2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
31 #include <linux/security.h>
35 #endif /* CONFIG_NFSD_V3 */
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms *p_next;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
68 } ____cacheline_aligned_in_smp;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
87 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
91 err = follow_down(&path);
95 exp2 = rqst_exp_get_by_name(rqstp, &path);
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
110 if (nfsd_v4client(rqstp) ||
111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
112 /* successfully crossed mount point */
114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
120 path.dentry = dentry;
130 static void follow_to_parent(struct path *path)
134 while (path->dentry == path->mnt->mnt_root && follow_up(path))
136 dp = dget_parent(path->dentry);
141 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
143 struct svc_export *exp2;
144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
145 .dentry = dget(dparent)};
147 follow_to_parent(&path);
149 exp2 = rqst_exp_parent(rqstp, &path);
150 if (PTR_ERR(exp2) == -ENOENT) {
151 *dentryp = dget(dparent);
152 } else if (IS_ERR(exp2)) {
154 return PTR_ERR(exp2);
156 *dentryp = dget(path.dentry);
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
168 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
170 if (d_mountpoint(dentry))
172 if (nfsd4_is_junction(dentry))
174 if (!(exp->ex_flags & NFSEXP_V4ROOT))
176 return dentry->d_inode != NULL;
180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
181 const char *name, unsigned int len,
182 struct svc_export **exp_ret, struct dentry **dentry_ret)
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
191 dparent = fhp->fh_dentry;
192 exp = fhp->fh_export;
195 /* Lookup the name, but don't follow links */
196 if (isdotent(name, len)) {
198 dentry = dget(dparent);
199 else if (dparent != exp->ex_path.dentry)
200 dentry = dget_parent(dparent);
201 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
202 dentry = dget(dparent); /* .. == . just like at / */
204 /* checking mountpoint crossing is very different when stepping up */
205 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
211 dentry = lookup_one_len(name, dparent, len);
212 host_err = PTR_ERR(dentry);
216 * check if we have crossed a mount point ...
218 if (nfsd_mountpoint(dentry, exp)) {
219 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
225 *dentry_ret = dentry;
231 return nfserrno(host_err);
235 * Look up one component of a pathname.
236 * N.B. After this call _both_ fhp and resfh need an fh_put
238 * If the lookup would cross a mountpoint, and the mounted filesystem
239 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
240 * accepted as it stands and the mounted directory is
241 * returned. Otherwise the covered directory is returned.
242 * NOTE: this mountpoint crossing is not supported properly by all
243 * clients and is explicitly disallowed for NFSv3
244 * NeilBrown <neilb@cse.unsw.edu.au>
247 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
248 unsigned int len, struct svc_fh *resfh)
250 struct svc_export *exp;
251 struct dentry *dentry;
254 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
257 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
260 err = check_nfsd_access(exp, rqstp);
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err = fh_compose(resfh, exp, dentry, fhp);
268 if (!err && !dentry->d_inode)
276 static int nfsd_break_lease(struct inode *inode)
278 if (!S_ISREG(inode->i_mode))
280 return break_lease(inode, O_WRONLY | O_NONBLOCK);
284 * Commit metadata changes to stable storage.
287 commit_metadata(struct svc_fh *fhp)
289 struct inode *inode = fhp->fh_dentry->d_inode;
290 const struct export_operations *export_ops = inode->i_sb->s_export_op;
292 if (!EX_ISSYNC(fhp->fh_export))
295 if (export_ops->commit_metadata)
296 return export_ops->commit_metadata(inode);
297 return sync_inode_metadata(inode, 1);
301 * Go over the attributes and take care of the small differences between
302 * NFS semantics and what Linux expects.
305 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
308 * NFSv2 does not differentiate between "set-[ac]time-to-now"
309 * which only requires access, and "set-[ac]time-to-X" which
310 * requires ownership.
311 * So if it looks like it might be "set both to the same time which
312 * is close to now", and if inode_change_ok fails, then we
313 * convert to "set to now" instead of "set to explicit time"
315 * We only call inode_change_ok as the last test as technically
316 * it is not an interface that we should be using.
318 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
319 #define MAX_TOUCH_TIME_ERROR (30*60)
320 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
321 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
325 * Now just make sure time is in the right ballpark.
326 * Solaris, at least, doesn't seem to care what the time
327 * request is. We require it be within 30 minutes of now.
329 time_t delta = iap->ia_atime.tv_sec - get_seconds();
332 if (delta < MAX_TOUCH_TIME_ERROR &&
333 inode_change_ok(inode, iap) != 0) {
335 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
336 * This will cause notify_change to set these times
339 iap->ia_valid &= ~BOTH_TIME_SET;
343 /* sanitize the mode change */
344 if (iap->ia_valid & ATTR_MODE) {
345 iap->ia_mode &= S_IALLUGO;
346 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
349 /* Revoke setuid/setgid on chown */
350 if (!S_ISDIR(inode->i_mode) &&
351 (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
352 ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) {
353 iap->ia_valid |= ATTR_KILL_PRIV;
354 if (iap->ia_valid & ATTR_MODE) {
355 /* we're setting mode too, just clear the s*id bits */
356 iap->ia_mode &= ~S_ISUID;
357 if (iap->ia_mode & S_IXGRP)
358 iap->ia_mode &= ~S_ISGID;
360 /* set ATTR_KILL_* bits and let VFS handle it */
361 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
367 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
370 struct inode *inode = fhp->fh_dentry->d_inode;
373 if (iap->ia_size < inode->i_size) {
376 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
377 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
382 host_err = get_write_access(inode);
386 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
388 goto out_put_write_access;
391 out_put_write_access:
392 put_write_access(inode);
394 return nfserrno(host_err);
398 * Set various file attributes. After this call fhp needs an fh_put.
401 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
402 int check_guard, time_t guardtime)
404 struct dentry *dentry;
406 int accmode = NFSD_MAY_SATTR;
412 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
413 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
414 if (iap->ia_valid & ATTR_SIZE)
418 err = fh_verify(rqstp, fhp, ftype, accmode);
422 dentry = fhp->fh_dentry;
423 inode = dentry->d_inode;
425 /* Ignore any mode updates on symlinks */
426 if (S_ISLNK(inode->i_mode))
427 iap->ia_valid &= ~ATTR_MODE;
432 nfsd_sanitize_attrs(inode, iap);
435 * The size case is special, it changes the file in addition to the
438 if (iap->ia_valid & ATTR_SIZE) {
439 err = nfsd_get_write_access(rqstp, fhp, iap);
445 iap->ia_valid |= ATTR_CTIME;
447 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
448 err = nfserr_notsync;
449 goto out_put_write_access;
452 host_err = nfsd_break_lease(inode);
454 goto out_put_write_access_nfserror;
457 host_err = notify_change(dentry, iap, NULL);
460 out_put_write_access_nfserror:
461 err = nfserrno(host_err);
462 out_put_write_access:
464 put_write_access(inode);
466 commit_metadata(fhp);
471 #if defined(CONFIG_NFSD_V4)
473 * NFS junction information is stored in an extended attribute.
475 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
478 * nfsd4_is_junction - Test if an object could be an NFS junction
480 * @dentry: object to test
482 * Returns 1 if "dentry" appears to contain NFS junction information.
483 * Otherwise 0 is returned.
485 int nfsd4_is_junction(struct dentry *dentry)
487 struct inode *inode = dentry->d_inode;
491 if (inode->i_mode & S_IXUGO)
493 if (!(inode->i_mode & S_ISVTX))
495 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
499 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
500 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
501 struct xdr_netobj *label)
505 struct dentry *dentry;
507 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
511 dentry = fhp->fh_dentry;
513 mutex_lock(&dentry->d_inode->i_mutex);
514 host_error = security_inode_setsecctx(dentry, label->data, label->len);
515 mutex_unlock(&dentry->d_inode->i_mutex);
516 return nfserrno(host_error);
519 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
520 struct xdr_netobj *label)
522 return nfserr_notsupp;
526 #endif /* defined(CONFIG_NFSD_V4) */
528 #ifdef CONFIG_NFSD_V3
530 * Check server access rights to a file system object
536 static struct accessmap nfs3_regaccess[] = {
537 { NFS3_ACCESS_READ, NFSD_MAY_READ },
538 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
539 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
540 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
545 static struct accessmap nfs3_diraccess[] = {
546 { NFS3_ACCESS_READ, NFSD_MAY_READ },
547 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
548 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
549 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
550 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
555 static struct accessmap nfs3_anyaccess[] = {
556 /* Some clients - Solaris 2.6 at least, make an access call
557 * to the server to check for access for things like /dev/null
558 * (which really, the server doesn't care about). So
559 * We provide simple access checking for them, looking
560 * mainly at mode bits, and we make sure to ignore read-only
563 { NFS3_ACCESS_READ, NFSD_MAY_READ },
564 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
565 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
566 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
572 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
574 struct accessmap *map;
575 struct svc_export *export;
576 struct dentry *dentry;
577 u32 query, result = 0, sresult = 0;
580 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
584 export = fhp->fh_export;
585 dentry = fhp->fh_dentry;
587 if (S_ISREG(dentry->d_inode->i_mode))
588 map = nfs3_regaccess;
589 else if (S_ISDIR(dentry->d_inode->i_mode))
590 map = nfs3_diraccess;
592 map = nfs3_anyaccess;
596 for (; map->access; map++) {
597 if (map->access & query) {
600 sresult |= map->access;
602 err2 = nfsd_permission(rqstp, export, dentry, map->how);
605 result |= map->access;
608 /* the following error codes just mean the access was not allowed,
609 * rather than an error occurred */
613 /* simply don't "or" in the access bit. */
623 *supported = sresult;
628 #endif /* CONFIG_NFSD_V3 */
630 static int nfsd_open_break_lease(struct inode *inode, int access)
634 if (access & NFSD_MAY_NOT_BREAK_LEASE)
636 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
637 return break_lease(inode, mode | O_NONBLOCK);
641 * Open an existing file or directory.
642 * The may_flags argument indicates the type of open (read/write/lock)
643 * and additional flags.
644 * N.B. After this call fhp needs an fh_put
647 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
648 int may_flags, struct file **filp)
652 int flags = O_RDONLY|O_LARGEFILE;
656 validate_process_creds();
659 * If we get here, then the client has already done an "open",
660 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
661 * in case a chmod has now revoked permission.
663 * Arguably we should also allow the owner override for
664 * directories, but we never have and it doesn't seem to have
665 * caused anyone a problem. If we were to change this, note
666 * also that our filldir callbacks would need a variant of
667 * lookup_one_len that doesn't check permissions.
670 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
671 err = fh_verify(rqstp, fhp, type, may_flags);
675 path.mnt = fhp->fh_export->ex_path.mnt;
676 path.dentry = fhp->fh_dentry;
677 inode = path.dentry->d_inode;
679 /* Disallow write access to files with the append-only bit set
680 * or any access when mandatory locking enabled
683 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
686 * We must ignore files (but only files) which might have mandatory
687 * locks on them because there is no way to know if the accesser has
690 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
696 host_err = nfsd_open_break_lease(inode, may_flags);
697 if (host_err) /* NOMEM or WOULDBLOCK */
700 if (may_flags & NFSD_MAY_WRITE) {
701 if (may_flags & NFSD_MAY_READ)
702 flags = O_RDWR|O_LARGEFILE;
704 flags = O_WRONLY|O_LARGEFILE;
706 *filp = dentry_open(&path, flags, current_cred());
708 host_err = PTR_ERR(*filp);
711 host_err = ima_file_check(*filp, may_flags);
713 if (may_flags & NFSD_MAY_64BIT_COOKIE)
714 (*filp)->f_mode |= FMODE_64BITHASH;
716 (*filp)->f_mode |= FMODE_32BITHASH;
720 err = nfserrno(host_err);
722 validate_process_creds();
730 nfsd_close(struct file *filp)
736 * Obtain the readahead parameters for the file
737 * specified by (dev, ino).
740 static inline struct raparms *
741 nfsd_get_raparms(dev_t dev, ino_t ino)
743 struct raparms *ra, **rap, **frap = NULL;
746 struct raparm_hbucket *rab;
748 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
749 rab = &raparm_hash[hash];
751 spin_lock(&rab->pb_lock);
752 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
753 if (ra->p_ino == ino && ra->p_dev == dev)
756 if (ra->p_count == 0)
759 depth = nfsdstats.ra_size;
761 spin_unlock(&rab->pb_lock);
771 if (rap != &rab->pb_head) {
773 ra->p_next = rab->pb_head;
777 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
778 spin_unlock(&rab->pb_lock);
783 * Grab and keep cached pages associated with a file in the svc_rqst
784 * so that they can be passed to the network sendmsg/sendpage routines
785 * directly. They will be released after the sending has completed.
788 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
789 struct splice_desc *sd)
791 struct svc_rqst *rqstp = sd->u.data;
792 struct page **pp = rqstp->rq_next_page;
793 struct page *page = buf->page;
798 if (rqstp->rq_res.page_len == 0) {
800 put_page(*rqstp->rq_next_page);
801 *(rqstp->rq_next_page++) = page;
802 rqstp->rq_res.page_base = buf->offset;
803 rqstp->rq_res.page_len = size;
804 } else if (page != pp[-1]) {
806 if (*rqstp->rq_next_page)
807 put_page(*rqstp->rq_next_page);
808 *(rqstp->rq_next_page++) = page;
809 rqstp->rq_res.page_len += size;
811 rqstp->rq_res.page_len += size;
816 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
817 struct splice_desc *sd)
819 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
823 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
824 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
832 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
833 struct splice_desc sd = {
840 rqstp->rq_next_page = rqstp->rq_respages + 1;
841 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
845 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
850 nfsdstats.io_read += host_err;
853 fsnotify_access(file);
855 err = nfserrno(host_err);
859 static void kill_suid(struct dentry *dentry)
862 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
864 mutex_lock(&dentry->d_inode->i_mutex);
866 * Note we call this on write, so notify_change will not
867 * encounter any conflicting delegations:
869 notify_change(dentry, &ia, NULL);
870 mutex_unlock(&dentry->d_inode->i_mutex);
874 * Gathered writes: If another process is currently writing to the file,
875 * there's a high chance this is another nfsd (triggered by a bulk write
876 * from a client's biod). Rather than syncing the file with each write
877 * request, we sleep for 10 msec.
879 * I don't know if this roughly approximates C. Juszak's idea of
880 * gathered writes, but it's a nice and simple solution (IMHO), and it
883 * Note: we do this only in the NFSv2 case, since v3 and higher have a
884 * better tool (separate unstable writes and commits) for solving this
887 static int wait_for_concurrent_writes(struct file *file)
889 struct inode *inode = file_inode(file);
890 static ino_t last_ino;
891 static dev_t last_dev;
894 if (atomic_read(&inode->i_writecount) > 1
895 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
896 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
898 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
901 if (inode->i_state & I_DIRTY) {
902 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
903 err = vfs_fsync(file, 0);
905 last_ino = inode->i_ino;
906 last_dev = inode->i_sb->s_dev;
911 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
912 loff_t offset, struct kvec *vec, int vlen,
913 unsigned long *cnt, int *stablep)
915 struct svc_export *exp;
916 struct dentry *dentry;
921 int stable = *stablep;
925 dentry = file->f_path.dentry;
926 inode = dentry->d_inode;
927 exp = fhp->fh_export;
929 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
934 /* Write the data. */
935 oldfs = get_fs(); set_fs(KERNEL_DS);
936 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
941 nfsdstats.io_write += host_err;
942 fsnotify_modify(file);
944 /* clear setuid/setgid flag after write */
945 if (inode->i_mode & (S_ISUID | S_ISGID))
950 host_err = wait_for_concurrent_writes(file);
952 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
956 dprintk("nfsd: write complete host_err=%d\n", host_err);
960 err = nfserrno(host_err);
965 * Read data from a file. count must contain the requested read count
966 * on entry. On return, *count contains the number of bytes actually read.
967 * N.B. After this call fhp needs an fh_put
969 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
970 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
977 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
981 inode = file_inode(file);
983 /* Get readahead parameters */
984 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
987 file->f_ra = ra->p_ra;
989 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
991 /* Write back readahead params */
993 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
994 spin_lock(&rab->pb_lock);
995 ra->p_ra = file->f_ra;
998 spin_unlock(&rab->pb_lock);
1005 /* As above, but use the provided file descriptor. */
1007 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1008 loff_t offset, struct kvec *vec, int vlen,
1009 unsigned long *count)
1014 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1015 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1018 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1019 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1020 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1026 * Write data to a file.
1027 * The stable flag requests synchronous writes.
1028 * N.B. After this call fhp needs an fh_put
1031 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1032 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1038 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1039 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1042 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1045 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1050 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1058 #ifdef CONFIG_NFSD_V3
1060 * Commit all pending writes to stable storage.
1062 * Note: we only guarantee that data that lies within the range specified
1063 * by the 'offset' and 'count' parameters will be synced.
1065 * Unfortunately we cannot lock the file to make sure we return full WCC
1066 * data to the client, as locking happens lower down in the filesystem.
1069 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1070 loff_t offset, unsigned long count)
1073 loff_t end = LLONG_MAX;
1074 __be32 err = nfserr_inval;
1079 end = offset + (loff_t)count - 1;
1084 err = nfsd_open(rqstp, fhp, S_IFREG,
1085 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1088 if (EX_ISSYNC(fhp->fh_export)) {
1089 int err2 = vfs_fsync_range(file, offset, end, 0);
1091 if (err2 != -EINVAL)
1092 err = nfserrno(err2);
1094 err = nfserr_notsupp;
1101 #endif /* CONFIG_NFSD_V3 */
1104 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1108 * Mode has already been set earlier in create:
1110 iap->ia_valid &= ~ATTR_MODE;
1112 * Setting uid/gid works only for root. Irix appears to
1113 * send along the gid on create when it tries to implement
1114 * setgid directories via NFS:
1116 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1117 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1119 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1123 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1124 * setting size to 0 may fail for some specific file systems by the permission
1125 * checking which requires WRITE permission but the mode is 000.
1126 * we ignore the resizing(to 0) on the just new created file, since the size is
1127 * 0 after file created.
1129 * call this only after vfs_create() is called.
1132 nfsd_check_ignore_resizing(struct iattr *iap)
1134 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1135 iap->ia_valid &= ~ATTR_SIZE;
1139 * Create a file (regular, directory, device, fifo); UNIX sockets
1140 * not yet implemented.
1141 * If the response fh has been verified, the parent directory should
1142 * already be locked. Note that the parent directory is left locked.
1144 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1147 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1148 char *fname, int flen, struct iattr *iap,
1149 int type, dev_t rdev, struct svc_fh *resfhp)
1151 struct dentry *dentry, *dchild = NULL;
1161 if (isdotent(fname, flen))
1164 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1168 dentry = fhp->fh_dentry;
1169 dirp = dentry->d_inode;
1171 err = nfserr_notdir;
1172 if (!dirp->i_op->lookup)
1175 * Check whether the response file handle has been verified yet.
1176 * If it has, the parent directory should already be locked.
1178 if (!resfhp->fh_dentry) {
1179 host_err = fh_want_write(fhp);
1183 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1184 fh_lock_nested(fhp, I_MUTEX_PARENT);
1185 dchild = lookup_one_len(fname, dentry, flen);
1186 host_err = PTR_ERR(dchild);
1189 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1193 /* called from nfsd_proc_create */
1194 dchild = dget(resfhp->fh_dentry);
1195 if (!fhp->fh_locked) {
1196 /* not actually possible */
1198 "nfsd_create: parent %pd2 not locked!\n",
1205 * Make sure the child dentry is still negative ...
1208 if (dchild->d_inode) {
1209 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1214 if (!(iap->ia_valid & ATTR_MODE))
1216 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1219 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1220 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1226 * Get the dir op function pointer.
1232 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1234 nfsd_check_ignore_resizing(iap);
1237 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1243 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1249 err = nfsd_create_setattr(rqstp, resfhp, iap);
1252 * nfsd_setattr already committed the child. Transactional filesystems
1253 * had a chance to commit changes for both parent and child
1254 * simultaneously making the following commit_metadata a noop.
1256 err2 = nfserrno(commit_metadata(fhp));
1260 * Update the file handle to get the new inode info.
1263 err = fh_update(resfhp);
1265 if (dchild && !IS_ERR(dchild))
1270 err = nfserrno(host_err);
1274 #ifdef CONFIG_NFSD_V3
1276 static inline int nfsd_create_is_exclusive(int createmode)
1278 return createmode == NFS3_CREATE_EXCLUSIVE
1279 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1283 * NFSv3 and NFSv4 version of nfsd_create
1286 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1287 char *fname, int flen, struct iattr *iap,
1288 struct svc_fh *resfhp, int createmode, u32 *verifier,
1289 bool *truncp, bool *created)
1291 struct dentry *dentry, *dchild = NULL;
1295 __u32 v_mtime=0, v_atime=0;
1301 if (isdotent(fname, flen))
1303 if (!(iap->ia_valid & ATTR_MODE))
1305 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1309 dentry = fhp->fh_dentry;
1310 dirp = dentry->d_inode;
1312 /* Get all the sanity checks out of the way before
1313 * we lock the parent. */
1314 err = nfserr_notdir;
1315 if (!dirp->i_op->lookup)
1318 host_err = fh_want_write(fhp);
1322 fh_lock_nested(fhp, I_MUTEX_PARENT);
1325 * Compose the response file handle.
1327 dchild = lookup_one_len(fname, dentry, flen);
1328 host_err = PTR_ERR(dchild);
1332 /* If file doesn't exist, check for permissions to create one */
1333 if (!dchild->d_inode) {
1334 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1339 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1343 if (nfsd_create_is_exclusive(createmode)) {
1344 /* solaris7 gets confused (bugid 4218508) if these have
1345 * the high bit set, so just clear the high bits. If this is
1346 * ever changed to use different attrs for storing the
1347 * verifier, then do_open_lookup() will also need to be fixed
1350 v_mtime = verifier[0]&0x7fffffff;
1351 v_atime = verifier[1]&0x7fffffff;
1354 if (dchild->d_inode) {
1357 switch (createmode) {
1358 case NFS3_CREATE_UNCHECKED:
1359 if (! S_ISREG(dchild->d_inode->i_mode))
1362 /* in nfsv4, we need to treat this case a little
1363 * differently. we don't want to truncate the
1364 * file now; this would be wrong if the OPEN
1365 * fails for some other reason. furthermore,
1366 * if the size is nonzero, we should ignore it
1367 * according to spec!
1369 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1372 iap->ia_valid &= ATTR_SIZE;
1376 case NFS3_CREATE_EXCLUSIVE:
1377 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1378 && dchild->d_inode->i_atime.tv_sec == v_atime
1379 && dchild->d_inode->i_size == 0 ) {
1384 case NFS4_CREATE_EXCLUSIVE4_1:
1385 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1386 && dchild->d_inode->i_atime.tv_sec == v_atime
1387 && dchild->d_inode->i_size == 0 ) {
1393 case NFS3_CREATE_GUARDED:
1400 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1408 nfsd_check_ignore_resizing(iap);
1410 if (nfsd_create_is_exclusive(createmode)) {
1411 /* Cram the verifier into atime/mtime */
1412 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1413 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1414 /* XXX someone who knows this better please fix it for nsec */
1415 iap->ia_mtime.tv_sec = v_mtime;
1416 iap->ia_atime.tv_sec = v_atime;
1417 iap->ia_mtime.tv_nsec = 0;
1418 iap->ia_atime.tv_nsec = 0;
1422 err = nfsd_create_setattr(rqstp, resfhp, iap);
1425 * nfsd_setattr already committed the child (and possibly also the parent).
1428 err = nfserrno(commit_metadata(fhp));
1431 * Update the filehandle to get the new inode info.
1434 err = fh_update(resfhp);
1438 if (dchild && !IS_ERR(dchild))
1444 err = nfserrno(host_err);
1447 #endif /* CONFIG_NFSD_V3 */
1450 * Read a symlink. On entry, *lenp must contain the maximum path length that
1451 * fits into the buffer. On return, it contains the true length.
1452 * N.B. After this call fhp needs an fh_put
1455 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1457 struct inode *inode;
1463 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1467 path.mnt = fhp->fh_export->ex_path.mnt;
1468 path.dentry = fhp->fh_dentry;
1469 inode = path.dentry->d_inode;
1472 if (!inode->i_op->readlink)
1476 /* N.B. Why does this call need a get_fs()??
1477 * Remove the set_fs and watch the fireworks:-) --okir
1480 oldfs = get_fs(); set_fs(KERNEL_DS);
1481 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1492 err = nfserrno(host_err);
1497 * Create a symlink and look up its inode
1498 * N.B. After this call _both_ fhp and resfhp need an fh_put
1501 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1502 char *fname, int flen,
1503 char *path, int plen,
1504 struct svc_fh *resfhp,
1507 struct dentry *dentry, *dnew;
1515 if (isdotent(fname, flen))
1518 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1522 host_err = fh_want_write(fhp);
1527 dentry = fhp->fh_dentry;
1528 dnew = lookup_one_len(fname, dentry, flen);
1529 host_err = PTR_ERR(dnew);
1533 if (unlikely(path[plen] != 0)) {
1534 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1535 if (path_alloced == NULL)
1538 strncpy(path_alloced, path, plen);
1539 path_alloced[plen] = 0;
1540 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1541 kfree(path_alloced);
1544 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1545 err = nfserrno(host_err);
1547 err = nfserrno(commit_metadata(fhp));
1552 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1554 if (err==0) err = cerr;
1559 err = nfserrno(host_err);
1565 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1568 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1569 char *name, int len, struct svc_fh *tfhp)
1571 struct dentry *ddir, *dnew, *dold;
1576 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1579 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1583 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1589 if (isdotent(name, len))
1592 host_err = fh_want_write(tfhp);
1594 err = nfserrno(host_err);
1598 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1599 ddir = ffhp->fh_dentry;
1600 dirp = ddir->d_inode;
1602 dnew = lookup_one_len(name, ddir, len);
1603 host_err = PTR_ERR(dnew);
1607 dold = tfhp->fh_dentry;
1612 host_err = nfsd_break_lease(dold->d_inode);
1614 err = nfserrno(host_err);
1617 host_err = vfs_link(dold, dirp, dnew, NULL);
1619 err = nfserrno(commit_metadata(ffhp));
1621 err = nfserrno(commit_metadata(tfhp));
1623 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1626 err = nfserrno(host_err);
1632 fh_drop_write(tfhp);
1637 err = nfserrno(host_err);
1643 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1646 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1647 struct svc_fh *tfhp, char *tname, int tlen)
1649 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1650 struct inode *fdir, *tdir;
1654 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1657 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1661 fdentry = ffhp->fh_dentry;
1662 fdir = fdentry->d_inode;
1664 tdentry = tfhp->fh_dentry;
1665 tdir = tdentry->d_inode;
1668 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1671 host_err = fh_want_write(ffhp);
1673 err = nfserrno(host_err);
1677 /* cannot use fh_lock as we need deadlock protective ordering
1678 * so do it by hand */
1679 trap = lock_rename(tdentry, fdentry);
1680 ffhp->fh_locked = tfhp->fh_locked = 1;
1684 odentry = lookup_one_len(fname, fdentry, flen);
1685 host_err = PTR_ERR(odentry);
1686 if (IS_ERR(odentry))
1690 if (!odentry->d_inode)
1693 if (odentry == trap)
1696 ndentry = lookup_one_len(tname, tdentry, tlen);
1697 host_err = PTR_ERR(ndentry);
1698 if (IS_ERR(ndentry))
1700 host_err = -ENOTEMPTY;
1701 if (ndentry == trap)
1705 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1707 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1710 host_err = nfsd_break_lease(odentry->d_inode);
1713 if (ndentry->d_inode) {
1714 host_err = nfsd_break_lease(ndentry->d_inode);
1718 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL);
1720 host_err = commit_metadata(tfhp);
1722 host_err = commit_metadata(ffhp);
1729 err = nfserrno(host_err);
1731 /* we cannot reply on fh_unlock on the two filehandles,
1732 * as that would do the wrong thing if the two directories
1733 * were the same, so again we do it by hand
1735 fill_post_wcc(ffhp);
1736 fill_post_wcc(tfhp);
1737 unlock_rename(tdentry, fdentry);
1738 ffhp->fh_locked = tfhp->fh_locked = 0;
1739 fh_drop_write(ffhp);
1746 * Unlink a file or directory
1747 * N.B. After this call fhp needs an fh_put
1750 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1751 char *fname, int flen)
1753 struct dentry *dentry, *rdentry;
1759 if (!flen || isdotent(fname, flen))
1761 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1765 host_err = fh_want_write(fhp);
1769 fh_lock_nested(fhp, I_MUTEX_PARENT);
1770 dentry = fhp->fh_dentry;
1771 dirp = dentry->d_inode;
1773 rdentry = lookup_one_len(fname, dentry, flen);
1774 host_err = PTR_ERR(rdentry);
1775 if (IS_ERR(rdentry))
1778 if (!rdentry->d_inode) {
1785 type = rdentry->d_inode->i_mode & S_IFMT;
1787 host_err = nfsd_break_lease(rdentry->d_inode);
1790 if (type != S_IFDIR)
1791 host_err = vfs_unlink(dirp, rdentry, NULL);
1793 host_err = vfs_rmdir(dirp, rdentry);
1795 host_err = commit_metadata(fhp);
1800 err = nfserrno(host_err);
1806 * We do this buffering because we must not call back into the file
1807 * system's ->lookup() method from the filldir callback. That may well
1808 * deadlock a number of file systems.
1810 * This is based heavily on the implementation of same in XFS.
1812 struct buffered_dirent {
1816 unsigned int d_type;
1820 struct readdir_data {
1821 struct dir_context ctx;
1827 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1828 loff_t offset, u64 ino, unsigned int d_type)
1830 struct readdir_data *buf = __buf;
1831 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1832 unsigned int reclen;
1834 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1835 if (buf->used + reclen > PAGE_SIZE) {
1840 de->namlen = namlen;
1841 de->offset = offset;
1843 de->d_type = d_type;
1844 memcpy(de->name, name, namlen);
1845 buf->used += reclen;
1850 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1851 struct readdir_cd *cdp, loff_t *offsetp)
1853 struct buffered_dirent *de;
1857 struct readdir_data buf = {
1858 .ctx.actor = nfsd_buffered_filldir,
1859 .dirent = (void *)__get_free_page(GFP_KERNEL)
1863 return nfserrno(-ENOMEM);
1868 struct inode *dir_inode = file_inode(file);
1869 unsigned int reclen;
1871 cdp->err = nfserr_eof; /* will be cleared on successful read */
1875 host_err = iterate_dir(file, &buf.ctx);
1888 * Various filldir functions may end up calling back into
1889 * lookup_one_len() and the file system's ->lookup() method.
1890 * These expect i_mutex to be held, as it would within readdir.
1892 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1896 de = (struct buffered_dirent *)buf.dirent;
1898 offset = de->offset;
1900 if (func(cdp, de->name, de->namlen, de->offset,
1901 de->ino, de->d_type))
1904 if (cdp->err != nfs_ok)
1907 reclen = ALIGN(sizeof(*de) + de->namlen,
1910 de = (struct buffered_dirent *)((char *)de + reclen);
1912 mutex_unlock(&dir_inode->i_mutex);
1913 if (size > 0) /* We bailed out early */
1916 offset = vfs_llseek(file, 0, SEEK_CUR);
1919 free_page((unsigned long)(buf.dirent));
1922 return nfserrno(host_err);
1929 * Read entries from a directory.
1930 * The NFSv3/4 verifier we ignore for now.
1933 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1934 struct readdir_cd *cdp, filldir_t func)
1938 loff_t offset = *offsetp;
1939 int may_flags = NFSD_MAY_READ;
1941 /* NFSv2 only supports 32 bit cookies */
1942 if (rqstp->rq_vers > 2)
1943 may_flags |= NFSD_MAY_64BIT_COOKIE;
1945 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1949 offset = vfs_llseek(file, offset, SEEK_SET);
1951 err = nfserrno((int)offset);
1955 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1957 if (err == nfserr_eof || err == nfserr_toosmall)
1958 err = nfs_ok; /* can still be found in ->err */
1966 * Get file system stats
1967 * N.B. After this call fhp needs an fh_put
1970 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1974 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1976 struct path path = {
1977 .mnt = fhp->fh_export->ex_path.mnt,
1978 .dentry = fhp->fh_dentry,
1980 if (vfs_statfs(&path, stat))
1986 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1988 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1992 * Check for a user's access permissions to this inode.
1995 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1996 struct dentry *dentry, int acc)
1998 struct inode *inode = dentry->d_inode;
2001 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2004 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2006 (acc & NFSD_MAY_READ)? " read" : "",
2007 (acc & NFSD_MAY_WRITE)? " write" : "",
2008 (acc & NFSD_MAY_EXEC)? " exec" : "",
2009 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2010 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2011 (acc & NFSD_MAY_LOCK)? " lock" : "",
2012 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2014 IS_IMMUTABLE(inode)? " immut" : "",
2015 IS_APPEND(inode)? " append" : "",
2016 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2017 dprintk(" owner %d/%d user %d/%d\n",
2018 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2021 /* Normally we reject any write/sattr etc access on a read-only file
2022 * system. But if it is IRIX doing check on write-access for a
2023 * device special file, we ignore rofs.
2025 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2026 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2027 if (exp_rdonly(rqstp, exp) ||
2028 __mnt_is_readonly(exp->ex_path.mnt))
2030 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2033 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2036 if (acc & NFSD_MAY_LOCK) {
2037 /* If we cannot rely on authentication in NLM requests,
2038 * just allow locks, otherwise require read permission, or
2041 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2044 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2047 * The file owner always gets access permission for accesses that
2048 * would normally be checked at open time. This is to make
2049 * file access work even when the client has done a fchmod(fd, 0).
2051 * However, `cp foo bar' should fail nevertheless when bar is
2052 * readonly. A sensible way to do this might be to reject all
2053 * attempts to truncate a read-only file, because a creat() call
2054 * always implies file truncation.
2055 * ... but this isn't really fair. A process may reasonably call
2056 * ftruncate on an open file descriptor on a file with perm 000.
2057 * We must trust the client to do permission checking - using "ACCESS"
2060 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2061 uid_eq(inode->i_uid, current_fsuid()))
2064 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2065 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2067 /* Allow read access to binaries even when mode 111 */
2068 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2069 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2070 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2071 err = inode_permission(inode, MAY_EXEC);
2073 return err? nfserrno(err) : 0;
2077 nfsd_racache_shutdown(void)
2079 struct raparms *raparm, *last_raparm;
2082 dprintk("nfsd: freeing readahead buffers.\n");
2084 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2085 raparm = raparm_hash[i].pb_head;
2087 last_raparm = raparm;
2088 raparm = raparm->p_next;
2091 raparm_hash[i].pb_head = NULL;
2095 * Initialize readahead param cache
2098 nfsd_racache_init(int cache_size)
2103 struct raparms **raparm = NULL;
2106 if (raparm_hash[0].pb_head)
2108 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2111 cache_size = nperbucket * RAPARM_HASH_SIZE;
2113 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2115 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2116 spin_lock_init(&raparm_hash[i].pb_lock);
2118 raparm = &raparm_hash[i].pb_head;
2119 for (j = 0; j < nperbucket; j++) {
2120 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2123 raparm = &(*raparm)->p_next;
2128 nfsdstats.ra_size = cache_size;
2132 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2133 nfsd_racache_shutdown();