1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.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 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
87 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
91 while (d_mountpoint(path.dentry) && 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 (!(exp->ex_flags & NFSEXP_V4ROOT))
173 return dentry->d_inode != NULL;
177 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
178 const char *name, unsigned int len,
179 struct svc_export **exp_ret, struct dentry **dentry_ret)
181 struct svc_export *exp;
182 struct dentry *dparent;
183 struct dentry *dentry;
187 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
189 /* Obtain dentry and export. */
190 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
194 dparent = fhp->fh_dentry;
195 exp = fhp->fh_export;
198 /* Lookup the name, but don't follow links */
199 if (isdotent(name, len)) {
201 dentry = dget(dparent);
202 else if (dparent != exp->ex_path.dentry)
203 dentry = dget_parent(dparent);
204 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
205 dentry = dget(dparent); /* .. == . just like at / */
207 /* checking mountpoint crossing is very different when stepping up */
208 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
214 dentry = lookup_one_len(name, dparent, len);
215 host_err = PTR_ERR(dentry);
219 * check if we have crossed a mount point ...
221 if (nfsd_mountpoint(dentry, exp)) {
222 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
228 *dentry_ret = dentry;
234 return nfserrno(host_err);
238 * Look up one component of a pathname.
239 * N.B. After this call _both_ fhp and resfh need an fh_put
241 * If the lookup would cross a mountpoint, and the mounted filesystem
242 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
243 * accepted as it stands and the mounted directory is
244 * returned. Otherwise the covered directory is returned.
245 * NOTE: this mountpoint crossing is not supported properly by all
246 * clients and is explicitly disallowed for NFSv3
247 * NeilBrown <neilb@cse.unsw.edu.au>
250 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
251 unsigned int len, struct svc_fh *resfh)
253 struct svc_export *exp;
254 struct dentry *dentry;
257 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
260 err = check_nfsd_access(exp, rqstp);
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err = fh_compose(resfh, exp, dentry, fhp);
268 if (!err && !dentry->d_inode)
277 * Commit metadata changes to stable storage.
280 commit_metadata(struct svc_fh *fhp)
282 struct inode *inode = fhp->fh_dentry->d_inode;
283 const struct export_operations *export_ops = inode->i_sb->s_export_op;
286 if (!EX_ISSYNC(fhp->fh_export))
289 if (export_ops->commit_metadata) {
290 error = export_ops->commit_metadata(inode);
292 struct writeback_control wbc = {
293 .sync_mode = WB_SYNC_ALL,
294 .nr_to_write = 0, /* metadata only */
297 error = sync_inode(inode, &wbc);
304 * Set various file attributes.
305 * N.B. After this call fhp needs an fh_put
308 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
309 int check_guard, time_t guardtime)
311 struct dentry *dentry;
313 int accmode = NFSD_MAY_SATTR;
319 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
320 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
321 if (iap->ia_valid & ATTR_SIZE)
325 err = fh_verify(rqstp, fhp, ftype, accmode);
329 dentry = fhp->fh_dentry;
330 inode = dentry->d_inode;
332 /* Ignore any mode updates on symlinks */
333 if (S_ISLNK(inode->i_mode))
334 iap->ia_valid &= ~ATTR_MODE;
340 * NFSv2 does not differentiate between "set-[ac]time-to-now"
341 * which only requires access, and "set-[ac]time-to-X" which
342 * requires ownership.
343 * So if it looks like it might be "set both to the same time which
344 * is close to now", and if inode_change_ok fails, then we
345 * convert to "set to now" instead of "set to explicit time"
347 * We only call inode_change_ok as the last test as technically
348 * it is not an interface that we should be using. It is only
349 * valid if the filesystem does not define it's own i_op->setattr.
351 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
352 #define MAX_TOUCH_TIME_ERROR (30*60)
353 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
354 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
358 * Now just make sure time is in the right ballpark.
359 * Solaris, at least, doesn't seem to care what the time
360 * request is. We require it be within 30 minutes of now.
362 time_t delta = iap->ia_atime.tv_sec - get_seconds();
365 if (delta < MAX_TOUCH_TIME_ERROR &&
366 inode_change_ok(inode, iap) != 0) {
368 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
369 * This will cause notify_change to set these times
372 iap->ia_valid &= ~BOTH_TIME_SET;
377 * The size case is special.
378 * It changes the file as well as the attributes.
380 if (iap->ia_valid & ATTR_SIZE) {
381 if (iap->ia_size < inode->i_size) {
382 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
383 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
389 * If we are changing the size of the file, then
390 * we need to break all leases.
392 host_err = break_lease(inode, O_WRONLY | O_NONBLOCK);
393 if (host_err == -EWOULDBLOCK)
394 host_err = -ETIMEDOUT;
395 if (host_err) /* ENOMEM or EWOULDBLOCK */
398 host_err = get_write_access(inode);
403 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
405 put_write_access(inode);
410 /* sanitize the mode change */
411 if (iap->ia_valid & ATTR_MODE) {
412 iap->ia_mode &= S_IALLUGO;
413 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
416 /* Revoke setuid/setgid on chown */
417 if (!S_ISDIR(inode->i_mode) &&
418 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
419 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
420 iap->ia_valid |= ATTR_KILL_PRIV;
421 if (iap->ia_valid & ATTR_MODE) {
422 /* we're setting mode too, just clear the s*id bits */
423 iap->ia_mode &= ~S_ISUID;
424 if (iap->ia_mode & S_IXGRP)
425 iap->ia_mode &= ~S_ISGID;
427 /* set ATTR_KILL_* bits and let VFS handle it */
428 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
432 /* Change the attributes. */
434 iap->ia_valid |= ATTR_CTIME;
436 err = nfserr_notsync;
437 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
439 host_err = notify_change(dentry, iap);
440 err = nfserrno(host_err);
444 put_write_access(inode);
446 if (EX_ISSYNC(fhp->fh_export))
447 write_inode_now(inode, 1);
452 err = nfserrno(host_err);
456 #if defined(CONFIG_NFSD_V2_ACL) || \
457 defined(CONFIG_NFSD_V3_ACL) || \
458 defined(CONFIG_NFSD_V4)
459 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
464 buflen = vfs_getxattr(dentry, key, NULL, 0);
468 *buf = kmalloc(buflen, GFP_KERNEL);
472 ret = vfs_getxattr(dentry, key, *buf, buflen);
479 #if defined(CONFIG_NFSD_V4)
481 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
488 buflen = posix_acl_xattr_size(pacl->a_count);
489 buf = kmalloc(buflen, GFP_KERNEL);
494 len = posix_acl_to_xattr(pacl, buf, buflen);
500 error = vfs_setxattr(dentry, key, buf, len, 0);
507 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
508 struct nfs4_acl *acl)
512 struct dentry *dentry;
514 struct posix_acl *pacl = NULL, *dpacl = NULL;
515 unsigned int flags = 0;
518 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
522 dentry = fhp->fh_dentry;
523 inode = dentry->d_inode;
524 if (S_ISDIR(inode->i_mode))
525 flags = NFS4_ACL_DIR;
527 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
528 if (host_error == -EINVAL) {
529 return nfserr_attrnotsupp;
530 } else if (host_error < 0)
533 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
537 if (S_ISDIR(inode->i_mode))
538 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
541 posix_acl_release(pacl);
542 posix_acl_release(dpacl);
544 if (host_error == -EOPNOTSUPP)
545 return nfserr_attrnotsupp;
547 return nfserrno(host_error);
550 static struct posix_acl *
551 _get_posix_acl(struct dentry *dentry, char *key)
554 struct posix_acl *pacl = NULL;
557 buflen = nfsd_getxattr(dentry, key, &buf);
561 return ERR_PTR(buflen);
563 pacl = posix_acl_from_xattr(buf, buflen);
569 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
571 struct inode *inode = dentry->d_inode;
573 struct posix_acl *pacl = NULL, *dpacl = NULL;
574 unsigned int flags = 0;
576 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
577 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
578 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
580 error = PTR_ERR(pacl);
585 if (S_ISDIR(inode->i_mode)) {
586 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
587 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
589 else if (IS_ERR(dpacl)) {
590 error = PTR_ERR(dpacl);
594 flags = NFS4_ACL_DIR;
597 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
599 error = PTR_ERR(*acl);
603 posix_acl_release(pacl);
604 posix_acl_release(dpacl);
608 #endif /* defined(CONFIG_NFS_V4) */
610 #ifdef CONFIG_NFSD_V3
612 * Check server access rights to a file system object
618 static struct accessmap nfs3_regaccess[] = {
619 { NFS3_ACCESS_READ, NFSD_MAY_READ },
620 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
621 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
622 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
627 static struct accessmap nfs3_diraccess[] = {
628 { NFS3_ACCESS_READ, NFSD_MAY_READ },
629 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
630 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
631 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
632 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
637 static struct accessmap nfs3_anyaccess[] = {
638 /* Some clients - Solaris 2.6 at least, make an access call
639 * to the server to check for access for things like /dev/null
640 * (which really, the server doesn't care about). So
641 * We provide simple access checking for them, looking
642 * mainly at mode bits, and we make sure to ignore read-only
645 { NFS3_ACCESS_READ, NFSD_MAY_READ },
646 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
647 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
648 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
654 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
656 struct accessmap *map;
657 struct svc_export *export;
658 struct dentry *dentry;
659 u32 query, result = 0, sresult = 0;
662 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
666 export = fhp->fh_export;
667 dentry = fhp->fh_dentry;
669 if (S_ISREG(dentry->d_inode->i_mode))
670 map = nfs3_regaccess;
671 else if (S_ISDIR(dentry->d_inode->i_mode))
672 map = nfs3_diraccess;
674 map = nfs3_anyaccess;
678 for (; map->access; map++) {
679 if (map->access & query) {
682 sresult |= map->access;
684 err2 = nfsd_permission(rqstp, export, dentry, map->how);
687 result |= map->access;
690 /* the following error codes just mean the access was not allowed,
691 * rather than an error occurred */
695 /* simply don't "or" in the access bit. */
705 *supported = sresult;
710 #endif /* CONFIG_NFSD_V3 */
715 * Open an existing file or directory.
716 * The access argument indicates the type of open (read/write/lock)
717 * N.B. After this call fhp needs an fh_put
720 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
721 int access, struct file **filp)
723 struct dentry *dentry;
725 int flags = O_RDONLY|O_LARGEFILE;
729 validate_process_creds();
732 * If we get here, then the client has already done an "open",
733 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
734 * in case a chmod has now revoked permission.
736 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
740 dentry = fhp->fh_dentry;
741 inode = dentry->d_inode;
743 /* Disallow write access to files with the append-only bit set
744 * or any access when mandatory locking enabled
747 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
750 * We must ignore files (but only files) which might have mandatory
751 * locks on them because there is no way to know if the accesser has
754 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
761 * Check to see if there are any leases on this file.
762 * This may block while leases are broken.
764 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
765 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
766 if (host_err == -EWOULDBLOCK)
767 host_err = -ETIMEDOUT;
768 if (host_err) /* NOMEM or WOULDBLOCK */
771 if (access & NFSD_MAY_WRITE) {
772 if (access & NFSD_MAY_READ)
773 flags = O_RDWR|O_LARGEFILE;
775 flags = O_WRONLY|O_LARGEFILE;
777 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
778 flags, current_cred());
780 host_err = PTR_ERR(*filp);
782 host_err = ima_file_check(*filp, access);
784 err = nfserrno(host_err);
786 validate_process_creds();
794 nfsd_close(struct file *filp)
800 * Obtain the readahead parameters for the file
801 * specified by (dev, ino).
804 static inline struct raparms *
805 nfsd_get_raparms(dev_t dev, ino_t ino)
807 struct raparms *ra, **rap, **frap = NULL;
810 struct raparm_hbucket *rab;
812 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
813 rab = &raparm_hash[hash];
815 spin_lock(&rab->pb_lock);
816 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
817 if (ra->p_ino == ino && ra->p_dev == dev)
820 if (ra->p_count == 0)
823 depth = nfsdstats.ra_size*11/10;
825 spin_unlock(&rab->pb_lock);
835 if (rap != &rab->pb_head) {
837 ra->p_next = rab->pb_head;
841 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
842 spin_unlock(&rab->pb_lock);
847 * Grab and keep cached pages associated with a file in the svc_rqst
848 * so that they can be passed to the network sendmsg/sendpage routines
849 * directly. They will be released after the sending has completed.
852 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
853 struct splice_desc *sd)
855 struct svc_rqst *rqstp = sd->u.data;
856 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
857 struct page *page = buf->page;
861 ret = buf->ops->confirm(pipe, buf);
867 if (rqstp->rq_res.page_len == 0) {
872 rqstp->rq_res.page_base = buf->offset;
873 rqstp->rq_res.page_len = size;
874 } else if (page != pp[-1]) {
880 rqstp->rq_res.page_len += size;
882 rqstp->rq_res.page_len += size;
887 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
888 struct splice_desc *sd)
890 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
893 static inline int svc_msnfs(struct svc_fh *ffhp)
896 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
903 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
904 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
913 inode = file->f_path.dentry->d_inode;
915 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
918 /* Get readahead parameters */
919 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
922 file->f_ra = ra->p_ra;
924 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
925 struct splice_desc sd = {
932 rqstp->rq_resused = 1;
933 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
937 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
941 /* Write back readahead params */
943 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
944 spin_lock(&rab->pb_lock);
945 ra->p_ra = file->f_ra;
948 spin_unlock(&rab->pb_lock);
952 nfsdstats.io_read += host_err;
955 fsnotify_access(file->f_path.dentry);
957 err = nfserrno(host_err);
962 static void kill_suid(struct dentry *dentry)
965 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
967 mutex_lock(&dentry->d_inode->i_mutex);
968 notify_change(dentry, &ia);
969 mutex_unlock(&dentry->d_inode->i_mutex);
973 * Gathered writes: If another process is currently writing to the file,
974 * there's a high chance this is another nfsd (triggered by a bulk write
975 * from a client's biod). Rather than syncing the file with each write
976 * request, we sleep for 10 msec.
978 * I don't know if this roughly approximates C. Juszak's idea of
979 * gathered writes, but it's a nice and simple solution (IMHO), and it
982 * Note: we do this only in the NFSv2 case, since v3 and higher have a
983 * better tool (separate unstable writes and commits) for solving this
986 static int wait_for_concurrent_writes(struct file *file)
988 struct inode *inode = file->f_path.dentry->d_inode;
989 static ino_t last_ino;
990 static dev_t last_dev;
993 if (atomic_read(&inode->i_writecount) > 1
994 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
995 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
997 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1000 if (inode->i_state & I_DIRTY) {
1001 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1002 err = vfs_fsync(file, file->f_path.dentry, 0);
1004 last_ino = inode->i_ino;
1005 last_dev = inode->i_sb->s_dev;
1010 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1011 loff_t offset, struct kvec *vec, int vlen,
1012 unsigned long *cnt, int *stablep)
1014 struct svc_export *exp;
1015 struct dentry *dentry;
1016 struct inode *inode;
1020 int stable = *stablep;
1026 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1027 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1031 dentry = file->f_path.dentry;
1032 inode = dentry->d_inode;
1033 exp = fhp->fh_export;
1036 * Request sync writes if
1037 * - the sync export option has been set, or
1038 * - the client requested O_SYNC behavior (NFSv3 feature).
1039 * - The file system doesn't support fsync().
1040 * When NFSv2 gathered writes have been configured for this volume,
1041 * flushing the data to disk is handled separately below.
1043 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1045 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1047 *stablep = 2; /* FILE_SYNC */
1050 if (!EX_ISSYNC(exp))
1052 if (stable && !use_wgather) {
1053 spin_lock(&file->f_lock);
1054 file->f_flags |= O_SYNC;
1055 spin_unlock(&file->f_lock);
1058 /* Write the data. */
1059 oldfs = get_fs(); set_fs(KERNEL_DS);
1060 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1065 nfsdstats.io_write += host_err;
1066 fsnotify_modify(file->f_path.dentry);
1068 /* clear setuid/setgid flag after write */
1069 if (inode->i_mode & (S_ISUID | S_ISGID))
1072 if (stable && use_wgather)
1073 host_err = wait_for_concurrent_writes(file);
1076 dprintk("nfsd: write complete host_err=%d\n", host_err);
1080 err = nfserrno(host_err);
1086 * Read data from a file. count must contain the requested read count
1087 * on entry. On return, *count contains the number of bytes actually read.
1088 * N.B. After this call fhp needs an fh_put
1091 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1092 loff_t offset, struct kvec *vec, int vlen,
1093 unsigned long *count)
1098 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1099 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1102 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1104 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1107 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1115 * Write data to a file.
1116 * The stable flag requests synchronous writes.
1117 * N.B. After this call fhp needs an fh_put
1120 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1121 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1127 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1128 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1131 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1134 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1139 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1147 #ifdef CONFIG_NFSD_V3
1149 * Commit all pending writes to stable storage.
1151 * Note: we only guarantee that data that lies within the range specified
1152 * by the 'offset' and 'count' parameters will be synced.
1154 * Unfortunately we cannot lock the file to make sure we return full WCC
1155 * data to the client, as locking happens lower down in the filesystem.
1158 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1159 loff_t offset, unsigned long count)
1162 loff_t end = LLONG_MAX;
1163 __be32 err = nfserr_inval;
1168 end = offset + (loff_t)count - 1;
1173 err = nfsd_open(rqstp, fhp, S_IFREG,
1174 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1177 if (EX_ISSYNC(fhp->fh_export)) {
1178 int err2 = vfs_fsync_range(file, file->f_path.dentry,
1181 if (err2 != -EINVAL)
1182 err = nfserrno(err2);
1184 err = nfserr_notsupp;
1191 #endif /* CONFIG_NFSD_V3 */
1194 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1198 * Mode has already been set earlier in create:
1200 iap->ia_valid &= ~ATTR_MODE;
1202 * Setting uid/gid works only for root. Irix appears to
1203 * send along the gid on create when it tries to implement
1204 * setgid directories via NFS:
1206 if (current_fsuid() != 0)
1207 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1209 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1213 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1214 * setting size to 0 may fail for some specific file systems by the permission
1215 * checking which requires WRITE permission but the mode is 000.
1216 * we ignore the resizing(to 0) on the just new created file, since the size is
1217 * 0 after file created.
1219 * call this only after vfs_create() is called.
1222 nfsd_check_ignore_resizing(struct iattr *iap)
1224 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1225 iap->ia_valid &= ~ATTR_SIZE;
1229 * Create a file (regular, directory, device, fifo); UNIX sockets
1230 * not yet implemented.
1231 * If the response fh has been verified, the parent directory should
1232 * already be locked. Note that the parent directory is left locked.
1234 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1237 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1238 char *fname, int flen, struct iattr *iap,
1239 int type, dev_t rdev, struct svc_fh *resfhp)
1241 struct dentry *dentry, *dchild = NULL;
1251 if (isdotent(fname, flen))
1254 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1258 dentry = fhp->fh_dentry;
1259 dirp = dentry->d_inode;
1261 err = nfserr_notdir;
1262 if (!dirp->i_op->lookup)
1265 * Check whether the response file handle has been verified yet.
1266 * If it has, the parent directory should already be locked.
1268 if (!resfhp->fh_dentry) {
1269 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1270 fh_lock_nested(fhp, I_MUTEX_PARENT);
1271 dchild = lookup_one_len(fname, dentry, flen);
1272 host_err = PTR_ERR(dchild);
1275 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1279 /* called from nfsd_proc_create */
1280 dchild = dget(resfhp->fh_dentry);
1281 if (!fhp->fh_locked) {
1282 /* not actually possible */
1284 "nfsd_create: parent %s/%s not locked!\n",
1285 dentry->d_parent->d_name.name,
1286 dentry->d_name.name);
1292 * Make sure the child dentry is still negative ...
1295 if (dchild->d_inode) {
1296 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1297 dentry->d_name.name, dchild->d_name.name);
1301 if (!(iap->ia_valid & ATTR_MODE))
1303 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1306 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1307 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1312 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1317 * Get the dir op function pointer.
1322 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1324 nfsd_check_ignore_resizing(iap);
1327 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1333 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1337 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1341 err = nfsd_create_setattr(rqstp, resfhp, iap);
1344 * nfsd_setattr already committed the child. Transactional filesystems
1345 * had a chance to commit changes for both parent and child
1346 * simultaneously making the following commit_metadata a noop.
1348 err2 = nfserrno(commit_metadata(fhp));
1351 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1353 * Update the file handle to get the new inode info.
1356 err = fh_update(resfhp);
1358 if (dchild && !IS_ERR(dchild))
1363 err = nfserrno(host_err);
1367 #ifdef CONFIG_NFSD_V3
1369 * NFSv3 version of nfsd_create
1372 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1373 char *fname, int flen, struct iattr *iap,
1374 struct svc_fh *resfhp, int createmode, u32 *verifier,
1375 int *truncp, int *created)
1377 struct dentry *dentry, *dchild = NULL;
1381 __u32 v_mtime=0, v_atime=0;
1387 if (isdotent(fname, flen))
1389 if (!(iap->ia_valid & ATTR_MODE))
1391 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1395 dentry = fhp->fh_dentry;
1396 dirp = dentry->d_inode;
1398 /* Get all the sanity checks out of the way before
1399 * we lock the parent. */
1400 err = nfserr_notdir;
1401 if (!dirp->i_op->lookup)
1403 fh_lock_nested(fhp, I_MUTEX_PARENT);
1406 * Compose the response file handle.
1408 dchild = lookup_one_len(fname, dentry, flen);
1409 host_err = PTR_ERR(dchild);
1413 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1417 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1418 /* solaris7 gets confused (bugid 4218508) if these have
1419 * the high bit set, so just clear the high bits. If this is
1420 * ever changed to use different attrs for storing the
1421 * verifier, then do_open_lookup() will also need to be fixed
1424 v_mtime = verifier[0]&0x7fffffff;
1425 v_atime = verifier[1]&0x7fffffff;
1428 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1431 if (dchild->d_inode) {
1434 switch (createmode) {
1435 case NFS3_CREATE_UNCHECKED:
1436 if (! S_ISREG(dchild->d_inode->i_mode))
1439 /* in nfsv4, we need to treat this case a little
1440 * differently. we don't want to truncate the
1441 * file now; this would be wrong if the OPEN
1442 * fails for some other reason. furthermore,
1443 * if the size is nonzero, we should ignore it
1444 * according to spec!
1446 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1449 iap->ia_valid &= ATTR_SIZE;
1453 case NFS3_CREATE_EXCLUSIVE:
1454 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1455 && dchild->d_inode->i_atime.tv_sec == v_atime
1456 && dchild->d_inode->i_size == 0 )
1459 case NFS3_CREATE_GUARDED:
1462 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1466 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1468 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1474 nfsd_check_ignore_resizing(iap);
1476 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1477 /* Cram the verifier into atime/mtime */
1478 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1479 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1480 /* XXX someone who knows this better please fix it for nsec */
1481 iap->ia_mtime.tv_sec = v_mtime;
1482 iap->ia_atime.tv_sec = v_atime;
1483 iap->ia_mtime.tv_nsec = 0;
1484 iap->ia_atime.tv_nsec = 0;
1488 err = nfsd_create_setattr(rqstp, resfhp, iap);
1491 * nfsd_setattr already committed the child (and possibly also the parent).
1494 err = nfserrno(commit_metadata(fhp));
1496 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1498 * Update the filehandle to get the new inode info.
1501 err = fh_update(resfhp);
1505 if (dchild && !IS_ERR(dchild))
1510 err = nfserrno(host_err);
1513 #endif /* CONFIG_NFSD_V3 */
1516 * Read a symlink. On entry, *lenp must contain the maximum path length that
1517 * fits into the buffer. On return, it contains the true length.
1518 * N.B. After this call fhp needs an fh_put
1521 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1523 struct dentry *dentry;
1524 struct inode *inode;
1529 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1533 dentry = fhp->fh_dentry;
1534 inode = dentry->d_inode;
1537 if (!inode->i_op->readlink)
1540 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1541 /* N.B. Why does this call need a get_fs()??
1542 * Remove the set_fs and watch the fireworks:-) --okir
1545 oldfs = get_fs(); set_fs(KERNEL_DS);
1546 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1557 err = nfserrno(host_err);
1562 * Create a symlink and look up its inode
1563 * N.B. After this call _both_ fhp and resfhp need an fh_put
1566 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1567 char *fname, int flen,
1568 char *path, int plen,
1569 struct svc_fh *resfhp,
1572 struct dentry *dentry, *dnew;
1580 if (isdotent(fname, flen))
1583 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1587 dentry = fhp->fh_dentry;
1588 dnew = lookup_one_len(fname, dentry, flen);
1589 host_err = PTR_ERR(dnew);
1593 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1597 if (unlikely(path[plen] != 0)) {
1598 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1599 if (path_alloced == NULL)
1602 strncpy(path_alloced, path, plen);
1603 path_alloced[plen] = 0;
1604 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1605 kfree(path_alloced);
1608 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1609 err = nfserrno(host_err);
1611 err = nfserrno(commit_metadata(fhp));
1614 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1616 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1618 if (err==0) err = cerr;
1623 err = nfserrno(host_err);
1629 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1632 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1633 char *name, int len, struct svc_fh *tfhp)
1635 struct dentry *ddir, *dnew, *dold;
1636 struct inode *dirp, *dest;
1640 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1643 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1651 if (isdotent(name, len))
1654 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1655 ddir = ffhp->fh_dentry;
1656 dirp = ddir->d_inode;
1658 dnew = lookup_one_len(name, ddir, len);
1659 host_err = PTR_ERR(dnew);
1663 dold = tfhp->fh_dentry;
1664 dest = dold->d_inode;
1666 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1668 err = nfserrno(host_err);
1671 host_err = vfs_link(dold, dirp, dnew);
1673 err = nfserrno(commit_metadata(ffhp));
1675 err = nfserrno(commit_metadata(tfhp));
1677 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1680 err = nfserrno(host_err);
1682 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1691 err = nfserrno(host_err);
1697 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1700 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1701 struct svc_fh *tfhp, char *tname, int tlen)
1703 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1704 struct inode *fdir, *tdir;
1708 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1711 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1715 fdentry = ffhp->fh_dentry;
1716 fdir = fdentry->d_inode;
1718 tdentry = tfhp->fh_dentry;
1719 tdir = tdentry->d_inode;
1721 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1722 if (ffhp->fh_export != tfhp->fh_export)
1726 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1729 /* cannot use fh_lock as we need deadlock protective ordering
1730 * so do it by hand */
1731 trap = lock_rename(tdentry, fdentry);
1732 ffhp->fh_locked = tfhp->fh_locked = 1;
1736 odentry = lookup_one_len(fname, fdentry, flen);
1737 host_err = PTR_ERR(odentry);
1738 if (IS_ERR(odentry))
1742 if (!odentry->d_inode)
1745 if (odentry == trap)
1748 ndentry = lookup_one_len(tname, tdentry, tlen);
1749 host_err = PTR_ERR(ndentry);
1750 if (IS_ERR(ndentry))
1752 host_err = -ENOTEMPTY;
1753 if (ndentry == trap)
1756 if (svc_msnfs(ffhp) &&
1757 ((atomic_read(&odentry->d_count) > 1)
1758 || (atomic_read(&ndentry->d_count) > 1))) {
1764 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1766 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1770 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1772 host_err = commit_metadata(tfhp);
1774 host_err = commit_metadata(ffhp);
1777 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1784 err = nfserrno(host_err);
1786 /* we cannot reply on fh_unlock on the two filehandles,
1787 * as that would do the wrong thing if the two directories
1788 * were the same, so again we do it by hand
1790 fill_post_wcc(ffhp);
1791 fill_post_wcc(tfhp);
1792 unlock_rename(tdentry, fdentry);
1793 ffhp->fh_locked = tfhp->fh_locked = 0;
1800 * Unlink a file or directory
1801 * N.B. After this call fhp needs an fh_put
1804 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1805 char *fname, int flen)
1807 struct dentry *dentry, *rdentry;
1813 if (!flen || isdotent(fname, flen))
1815 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1819 fh_lock_nested(fhp, I_MUTEX_PARENT);
1820 dentry = fhp->fh_dentry;
1821 dirp = dentry->d_inode;
1823 rdentry = lookup_one_len(fname, dentry, flen);
1824 host_err = PTR_ERR(rdentry);
1825 if (IS_ERR(rdentry))
1828 if (!rdentry->d_inode) {
1835 type = rdentry->d_inode->i_mode & S_IFMT;
1837 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1841 if (type != S_IFDIR) { /* It's UNLINK */
1843 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1844 (atomic_read(&rdentry->d_count) > 1)) {
1848 host_err = vfs_unlink(dirp, rdentry);
1849 } else { /* It's RMDIR */
1850 host_err = vfs_rmdir(dirp, rdentry);
1856 host_err = commit_metadata(fhp);
1858 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1860 err = nfserrno(host_err);
1866 * We do this buffering because we must not call back into the file
1867 * system's ->lookup() method from the filldir callback. That may well
1868 * deadlock a number of file systems.
1870 * This is based heavily on the implementation of same in XFS.
1872 struct buffered_dirent {
1876 unsigned int d_type;
1880 struct readdir_data {
1886 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1887 loff_t offset, u64 ino, unsigned int d_type)
1889 struct readdir_data *buf = __buf;
1890 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1891 unsigned int reclen;
1893 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1894 if (buf->used + reclen > PAGE_SIZE) {
1899 de->namlen = namlen;
1900 de->offset = offset;
1902 de->d_type = d_type;
1903 memcpy(de->name, name, namlen);
1904 buf->used += reclen;
1909 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1910 struct readdir_cd *cdp, loff_t *offsetp)
1912 struct readdir_data buf;
1913 struct buffered_dirent *de;
1918 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1920 return nfserrno(-ENOMEM);
1925 struct inode *dir_inode = file->f_path.dentry->d_inode;
1926 unsigned int reclen;
1928 cdp->err = nfserr_eof; /* will be cleared on successful read */
1932 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1945 * Various filldir functions may end up calling back into
1946 * lookup_one_len() and the file system's ->lookup() method.
1947 * These expect i_mutex to be held, as it would within readdir.
1949 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1953 de = (struct buffered_dirent *)buf.dirent;
1955 offset = de->offset;
1957 if (func(cdp, de->name, de->namlen, de->offset,
1958 de->ino, de->d_type))
1961 if (cdp->err != nfs_ok)
1964 reclen = ALIGN(sizeof(*de) + de->namlen,
1967 de = (struct buffered_dirent *)((char *)de + reclen);
1969 mutex_unlock(&dir_inode->i_mutex);
1970 if (size > 0) /* We bailed out early */
1973 offset = vfs_llseek(file, 0, SEEK_CUR);
1976 free_page((unsigned long)(buf.dirent));
1979 return nfserrno(host_err);
1986 * Read entries from a directory.
1987 * The NFSv3/4 verifier we ignore for now.
1990 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1991 struct readdir_cd *cdp, filldir_t func)
1995 loff_t offset = *offsetp;
1997 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2001 offset = vfs_llseek(file, offset, 0);
2003 err = nfserrno((int)offset);
2007 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2009 if (err == nfserr_eof || err == nfserr_toosmall)
2010 err = nfs_ok; /* can still be found in ->err */
2018 * Get file system stats
2019 * N.B. After this call fhp needs an fh_put
2022 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2024 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2025 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2030 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2032 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2036 * Check for a user's access permissions to this inode.
2039 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2040 struct dentry *dentry, int acc)
2042 struct inode *inode = dentry->d_inode;
2046 if (acc == NFSD_MAY_NOP)
2049 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2051 (acc & NFSD_MAY_READ)? " read" : "",
2052 (acc & NFSD_MAY_WRITE)? " write" : "",
2053 (acc & NFSD_MAY_EXEC)? " exec" : "",
2054 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2055 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2056 (acc & NFSD_MAY_LOCK)? " lock" : "",
2057 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2059 IS_IMMUTABLE(inode)? " immut" : "",
2060 IS_APPEND(inode)? " append" : "",
2061 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2062 dprintk(" owner %d/%d user %d/%d\n",
2063 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2066 /* Normally we reject any write/sattr etc access on a read-only file
2067 * system. But if it is IRIX doing check on write-access for a
2068 * device special file, we ignore rofs.
2070 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2071 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2072 if (exp_rdonly(rqstp, exp) ||
2073 __mnt_is_readonly(exp->ex_path.mnt))
2075 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2078 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2081 if (acc & NFSD_MAY_LOCK) {
2082 /* If we cannot rely on authentication in NLM requests,
2083 * just allow locks, otherwise require read permission, or
2086 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2089 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2092 * The file owner always gets access permission for accesses that
2093 * would normally be checked at open time. This is to make
2094 * file access work even when the client has done a fchmod(fd, 0).
2096 * However, `cp foo bar' should fail nevertheless when bar is
2097 * readonly. A sensible way to do this might be to reject all
2098 * attempts to truncate a read-only file, because a creat() call
2099 * always implies file truncation.
2100 * ... but this isn't really fair. A process may reasonably call
2101 * ftruncate on an open file descriptor on a file with perm 000.
2102 * We must trust the client to do permission checking - using "ACCESS"
2105 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2106 inode->i_uid == current_fsuid())
2109 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2110 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2112 /* Allow read access to binaries even when mode 111 */
2113 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2114 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2115 err = inode_permission(inode, MAY_EXEC);
2119 /* Do integrity (permission) checking now, but defer incrementing
2120 * IMA counts to the actual file open.
2122 path.mnt = exp->ex_path.mnt;
2123 path.dentry = dentry;
2125 return err? nfserrno(err) : 0;
2129 nfsd_racache_shutdown(void)
2131 struct raparms *raparm, *last_raparm;
2134 dprintk("nfsd: freeing readahead buffers.\n");
2136 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2137 raparm = raparm_hash[i].pb_head;
2139 last_raparm = raparm;
2140 raparm = raparm->p_next;
2143 raparm_hash[i].pb_head = NULL;
2147 * Initialize readahead param cache
2150 nfsd_racache_init(int cache_size)
2155 struct raparms **raparm = NULL;
2158 if (raparm_hash[0].pb_head)
2160 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2163 cache_size = nperbucket * RAPARM_HASH_SIZE;
2165 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2167 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2168 spin_lock_init(&raparm_hash[i].pb_lock);
2170 raparm = &raparm_hash[i].pb_head;
2171 for (j = 0; j < nperbucket; j++) {
2172 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2175 raparm = &(*raparm)->p_next;
2180 nfsdstats.ra_size = cache_size;
2184 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2185 nfsd_racache_shutdown();
2189 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2191 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2193 struct inode *inode = fhp->fh_dentry->d_inode;
2197 struct posix_acl *acl;
2199 if (!IS_POSIXACL(inode))
2200 return ERR_PTR(-EOPNOTSUPP);
2203 case ACL_TYPE_ACCESS:
2204 name = POSIX_ACL_XATTR_ACCESS;
2206 case ACL_TYPE_DEFAULT:
2207 name = POSIX_ACL_XATTR_DEFAULT;
2210 return ERR_PTR(-EOPNOTSUPP);
2213 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2215 return ERR_PTR(size);
2217 acl = posix_acl_from_xattr(value, size);
2223 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2225 struct inode *inode = fhp->fh_dentry->d_inode;
2231 if (!IS_POSIXACL(inode) ||
2232 !inode->i_op->setxattr || !inode->i_op->removexattr)
2235 case ACL_TYPE_ACCESS:
2236 name = POSIX_ACL_XATTR_ACCESS;
2238 case ACL_TYPE_DEFAULT:
2239 name = POSIX_ACL_XATTR_DEFAULT;
2245 if (acl && acl->a_count) {
2246 size = posix_acl_xattr_size(acl->a_count);
2247 value = kmalloc(size, GFP_KERNEL);
2250 error = posix_acl_to_xattr(acl, value, size);
2257 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2261 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2263 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2266 error = vfs_removexattr(fhp->fh_dentry, name);
2267 if (error == -ENODATA)
2271 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2277 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */