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>
19 #include <linux/config.h>
20 #include <linux/string.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
24 #include <linux/file.h>
25 #include <linux/mount.h>
26 #include <linux/major.h>
27 #include <linux/ext2_fs.h>
28 #include <linux/proc_fs.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/net.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/pagemap.h>
36 #include <linux/module.h>
37 #include <linux/namei.h>
38 #include <linux/vfs.h>
39 #include <linux/delay.h>
40 #include <linux/sunrpc/svc.h>
41 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfs3.h>
44 #include <linux/nfsd/xdr3.h>
45 #endif /* CONFIG_NFSD_V3 */
46 #include <linux/nfsd/nfsfh.h>
47 #include <linux/quotaops.h>
48 #include <linux/fsnotify.h>
49 #include <linux/posix_acl.h>
50 #include <linux/posix_acl_xattr.h>
51 #include <linux/xattr.h>
53 #include <linux/nfs4.h>
54 #include <linux/nfs4_acl.h>
55 #include <linux/nfsd_idmap.h>
56 #include <linux/security.h>
57 #endif /* CONFIG_NFSD_V4 */
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
65 /* We must ignore files (but only files) which might have mandatory
66 * locks on them because there is no way to know if the accesser has
69 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
72 * This is a cache of readahead params that help us choose the proper
73 * readahead strategy. Initially, we set all readahead parameters to 0
74 * and let the VFS handle things.
75 * If you increase the number of cached files very much, you'll need to
76 * add a hash table here.
79 struct raparms *p_next;
84 struct file_ra_state p_ra;
87 static struct raparms * raparml;
88 static struct raparms * raparm_cache;
91 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
93 * Returns -EAGAIN leaving *dpp and *expp unchanged,
94 * or nfs_ok having possibly changed *dpp and *expp
97 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
98 struct svc_export **expp)
100 struct svc_export *exp = *expp, *exp2 = NULL;
101 struct dentry *dentry = *dpp;
102 struct vfsmount *mnt = mntget(exp->ex_mnt);
103 struct dentry *mounts = dget(dentry);
106 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
108 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
115 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
116 /* successfully crossed mount point */
122 if (exp2) exp_put(exp2);
131 * Look up one component of a pathname.
132 * N.B. After this call _both_ fhp and resfh need an fh_put
134 * If the lookup would cross a mountpoint, and the mounted filesystem
135 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
136 * accepted as it stands and the mounted directory is
137 * returned. Otherwise the covered directory is returned.
138 * NOTE: this mountpoint crossing is not supported properly by all
139 * clients and is explicitly disallowed for NFSv3
140 * NeilBrown <neilb@cse.unsw.edu.au>
143 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
144 int len, struct svc_fh *resfh)
146 struct svc_export *exp;
147 struct dentry *dparent;
148 struct dentry *dentry;
151 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
153 /* Obtain dentry and export. */
154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
158 dparent = fhp->fh_dentry;
159 exp = fhp->fh_export;
164 /* Lookup the name, but don't follow links */
165 if (isdotent(name, len)) {
167 dentry = dget(dparent);
168 else if (dparent != exp->ex_dentry) {
169 dentry = dget_parent(dparent);
170 } else if (!EX_NOHIDE(exp))
171 dentry = dget(dparent); /* .. == . just like at / */
173 /* checking mountpoint crossing is very different when stepping up */
174 struct svc_export *exp2 = NULL;
176 struct vfsmount *mnt = mntget(exp->ex_mnt);
177 dentry = dget(dparent);
178 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
180 dp = dget_parent(dentry);
184 exp2 = exp_parent(exp->ex_client, mnt, dentry,
194 dentry = dget(dparent);
203 dentry = lookup_one_len(name, dparent, len);
204 err = PTR_ERR(dentry);
208 * check if we have crossed a mount point ...
210 if (d_mountpoint(dentry)) {
211 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
218 * Note: we compose the file handle now, but as the
219 * dentry may be negative, it may need to be updated.
221 err = fh_compose(resfh, exp, dentry, fhp);
222 if (!err && !dentry->d_inode)
235 * Set various file attributes.
236 * N.B. After this call fhp needs an fh_put
239 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
240 int check_guard, time_t guardtime)
242 struct dentry *dentry;
244 int accmode = MAY_SATTR;
250 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
251 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
252 if (iap->ia_valid & ATTR_SIZE)
256 err = fh_verify(rqstp, fhp, ftype, accmode);
260 dentry = fhp->fh_dentry;
261 inode = dentry->d_inode;
263 /* Ignore any mode updates on symlinks */
264 if (S_ISLNK(inode->i_mode))
265 iap->ia_valid &= ~ATTR_MODE;
270 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
271 * which only requires access, and "set-[ac]time-to-X" which
272 * requires ownership.
273 * So if it looks like it might be "set both to the same time which
274 * is close to now", and if inode_change_ok fails, then we
275 * convert to "set to now" instead of "set to explicit time"
277 * We only call inode_change_ok as the last test as technically
278 * it is not an interface that we should be using. It is only
279 * valid if the filesystem does not define it's own i_op->setattr.
281 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
282 #define MAX_TOUCH_TIME_ERROR (30*60)
283 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
284 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
286 /* Looks probable. Now just make sure time is in the right ballpark.
287 * Solaris, at least, doesn't seem to care what the time request is.
288 * We require it be within 30 minutes of now.
290 time_t delta = iap->ia_atime.tv_sec - get_seconds();
291 if (delta<0) delta = -delta;
292 if (delta < MAX_TOUCH_TIME_ERROR &&
293 inode_change_ok(inode, iap) != 0) {
294 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
295 * this will cause notify_change to set these times to "now"
297 iap->ia_valid &= ~BOTH_TIME_SET;
301 /* The size case is special. It changes the file as well as the attributes. */
302 if (iap->ia_valid & ATTR_SIZE) {
303 if (iap->ia_size < inode->i_size) {
304 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
310 * If we are changing the size of the file, then
311 * we need to break all leases.
313 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
314 if (err == -EWOULDBLOCK)
316 if (err) /* ENOMEM or EWOULDBLOCK */
319 err = get_write_access(inode);
324 err = locks_verify_truncate(inode, NULL, iap->ia_size);
326 put_write_access(inode);
332 imode = inode->i_mode;
333 if (iap->ia_valid & ATTR_MODE) {
334 iap->ia_mode &= S_IALLUGO;
335 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
338 /* Revoke setuid/setgid bit on chown/chgrp */
339 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
340 iap->ia_valid |= ATTR_KILL_SUID;
341 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
342 iap->ia_valid |= ATTR_KILL_SGID;
344 /* Change the attributes. */
346 iap->ia_valid |= ATTR_CTIME;
348 err = nfserr_notsync;
349 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
351 err = notify_change(dentry, iap);
356 put_write_access(inode);
358 if (EX_ISSYNC(fhp->fh_export))
359 write_inode_now(inode, 1);
368 #if defined(CONFIG_NFSD_V2_ACL) || \
369 defined(CONFIG_NFSD_V3_ACL) || \
370 defined(CONFIG_NFSD_V4)
371 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
375 buflen = vfs_getxattr(dentry, key, NULL, 0);
379 *buf = kmalloc(buflen, GFP_KERNEL);
383 return vfs_getxattr(dentry, key, *buf, buflen);
387 #if defined(CONFIG_NFSD_V4)
389 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
396 buflen = posix_acl_xattr_size(pacl->a_count);
397 buf = kmalloc(buflen, GFP_KERNEL);
402 len = posix_acl_to_xattr(pacl, buf, buflen);
408 error = vfs_setxattr(dentry, key, buf, len, 0);
415 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
416 struct nfs4_acl *acl)
419 struct dentry *dentry;
421 struct posix_acl *pacl = NULL, *dpacl = NULL;
422 unsigned int flags = 0;
425 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
429 dentry = fhp->fh_dentry;
430 inode = dentry->d_inode;
431 if (S_ISDIR(inode->i_mode))
432 flags = NFS4_ACL_DIR;
434 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
435 if (error == -EINVAL) {
436 error = nfserr_attrnotsupp;
438 } else if (error < 0)
442 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
448 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
456 posix_acl_release(pacl);
457 posix_acl_release(dpacl);
460 error = nfserrno(error);
464 static struct posix_acl *
465 _get_posix_acl(struct dentry *dentry, char *key)
468 struct posix_acl *pacl = NULL;
471 buflen = nfsd_getxattr(dentry, key, &buf);
475 return ERR_PTR(buflen);
477 pacl = posix_acl_from_xattr(buf, buflen);
483 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
485 struct inode *inode = dentry->d_inode;
487 struct posix_acl *pacl = NULL, *dpacl = NULL;
488 unsigned int flags = 0;
490 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
491 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
492 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
494 error = PTR_ERR(pacl);
499 if (S_ISDIR(inode->i_mode)) {
500 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
501 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
503 else if (IS_ERR(dpacl)) {
504 error = PTR_ERR(dpacl);
508 flags = NFS4_ACL_DIR;
511 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
513 error = PTR_ERR(*acl);
517 posix_acl_release(pacl);
518 posix_acl_release(dpacl);
522 #endif /* defined(CONFIG_NFS_V4) */
524 #ifdef CONFIG_NFSD_V3
526 * Check server access rights to a file system object
532 static struct accessmap nfs3_regaccess[] = {
533 { NFS3_ACCESS_READ, MAY_READ },
534 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
535 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
536 { NFS3_ACCESS_EXTEND, MAY_WRITE },
541 static struct accessmap nfs3_diraccess[] = {
542 { NFS3_ACCESS_READ, MAY_READ },
543 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
544 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
545 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
546 { NFS3_ACCESS_DELETE, MAY_REMOVE },
551 static struct accessmap nfs3_anyaccess[] = {
552 /* Some clients - Solaris 2.6 at least, make an access call
553 * to the server to check for access for things like /dev/null
554 * (which really, the server doesn't care about). So
555 * We provide simple access checking for them, looking
556 * mainly at mode bits, and we make sure to ignore read-only
559 { NFS3_ACCESS_READ, MAY_READ },
560 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
561 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
562 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
568 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
570 struct accessmap *map;
571 struct svc_export *export;
572 struct dentry *dentry;
573 u32 query, result = 0, sresult = 0;
576 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
580 export = fhp->fh_export;
581 dentry = fhp->fh_dentry;
583 if (S_ISREG(dentry->d_inode->i_mode))
584 map = nfs3_regaccess;
585 else if (S_ISDIR(dentry->d_inode->i_mode))
586 map = nfs3_diraccess;
588 map = nfs3_anyaccess;
592 for (; map->access; map++) {
593 if (map->access & query) {
596 sresult |= map->access;
598 err2 = nfsd_permission(export, dentry, map->how);
601 result |= map->access;
604 /* the following error codes just mean the access was not allowed,
605 * rather than an error occurred */
609 /* simply don't "or" in the access bit. */
619 *supported = sresult;
624 #endif /* CONFIG_NFSD_V3 */
629 * Open an existing file or directory.
630 * The access argument indicates the type of open (read/write/lock)
631 * N.B. After this call fhp needs an fh_put
634 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
635 int access, struct file **filp)
637 struct dentry *dentry;
639 int flags = O_RDONLY|O_LARGEFILE, err;
642 * If we get here, then the client has already done an "open",
643 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
644 * in case a chmod has now revoked permission.
646 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
650 dentry = fhp->fh_dentry;
651 inode = dentry->d_inode;
653 /* Disallow write access to files with the append-only bit set
654 * or any access when mandatory locking enabled
657 if (IS_APPEND(inode) && (access & MAY_WRITE))
659 if (IS_ISMNDLK(inode))
666 * Check to see if there are any leases on this file.
667 * This may block while leases are broken.
669 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
670 if (err == -EWOULDBLOCK)
672 if (err) /* NOMEM or WOULDBLOCK */
675 if (access & MAY_WRITE) {
676 flags = O_WRONLY|O_LARGEFILE;
680 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
682 err = PTR_ERR(*filp);
694 nfsd_close(struct file *filp)
701 * As this calls fsync (not fdatasync) there is no need for a write_inode
704 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
705 const struct file_operations *fop)
707 struct inode *inode = dp->d_inode;
708 int (*fsync) (struct file *, struct dentry *, int);
711 err = filemap_fdatawrite(inode->i_mapping);
712 if (err == 0 && fop && (fsync = fop->fsync))
713 err = fsync(filp, dp, 0);
715 err = filemap_fdatawait(inode->i_mapping);
722 nfsd_sync(struct file *filp)
725 struct inode *inode = filp->f_dentry->d_inode;
726 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
727 mutex_lock(&inode->i_mutex);
728 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
729 mutex_unlock(&inode->i_mutex);
735 nfsd_sync_dir(struct dentry *dp)
737 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
741 * Obtain the readahead parameters for the file
742 * specified by (dev, ino).
744 static DEFINE_SPINLOCK(ra_lock);
746 static inline struct raparms *
747 nfsd_get_raparms(dev_t dev, ino_t ino)
749 struct raparms *ra, **rap, **frap = NULL;
753 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
754 if (ra->p_ino == ino && ra->p_dev == dev)
757 if (ra->p_count == 0)
760 depth = nfsdstats.ra_size*11/10;
762 spin_unlock(&ra_lock);
771 if (rap != &raparm_cache) {
773 ra->p_next = raparm_cache;
777 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
778 spin_unlock(&ra_lock);
783 * Grab and keep cached pages assosiated with a file in the svc_rqst
784 * so that they can be passed to the netowork sendmsg/sendpage routines
785 * directrly. They will be released after the sending has completed.
788 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
790 unsigned long count = desc->count;
791 struct svc_rqst *rqstp = desc->arg.data;
796 if (rqstp->rq_res.page_len == 0) {
798 rqstp->rq_respages[rqstp->rq_resused++] = page;
799 rqstp->rq_res.page_base = offset;
800 rqstp->rq_res.page_len = size;
801 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
803 rqstp->rq_respages[rqstp->rq_resused++] = page;
804 rqstp->rq_res.page_len += size;
806 rqstp->rq_res.page_len += size;
809 desc->count = count - size;
810 desc->written += size;
815 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
816 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
824 inode = file->f_dentry->d_inode;
826 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
827 (!lock_may_read(inode, offset, *count)))
831 /* Get readahead parameters */
832 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
835 file->f_ra = ra->p_ra;
837 if (file->f_op->sendfile) {
838 svc_pushback_unused_pages(rqstp);
839 err = file->f_op->sendfile(file, &offset, *count,
840 nfsd_read_actor, rqstp);
844 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
848 /* Write back readahead params */
851 ra->p_ra = file->f_ra;
854 spin_unlock(&ra_lock);
858 nfsdstats.io_read += err;
861 fsnotify_access(file->f_dentry);
868 static void kill_suid(struct dentry *dentry)
871 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
873 mutex_lock(&dentry->d_inode->i_mutex);
874 notify_change(dentry, &ia);
875 mutex_unlock(&dentry->d_inode->i_mutex);
879 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
880 loff_t offset, struct kvec *vec, int vlen,
881 unsigned long cnt, int *stablep)
883 struct svc_export *exp;
884 struct dentry *dentry;
888 int stable = *stablep;
893 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
894 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
898 dentry = file->f_dentry;
899 inode = dentry->d_inode;
900 exp = fhp->fh_export;
903 * Request sync writes if
904 * - the sync export option has been set, or
905 * - the client requested O_SYNC behavior (NFSv3 feature).
906 * - The file system doesn't support fsync().
907 * When gathered writes have been configured for this volume,
908 * flushing the data to disk is handled separately below.
911 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
913 *stablep = 2; /* FILE_SYNC */
918 if (stable && !EX_WGATHER(exp))
919 file->f_flags |= O_SYNC;
921 /* Write the data. */
922 oldfs = get_fs(); set_fs(KERNEL_DS);
923 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
926 nfsdstats.io_write += cnt;
927 fsnotify_modify(file->f_dentry);
930 /* clear setuid/setgid flag after write */
931 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
934 if (err >= 0 && stable) {
935 static ino_t last_ino;
936 static dev_t last_dev;
939 * Gathered writes: If another process is currently
940 * writing to the file, there's a high chance
941 * this is another nfsd (triggered by a bulk write
942 * from a client's biod). Rather than syncing the
943 * file with each write request, we sleep for 10 msec.
945 * I don't know if this roughly approximates
946 * C. Juszak's idea of gathered writes, but it's a
947 * nice and simple solution (IMHO), and it seems to
950 if (EX_WGATHER(exp)) {
951 if (atomic_read(&inode->i_writecount) > 1
952 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
953 dprintk("nfsd: write defer %d\n", current->pid);
955 dprintk("nfsd: write resume %d\n", current->pid);
958 if (inode->i_state & I_DIRTY) {
959 dprintk("nfsd: write sync %d\n", current->pid);
963 wake_up(&inode->i_wait);
966 last_ino = inode->i_ino;
967 last_dev = inode->i_sb->s_dev;
970 dprintk("nfsd: write complete err=%d\n", err);
980 * Read data from a file. count must contain the requested read count
981 * on entry. On return, *count contains the number of bytes actually read.
982 * N.B. After this call fhp needs an fh_put
985 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
986 loff_t offset, struct kvec *vec, int vlen,
987 unsigned long *count)
992 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
993 MAY_READ|MAY_OWNER_OVERRIDE);
996 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
998 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1001 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1009 * Write data to a file.
1010 * The stable flag requests synchronous writes.
1011 * N.B. After this call fhp needs an fh_put
1014 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1015 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1021 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1022 MAY_WRITE|MAY_OWNER_OVERRIDE);
1025 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1028 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1033 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1041 #ifdef CONFIG_NFSD_V3
1043 * Commit all pending writes to stable storage.
1044 * Strictly speaking, we could sync just the indicated file region here,
1045 * but there's currently no way we can ask the VFS to do so.
1047 * Unfortunately we cannot lock the file to make sure we return full WCC
1048 * data to the client, as locking happens lower down in the filesystem.
1051 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1052 loff_t offset, unsigned long count)
1057 if ((u64)count > ~(u64)offset)
1058 return nfserr_inval;
1060 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1062 if (EX_ISSYNC(fhp->fh_export)) {
1063 if (file->f_op && file->f_op->fsync) {
1064 err = nfserrno(nfsd_sync(file));
1066 err = nfserr_notsupp;
1073 #endif /* CONFIG_NFSD_V3 */
1076 * Create a file (regular, directory, device, fifo); UNIX sockets
1077 * not yet implemented.
1078 * If the response fh has been verified, the parent directory should
1079 * already be locked. Note that the parent directory is left locked.
1081 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1084 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1085 char *fname, int flen, struct iattr *iap,
1086 int type, dev_t rdev, struct svc_fh *resfhp)
1088 struct dentry *dentry, *dchild = NULL;
1096 if (isdotent(fname, flen))
1099 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1103 dentry = fhp->fh_dentry;
1104 dirp = dentry->d_inode;
1106 err = nfserr_notdir;
1107 if(!dirp->i_op || !dirp->i_op->lookup)
1110 * Check whether the response file handle has been verified yet.
1111 * If it has, the parent directory should already be locked.
1113 if (!resfhp->fh_dentry) {
1114 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1116 dchild = lookup_one_len(fname, dentry, flen);
1117 err = PTR_ERR(dchild);
1120 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1124 /* called from nfsd_proc_create */
1125 dchild = dget(resfhp->fh_dentry);
1126 if (!fhp->fh_locked) {
1127 /* not actually possible */
1129 "nfsd_create: parent %s/%s not locked!\n",
1130 dentry->d_parent->d_name.name,
1131 dentry->d_name.name);
1137 * Make sure the child dentry is still negative ...
1140 if (dchild->d_inode) {
1141 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1142 dentry->d_name.name, dchild->d_name.name);
1146 if (!(iap->ia_valid & ATTR_MODE))
1148 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1151 * Get the dir op function pointer.
1156 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1159 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1165 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1168 printk("nfsd: bad file type %o in nfsd_create\n", type);
1174 if (EX_ISSYNC(fhp->fh_export)) {
1175 err = nfserrno(nfsd_sync_dir(dentry));
1176 write_inode_now(dchild->d_inode, 1);
1180 /* Set file attributes. Mode has already been set and
1181 * setting uid/gid works only for root. Irix appears to
1182 * send along the gid when it tries to implement setgid
1183 * directories via NFS.
1185 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1186 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1191 * Update the file handle to get the new inode info.
1194 err = fh_update(resfhp);
1196 if (dchild && !IS_ERR(dchild))
1201 err = nfserrno(err);
1205 #ifdef CONFIG_NFSD_V3
1207 * NFSv3 version of nfsd_create
1210 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1211 char *fname, int flen, struct iattr *iap,
1212 struct svc_fh *resfhp, int createmode, u32 *verifier,
1215 struct dentry *dentry, *dchild = NULL;
1218 __u32 v_mtime=0, v_atime=0;
1225 if (isdotent(fname, flen))
1227 if (!(iap->ia_valid & ATTR_MODE))
1229 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1233 dentry = fhp->fh_dentry;
1234 dirp = dentry->d_inode;
1236 /* Get all the sanity checks out of the way before
1237 * we lock the parent. */
1238 err = nfserr_notdir;
1239 if(!dirp->i_op || !dirp->i_op->lookup)
1244 * Compose the response file handle.
1246 dchild = lookup_one_len(fname, dentry, flen);
1247 err = PTR_ERR(dchild);
1251 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1255 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1256 /* while the verifier would fit in mtime+atime,
1257 * solaris7 gets confused (bugid 4218508) if these have
1258 * the high bit set, so we use the mode as well
1260 v_mtime = verifier[0]&0x7fffffff;
1261 v_atime = verifier[1]&0x7fffffff;
1263 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1264 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1268 if (dchild->d_inode) {
1271 switch (createmode) {
1272 case NFS3_CREATE_UNCHECKED:
1273 if (! S_ISREG(dchild->d_inode->i_mode))
1276 /* in nfsv4, we need to treat this case a little
1277 * differently. we don't want to truncate the
1278 * file now; this would be wrong if the OPEN
1279 * fails for some other reason. furthermore,
1280 * if the size is nonzero, we should ignore it
1281 * according to spec!
1283 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1286 iap->ia_valid &= ATTR_SIZE;
1290 case NFS3_CREATE_EXCLUSIVE:
1291 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1292 && dchild->d_inode->i_atime.tv_sec == v_atime
1293 && dchild->d_inode->i_mode == v_mode
1294 && dchild->d_inode->i_size == 0 )
1297 case NFS3_CREATE_GUARDED:
1303 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1307 if (EX_ISSYNC(fhp->fh_export)) {
1308 err = nfserrno(nfsd_sync_dir(dentry));
1309 /* setattr will sync the child (or not) */
1312 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1313 /* Cram the verifier into atime/mtime/mode */
1314 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1315 | ATTR_MTIME_SET|ATTR_ATIME_SET
1317 /* XXX someone who knows this better please fix it for nsec */
1318 iap->ia_mtime.tv_sec = v_mtime;
1319 iap->ia_atime.tv_sec = v_atime;
1320 iap->ia_mtime.tv_nsec = 0;
1321 iap->ia_atime.tv_nsec = 0;
1322 iap->ia_mode = v_mode;
1325 /* Set file attributes.
1326 * Mode has already been set but we might need to reset it
1327 * for CREATE_EXCLUSIVE
1328 * Irix appears to send along the gid when it tries to
1329 * implement setgid directories via NFS. Clear out all that cruft.
1332 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1333 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1339 * Update the filehandle to get the new inode info.
1342 err = fh_update(resfhp);
1346 if (dchild && !IS_ERR(dchild))
1351 err = nfserrno(err);
1354 #endif /* CONFIG_NFSD_V3 */
1357 * Read a symlink. On entry, *lenp must contain the maximum path length that
1358 * fits into the buffer. On return, it contains the true length.
1359 * N.B. After this call fhp needs an fh_put
1362 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1364 struct dentry *dentry;
1365 struct inode *inode;
1369 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1373 dentry = fhp->fh_dentry;
1374 inode = dentry->d_inode;
1377 if (!inode->i_op || !inode->i_op->readlink)
1380 touch_atime(fhp->fh_export->ex_mnt, dentry);
1381 /* N.B. Why does this call need a get_fs()??
1382 * Remove the set_fs and watch the fireworks:-) --okir
1385 oldfs = get_fs(); set_fs(KERNEL_DS);
1386 err = inode->i_op->readlink(dentry, buf, *lenp);
1397 err = nfserrno(err);
1402 * Create a symlink and look up its inode
1403 * N.B. After this call _both_ fhp and resfhp need an fh_put
1406 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1407 char *fname, int flen,
1408 char *path, int plen,
1409 struct svc_fh *resfhp,
1412 struct dentry *dentry, *dnew;
1420 if (isdotent(fname, flen))
1423 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1427 dentry = fhp->fh_dentry;
1428 dnew = lookup_one_len(fname, dentry, flen);
1429 err = PTR_ERR(dnew);
1434 /* Only the MODE ATTRibute is even vaguely meaningful */
1435 if (iap && (iap->ia_valid & ATTR_MODE))
1436 mode = iap->ia_mode & S_IALLUGO;
1438 if (unlikely(path[plen] != 0)) {
1439 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1440 if (path_alloced == NULL)
1443 strncpy(path_alloced, path, plen);
1444 path_alloced[plen] = 0;
1445 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1446 kfree(path_alloced);
1449 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1452 if (EX_ISSYNC(fhp->fh_export))
1453 err = nfsd_sync_dir(dentry);
1455 err = nfserrno(err);
1458 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1460 if (err==0) err = cerr;
1465 err = nfserrno(err);
1471 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1474 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1475 char *name, int len, struct svc_fh *tfhp)
1477 struct dentry *ddir, *dnew, *dold;
1478 struct inode *dirp, *dest;
1481 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1484 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1492 if (isdotent(name, len))
1496 ddir = ffhp->fh_dentry;
1497 dirp = ddir->d_inode;
1499 dnew = lookup_one_len(name, ddir, len);
1500 err = PTR_ERR(dnew);
1504 dold = tfhp->fh_dentry;
1505 dest = dold->d_inode;
1507 err = vfs_link(dold, dirp, dnew);
1509 if (EX_ISSYNC(ffhp->fh_export)) {
1510 err = nfserrno(nfsd_sync_dir(ddir));
1511 write_inode_now(dest, 1);
1514 if (err == -EXDEV && rqstp->rq_vers == 2)
1517 err = nfserrno(err);
1526 err = nfserrno(err);
1532 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1535 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1536 struct svc_fh *tfhp, char *tname, int tlen)
1538 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1539 struct inode *fdir, *tdir;
1542 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1545 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1549 fdentry = ffhp->fh_dentry;
1550 fdir = fdentry->d_inode;
1552 tdentry = tfhp->fh_dentry;
1553 tdir = tdentry->d_inode;
1555 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1556 if (fdir->i_sb != tdir->i_sb)
1560 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1563 /* cannot use fh_lock as we need deadlock protective ordering
1564 * so do it by hand */
1565 trap = lock_rename(tdentry, fdentry);
1566 ffhp->fh_locked = tfhp->fh_locked = 1;
1570 odentry = lookup_one_len(fname, fdentry, flen);
1571 err = PTR_ERR(odentry);
1572 if (IS_ERR(odentry))
1576 if (!odentry->d_inode)
1579 if (odentry == trap)
1582 ndentry = lookup_one_len(tname, tdentry, tlen);
1583 err = PTR_ERR(ndentry);
1584 if (IS_ERR(ndentry))
1587 if (ndentry == trap)
1591 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1592 ((atomic_read(&odentry->d_count) > 1)
1593 || (atomic_read(&ndentry->d_count) > 1))) {
1597 err = vfs_rename(fdir, odentry, tdir, ndentry);
1598 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1599 err = nfsd_sync_dir(tdentry);
1601 err = nfsd_sync_dir(fdentry);
1610 err = nfserrno(err);
1612 /* we cannot reply on fh_unlock on the two filehandles,
1613 * as that would do the wrong thing if the two directories
1614 * were the same, so again we do it by hand
1616 fill_post_wcc(ffhp);
1617 fill_post_wcc(tfhp);
1618 unlock_rename(tdentry, fdentry);
1619 ffhp->fh_locked = tfhp->fh_locked = 0;
1626 * Unlink a file or directory
1627 * N.B. After this call fhp needs an fh_put
1630 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1631 char *fname, int flen)
1633 struct dentry *dentry, *rdentry;
1638 if (!flen || isdotent(fname, flen))
1640 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1645 dentry = fhp->fh_dentry;
1646 dirp = dentry->d_inode;
1648 rdentry = lookup_one_len(fname, dentry, flen);
1649 err = PTR_ERR(rdentry);
1650 if (IS_ERR(rdentry))
1653 if (!rdentry->d_inode) {
1660 type = rdentry->d_inode->i_mode & S_IFMT;
1662 if (type != S_IFDIR) { /* It's UNLINK */
1664 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1665 (atomic_read(&rdentry->d_count) > 1)) {
1669 err = vfs_unlink(dirp, rdentry);
1670 } else { /* It's RMDIR */
1671 err = vfs_rmdir(dirp, rdentry);
1677 EX_ISSYNC(fhp->fh_export))
1678 err = nfsd_sync_dir(dentry);
1681 err = nfserrno(err);
1687 * Read entries from a directory.
1688 * The NFSv3/4 verifier we ignore for now.
1691 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1692 struct readdir_cd *cdp, encode_dent_fn func)
1696 loff_t offset = *offsetp;
1698 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1702 offset = vfs_llseek(file, offset, 0);
1704 err = nfserrno((int)offset);
1709 * Read the directory entries. This silly loop is necessary because
1710 * readdir() is not guaranteed to fill up the entire buffer, but
1711 * may choose to do less.
1715 cdp->err = nfserr_eof; /* will be cleared on successful read */
1716 err = vfs_readdir(file, (filldir_t) func, cdp);
1717 } while (err >=0 && cdp->err == nfs_ok);
1719 err = nfserrno(err);
1722 *offsetp = vfs_llseek(file, 0, 1);
1724 if (err == nfserr_eof || err == nfserr_toosmall)
1725 err = nfs_ok; /* can still be found in ->err */
1733 * Get file system stats
1734 * N.B. After this call fhp needs an fh_put
1737 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1739 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1740 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1746 * Check for a user's access permissions to this inode.
1749 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1751 struct inode *inode = dentry->d_inode;
1757 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1759 (acc & MAY_READ)? " read" : "",
1760 (acc & MAY_WRITE)? " write" : "",
1761 (acc & MAY_EXEC)? " exec" : "",
1762 (acc & MAY_SATTR)? " sattr" : "",
1763 (acc & MAY_TRUNC)? " trunc" : "",
1764 (acc & MAY_LOCK)? " lock" : "",
1765 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1767 IS_IMMUTABLE(inode)? " immut" : "",
1768 IS_APPEND(inode)? " append" : "",
1769 IS_RDONLY(inode)? " ro" : "");
1770 dprintk(" owner %d/%d user %d/%d\n",
1771 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1774 /* Normally we reject any write/sattr etc access on a read-only file
1775 * system. But if it is IRIX doing check on write-access for a
1776 * device special file, we ignore rofs.
1778 if (!(acc & MAY_LOCAL_ACCESS))
1779 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1780 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1782 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1785 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1788 if (acc & MAY_LOCK) {
1789 /* If we cannot rely on authentication in NLM requests,
1790 * just allow locks, otherwise require read permission, or
1793 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1796 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1799 * The file owner always gets access permission for accesses that
1800 * would normally be checked at open time. This is to make
1801 * file access work even when the client has done a fchmod(fd, 0).
1803 * However, `cp foo bar' should fail nevertheless when bar is
1804 * readonly. A sensible way to do this might be to reject all
1805 * attempts to truncate a read-only file, because a creat() call
1806 * always implies file truncation.
1807 * ... but this isn't really fair. A process may reasonably call
1808 * ftruncate on an open file descriptor on a file with perm 000.
1809 * We must trust the client to do permission checking - using "ACCESS"
1812 if ((acc & MAY_OWNER_OVERRIDE) &&
1813 inode->i_uid == current->fsuid)
1816 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1818 /* Allow read access to binaries even when mode 111 */
1819 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1820 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1821 err = permission(inode, MAY_EXEC, NULL);
1823 return err? nfserrno(err) : 0;
1827 nfsd_racache_shutdown(void)
1831 dprintk("nfsd: freeing readahead buffers.\n");
1833 raparm_cache = raparml = NULL;
1836 * Initialize readahead param cache
1839 nfsd_racache_init(int cache_size)
1845 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1847 if (raparml != NULL) {
1848 dprintk("nfsd: allocating %d readahead buffers.\n",
1850 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1851 for (i = 0; i < cache_size - 1; i++) {
1852 raparml[i].p_next = raparml + i + 1;
1854 raparm_cache = raparml;
1857 "nfsd: Could not allocate memory read-ahead cache.\n");
1860 nfsdstats.ra_size = cache_size;
1864 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1866 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1868 struct inode *inode = fhp->fh_dentry->d_inode;
1872 struct posix_acl *acl;
1874 if (!IS_POSIXACL(inode))
1875 return ERR_PTR(-EOPNOTSUPP);
1878 case ACL_TYPE_ACCESS:
1879 name = POSIX_ACL_XATTR_ACCESS;
1881 case ACL_TYPE_DEFAULT:
1882 name = POSIX_ACL_XATTR_DEFAULT;
1885 return ERR_PTR(-EOPNOTSUPP);
1888 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1890 return ERR_PTR(size);
1892 acl = posix_acl_from_xattr(value, size);
1898 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1900 struct inode *inode = fhp->fh_dentry->d_inode;
1906 if (!IS_POSIXACL(inode) || !inode->i_op ||
1907 !inode->i_op->setxattr || !inode->i_op->removexattr)
1910 case ACL_TYPE_ACCESS:
1911 name = POSIX_ACL_XATTR_ACCESS;
1913 case ACL_TYPE_DEFAULT:
1914 name = POSIX_ACL_XATTR_DEFAULT;
1920 if (acl && acl->a_count) {
1921 size = posix_acl_xattr_size(acl->a_count);
1922 value = kmalloc(size, GFP_KERNEL);
1925 size = posix_acl_to_xattr(acl, value, size);
1934 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1936 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1939 error = vfs_removexattr(fhp->fh_dentry, name);
1940 if (error == -ENODATA)
1949 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */