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/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/ext2_fs.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
58 #include <asm/uaccess.h>
60 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
64 /* We must ignore files (but only files) which might have mandatory
65 * locks on them because there is no way to know if the accesser has
68 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
71 * This is a cache of readahead params that help us choose the proper
72 * readahead strategy. Initially, we set all readahead parameters to 0
73 * and let the VFS handle things.
74 * If you increase the number of cached files very much, you'll need to
75 * add a hash table here.
78 struct raparms *p_next;
83 struct file_ra_state p_ra;
86 static struct raparms * raparml;
87 static struct raparms * raparm_cache;
90 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
92 * Returns -EAGAIN leaving *dpp and *expp unchanged,
93 * or nfs_ok having possibly changed *dpp and *expp
96 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
97 struct svc_export **expp)
99 struct svc_export *exp = *expp, *exp2 = NULL;
100 struct dentry *dentry = *dpp;
101 struct vfsmount *mnt = mntget(exp->ex_mnt);
102 struct dentry *mounts = dget(dentry);
105 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
107 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
114 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
115 /* successfully crossed mount point */
121 if (exp2) exp_put(exp2);
130 * Look up one component of a pathname.
131 * N.B. After this call _both_ fhp and resfh need an fh_put
133 * If the lookup would cross a mountpoint, and the mounted filesystem
134 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
135 * accepted as it stands and the mounted directory is
136 * returned. Otherwise the covered directory is returned.
137 * NOTE: this mountpoint crossing is not supported properly by all
138 * clients and is explicitly disallowed for NFSv3
139 * NeilBrown <neilb@cse.unsw.edu.au>
142 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
143 int len, struct svc_fh *resfh)
145 struct svc_export *exp;
146 struct dentry *dparent;
147 struct dentry *dentry;
150 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
152 /* Obtain dentry and export. */
153 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
157 dparent = fhp->fh_dentry;
158 exp = fhp->fh_export;
163 /* Lookup the name, but don't follow links */
164 if (isdotent(name, len)) {
166 dentry = dget(dparent);
167 else if (dparent != exp->ex_dentry) {
168 dentry = dget_parent(dparent);
169 } else if (!EX_NOHIDE(exp))
170 dentry = dget(dparent); /* .. == . just like at / */
172 /* checking mountpoint crossing is very different when stepping up */
173 struct svc_export *exp2 = NULL;
175 struct vfsmount *mnt = mntget(exp->ex_mnt);
176 dentry = dget(dparent);
177 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
179 dp = dget_parent(dentry);
183 exp2 = exp_parent(exp->ex_client, mnt, dentry,
193 dentry = dget(dparent);
202 dentry = lookup_one_len(name, dparent, len);
203 err = PTR_ERR(dentry);
207 * check if we have crossed a mount point ...
209 if (d_mountpoint(dentry)) {
210 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
217 * Note: we compose the file handle now, but as the
218 * dentry may be negative, it may need to be updated.
220 err = fh_compose(resfh, exp, dentry, fhp);
221 if (!err && !dentry->d_inode)
234 * Set various file attributes.
235 * N.B. After this call fhp needs an fh_put
238 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
239 int check_guard, time_t guardtime)
241 struct dentry *dentry;
243 int accmode = MAY_SATTR;
249 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
250 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
251 if (iap->ia_valid & ATTR_SIZE)
255 err = fh_verify(rqstp, fhp, ftype, accmode);
259 dentry = fhp->fh_dentry;
260 inode = dentry->d_inode;
262 /* Ignore any mode updates on symlinks */
263 if (S_ISLNK(inode->i_mode))
264 iap->ia_valid &= ~ATTR_MODE;
269 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
270 * which only requires access, and "set-[ac]time-to-X" which
271 * requires ownership.
272 * So if it looks like it might be "set both to the same time which
273 * is close to now", and if inode_change_ok fails, then we
274 * convert to "set to now" instead of "set to explicit time"
276 * We only call inode_change_ok as the last test as technically
277 * it is not an interface that we should be using. It is only
278 * valid if the filesystem does not define it's own i_op->setattr.
280 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
281 #define MAX_TOUCH_TIME_ERROR (30*60)
282 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
283 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
285 /* Looks probable. Now just make sure time is in the right ballpark.
286 * Solaris, at least, doesn't seem to care what the time request is.
287 * We require it be within 30 minutes of now.
289 time_t delta = iap->ia_atime.tv_sec - get_seconds();
290 if (delta<0) delta = -delta;
291 if (delta < MAX_TOUCH_TIME_ERROR &&
292 inode_change_ok(inode, iap) != 0) {
293 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
294 * this will cause notify_change to set these times to "now"
296 iap->ia_valid &= ~BOTH_TIME_SET;
300 /* The size case is special. It changes the file as well as the attributes. */
301 if (iap->ia_valid & ATTR_SIZE) {
302 if (iap->ia_size < inode->i_size) {
303 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
309 * If we are changing the size of the file, then
310 * we need to break all leases.
312 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
313 if (err == -EWOULDBLOCK)
315 if (err) /* ENOMEM or EWOULDBLOCK */
318 err = get_write_access(inode);
323 err = locks_verify_truncate(inode, NULL, iap->ia_size);
325 put_write_access(inode);
331 imode = inode->i_mode;
332 if (iap->ia_valid & ATTR_MODE) {
333 iap->ia_mode &= S_IALLUGO;
334 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
337 /* Revoke setuid/setgid bit on chown/chgrp */
338 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
339 iap->ia_valid |= ATTR_KILL_SUID;
340 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
341 iap->ia_valid |= ATTR_KILL_SGID;
343 /* Change the attributes. */
345 iap->ia_valid |= ATTR_CTIME;
347 err = nfserr_notsync;
348 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
350 err = notify_change(dentry, iap);
355 put_write_access(inode);
357 if (EX_ISSYNC(fhp->fh_export))
358 write_inode_now(inode, 1);
367 #if defined(CONFIG_NFSD_V2_ACL) || \
368 defined(CONFIG_NFSD_V3_ACL) || \
369 defined(CONFIG_NFSD_V4)
370 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
374 buflen = vfs_getxattr(dentry, key, NULL, 0);
378 *buf = kmalloc(buflen, GFP_KERNEL);
382 return vfs_getxattr(dentry, key, *buf, buflen);
386 #if defined(CONFIG_NFSD_V4)
388 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
395 buflen = posix_acl_xattr_size(pacl->a_count);
396 buf = kmalloc(buflen, GFP_KERNEL);
401 len = posix_acl_to_xattr(pacl, buf, buflen);
407 error = vfs_setxattr(dentry, key, buf, len, 0);
414 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
415 struct nfs4_acl *acl)
418 struct dentry *dentry;
420 struct posix_acl *pacl = NULL, *dpacl = NULL;
421 unsigned int flags = 0;
424 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
428 dentry = fhp->fh_dentry;
429 inode = dentry->d_inode;
430 if (S_ISDIR(inode->i_mode))
431 flags = NFS4_ACL_DIR;
433 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
434 if (error == -EINVAL) {
435 error = nfserr_attrnotsupp;
437 } else if (error < 0)
441 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
447 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
455 posix_acl_release(pacl);
456 posix_acl_release(dpacl);
459 error = nfserrno(error);
463 static struct posix_acl *
464 _get_posix_acl(struct dentry *dentry, char *key)
467 struct posix_acl *pacl = NULL;
470 buflen = nfsd_getxattr(dentry, key, &buf);
474 return ERR_PTR(buflen);
476 pacl = posix_acl_from_xattr(buf, buflen);
482 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
484 struct inode *inode = dentry->d_inode;
486 struct posix_acl *pacl = NULL, *dpacl = NULL;
487 unsigned int flags = 0;
489 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
490 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
491 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
493 error = PTR_ERR(pacl);
498 if (S_ISDIR(inode->i_mode)) {
499 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
500 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
502 else if (IS_ERR(dpacl)) {
503 error = PTR_ERR(dpacl);
507 flags = NFS4_ACL_DIR;
510 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
512 error = PTR_ERR(*acl);
516 posix_acl_release(pacl);
517 posix_acl_release(dpacl);
521 #endif /* defined(CONFIG_NFS_V4) */
523 #ifdef CONFIG_NFSD_V3
525 * Check server access rights to a file system object
531 static struct accessmap nfs3_regaccess[] = {
532 { NFS3_ACCESS_READ, MAY_READ },
533 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
534 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
535 { NFS3_ACCESS_EXTEND, MAY_WRITE },
540 static struct accessmap nfs3_diraccess[] = {
541 { NFS3_ACCESS_READ, MAY_READ },
542 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
543 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
544 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
545 { NFS3_ACCESS_DELETE, MAY_REMOVE },
550 static struct accessmap nfs3_anyaccess[] = {
551 /* Some clients - Solaris 2.6 at least, make an access call
552 * to the server to check for access for things like /dev/null
553 * (which really, the server doesn't care about). So
554 * We provide simple access checking for them, looking
555 * mainly at mode bits, and we make sure to ignore read-only
558 { NFS3_ACCESS_READ, MAY_READ },
559 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
560 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
561 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
567 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
569 struct accessmap *map;
570 struct svc_export *export;
571 struct dentry *dentry;
572 u32 query, result = 0, sresult = 0;
575 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
579 export = fhp->fh_export;
580 dentry = fhp->fh_dentry;
582 if (S_ISREG(dentry->d_inode->i_mode))
583 map = nfs3_regaccess;
584 else if (S_ISDIR(dentry->d_inode->i_mode))
585 map = nfs3_diraccess;
587 map = nfs3_anyaccess;
591 for (; map->access; map++) {
592 if (map->access & query) {
595 sresult |= map->access;
597 err2 = nfsd_permission(export, dentry, map->how);
600 result |= map->access;
603 /* the following error codes just mean the access was not allowed,
604 * rather than an error occurred */
608 /* simply don't "or" in the access bit. */
618 *supported = sresult;
623 #endif /* CONFIG_NFSD_V3 */
628 * Open an existing file or directory.
629 * The access argument indicates the type of open (read/write/lock)
630 * N.B. After this call fhp needs an fh_put
633 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
634 int access, struct file **filp)
636 struct dentry *dentry;
638 int flags = O_RDONLY|O_LARGEFILE, err;
641 * If we get here, then the client has already done an "open",
642 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
643 * in case a chmod has now revoked permission.
645 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
649 dentry = fhp->fh_dentry;
650 inode = dentry->d_inode;
652 /* Disallow write access to files with the append-only bit set
653 * or any access when mandatory locking enabled
656 if (IS_APPEND(inode) && (access & MAY_WRITE))
658 if (IS_ISMNDLK(inode))
665 * Check to see if there are any leases on this file.
666 * This may block while leases are broken.
668 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
669 if (err == -EWOULDBLOCK)
671 if (err) /* NOMEM or WOULDBLOCK */
674 if (access & MAY_WRITE) {
675 if (access & MAY_READ)
676 flags = O_RDWR|O_LARGEFILE;
678 flags = O_WRONLY|O_LARGEFILE;
682 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
684 err = PTR_ERR(*filp);
696 nfsd_close(struct file *filp)
703 * As this calls fsync (not fdatasync) there is no need for a write_inode
706 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
707 const struct file_operations *fop)
709 struct inode *inode = dp->d_inode;
710 int (*fsync) (struct file *, struct dentry *, int);
713 err = filemap_fdatawrite(inode->i_mapping);
714 if (err == 0 && fop && (fsync = fop->fsync))
715 err = fsync(filp, dp, 0);
717 err = filemap_fdatawait(inode->i_mapping);
724 nfsd_sync(struct file *filp)
727 struct inode *inode = filp->f_dentry->d_inode;
728 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
729 mutex_lock(&inode->i_mutex);
730 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
731 mutex_unlock(&inode->i_mutex);
737 nfsd_sync_dir(struct dentry *dp)
739 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
743 * Obtain the readahead parameters for the file
744 * specified by (dev, ino).
746 static DEFINE_SPINLOCK(ra_lock);
748 static inline struct raparms *
749 nfsd_get_raparms(dev_t dev, ino_t ino)
751 struct raparms *ra, **rap, **frap = NULL;
755 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
756 if (ra->p_ino == ino && ra->p_dev == dev)
759 if (ra->p_count == 0)
762 depth = nfsdstats.ra_size*11/10;
764 spin_unlock(&ra_lock);
773 if (rap != &raparm_cache) {
775 ra->p_next = raparm_cache;
779 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
780 spin_unlock(&ra_lock);
785 * Grab and keep cached pages assosiated with a file in the svc_rqst
786 * so that they can be passed to the netowork sendmsg/sendpage routines
787 * directrly. They will be released after the sending has completed.
790 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
792 unsigned long count = desc->count;
793 struct svc_rqst *rqstp = desc->arg.data;
798 if (rqstp->rq_res.page_len == 0) {
800 rqstp->rq_respages[rqstp->rq_resused++] = page;
801 rqstp->rq_res.page_base = offset;
802 rqstp->rq_res.page_len = size;
803 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
805 rqstp->rq_respages[rqstp->rq_resused++] = page;
806 rqstp->rq_res.page_len += size;
808 rqstp->rq_res.page_len += size;
811 desc->count = count - size;
812 desc->written += size;
817 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
818 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
826 inode = file->f_dentry->d_inode;
828 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
829 (!lock_may_read(inode, offset, *count)))
833 /* Get readahead parameters */
834 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
837 file->f_ra = ra->p_ra;
839 if (file->f_op->sendfile && rqstp->rq_sendfile_ok) {
840 svc_pushback_unused_pages(rqstp);
841 err = file->f_op->sendfile(file, &offset, *count,
842 nfsd_read_actor, rqstp);
846 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
850 /* Write back readahead params */
853 ra->p_ra = file->f_ra;
856 spin_unlock(&ra_lock);
860 nfsdstats.io_read += err;
863 fsnotify_access(file->f_dentry);
870 static void kill_suid(struct dentry *dentry)
873 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
875 mutex_lock(&dentry->d_inode->i_mutex);
876 notify_change(dentry, &ia);
877 mutex_unlock(&dentry->d_inode->i_mutex);
881 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
882 loff_t offset, struct kvec *vec, int vlen,
883 unsigned long cnt, int *stablep)
885 struct svc_export *exp;
886 struct dentry *dentry;
890 int stable = *stablep;
895 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
896 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
900 dentry = file->f_dentry;
901 inode = dentry->d_inode;
902 exp = fhp->fh_export;
905 * Request sync writes if
906 * - the sync export option has been set, or
907 * - the client requested O_SYNC behavior (NFSv3 feature).
908 * - The file system doesn't support fsync().
909 * When gathered writes have been configured for this volume,
910 * flushing the data to disk is handled separately below.
913 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
915 *stablep = 2; /* FILE_SYNC */
920 if (stable && !EX_WGATHER(exp))
921 file->f_flags |= O_SYNC;
923 /* Write the data. */
924 oldfs = get_fs(); set_fs(KERNEL_DS);
925 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
928 nfsdstats.io_write += cnt;
929 fsnotify_modify(file->f_dentry);
932 /* clear setuid/setgid flag after write */
933 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
936 if (err >= 0 && stable) {
937 static ino_t last_ino;
938 static dev_t last_dev;
941 * Gathered writes: If another process is currently
942 * writing to the file, there's a high chance
943 * this is another nfsd (triggered by a bulk write
944 * from a client's biod). Rather than syncing the
945 * file with each write request, we sleep for 10 msec.
947 * I don't know if this roughly approximates
948 * C. Juszak's idea of gathered writes, but it's a
949 * nice and simple solution (IMHO), and it seems to
952 if (EX_WGATHER(exp)) {
953 if (atomic_read(&inode->i_writecount) > 1
954 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
955 dprintk("nfsd: write defer %d\n", current->pid);
957 dprintk("nfsd: write resume %d\n", current->pid);
960 if (inode->i_state & I_DIRTY) {
961 dprintk("nfsd: write sync %d\n", current->pid);
965 wake_up(&inode->i_wait);
968 last_ino = inode->i_ino;
969 last_dev = inode->i_sb->s_dev;
972 dprintk("nfsd: write complete err=%d\n", err);
982 * Read data from a file. count must contain the requested read count
983 * on entry. On return, *count contains the number of bytes actually read.
984 * N.B. After this call fhp needs an fh_put
987 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
988 loff_t offset, struct kvec *vec, int vlen,
989 unsigned long *count)
994 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
995 MAY_READ|MAY_OWNER_OVERRIDE);
998 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1000 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1003 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1011 * Write data to a file.
1012 * The stable flag requests synchronous writes.
1013 * N.B. After this call fhp needs an fh_put
1016 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1017 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1023 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1024 MAY_WRITE|MAY_OWNER_OVERRIDE);
1027 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1030 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1035 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1043 #ifdef CONFIG_NFSD_V3
1045 * Commit all pending writes to stable storage.
1046 * Strictly speaking, we could sync just the indicated file region here,
1047 * but there's currently no way we can ask the VFS to do so.
1049 * Unfortunately we cannot lock the file to make sure we return full WCC
1050 * data to the client, as locking happens lower down in the filesystem.
1053 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1054 loff_t offset, unsigned long count)
1059 if ((u64)count > ~(u64)offset)
1060 return nfserr_inval;
1062 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1064 if (EX_ISSYNC(fhp->fh_export)) {
1065 if (file->f_op && file->f_op->fsync) {
1066 err = nfserrno(nfsd_sync(file));
1068 err = nfserr_notsupp;
1075 #endif /* CONFIG_NFSD_V3 */
1078 * Create a file (regular, directory, device, fifo); UNIX sockets
1079 * not yet implemented.
1080 * If the response fh has been verified, the parent directory should
1081 * already be locked. Note that the parent directory is left locked.
1083 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1086 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1087 char *fname, int flen, struct iattr *iap,
1088 int type, dev_t rdev, struct svc_fh *resfhp)
1090 struct dentry *dentry, *dchild = NULL;
1098 if (isdotent(fname, flen))
1101 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1105 dentry = fhp->fh_dentry;
1106 dirp = dentry->d_inode;
1108 err = nfserr_notdir;
1109 if(!dirp->i_op || !dirp->i_op->lookup)
1112 * Check whether the response file handle has been verified yet.
1113 * If it has, the parent directory should already be locked.
1115 if (!resfhp->fh_dentry) {
1116 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1117 fh_lock_nested(fhp, I_MUTEX_PARENT);
1118 dchild = lookup_one_len(fname, dentry, flen);
1119 err = PTR_ERR(dchild);
1122 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1126 /* called from nfsd_proc_create */
1127 dchild = dget(resfhp->fh_dentry);
1128 if (!fhp->fh_locked) {
1129 /* not actually possible */
1131 "nfsd_create: parent %s/%s not locked!\n",
1132 dentry->d_parent->d_name.name,
1133 dentry->d_name.name);
1139 * Make sure the child dentry is still negative ...
1142 if (dchild->d_inode) {
1143 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1144 dentry->d_name.name, dchild->d_name.name);
1148 if (!(iap->ia_valid & ATTR_MODE))
1150 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1153 * Get the dir op function pointer.
1158 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1161 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1167 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1170 printk("nfsd: bad file type %o in nfsd_create\n", type);
1176 if (EX_ISSYNC(fhp->fh_export)) {
1177 err = nfserrno(nfsd_sync_dir(dentry));
1178 write_inode_now(dchild->d_inode, 1);
1182 /* Set file attributes. Mode has already been set and
1183 * setting uid/gid works only for root. Irix appears to
1184 * send along the gid when it tries to implement setgid
1185 * directories via NFS.
1187 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1188 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1193 * Update the file handle to get the new inode info.
1196 err = fh_update(resfhp);
1198 if (dchild && !IS_ERR(dchild))
1203 err = nfserrno(err);
1207 #ifdef CONFIG_NFSD_V3
1209 * NFSv3 version of nfsd_create
1212 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1213 char *fname, int flen, struct iattr *iap,
1214 struct svc_fh *resfhp, int createmode, u32 *verifier,
1217 struct dentry *dentry, *dchild = NULL;
1220 __u32 v_mtime=0, v_atime=0;
1227 if (isdotent(fname, flen))
1229 if (!(iap->ia_valid & ATTR_MODE))
1231 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1235 dentry = fhp->fh_dentry;
1236 dirp = dentry->d_inode;
1238 /* Get all the sanity checks out of the way before
1239 * we lock the parent. */
1240 err = nfserr_notdir;
1241 if(!dirp->i_op || !dirp->i_op->lookup)
1243 fh_lock_nested(fhp, I_MUTEX_PARENT);
1246 * Compose the response file handle.
1248 dchild = lookup_one_len(fname, dentry, flen);
1249 err = PTR_ERR(dchild);
1253 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1257 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1258 /* while the verifier would fit in mtime+atime,
1259 * solaris7 gets confused (bugid 4218508) if these have
1260 * the high bit set, so we use the mode as well
1262 v_mtime = verifier[0]&0x7fffffff;
1263 v_atime = verifier[1]&0x7fffffff;
1265 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1266 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1270 if (dchild->d_inode) {
1273 switch (createmode) {
1274 case NFS3_CREATE_UNCHECKED:
1275 if (! S_ISREG(dchild->d_inode->i_mode))
1278 /* in nfsv4, we need to treat this case a little
1279 * differently. we don't want to truncate the
1280 * file now; this would be wrong if the OPEN
1281 * fails for some other reason. furthermore,
1282 * if the size is nonzero, we should ignore it
1283 * according to spec!
1285 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1288 iap->ia_valid &= ATTR_SIZE;
1292 case NFS3_CREATE_EXCLUSIVE:
1293 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1294 && dchild->d_inode->i_atime.tv_sec == v_atime
1295 && dchild->d_inode->i_mode == v_mode
1296 && dchild->d_inode->i_size == 0 )
1299 case NFS3_CREATE_GUARDED:
1305 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1309 if (EX_ISSYNC(fhp->fh_export)) {
1310 err = nfserrno(nfsd_sync_dir(dentry));
1311 /* setattr will sync the child (or not) */
1314 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1315 /* Cram the verifier into atime/mtime/mode */
1316 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1317 | ATTR_MTIME_SET|ATTR_ATIME_SET
1319 /* XXX someone who knows this better please fix it for nsec */
1320 iap->ia_mtime.tv_sec = v_mtime;
1321 iap->ia_atime.tv_sec = v_atime;
1322 iap->ia_mtime.tv_nsec = 0;
1323 iap->ia_atime.tv_nsec = 0;
1324 iap->ia_mode = v_mode;
1327 /* Set file attributes.
1328 * Mode has already been set but we might need to reset it
1329 * for CREATE_EXCLUSIVE
1330 * Irix appears to send along the gid when it tries to
1331 * implement setgid directories via NFS. Clear out all that cruft.
1334 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1335 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1341 * Update the filehandle to get the new inode info.
1344 err = fh_update(resfhp);
1348 if (dchild && !IS_ERR(dchild))
1353 err = nfserrno(err);
1356 #endif /* CONFIG_NFSD_V3 */
1359 * Read a symlink. On entry, *lenp must contain the maximum path length that
1360 * fits into the buffer. On return, it contains the true length.
1361 * N.B. After this call fhp needs an fh_put
1364 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1366 struct dentry *dentry;
1367 struct inode *inode;
1371 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1375 dentry = fhp->fh_dentry;
1376 inode = dentry->d_inode;
1379 if (!inode->i_op || !inode->i_op->readlink)
1382 touch_atime(fhp->fh_export->ex_mnt, dentry);
1383 /* N.B. Why does this call need a get_fs()??
1384 * Remove the set_fs and watch the fireworks:-) --okir
1387 oldfs = get_fs(); set_fs(KERNEL_DS);
1388 err = inode->i_op->readlink(dentry, buf, *lenp);
1399 err = nfserrno(err);
1404 * Create a symlink and look up its inode
1405 * N.B. After this call _both_ fhp and resfhp need an fh_put
1408 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1409 char *fname, int flen,
1410 char *path, int plen,
1411 struct svc_fh *resfhp,
1414 struct dentry *dentry, *dnew;
1422 if (isdotent(fname, flen))
1425 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1429 dentry = fhp->fh_dentry;
1430 dnew = lookup_one_len(fname, dentry, flen);
1431 err = PTR_ERR(dnew);
1436 /* Only the MODE ATTRibute is even vaguely meaningful */
1437 if (iap && (iap->ia_valid & ATTR_MODE))
1438 mode = iap->ia_mode & S_IALLUGO;
1440 if (unlikely(path[plen] != 0)) {
1441 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1442 if (path_alloced == NULL)
1445 strncpy(path_alloced, path, plen);
1446 path_alloced[plen] = 0;
1447 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1448 kfree(path_alloced);
1451 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1454 if (EX_ISSYNC(fhp->fh_export))
1455 err = nfsd_sync_dir(dentry);
1457 err = nfserrno(err);
1460 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1462 if (err==0) err = cerr;
1467 err = nfserrno(err);
1473 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1476 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1477 char *name, int len, struct svc_fh *tfhp)
1479 struct dentry *ddir, *dnew, *dold;
1480 struct inode *dirp, *dest;
1483 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1486 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1494 if (isdotent(name, len))
1497 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1498 ddir = ffhp->fh_dentry;
1499 dirp = ddir->d_inode;
1501 dnew = lookup_one_len(name, ddir, len);
1502 err = PTR_ERR(dnew);
1506 dold = tfhp->fh_dentry;
1507 dest = dold->d_inode;
1509 err = vfs_link(dold, dirp, dnew);
1511 if (EX_ISSYNC(ffhp->fh_export)) {
1512 err = nfserrno(nfsd_sync_dir(ddir));
1513 write_inode_now(dest, 1);
1516 if (err == -EXDEV && rqstp->rq_vers == 2)
1519 err = nfserrno(err);
1529 err = nfserrno(err);
1535 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1538 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1539 struct svc_fh *tfhp, char *tname, int tlen)
1541 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1542 struct inode *fdir, *tdir;
1545 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1548 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1552 fdentry = ffhp->fh_dentry;
1553 fdir = fdentry->d_inode;
1555 tdentry = tfhp->fh_dentry;
1556 tdir = tdentry->d_inode;
1558 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1559 if (ffhp->fh_export != tfhp->fh_export)
1563 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1566 /* cannot use fh_lock as we need deadlock protective ordering
1567 * so do it by hand */
1568 trap = lock_rename(tdentry, fdentry);
1569 ffhp->fh_locked = tfhp->fh_locked = 1;
1573 odentry = lookup_one_len(fname, fdentry, flen);
1574 err = PTR_ERR(odentry);
1575 if (IS_ERR(odentry))
1579 if (!odentry->d_inode)
1582 if (odentry == trap)
1585 ndentry = lookup_one_len(tname, tdentry, tlen);
1586 err = PTR_ERR(ndentry);
1587 if (IS_ERR(ndentry))
1590 if (ndentry == trap)
1594 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1595 ((atomic_read(&odentry->d_count) > 1)
1596 || (atomic_read(&ndentry->d_count) > 1))) {
1600 err = vfs_rename(fdir, odentry, tdir, ndentry);
1601 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1602 err = nfsd_sync_dir(tdentry);
1604 err = nfsd_sync_dir(fdentry);
1613 err = nfserrno(err);
1615 /* we cannot reply on fh_unlock on the two filehandles,
1616 * as that would do the wrong thing if the two directories
1617 * were the same, so again we do it by hand
1619 fill_post_wcc(ffhp);
1620 fill_post_wcc(tfhp);
1621 unlock_rename(tdentry, fdentry);
1622 ffhp->fh_locked = tfhp->fh_locked = 0;
1629 * Unlink a file or directory
1630 * N.B. After this call fhp needs an fh_put
1633 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1634 char *fname, int flen)
1636 struct dentry *dentry, *rdentry;
1641 if (!flen || isdotent(fname, flen))
1643 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1647 fh_lock_nested(fhp, I_MUTEX_PARENT);
1648 dentry = fhp->fh_dentry;
1649 dirp = dentry->d_inode;
1651 rdentry = lookup_one_len(fname, dentry, flen);
1652 err = PTR_ERR(rdentry);
1653 if (IS_ERR(rdentry))
1656 if (!rdentry->d_inode) {
1663 type = rdentry->d_inode->i_mode & S_IFMT;
1665 if (type != S_IFDIR) { /* It's UNLINK */
1667 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1668 (atomic_read(&rdentry->d_count) > 1)) {
1672 err = vfs_unlink(dirp, rdentry);
1673 } else { /* It's RMDIR */
1674 err = vfs_rmdir(dirp, rdentry);
1680 EX_ISSYNC(fhp->fh_export))
1681 err = nfsd_sync_dir(dentry);
1684 err = nfserrno(err);
1690 * Read entries from a directory.
1691 * The NFSv3/4 verifier we ignore for now.
1694 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1695 struct readdir_cd *cdp, encode_dent_fn func)
1699 loff_t offset = *offsetp;
1701 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1705 offset = vfs_llseek(file, offset, 0);
1707 err = nfserrno((int)offset);
1712 * Read the directory entries. This silly loop is necessary because
1713 * readdir() is not guaranteed to fill up the entire buffer, but
1714 * may choose to do less.
1718 cdp->err = nfserr_eof; /* will be cleared on successful read */
1719 err = vfs_readdir(file, (filldir_t) func, cdp);
1720 } while (err >=0 && cdp->err == nfs_ok);
1722 err = nfserrno(err);
1725 *offsetp = vfs_llseek(file, 0, 1);
1727 if (err == nfserr_eof || err == nfserr_toosmall)
1728 err = nfs_ok; /* can still be found in ->err */
1736 * Get file system stats
1737 * N.B. After this call fhp needs an fh_put
1740 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1742 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1743 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1749 * Check for a user's access permissions to this inode.
1752 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1754 struct inode *inode = dentry->d_inode;
1760 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1762 (acc & MAY_READ)? " read" : "",
1763 (acc & MAY_WRITE)? " write" : "",
1764 (acc & MAY_EXEC)? " exec" : "",
1765 (acc & MAY_SATTR)? " sattr" : "",
1766 (acc & MAY_TRUNC)? " trunc" : "",
1767 (acc & MAY_LOCK)? " lock" : "",
1768 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1770 IS_IMMUTABLE(inode)? " immut" : "",
1771 IS_APPEND(inode)? " append" : "",
1772 IS_RDONLY(inode)? " ro" : "");
1773 dprintk(" owner %d/%d user %d/%d\n",
1774 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1777 /* Normally we reject any write/sattr etc access on a read-only file
1778 * system. But if it is IRIX doing check on write-access for a
1779 * device special file, we ignore rofs.
1781 if (!(acc & MAY_LOCAL_ACCESS))
1782 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1783 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1785 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1788 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1791 if (acc & MAY_LOCK) {
1792 /* If we cannot rely on authentication in NLM requests,
1793 * just allow locks, otherwise require read permission, or
1796 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1799 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1802 * The file owner always gets access permission for accesses that
1803 * would normally be checked at open time. This is to make
1804 * file access work even when the client has done a fchmod(fd, 0).
1806 * However, `cp foo bar' should fail nevertheless when bar is
1807 * readonly. A sensible way to do this might be to reject all
1808 * attempts to truncate a read-only file, because a creat() call
1809 * always implies file truncation.
1810 * ... but this isn't really fair. A process may reasonably call
1811 * ftruncate on an open file descriptor on a file with perm 000.
1812 * We must trust the client to do permission checking - using "ACCESS"
1815 if ((acc & MAY_OWNER_OVERRIDE) &&
1816 inode->i_uid == current->fsuid)
1819 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1821 /* Allow read access to binaries even when mode 111 */
1822 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1823 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1824 err = permission(inode, MAY_EXEC, NULL);
1826 return err? nfserrno(err) : 0;
1830 nfsd_racache_shutdown(void)
1834 dprintk("nfsd: freeing readahead buffers.\n");
1836 raparm_cache = raparml = NULL;
1839 * Initialize readahead param cache
1842 nfsd_racache_init(int cache_size)
1848 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1850 if (raparml != NULL) {
1851 dprintk("nfsd: allocating %d readahead buffers.\n",
1853 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1854 for (i = 0; i < cache_size - 1; i++) {
1855 raparml[i].p_next = raparml + i + 1;
1857 raparm_cache = raparml;
1860 "nfsd: Could not allocate memory read-ahead cache.\n");
1863 nfsdstats.ra_size = cache_size;
1867 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1869 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1871 struct inode *inode = fhp->fh_dentry->d_inode;
1875 struct posix_acl *acl;
1877 if (!IS_POSIXACL(inode))
1878 return ERR_PTR(-EOPNOTSUPP);
1881 case ACL_TYPE_ACCESS:
1882 name = POSIX_ACL_XATTR_ACCESS;
1884 case ACL_TYPE_DEFAULT:
1885 name = POSIX_ACL_XATTR_DEFAULT;
1888 return ERR_PTR(-EOPNOTSUPP);
1891 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1893 return ERR_PTR(size);
1895 acl = posix_acl_from_xattr(value, size);
1901 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1903 struct inode *inode = fhp->fh_dentry->d_inode;
1909 if (!IS_POSIXACL(inode) || !inode->i_op ||
1910 !inode->i_op->setxattr || !inode->i_op->removexattr)
1913 case ACL_TYPE_ACCESS:
1914 name = POSIX_ACL_XATTR_ACCESS;
1916 case ACL_TYPE_DEFAULT:
1917 name = POSIX_ACL_XATTR_DEFAULT;
1923 if (acl && acl->a_count) {
1924 size = posix_acl_xattr_size(acl->a_count);
1925 value = kmalloc(size, GFP_KERNEL);
1928 error = posix_acl_to_xattr(acl, value, size);
1936 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1938 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1941 error = vfs_removexattr(fhp->fh_dentry, name);
1942 if (error == -ENODATA)
1951 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */