_dir(dentry));
1486 /* setattr will sync the child (or not) */
1489 nfsd_check_ignore_resizing(iap);
1491 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1492 /* Cram the verifier into atime/mtime */
1493 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1494 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1495 /* XXX someone who knows this better please fix it for nsec */
1496 iap->ia_mtime.tv_sec = v_mtime;
1497 iap->ia_atime.tv_sec = v_atime;
1498 iap->ia_mtime.tv_nsec = 0;
1499 iap->ia_atime.tv_nsec = 0;
1503 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1507 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1509 * Update the filehandle to get the new inode info.
1512 err = fh_update(resfhp);
1516 if (dchild && !IS_ERR(dchild))
1521 err = nfserrno(host_err);
1524 #endif /* CONFIG_NFSD_V3 */
1527 * Read a symlink. On entry, *lenp must contain the maximum path length that
1528 * fits into the buffer. On return, it contains the true length.
1529 * N.B. After this call fhp needs an fh_put
1532 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1534 struct dentry *dentry;
1535 struct inode *inode;
1540 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1544 dentry = fhp->fh_dentry;
1545 inode = dentry->d_inode;
1548 if (!inode->i_op->readlink)
1551 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1552 /* N.B. Why does this call need a get_fs()??
1553 * Remove the set_fs and watch the fireworks:-) --okir
1556 oldfs = get_fs(); set_fs(KERNEL_DS);
1557 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1568 err = nfserrno(host_err);
1573 * Create a symlink and look up its inode
1574 * N.B. After this call _both_ fhp and resfhp need an fh_put
1577 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1578 char *fname, int flen,
1579 char *path, int plen,
1580 struct svc_fh *resfhp,
1583 struct dentry *dentry, *dnew;
1591 if (isdotent(fname, flen))
1594 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1598 dentry = fhp->fh_dentry;
1599 dnew = lookup_one_len(fname, dentry, flen);
1600 host_err = PTR_ERR(dnew);
1604 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1608 if (unlikely(path[plen] != 0)) {
1609 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1610 if (path_alloced == NULL)
1613 strncpy(path_alloced, path, plen);
1614 path_alloced[plen] = 0;
1615 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1616 kfree(path_alloced);
1619 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1622 if (EX_ISSYNC(fhp->fh_export))
1623 host_err = nfsd_sync_dir(dentry);
1625 err = nfserrno(host_err);
1628 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1630 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1632 if (err==0) err = cerr;
1637 err = nfserrno(host_err);
1643 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1646 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1647 char *name, int len, struct svc_fh *tfhp)
1649 struct dentry *ddir, *dnew, *dold;
1650 struct inode *dirp, *dest;
1654 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1657 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1665 if (isdotent(name, len))
1668 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1669 ddir = ffhp->fh_dentry;
1670 dirp = ddir->d_inode;
1672 dnew = lookup_one_len(name, ddir, len);
1673 host_err = PTR_ERR(dnew);
1677 dold = tfhp->fh_dentry;
1678 dest = dold->d_inode;
1680 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1682 err = nfserrno(host_err);
1685 host_err = vfs_link(dold, dirp, dnew);
1687 if (EX_ISSYNC(ffhp->fh_export)) {
1688 err = nfserrno(nfsd_sync_dir(ddir));
1689 write_inode_now(dest, 1);
1693 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1696 err = nfserrno(host_err);
1698 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1707 err = nfserrno(host_err);
1713 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1716 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1717 struct svc_fh *tfhp, char *tname, int tlen)
1719 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1720 struct inode *fdir, *tdir;
1724 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1727 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1731 fdentry = ffhp->fh_dentry;
1732 fdir = fdentry->d_inode;
1734 tdentry = tfhp->fh_dentry;
1735 tdir = tdentry->d_inode;
1737 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1738 if (ffhp->fh_export != tfhp->fh_export)
1742 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1745 /* cannot use fh_lock as we need deadlock protective ordering
1746 * so do it by hand */
1747 trap = lock_rename(tdentry, fdentry);
1748 ffhp->fh_locked = tfhp->fh_locked = 1;
1752 odentry = lookup_one_len(fname, fdentry, flen);
1753 host_err = PTR_ERR(odentry);
1754 if (IS_ERR(odentry))
1758 if (!odentry->d_inode)
1761 if (odentry == trap)
1764 ndentry = lookup_one_len(tname, tdentry, tlen);
1765 host_err = PTR_ERR(ndentry);
1766 if (IS_ERR(ndentry))
1768 host_err = -ENOTEMPTY;
1769 if (ndentry == trap)
1772 if (svc_msnfs(ffhp) &&
1773 ((atomic_read(&odentry->d_count) > 1)
1774 || (atomic_read(&ndentry->d_count) > 1))) {
1780 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1782 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1786 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1787 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1788 host_err = nfsd_sync_dir(tdentry);
1790 host_err = nfsd_sync_dir(fdentry);
1793 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1800 err = nfserrno(host_err);
1802 /* we cannot reply on fh_unlock on the two filehandles,
1803 * as that would do the wrong thing if the two directories
1804 * were the same, so again we do it by hand
1806 fill_post_wcc(ffhp);
1807 fill_post_wcc(tfhp);
1808 unlock_rename(tdentry, fdentry);
1809 ffhp->fh_locked = tfhp->fh_locked = 0;
1816 * Unlink a file or directory
1817 * N.B. After this call fhp needs an fh_put
1820 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1821 char *fname, int flen)
1823 struct dentry *dentry, *rdentry;
1829 if (!flen || isdotent(fname, flen))
1831 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1835 fh_lock_nested(fhp, I_MUTEX_PARENT);
1836 dentry = fhp->fh_dentry;
1837 dirp = dentry->d_inode;
1839 rdentry = lookup_one_len(fname, dentry, flen);
1840 host_err = PTR_ERR(rdentry);
1841 if (IS_ERR(rdentry))
1844 if (!rdentry->d_inode) {
1851 type = rdentry->d_inode->i_mode & S_IFMT;
1853 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1857 if (type != S_IFDIR) { /* It's UNLINK */
1859 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1860 (atomic_read(&rdentry->d_count) > 1)) {
1864 host_err = vfs_unlink(dirp, rdentry);
1865 } else { /* It's RMDIR */
1866 host_err = vfs_rmdir(dirp, rdentry);
1873 if (EX_ISSYNC(fhp->fh_export))
1874 host_err = nfsd_sync_dir(dentry);
1877 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1879 err = nfserrno(host_err);
1885 * We do this buffering because we must not call back into the file
1886 * system's ->lookup() method from the filldir callback. That may well
1887 * deadlock a number of file systems.
1889 * This is based heavily on the implementation of same in XFS.
1891 struct buffered_dirent {
1895 unsigned int d_type;
1899 struct readdir_data {
1905 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1906 loff_t offset, u64 ino, unsigned int d_type)
1908 struct readdir_data *buf = __buf;
1909 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1910 unsigned int reclen;
1912 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1913 if (buf->used + reclen > PAGE_SIZE) {
1918 de->namlen = namlen;
1919 de->offset = offset;
1921 de->d_type = d_type;
1922 memcpy(de->name, name, namlen);
1923 buf->used += reclen;
1928 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1929 struct readdir_cd *cdp, loff_t *offsetp)
1931 struct readdir_data buf;
1932 struct buffered_dirent *de;
1937 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1939 return nfserrno(-ENOMEM);
1944 struct inode *dir_inode = file->f_path.dentry->d_inode;
1945 unsigned int reclen;
1947 cdp->err = nfserr_eof; /* will be cleared on successful read */
1951 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1964 * Various filldir functions may end up calling back into
1965 * lookup_one_len() and the file system's ->lookup() method.
1966 * These expect i_mutex to be held, as it would within readdir.
1968 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1972 de = (struct buffered_dirent *)buf.dirent;
1974 offset = de->offset;
1976 if (func(cdp, de->name, de->namlen, de->offset,
1977 de->ino, de->d_type))
1980 if (cdp->err != nfs_ok)
1983 reclen = ALIGN(sizeof(*de) + de->namlen,
1986 de = (struct buffered_dirent *)((char *)de + reclen);
1988 mutex_unlock(&dir_inode->i_mutex);
1989 if (size > 0) /* We bailed out early */
1992 offset = vfs_llseek(file, 0, SEEK_CUR);
1995 free_page((unsigned long)(buf.dirent));
1998 return nfserrno(host_err);
2005 * Read entries from a directory.
2006 * The NFSv3/4 verifier we ignore for now.
2009 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2010 struct readdir_cd *cdp, filldir_t func)
2014 loff_t offset = *offsetp;
2016 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2020 offset = vfs_llseek(file, offset, 0);
2022 err = nfserrno((int)offset);
2026 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2028 if (err == nfserr_eof || err == nfserr_toosmall)
2029 err = nfs_ok; /* can still be found in ->err */
2037 * Get file system stats
2038 * N.B. After this call fhp needs an fh_put
2041 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2043 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2044 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2049 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2051 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2055 * Check for a user's access permissions to this inode.
2058 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2059 struct dentry *dentry, int acc)
2061 struct inode *inode = dentry->d_inode;
2065 if (acc == NFSD_MAY_NOP)
2068 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2070 (acc & NFSD_MAY_READ)? " read" : "",
2071 (acc & NFSD_MAY_WRITE)? " write" : "",
2072 (acc & NFSD_MAY_EXEC)? " exec" : "",
2073 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2074 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2075 (acc & NFSD_MAY_LOCK)? " lock" : "",
2076 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2078 IS_IMMUTABLE(inode)? " immut" : "",
2079 IS_APPEND(inode)? " append" : "",
2080 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2081 dprintk(" owner %d/%d user %d/%d\n",
2082 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2085 /* Normally we reject any write/sattr etc access on a read-only file
2086 * system. But if it is IRIX doing check on write-access for a
2087 * device special file, we ignore rofs.
2089 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2090 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2091 if (exp_rdonly(rqstp, exp) ||
2092 __mnt_is_readonly(exp->ex_path.mnt))
2094 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2097 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2100 if (acc & NFSD_MAY_LOCK) {
2101 /* If we cannot rely on authentication in NLM requests,
2102 * just allow locks, otherwise require read permission, or
2105 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2108 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2111 * The file owner always gets access permission for accesses that
2112 * would normally be checked at open time. This is to make
2113 * file access work even when the client has done a fchmod(fd, 0).
2115 * However, `cp foo bar' should fail nevertheless when bar is
2116 * readonly. A sensible way to do this might be to reject all
2117 * attempts to truncate a read-only file, because a creat() call
2118 * always implies file truncation.
2119 * ... but this isn't really fair. A process may reasonably call
2120 * ftruncate on an open file descriptor on a file with perm 000.
2121 * We must trust the client to do permission checking - using "ACCESS"
2124 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2125 inode->i_uid == current_fsuid())
2128 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2129 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2131 /* Allow read access to binaries even when mode 111 */
2132 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2133 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2134 err = inode_permission(inode, MAY_EXEC);
2138 /* Do integrity (permission) checking now, but defer incrementing
2139 * IMA counts to the actual file open.
2141 path.mnt = exp->ex_path.mnt;
2142 path.dentry = dentry;
2143 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2146 return err? nfserrno(err) : 0;
2150 nfsd_racache_shutdown(void)
2152 struct raparms *raparm, *last_raparm;
2155 dprintk("nfsd: freeing readahead buffers.\n");
2157 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2158 raparm = raparm_hash[i].pb_head;
2160 last_raparm = raparm;
2161 raparm = raparm->p_next;
2164 raparm_hash[i].pb_head = NULL;
2168 * Initialize readahead param cache
2171 nfsd_racache_init(int cache_size)
2176 struct raparms **raparm = NULL;
2179 if (raparm_hash[0].pb_head)
2181 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2184 cache_size = nperbucket * RAPARM_HASH_SIZE;
2186 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2188 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2189 spin_lock_init(&raparm_hash[i].pb_lock);
2191 raparm = &raparm_hash[i].pb_head;
2192 for (j = 0; j < nperbucket; j++) {
2193 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2196 raparm = &(*raparm)->p_next;
2201 nfsdstats.ra_size = cache_size;
2205 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2206 nfsd_racache_shutdown();
2210 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2212 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2214 struct inode *inode = fhp->fh_dentry->d_inode;
2218 struct posix_acl *acl;
2220 if (!IS_POSIXACL(inode))
2221 return ERR_PTR(-EOPNOTSUPP);
2224 case ACL_TYPE_ACCESS:
2225 name = POSIX_ACL_XATTR_ACCESS;
2227 case ACL_TYPE_DEFAULT:
2228 name = POSIX_ACL_XATTR_DEFAULT;
2231 return ERR_PTR(-EOPNOTSUPP);
2234 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2236 return ERR_PTR(size);
2238 acl = posix_acl_from_xattr(value, size);
2244 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2246 struct inode *inode = fhp->fh_dentry->d_inode;
2252 if (!IS_POSIXACL(inode) ||
2253 !inode->i_op->setxattr || !inode->i_op->removexattr)
2256 case ACL_TYPE_ACCESS:
2257 name = POSIX_ACL_XATTR_ACCESS;
2259 case ACL_TYPE_DEFAULT:
2260 name = POSIX_ACL_XATTR_DEFAULT;
2266 if (acl && acl->a_count) {
2267 size = posix_acl_xattr_size(acl->a_count);
2268 value = kmalloc(size, GFP_KERNEL);
2271 error = posix_acl_to_xattr(acl, value, size);
2278 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2282 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2284 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2287 error = vfs_removexattr(fhp->fh_dentry, name);
2288 if (error == -ENODATA)
2292 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2298 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */