2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/pagemap.h>
15 #include <linux/uio.h>
16 #include <linux/blkdev.h>
18 #include <linux/smp_lock.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/ext2_fs.h>
22 #include <linux/crc32.h>
23 #include <linux/lm_interface.h>
24 #include <linux/writeback.h>
25 #include <asm/uaccess.h>
46 * Most fields left uninitialised to catch anybody who tries to
47 * use them. f_flags set to prevent file_accessed() from touching
48 * any other part of this. Its use is purely as a flag so that we
49 * know (in readpage()) whether or not do to locking.
51 struct file gfs2_internal_file_sentinel = {
52 .f_flags = O_NOATIME|O_RDONLY,
55 static int gfs2_read_actor(read_descriptor_t *desc, struct page *page,
56 unsigned long offset, unsigned long size)
59 unsigned long count = desc->count;
65 memcpy(desc->arg.data, kaddr + offset, size);
68 desc->count = count - size;
69 desc->written += size;
70 desc->arg.buf += size;
74 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
75 char *buf, loff_t *pos, unsigned size)
77 struct inode *inode = &ip->i_inode;
78 read_descriptor_t desc;
83 do_generic_mapping_read(inode->i_mapping, ra_state,
84 &gfs2_internal_file_sentinel, pos, &desc,
86 return desc.written ? desc.written : desc.error;
90 * gfs2_llseek - seek to a location in a file
93 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
95 * SEEK_END requires the glock for the file because it references the
98 * Returns: The new offset, or errno
101 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
103 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
104 struct gfs2_holder i_gh;
108 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
111 error = remote_llseek(file, offset, origin);
112 gfs2_glock_dq_uninit(&i_gh);
115 error = remote_llseek(file, offset, origin);
121 * gfs2_readdir - Read directory entries from a directory
122 * @file: The directory to read from
123 * @dirent: Buffer for dirents
124 * @filldir: Function used to do the copying
129 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
131 struct inode *dir = file->f_mapping->host;
132 struct gfs2_inode *dip = GFS2_I(dir);
133 struct gfs2_holder d_gh;
134 u64 offset = file->f_pos;
137 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh);
138 error = gfs2_glock_nq_atime(&d_gh);
140 gfs2_holder_uninit(&d_gh);
144 error = gfs2_dir_read(dir, &offset, dirent, filldir);
146 gfs2_glock_dq_uninit(&d_gh);
148 file->f_pos = offset;
155 * @table: A table of 32 u32 flags
156 * @val: a 32 bit value to convert
158 * This function can be used to convert between fsflags values and
159 * GFS2's own flags values.
161 * Returns: the converted flags
163 static u32 fsflags_cvt(const u32 *table, u32 val)
175 static const u32 fsflags_to_gfs2[32] = {
177 [4] = GFS2_DIF_IMMUTABLE,
178 [5] = GFS2_DIF_APPENDONLY,
179 [7] = GFS2_DIF_NOATIME,
180 [12] = GFS2_DIF_EXHASH,
181 [14] = GFS2_DIF_JDATA,
182 [20] = GFS2_DIF_DIRECTIO,
185 static const u32 gfs2_to_fsflags[32] = {
186 [gfs2fl_Sync] = FS_SYNC_FL,
187 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
188 [gfs2fl_AppendOnly] = FS_APPEND_FL,
189 [gfs2fl_NoAtime] = FS_NOATIME_FL,
190 [gfs2fl_ExHash] = FS_INDEX_FL,
191 [gfs2fl_Jdata] = FS_JOURNAL_DATA_FL,
192 [gfs2fl_Directio] = FS_DIRECTIO_FL,
193 [gfs2fl_InheritDirectio] = FS_DIRECTIO_FL,
194 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
197 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
199 struct inode *inode = filp->f_path.dentry->d_inode;
200 struct gfs2_inode *ip = GFS2_I(inode);
201 struct gfs2_holder gh;
205 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
206 error = gfs2_glock_nq_atime(&gh);
210 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_di.di_flags);
211 if (put_user(fsflags, ptr))
214 gfs2_glock_dq_m(1, &gh);
215 gfs2_holder_uninit(&gh);
219 void gfs2_set_inode_flags(struct inode *inode)
221 struct gfs2_inode *ip = GFS2_I(inode);
222 struct gfs2_dinode_host *di = &ip->i_di;
223 unsigned int flags = inode->i_flags;
225 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
226 if (di->di_flags & GFS2_DIF_IMMUTABLE)
227 flags |= S_IMMUTABLE;
228 if (di->di_flags & GFS2_DIF_APPENDONLY)
230 if (di->di_flags & GFS2_DIF_NOATIME)
232 if (di->di_flags & GFS2_DIF_SYNC)
234 inode->i_flags = flags;
237 /* Flags that can be set by user space */
238 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
240 GFS2_DIF_IMMUTABLE| \
241 GFS2_DIF_APPENDONLY| \
245 GFS2_DIF_INHERIT_DIRECTIO| \
246 GFS2_DIF_INHERIT_JDATA)
249 * gfs2_set_flags - set flags on an inode
251 * @flags: The flags to set
252 * @mask: Indicates which flags are valid
255 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
257 struct inode *inode = filp->f_path.dentry->d_inode;
258 struct gfs2_inode *ip = GFS2_I(inode);
259 struct gfs2_sbd *sdp = GFS2_SB(inode);
260 struct buffer_head *bh;
261 struct gfs2_holder gh;
263 u32 new_flags, flags;
265 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
269 flags = ip->i_di.di_flags;
270 new_flags = (flags & ~mask) | (reqflags & mask);
271 if ((new_flags ^ flags) == 0)
274 if (S_ISDIR(inode->i_mode)) {
275 if ((new_flags ^ flags) & GFS2_DIF_JDATA)
276 new_flags ^= (GFS2_DIF_JDATA|GFS2_DIF_INHERIT_JDATA);
277 if ((new_flags ^ flags) & GFS2_DIF_DIRECTIO)
278 new_flags ^= (GFS2_DIF_DIRECTIO|GFS2_DIF_INHERIT_DIRECTIO);
282 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
286 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
288 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
290 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
291 !capable(CAP_LINUX_IMMUTABLE))
293 if (!IS_IMMUTABLE(inode)) {
294 error = permission(inode, MAY_WRITE, NULL);
299 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
302 error = gfs2_meta_inode_buffer(ip, &bh);
305 gfs2_trans_add_bh(ip->i_gl, bh, 1);
306 ip->i_di.di_flags = new_flags;
307 gfs2_dinode_out(ip, bh->b_data);
309 gfs2_set_inode_flags(inode);
313 gfs2_glock_dq_uninit(&gh);
317 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
319 u32 fsflags, gfsflags;
320 if (get_user(fsflags, ptr))
322 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
323 return do_gfs2_set_flags(filp, gfsflags, ~0);
326 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
329 case FS_IOC_GETFLAGS:
330 return gfs2_get_flags(filp, (u32 __user *)arg);
331 case FS_IOC_SETFLAGS:
332 return gfs2_set_flags(filp, (u32 __user *)arg);
340 * @file: The file to map
341 * @vma: The VMA which described the mapping
343 * Returns: 0 or error code
346 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
348 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
349 struct gfs2_holder i_gh;
352 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
353 error = gfs2_glock_nq_atime(&i_gh);
355 gfs2_holder_uninit(&i_gh);
359 /* This is VM_MAYWRITE instead of VM_WRITE because a call
360 to mprotect() can turn on VM_WRITE later. */
362 if ((vma->vm_flags & (VM_MAYSHARE | VM_MAYWRITE)) ==
363 (VM_MAYSHARE | VM_MAYWRITE))
364 vma->vm_ops = &gfs2_vm_ops_sharewrite;
366 vma->vm_ops = &gfs2_vm_ops_private;
368 gfs2_glock_dq_uninit(&i_gh);
374 * gfs2_open - open a file
375 * @inode: the inode to open
376 * @file: the struct file for this opening
381 static int gfs2_open(struct inode *inode, struct file *file)
383 struct gfs2_inode *ip = GFS2_I(inode);
384 struct gfs2_holder i_gh;
385 struct gfs2_file *fp;
388 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
392 mutex_init(&fp->f_fl_mutex);
394 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
395 file->private_data = fp;
397 if (S_ISREG(ip->i_inode.i_mode)) {
398 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
403 if (!(file->f_flags & O_LARGEFILE) &&
404 ip->i_di.di_size > MAX_NON_LFS) {
409 /* Listen to the Direct I/O flag */
411 if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO)
412 file->f_flags |= O_DIRECT;
414 gfs2_glock_dq_uninit(&i_gh);
420 gfs2_glock_dq_uninit(&i_gh);
422 file->private_data = NULL;
428 * gfs2_close - called to close a struct file
429 * @inode: the inode the struct file belongs to
430 * @file: the struct file being closed
435 static int gfs2_close(struct inode *inode, struct file *file)
437 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
438 struct gfs2_file *fp;
440 fp = file->private_data;
441 file->private_data = NULL;
443 if (gfs2_assert_warn(sdp, fp))
452 * gfs2_fsync - sync the dirty data for a file (across the cluster)
453 * @file: the file that points to the dentry (we ignore this)
454 * @dentry: the dentry that points to the inode to sync
456 * The VFS will flush "normal" data for us. We only need to worry
457 * about metadata here. For journaled data, we just do a log flush
458 * as we can't avoid it. Otherwise we can just bale out if datasync
459 * is set. For stuffed inodes we must flush the log in order to
460 * ensure that all data is on disk.
462 * The call to write_inode_now() is there to write back metadata and
463 * the inode itself. It does also try and write the data, but thats
464 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
470 static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
472 struct inode *inode = dentry->d_inode;
473 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
476 if (gfs2_is_jdata(GFS2_I(inode))) {
477 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
481 if (sync_state != 0) {
483 ret = write_inode_now(inode, 0);
485 if (gfs2_is_stuffed(GFS2_I(inode)))
486 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
493 * gfs2_lock - acquire/release a posix lock on a file
494 * @file: the file pointer
495 * @cmd: either modify or retrieve lock state, possibly wait
496 * @fl: type and range of lock
501 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
503 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
504 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
505 struct lm_lockname name =
506 { .ln_number = ip->i_num.no_addr,
507 .ln_type = LM_TYPE_PLOCK };
509 if (!(fl->fl_flags & FL_POSIX))
511 if ((ip->i_inode.i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
514 if (sdp->sd_args.ar_localflocks) {
516 struct file_lock tmp;
518 ret = posix_test_lock(file, fl, &tmp);
519 fl->fl_type = F_UNLCK;
521 memcpy(fl, &tmp, sizeof(struct file_lock));
524 return posix_lock_file_wait(file, fl);
529 return gfs2_lm_plock_get(sdp, &name, file, fl);
530 else if (fl->fl_type == F_UNLCK)
531 return gfs2_lm_punlock(sdp, &name, file, fl);
533 return gfs2_lm_plock(sdp, &name, file, cmd, fl);
536 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
538 struct gfs2_file *fp = file->private_data;
539 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
540 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
541 struct gfs2_glock *gl;
546 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
547 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
549 mutex_lock(&fp->f_fl_mutex);
553 if (fl_gh->gh_state == state)
556 flock_lock_file_wait(file,
557 &(struct file_lock){.fl_type = F_UNLCK});
558 gfs2_glock_dq_uninit(fl_gh);
560 error = gfs2_glock_get(GFS2_SB(&ip->i_inode),
561 ip->i_num.no_addr, &gfs2_flock_glops,
567 gfs2_holder_init(gl, state, flags, fl_gh);
570 error = gfs2_glock_nq(fl_gh);
572 gfs2_holder_uninit(fl_gh);
573 if (error == GLR_TRYFAILED)
576 error = flock_lock_file_wait(file, fl);
577 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
581 mutex_unlock(&fp->f_fl_mutex);
585 static void do_unflock(struct file *file, struct file_lock *fl)
587 struct gfs2_file *fp = file->private_data;
588 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
590 mutex_lock(&fp->f_fl_mutex);
591 flock_lock_file_wait(file, fl);
593 gfs2_glock_dq_uninit(fl_gh);
594 mutex_unlock(&fp->f_fl_mutex);
598 * gfs2_flock - acquire/release a flock lock on a file
599 * @file: the file pointer
600 * @cmd: either modify or retrieve lock state, possibly wait
601 * @fl: type and range of lock
606 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
608 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
609 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
611 if (!(fl->fl_flags & FL_FLOCK))
613 if ((ip->i_inode.i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
616 if (sdp->sd_args.ar_localflocks)
617 return flock_lock_file_wait(file, fl);
619 if (fl->fl_type == F_UNLCK) {
620 do_unflock(file, fl);
623 return do_flock(file, cmd, fl);
627 const struct file_operations gfs2_file_fops = {
628 .llseek = gfs2_llseek,
629 .read = do_sync_read,
630 .aio_read = generic_file_aio_read,
631 .write = do_sync_write,
632 .aio_write = generic_file_aio_write,
633 .unlocked_ioctl = gfs2_ioctl,
636 .release = gfs2_close,
639 .sendfile = generic_file_sendfile,
641 .splice_read = generic_file_splice_read,
642 .splice_write = generic_file_splice_write,
645 const struct file_operations gfs2_dir_fops = {
646 .readdir = gfs2_readdir,
647 .unlocked_ioctl = gfs2_ioctl,
649 .release = gfs2_close,