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/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/dnotify.h>
48 #include <linux/xattr_acl.h>
50 #include <linux/posix_acl.h>
51 #include <linux/posix_acl_xattr.h>
52 #include <linux/xattr_acl.h>
53 #include <linux/xattr.h>
54 #include <linux/nfs4.h>
55 #include <linux/nfs4_acl.h>
56 #include <linux/nfsd_idmap.h>
57 #include <linux/security.h>
58 #endif /* CONFIG_NFSD_V4 */
60 #include <asm/uaccess.h>
62 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
66 /* We must ignore files (but only files) which might have mandatory
67 * locks on them because there is no way to know if the accesser has
70 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
73 * This is a cache of readahead params that help us choose the proper
74 * readahead strategy. Initially, we set all readahead parameters to 0
75 * and let the VFS handle things.
76 * If you increase the number of cached files very much, you'll need to
77 * add a hash table here.
80 struct raparms *p_next;
85 struct file_ra_state p_ra;
88 static struct raparms * raparml;
89 static struct raparms * raparm_cache;
92 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94 * Returns -EAGAIN leaving *dpp and *expp unchanged,
95 * or nfs_ok having possibly changed *dpp and *expp
98 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
99 struct svc_export **expp)
101 struct svc_export *exp = *expp, *exp2 = NULL;
102 struct dentry *dentry = *dpp;
103 struct vfsmount *mnt = mntget(exp->ex_mnt);
104 struct dentry *mounts = dget(dentry);
107 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
109 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
116 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
117 /* successfully crossed mount point */
123 if (exp2) exp_put(exp2);
132 * Look up one component of a pathname.
133 * N.B. After this call _both_ fhp and resfh need an fh_put
135 * If the lookup would cross a mountpoint, and the mounted filesystem
136 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
137 * accepted as it stands and the mounted directory is
138 * returned. Otherwise the covered directory is returned.
139 * NOTE: this mountpoint crossing is not supported properly by all
140 * clients and is explicitly disallowed for NFSv3
141 * NeilBrown <neilb@cse.unsw.edu.au>
144 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
145 int len, struct svc_fh *resfh)
147 struct svc_export *exp;
148 struct dentry *dparent;
149 struct dentry *dentry;
152 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
154 /* Obtain dentry and export. */
155 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
159 dparent = fhp->fh_dentry;
160 exp = fhp->fh_export;
165 /* Lookup the name, but don't follow links */
166 if (isdotent(name, len)) {
168 dentry = dget(dparent);
169 else if (dparent != exp->ex_dentry) {
170 dentry = dget_parent(dparent);
171 } else if (!EX_NOHIDE(exp))
172 dentry = dget(dparent); /* .. == . just like at / */
174 /* checking mountpoint crossing is very different when stepping up */
175 struct svc_export *exp2 = NULL;
177 struct vfsmount *mnt = mntget(exp->ex_mnt);
178 dentry = dget(dparent);
179 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
181 dp = dget_parent(dentry);
185 exp2 = exp_parent(exp->ex_client, mnt, dentry,
195 dentry = dget(dparent);
204 dentry = lookup_one_len(name, dparent, len);
205 err = PTR_ERR(dentry);
209 * check if we have crossed a mount point ...
211 if (d_mountpoint(dentry)) {
212 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
219 * Note: we compose the file handle now, but as the
220 * dentry may be negative, it may need to be updated.
222 err = fh_compose(resfh, exp, dentry, fhp);
223 if (!err && !dentry->d_inode)
236 * Set various file attributes.
237 * N.B. After this call fhp needs an fh_put
240 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
241 int check_guard, time_t guardtime)
243 struct dentry *dentry;
245 int accmode = MAY_SATTR;
251 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
252 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
253 if (iap->ia_valid & ATTR_SIZE)
257 err = fh_verify(rqstp, fhp, ftype, accmode);
261 dentry = fhp->fh_dentry;
262 inode = dentry->d_inode;
264 /* Ignore any mode updates on symlinks */
265 if (S_ISLNK(inode->i_mode))
266 iap->ia_valid &= ~ATTR_MODE;
271 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
272 * which only requires access, and "set-[ac]time-to-X" which
273 * requires ownership.
274 * So if it looks like it might be "set both to the same time which
275 * is close to now", and if inode_change_ok fails, then we
276 * convert to "set to now" instead of "set to explicit time"
278 * We only call inode_change_ok as the last test as technically
279 * it is not an interface that we should be using. It is only
280 * valid if the filesystem does not define it's own i_op->setattr.
282 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
283 #define MAX_TOUCH_TIME_ERROR (30*60)
284 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
285 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
287 /* Looks probable. Now just make sure time is in the right ballpark.
288 * Solaris, at least, doesn't seem to care what the time request is.
289 * We require it be within 30 minutes of now.
291 time_t delta = iap->ia_atime.tv_sec - get_seconds();
292 if (delta<0) delta = -delta;
293 if (delta < MAX_TOUCH_TIME_ERROR &&
294 inode_change_ok(inode, iap) != 0) {
295 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
296 * this will cause notify_change to set these times to "now"
298 iap->ia_valid &= ~BOTH_TIME_SET;
302 /* The size case is special. It changes the file as well as the attributes. */
303 if (iap->ia_valid & ATTR_SIZE) {
304 if (iap->ia_size < inode->i_size) {
305 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
311 * If we are changing the size of the file, then
312 * we need to break all leases.
314 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
315 if (err == -EWOULDBLOCK)
317 if (err) /* ENOMEM or EWOULDBLOCK */
320 err = get_write_access(inode);
325 err = locks_verify_truncate(inode, NULL, iap->ia_size);
327 put_write_access(inode);
333 imode = inode->i_mode;
334 if (iap->ia_valid & ATTR_MODE) {
335 iap->ia_mode &= S_IALLUGO;
336 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
339 /* Revoke setuid/setgid bit on chown/chgrp */
340 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
341 iap->ia_valid |= ATTR_KILL_SUID;
342 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
343 iap->ia_valid |= ATTR_KILL_SGID;
345 /* Change the attributes. */
347 iap->ia_valid |= ATTR_CTIME;
349 err = nfserr_notsync;
350 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
352 err = notify_change(dentry, iap);
357 put_write_access(inode);
359 if (EX_ISSYNC(fhp->fh_export))
360 write_inode_now(inode, 1);
369 #if defined(CONFIG_NFSD_V4)
372 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
378 struct inode *inode = dentry->d_inode;
380 buflen = posix_acl_xattr_size(pacl->a_count);
381 buf = kmalloc(buflen, GFP_KERNEL);
386 len = posix_acl_to_xattr(pacl, buf, buflen);
393 if (inode->i_op && inode->i_op->setxattr) {
395 security_inode_setxattr(dentry, key, buf, len, 0);
396 error = inode->i_op->setxattr(dentry, key, buf, len, 0);
398 security_inode_post_setxattr(dentry, key, buf, len, 0);
407 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
408 struct nfs4_acl *acl)
411 struct dentry *dentry;
413 struct posix_acl *pacl = NULL, *dpacl = NULL;
414 unsigned int flags = 0;
417 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
421 dentry = fhp->fh_dentry;
422 inode = dentry->d_inode;
423 if (S_ISDIR(inode->i_mode))
424 flags = NFS4_ACL_DIR;
426 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
431 error = set_nfsv4_acl_one(dentry, pacl, XATTR_NAME_ACL_ACCESS);
437 error = set_nfsv4_acl_one(dentry, dpacl, XATTR_NAME_ACL_DEFAULT);
445 posix_acl_release(pacl);
446 posix_acl_release(dpacl);
449 error = nfserrno(error);
453 static struct posix_acl *
454 _get_posix_acl(struct dentry *dentry, char *key)
456 struct inode *inode = dentry->d_inode;
458 int buflen, error = 0;
459 struct posix_acl *pacl = NULL;
462 if (inode->i_op == NULL)
464 if (inode->i_op->getxattr == NULL)
467 error = security_inode_getxattr(dentry, key);
471 buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
473 error = buflen < 0 ? buflen : -ENODATA;
477 buf = kmalloc(buflen, GFP_KERNEL);
483 error = inode->i_op->getxattr(dentry, key, buf, buflen);
487 pacl = posix_acl_from_xattr(buf, buflen);
492 pacl = ERR_PTR(error);
497 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
499 struct inode *inode = dentry->d_inode;
501 struct posix_acl *pacl = NULL, *dpacl = NULL;
502 unsigned int flags = 0;
504 pacl = _get_posix_acl(dentry, XATTR_NAME_ACL_ACCESS);
505 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
506 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
508 error = PTR_ERR(pacl);
513 if (S_ISDIR(inode->i_mode)) {
514 dpacl = _get_posix_acl(dentry, XATTR_NAME_ACL_DEFAULT);
515 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
517 else if (IS_ERR(dpacl)) {
518 error = PTR_ERR(dpacl);
522 flags = NFS4_ACL_DIR;
525 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
527 error = PTR_ERR(*acl);
531 posix_acl_release(pacl);
532 posix_acl_release(dpacl);
536 #endif /* defined(CONFIG_NFS_V4) */
538 #ifdef CONFIG_NFSD_V3
540 * Check server access rights to a file system object
546 static struct accessmap nfs3_regaccess[] = {
547 { NFS3_ACCESS_READ, MAY_READ },
548 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
549 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
550 { NFS3_ACCESS_EXTEND, MAY_WRITE },
555 static struct accessmap nfs3_diraccess[] = {
556 { NFS3_ACCESS_READ, MAY_READ },
557 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
558 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
559 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
560 { NFS3_ACCESS_DELETE, MAY_REMOVE },
565 static struct accessmap nfs3_anyaccess[] = {
566 /* Some clients - Solaris 2.6 at least, make an access call
567 * to the server to check for access for things like /dev/null
568 * (which really, the server doesn't care about). So
569 * We provide simple access checking for them, looking
570 * mainly at mode bits, and we make sure to ignore read-only
573 { NFS3_ACCESS_READ, MAY_READ },
574 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
575 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
576 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
582 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
584 struct accessmap *map;
585 struct svc_export *export;
586 struct dentry *dentry;
587 u32 query, result = 0, sresult = 0;
590 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
594 export = fhp->fh_export;
595 dentry = fhp->fh_dentry;
597 if (S_ISREG(dentry->d_inode->i_mode))
598 map = nfs3_regaccess;
599 else if (S_ISDIR(dentry->d_inode->i_mode))
600 map = nfs3_diraccess;
602 map = nfs3_anyaccess;
606 for (; map->access; map++) {
607 if (map->access & query) {
610 sresult |= map->access;
612 err2 = nfsd_permission(export, dentry, map->how);
615 result |= map->access;
618 /* the following error codes just mean the access was not allowed,
619 * rather than an error occurred */
623 /* simply don't "or" in the access bit. */
633 *supported = sresult;
638 #endif /* CONFIG_NFSD_V3 */
643 * Open an existing file or directory.
644 * The access argument indicates the type of open (read/write/lock)
645 * N.B. After this call fhp needs an fh_put
648 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
649 int access, struct file **filp)
651 struct dentry *dentry;
653 int flags = O_RDONLY|O_LARGEFILE, err;
656 * If we get here, then the client has already done an "open",
657 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
658 * in case a chmod has now revoked permission.
660 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
664 dentry = fhp->fh_dentry;
665 inode = dentry->d_inode;
667 /* Disallow access to files with the append-only bit set or
668 * with mandatory locking enabled
671 if (IS_APPEND(inode) || IS_ISMNDLK(inode))
677 * Check to see if there are any leases on this file.
678 * This may block while leases are broken.
680 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
681 if (err == -EWOULDBLOCK)
683 if (err) /* NOMEM or WOULDBLOCK */
686 if (access & MAY_WRITE) {
687 flags = O_WRONLY|O_LARGEFILE;
691 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
693 err = PTR_ERR(*filp);
705 nfsd_close(struct file *filp)
712 * As this calls fsync (not fdatasync) there is no need for a write_inode
715 inline void nfsd_dosync(struct file *filp, struct dentry *dp,
716 struct file_operations *fop)
718 struct inode *inode = dp->d_inode;
719 int (*fsync) (struct file *, struct dentry *, int);
721 filemap_fdatawrite(inode->i_mapping);
722 if (fop && (fsync = fop->fsync))
724 filemap_fdatawait(inode->i_mapping);
729 nfsd_sync(struct file *filp)
731 struct inode *inode = filp->f_dentry->d_inode;
732 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
734 nfsd_dosync(filp, filp->f_dentry, filp->f_op);
739 nfsd_sync_dir(struct dentry *dp)
741 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
745 * Obtain the readahead parameters for the file
746 * specified by (dev, ino).
748 static spinlock_t ra_lock = SPIN_LOCK_UNLOCKED;
750 static inline struct raparms *
751 nfsd_get_raparms(dev_t dev, ino_t ino)
753 struct raparms *ra, **rap, **frap = NULL;
757 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
758 if (ra->p_ino == ino && ra->p_dev == dev)
761 if (ra->p_count == 0)
764 depth = nfsdstats.ra_size*11/10;
766 spin_unlock(&ra_lock);
775 if (rap != &raparm_cache) {
777 ra->p_next = raparm_cache;
781 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
782 spin_unlock(&ra_lock);
787 * Grab and keep cached pages assosiated with a file in the svc_rqst
788 * so that they can be passed to the netowork sendmsg/sendpage routines
789 * directrly. They will be released after the sending has completed.
792 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
794 unsigned long count = desc->count;
795 struct svc_rqst *rqstp = desc->arg.data;
800 if (rqstp->rq_res.page_len == 0) {
802 rqstp->rq_respages[rqstp->rq_resused++] = page;
803 rqstp->rq_res.page_base = offset;
804 rqstp->rq_res.page_len = size;
805 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
807 rqstp->rq_respages[rqstp->rq_resused++] = page;
808 rqstp->rq_res.page_len += size;
810 rqstp->rq_res.page_len += size;
813 desc->count = count - size;
814 desc->written += size;
819 * Read data from a file. count must contain the requested read count
820 * on entry. On return, *count contains the number of bytes actually read.
821 * N.B. After this call fhp needs an fh_put
824 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
825 struct kvec *vec, int vlen, unsigned long *count)
833 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
837 inode = file->f_dentry->d_inode;
839 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
840 (!lock_may_read(inode, offset, *count)))
844 /* Get readahead parameters */
845 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
848 file->f_ra = ra->p_ra;
850 if (file->f_op->sendfile) {
851 svc_pushback_unused_pages(rqstp);
852 err = file->f_op->sendfile(file, &offset, *count,
853 nfsd_read_actor, rqstp);
857 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
861 /* Write back readahead params */
864 ra->p_ra = file->f_ra;
867 spin_unlock(&ra_lock);
871 nfsdstats.io_read += err;
874 dnotify_parent(file->f_dentry, DN_ACCESS);
884 * Write data to a file.
885 * The stable flag requests synchronous writes.
886 * N.B. After this call fhp needs an fh_put
889 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t offset,
890 struct kvec *vec, int vlen,
891 unsigned long cnt, int *stablep)
893 struct svc_export *exp;
895 struct dentry *dentry;
899 int stable = *stablep;
901 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
909 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
910 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
914 dentry = file->f_dentry;
915 inode = dentry->d_inode;
916 exp = fhp->fh_export;
919 * Request sync writes if
920 * - the sync export option has been set, or
921 * - the client requested O_SYNC behavior (NFSv3 feature).
922 * - The file system doesn't support fsync().
923 * When gathered writes have been configured for this volume,
924 * flushing the data to disk is handled separately below.
927 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
929 *stablep = 2; /* FILE_SYNC */
934 if (stable && !EX_WGATHER(exp))
935 file->f_flags |= O_SYNC;
937 /* Write the data. */
938 oldfs = get_fs(); set_fs(KERNEL_DS);
939 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
942 nfsdstats.io_write += cnt;
943 dnotify_parent(file->f_dentry, DN_MODIFY);
946 /* clear setuid/setgid flag after write */
947 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
949 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
952 notify_change(dentry, &ia);
956 if (err >= 0 && stable) {
957 static ino_t last_ino;
958 static dev_t last_dev;
961 * Gathered writes: If another process is currently
962 * writing to the file, there's a high chance
963 * this is another nfsd (triggered by a bulk write
964 * from a client's biod). Rather than syncing the
965 * file with each write request, we sleep for 10 msec.
967 * I don't know if this roughly approximates
968 * C. Juszak's idea of gathered writes, but it's a
969 * nice and simple solution (IMHO), and it seems to
972 if (EX_WGATHER(exp)) {
973 if (atomic_read(&inode->i_writecount) > 1
974 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
975 dprintk("nfsd: write defer %d\n", current->pid);
976 set_current_state(TASK_UNINTERRUPTIBLE);
977 schedule_timeout((HZ+99)/100);
978 dprintk("nfsd: write resume %d\n", current->pid);
981 if (inode->i_state & I_DIRTY) {
982 dprintk("nfsd: write sync %d\n", current->pid);
986 wake_up(&inode->i_wait);
989 last_ino = inode->i_ino;
990 last_dev = inode->i_sb->s_dev;
993 dprintk("nfsd: write complete err=%d\n", err);
1005 #ifdef CONFIG_NFSD_V3
1007 * Commit all pending writes to stable storage.
1008 * Strictly speaking, we could sync just the indicated file region here,
1009 * but there's currently no way we can ask the VFS to do so.
1011 * Unfortunately we cannot lock the file to make sure we return full WCC
1012 * data to the client, as locking happens lower down in the filesystem.
1015 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1016 loff_t offset, unsigned long count)
1021 if ((u64)count > ~(u64)offset)
1022 return nfserr_inval;
1024 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1026 if (EX_ISSYNC(fhp->fh_export)) {
1027 if (file->f_op && file->f_op->fsync) {
1030 err = nfserr_notsupp;
1037 #endif /* CONFIG_NFSD_V3 */
1040 * Create a file (regular, directory, device, fifo); UNIX sockets
1041 * not yet implemented.
1042 * If the response fh has been verified, the parent directory should
1043 * already be locked. Note that the parent directory is left locked.
1045 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1048 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1049 char *fname, int flen, struct iattr *iap,
1050 int type, dev_t rdev, struct svc_fh *resfhp)
1052 struct dentry *dentry, *dchild = NULL;
1060 if (isdotent(fname, flen))
1063 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1067 dentry = fhp->fh_dentry;
1068 dirp = dentry->d_inode;
1070 err = nfserr_notdir;
1071 if(!dirp->i_op || !dirp->i_op->lookup)
1074 * Check whether the response file handle has been verified yet.
1075 * If it has, the parent directory should already be locked.
1077 if (!resfhp->fh_dentry) {
1078 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1080 dchild = lookup_one_len(fname, dentry, flen);
1081 err = PTR_ERR(dchild);
1084 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1088 /* called from nfsd_proc_create */
1089 dchild = dget(resfhp->fh_dentry);
1090 if (!fhp->fh_locked) {
1091 /* not actually possible */
1093 "nfsd_create: parent %s/%s not locked!\n",
1094 dentry->d_parent->d_name.name,
1095 dentry->d_name.name);
1101 * Make sure the child dentry is still negative ...
1104 if (dchild->d_inode) {
1105 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1106 dentry->d_name.name, dchild->d_name.name);
1110 if (!(iap->ia_valid & ATTR_MODE))
1112 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1115 * Get the dir op function pointer.
1120 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1123 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1129 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1132 printk("nfsd: bad file type %o in nfsd_create\n", type);
1138 if (EX_ISSYNC(fhp->fh_export)) {
1139 nfsd_sync_dir(dentry);
1140 write_inode_now(dchild->d_inode, 1);
1144 /* Set file attributes. Mode has already been set and
1145 * setting uid/gid works only for root. Irix appears to
1146 * send along the gid when it tries to implement setgid
1147 * directories via NFS.
1150 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1151 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1153 * Update the file handle to get the new inode info.
1156 err = fh_update(resfhp);
1158 if (dchild && !IS_ERR(dchild))
1163 err = nfserrno(err);
1167 #ifdef CONFIG_NFSD_V3
1169 * NFSv3 version of nfsd_create
1172 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1173 char *fname, int flen, struct iattr *iap,
1174 struct svc_fh *resfhp, int createmode, u32 *verifier,
1177 struct dentry *dentry, *dchild = NULL;
1180 __u32 v_mtime=0, v_atime=0;
1187 if (isdotent(fname, flen))
1189 if (!(iap->ia_valid & ATTR_MODE))
1191 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1195 dentry = fhp->fh_dentry;
1196 dirp = dentry->d_inode;
1198 /* Get all the sanity checks out of the way before
1199 * we lock the parent. */
1200 err = nfserr_notdir;
1201 if(!dirp->i_op || !dirp->i_op->lookup)
1206 * Compose the response file handle.
1208 dchild = lookup_one_len(fname, dentry, flen);
1209 err = PTR_ERR(dchild);
1213 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1217 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1218 /* while the verifier would fit in mtime+atime,
1219 * solaris7 gets confused (bugid 4218508) if these have
1220 * the high bit set, so we use the mode as well
1222 v_mtime = verifier[0]&0x7fffffff;
1223 v_atime = verifier[1]&0x7fffffff;
1225 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1226 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1230 if (dchild->d_inode) {
1233 switch (createmode) {
1234 case NFS3_CREATE_UNCHECKED:
1235 if (! S_ISREG(dchild->d_inode->i_mode))
1238 /* in nfsv4, we need to treat this case a little
1239 * differently. we don't want to truncate the
1240 * file now; this would be wrong if the OPEN
1241 * fails for some other reason. furthermore,
1242 * if the size is nonzero, we should ignore it
1243 * according to spec!
1245 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1248 iap->ia_valid &= ATTR_SIZE;
1252 case NFS3_CREATE_EXCLUSIVE:
1253 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1254 && dchild->d_inode->i_atime.tv_sec == v_atime
1255 && dchild->d_inode->i_mode == v_mode
1256 && dchild->d_inode->i_size == 0 )
1259 case NFS3_CREATE_GUARDED:
1265 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1269 if (EX_ISSYNC(fhp->fh_export)) {
1270 nfsd_sync_dir(dentry);
1271 /* setattr will sync the child (or not) */
1275 * Update the filehandle to get the new inode info.
1277 err = fh_update(resfhp);
1281 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1282 /* Cram the verifier into atime/mtime/mode */
1283 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1284 | ATTR_MTIME_SET|ATTR_ATIME_SET
1286 /* XXX someone who knows this better please fix it for nsec */
1287 iap->ia_mtime.tv_sec = v_mtime;
1288 iap->ia_atime.tv_sec = v_atime;
1289 iap->ia_mtime.tv_nsec = 0;
1290 iap->ia_atime.tv_nsec = 0;
1291 iap->ia_mode = v_mode;
1294 /* Set file attributes.
1295 * Mode has already been set but we might need to reset it
1296 * for CREATE_EXCLUSIVE
1297 * Irix appears to send along the gid when it tries to
1298 * implement setgid directories via NFS. Clear out all that cruft.
1301 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1302 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1306 if (dchild && !IS_ERR(dchild))
1311 err = nfserrno(err);
1314 #endif /* CONFIG_NFSD_V3 */
1317 * Read a symlink. On entry, *lenp must contain the maximum path length that
1318 * fits into the buffer. On return, it contains the true length.
1319 * N.B. After this call fhp needs an fh_put
1322 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1324 struct dentry *dentry;
1325 struct inode *inode;
1329 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1333 dentry = fhp->fh_dentry;
1334 inode = dentry->d_inode;
1337 if (!inode->i_op || !inode->i_op->readlink)
1340 touch_atime(fhp->fh_export->ex_mnt, dentry);
1341 /* N.B. Why does this call need a get_fs()??
1342 * Remove the set_fs and watch the fireworks:-) --okir
1345 oldfs = get_fs(); set_fs(KERNEL_DS);
1346 err = inode->i_op->readlink(dentry, buf, *lenp);
1357 err = nfserrno(err);
1362 * Create a symlink and look up its inode
1363 * N.B. After this call _both_ fhp and resfhp need an fh_put
1366 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1367 char *fname, int flen,
1368 char *path, int plen,
1369 struct svc_fh *resfhp,
1372 struct dentry *dentry, *dnew;
1380 if (isdotent(fname, flen))
1383 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1387 dentry = fhp->fh_dentry;
1388 dnew = lookup_one_len(fname, dentry, flen);
1389 err = PTR_ERR(dnew);
1394 /* Only the MODE ATTRibute is even vaguely meaningful */
1395 if (iap && (iap->ia_valid & ATTR_MODE))
1396 mode = iap->ia_mode & S_IALLUGO;
1398 if (unlikely(path[plen] != 0)) {
1399 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1400 if (path_alloced == NULL)
1403 strncpy(path_alloced, path, plen);
1404 path_alloced[plen] = 0;
1405 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1406 kfree(path_alloced);
1409 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1412 if (EX_ISSYNC(fhp->fh_export))
1413 nfsd_sync_dir(dentry);
1415 err = nfserrno(err);
1418 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1420 if (err==0) err = cerr;
1425 err = nfserrno(err);
1431 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1434 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1435 char *name, int len, struct svc_fh *tfhp)
1437 struct dentry *ddir, *dnew, *dold;
1438 struct inode *dirp, *dest;
1441 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1444 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1452 if (isdotent(name, len))
1456 ddir = ffhp->fh_dentry;
1457 dirp = ddir->d_inode;
1459 dnew = lookup_one_len(name, ddir, len);
1460 err = PTR_ERR(dnew);
1464 dold = tfhp->fh_dentry;
1465 dest = dold->d_inode;
1467 err = vfs_link(dold, dirp, dnew);
1469 if (EX_ISSYNC(ffhp->fh_export)) {
1470 nfsd_sync_dir(ddir);
1471 write_inode_now(dest, 1);
1474 if (err == -EXDEV && rqstp->rq_vers == 2)
1477 err = nfserrno(err);
1486 err = nfserrno(err);
1492 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1495 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1496 struct svc_fh *tfhp, char *tname, int tlen)
1498 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1499 struct inode *fdir, *tdir;
1502 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1505 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1509 fdentry = ffhp->fh_dentry;
1510 fdir = fdentry->d_inode;
1512 tdentry = tfhp->fh_dentry;
1513 tdir = tdentry->d_inode;
1515 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1516 if (fdir->i_sb != tdir->i_sb)
1520 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1523 /* cannot use fh_lock as we need deadlock protective ordering
1524 * so do it by hand */
1525 trap = lock_rename(tdentry, fdentry);
1526 ffhp->fh_locked = tfhp->fh_locked = 1;
1530 odentry = lookup_one_len(fname, fdentry, flen);
1531 err = PTR_ERR(odentry);
1532 if (IS_ERR(odentry))
1536 if (!odentry->d_inode)
1539 if (odentry == trap)
1542 ndentry = lookup_one_len(tname, tdentry, tlen);
1543 err = PTR_ERR(ndentry);
1544 if (IS_ERR(ndentry))
1547 if (ndentry == trap)
1551 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1552 ((atomic_read(&odentry->d_count) > 1)
1553 || (atomic_read(&ndentry->d_count) > 1))) {
1557 err = vfs_rename(fdir, odentry, tdir, ndentry);
1558 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1559 nfsd_sync_dir(tdentry);
1560 nfsd_sync_dir(fdentry);
1569 err = nfserrno(err);
1571 /* we cannot reply on fh_unlock on the two filehandles,
1572 * as that would do the wrong thing if the two directories
1573 * were the same, so again we do it by hand
1575 fill_post_wcc(ffhp);
1576 fill_post_wcc(tfhp);
1577 unlock_rename(tdentry, fdentry);
1578 ffhp->fh_locked = tfhp->fh_locked = 0;
1585 * Unlink a file or directory
1586 * N.B. After this call fhp needs an fh_put
1589 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1590 char *fname, int flen)
1592 struct dentry *dentry, *rdentry;
1597 if (!flen || isdotent(fname, flen))
1599 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1604 dentry = fhp->fh_dentry;
1605 dirp = dentry->d_inode;
1607 rdentry = lookup_one_len(fname, dentry, flen);
1608 err = PTR_ERR(rdentry);
1609 if (IS_ERR(rdentry))
1612 if (!rdentry->d_inode) {
1619 type = rdentry->d_inode->i_mode & S_IFMT;
1621 if (type != S_IFDIR) { /* It's UNLINK */
1623 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1624 (atomic_read(&rdentry->d_count) > 1)) {
1628 err = vfs_unlink(dirp, rdentry);
1629 } else { /* It's RMDIR */
1630 err = vfs_rmdir(dirp, rdentry);
1637 if (EX_ISSYNC(fhp->fh_export))
1638 nfsd_sync_dir(dentry);
1644 err = nfserrno(err);
1649 * Read entries from a directory.
1650 * The NFSv3/4 verifier we ignore for now.
1653 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1654 struct readdir_cd *cdp, encode_dent_fn func)
1658 loff_t offset = *offsetp;
1660 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1664 offset = vfs_llseek(file, offset, 0);
1666 err = nfserrno((int)offset);
1671 * Read the directory entries. This silly loop is necessary because
1672 * readdir() is not guaranteed to fill up the entire buffer, but
1673 * may choose to do less.
1677 cdp->err = nfserr_eof; /* will be cleared on successful read */
1678 err = vfs_readdir(file, (filldir_t) func, cdp);
1679 } while (err >=0 && cdp->err == nfs_ok);
1681 err = nfserrno(err);
1684 *offsetp = vfs_llseek(file, 0, 1);
1686 if (err == nfserr_eof || err == nfserr_toosmall)
1687 err = nfs_ok; /* can still be found in ->err */
1695 * Get file system stats
1696 * N.B. After this call fhp needs an fh_put
1699 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1701 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1702 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1708 * Check for a user's access permissions to this inode.
1711 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1713 struct inode *inode = dentry->d_inode;
1719 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1721 (acc & MAY_READ)? " read" : "",
1722 (acc & MAY_WRITE)? " write" : "",
1723 (acc & MAY_EXEC)? " exec" : "",
1724 (acc & MAY_SATTR)? " sattr" : "",
1725 (acc & MAY_TRUNC)? " trunc" : "",
1726 (acc & MAY_LOCK)? " lock" : "",
1727 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1729 IS_IMMUTABLE(inode)? " immut" : "",
1730 IS_APPEND(inode)? " append" : "",
1731 IS_RDONLY(inode)? " ro" : "");
1732 dprintk(" owner %d/%d user %d/%d\n",
1733 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1736 /* Normally we reject any write/sattr etc access on a read-only file
1737 * system. But if it is IRIX doing check on write-access for a
1738 * device special file, we ignore rofs.
1740 if (!(acc & MAY_LOCAL_ACCESS))
1741 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1742 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1744 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1747 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1750 if (acc & MAY_LOCK) {
1751 /* If we cannot rely on authentication in NLM requests,
1752 * just allow locks, otherwise require read permission, or
1755 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1758 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1761 * The file owner always gets access permission for accesses that
1762 * would normally be checked at open time. This is to make
1763 * file access work even when the client has done a fchmod(fd, 0).
1765 * However, `cp foo bar' should fail nevertheless when bar is
1766 * readonly. A sensible way to do this might be to reject all
1767 * attempts to truncate a read-only file, because a creat() call
1768 * always implies file truncation.
1769 * ... but this isn't really fair. A process may reasonably call
1770 * ftruncate on an open file descriptor on a file with perm 000.
1771 * We must trust the client to do permission checking - using "ACCESS"
1774 if ((acc & MAY_OWNER_OVERRIDE) &&
1775 inode->i_uid == current->fsuid)
1778 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1780 /* Allow read access to binaries even when mode 111 */
1781 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1782 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1783 err = permission(inode, MAY_EXEC, NULL);
1785 return err? nfserrno(err) : 0;
1789 nfsd_racache_shutdown(void)
1793 dprintk("nfsd: freeing readahead buffers.\n");
1795 raparm_cache = raparml = NULL;
1798 * Initialize readahead param cache
1801 nfsd_racache_init(int cache_size)
1807 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1809 if (raparml != NULL) {
1810 dprintk("nfsd: allocating %d readahead buffers.\n",
1812 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1813 for (i = 0; i < cache_size - 1; i++) {
1814 raparml[i].p_next = raparml + i + 1;
1816 raparm_cache = raparml;
1819 "nfsd: Could not allocate memory read-ahead cache.\n");
1822 nfsdstats.ra_size = cache_size;
1826 #ifdef CONFIG_NFSD_ACL
1828 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1830 struct inode *inode = fhp->fh_dentry->d_inode;
1834 struct posix_acl *acl;
1836 if (!IS_POSIXACL(inode) || !inode->i_op || !inode->i_op->getxattr)
1837 return ERR_PTR(-EOPNOTSUPP);
1839 case ACL_TYPE_ACCESS:
1840 name = XATTR_NAME_ACL_ACCESS;
1842 case ACL_TYPE_DEFAULT:
1843 name = XATTR_NAME_ACL_DEFAULT;
1846 return ERR_PTR(-EOPNOTSUPP);
1849 size = inode->i_op->getxattr(fhp->fh_dentry, name, NULL, 0);
1852 acl = ERR_PTR(size);
1854 } else if (size > 0) {
1855 value = kmalloc(size, GFP_KERNEL);
1857 acl = ERR_PTR(-ENOMEM);
1860 size = inode->i_op->getxattr(fhp->fh_dentry, name, value, size);
1862 acl = ERR_PTR(size);
1866 acl = posix_acl_from_xattr(value, size);
1872 #endif /* CONFIG_NFSD_ACL */
1874 #ifdef CONFIG_NFSD_ACL
1876 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1878 struct inode *inode = fhp->fh_dentry->d_inode;
1884 if (!IS_POSIXACL(inode) || !inode->i_op ||
1885 !inode->i_op->setxattr || !inode->i_op->removexattr)
1888 case ACL_TYPE_ACCESS:
1889 name = XATTR_NAME_ACL_ACCESS;
1891 case ACL_TYPE_DEFAULT:
1892 name = XATTR_NAME_ACL_DEFAULT;
1898 if (acl && acl->a_count) {
1899 size = xattr_acl_size(acl->a_count);
1900 value = kmalloc(size, GFP_KERNEL);
1903 size = posix_acl_to_xattr(acl, value, size);
1911 if (!fhp->fh_locked)
1912 fh_lock(fhp); /* unlocking is done automatically */
1914 error = inode->i_op->setxattr(fhp->fh_dentry, name,
1917 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1920 error = inode->i_op->removexattr(fhp->fh_dentry, name);
1921 if (error == -ENODATA)
1930 #endif /* CONFIG_NFSD_ACL */