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>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/ext2_fs.h>
21 #include <linux/crc32.h>
22 #include <linux/lm_interface.h>
23 #include <linux/writeback.h>
24 #include <asm/uaccess.h>
40 #include "ops_address.h"
43 * gfs2_llseek - seek to a location in a file
46 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
48 * SEEK_END requires the glock for the file because it references the
51 * Returns: The new offset, or errno
54 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
56 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
57 struct gfs2_holder i_gh;
61 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
64 error = remote_llseek(file, offset, origin);
65 gfs2_glock_dq_uninit(&i_gh);
68 error = remote_llseek(file, offset, origin);
74 * gfs2_readdir - Read directory entries from a directory
75 * @file: The directory to read from
76 * @dirent: Buffer for dirents
77 * @filldir: Function used to do the copying
82 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
84 struct inode *dir = file->f_mapping->host;
85 struct gfs2_inode *dip = GFS2_I(dir);
86 struct gfs2_holder d_gh;
87 u64 offset = file->f_pos;
90 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh);
91 error = gfs2_glock_nq_atime(&d_gh);
93 gfs2_holder_uninit(&d_gh);
97 error = gfs2_dir_read(dir, &offset, dirent, filldir);
99 gfs2_glock_dq_uninit(&d_gh);
101 file->f_pos = offset;
108 * @table: A table of 32 u32 flags
109 * @val: a 32 bit value to convert
111 * This function can be used to convert between fsflags values and
112 * GFS2's own flags values.
114 * Returns: the converted flags
116 static u32 fsflags_cvt(const u32 *table, u32 val)
128 static const u32 fsflags_to_gfs2[32] = {
130 [4] = GFS2_DIF_IMMUTABLE,
131 [5] = GFS2_DIF_APPENDONLY,
132 [7] = GFS2_DIF_NOATIME,
133 [12] = GFS2_DIF_EXHASH,
134 [14] = GFS2_DIF_INHERIT_JDATA,
135 [20] = GFS2_DIF_INHERIT_DIRECTIO,
138 static const u32 gfs2_to_fsflags[32] = {
139 [gfs2fl_Sync] = FS_SYNC_FL,
140 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
141 [gfs2fl_AppendOnly] = FS_APPEND_FL,
142 [gfs2fl_NoAtime] = FS_NOATIME_FL,
143 [gfs2fl_ExHash] = FS_INDEX_FL,
144 [gfs2fl_InheritDirectio] = FS_DIRECTIO_FL,
145 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
148 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
150 struct inode *inode = filp->f_path.dentry->d_inode;
151 struct gfs2_inode *ip = GFS2_I(inode);
152 struct gfs2_holder gh;
156 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
157 error = gfs2_glock_nq_atime(&gh);
161 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_di.di_flags);
162 if (!S_ISDIR(inode->i_mode)) {
163 if (ip->i_di.di_flags & GFS2_DIF_JDATA)
164 fsflags |= FS_JOURNAL_DATA_FL;
165 if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO)
166 fsflags |= FS_DIRECTIO_FL;
168 if (put_user(fsflags, ptr))
172 gfs2_holder_uninit(&gh);
176 void gfs2_set_inode_flags(struct inode *inode)
178 struct gfs2_inode *ip = GFS2_I(inode);
179 struct gfs2_dinode_host *di = &ip->i_di;
180 unsigned int flags = inode->i_flags;
182 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
183 if (di->di_flags & GFS2_DIF_IMMUTABLE)
184 flags |= S_IMMUTABLE;
185 if (di->di_flags & GFS2_DIF_APPENDONLY)
187 if (di->di_flags & GFS2_DIF_NOATIME)
189 if (di->di_flags & GFS2_DIF_SYNC)
191 inode->i_flags = flags;
194 /* Flags that can be set by user space */
195 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
197 GFS2_DIF_IMMUTABLE| \
198 GFS2_DIF_APPENDONLY| \
202 GFS2_DIF_INHERIT_DIRECTIO| \
203 GFS2_DIF_INHERIT_JDATA)
206 * gfs2_set_flags - set flags on an inode
208 * @flags: The flags to set
209 * @mask: Indicates which flags are valid
212 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
214 struct inode *inode = filp->f_path.dentry->d_inode;
215 struct gfs2_inode *ip = GFS2_I(inode);
216 struct gfs2_sbd *sdp = GFS2_SB(inode);
217 struct buffer_head *bh;
218 struct gfs2_holder gh;
220 u32 new_flags, flags;
222 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
226 flags = ip->i_di.di_flags;
227 new_flags = (flags & ~mask) | (reqflags & mask);
228 if ((new_flags ^ flags) == 0)
232 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
236 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
238 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
240 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
241 !capable(CAP_LINUX_IMMUTABLE))
243 if (!IS_IMMUTABLE(inode)) {
244 error = permission(inode, MAY_WRITE, NULL);
248 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
249 if (flags & GFS2_DIF_JDATA)
250 gfs2_log_flush(sdp, ip->i_gl);
251 error = filemap_fdatawrite(inode->i_mapping);
254 error = filemap_fdatawait(inode->i_mapping);
258 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
261 error = gfs2_meta_inode_buffer(ip, &bh);
264 gfs2_trans_add_bh(ip->i_gl, bh, 1);
265 ip->i_di.di_flags = new_flags;
266 gfs2_dinode_out(ip, bh->b_data);
268 gfs2_set_inode_flags(inode);
269 gfs2_set_aops(inode);
273 gfs2_glock_dq_uninit(&gh);
277 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
279 struct inode *inode = filp->f_path.dentry->d_inode;
280 u32 fsflags, gfsflags;
281 if (get_user(fsflags, ptr))
283 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
284 if (!S_ISDIR(inode->i_mode)) {
285 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
286 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
287 if (gfsflags & GFS2_DIF_INHERIT_DIRECTIO)
288 gfsflags ^= (GFS2_DIF_DIRECTIO | GFS2_DIF_INHERIT_DIRECTIO);
289 return do_gfs2_set_flags(filp, gfsflags, ~0);
291 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
294 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
297 case FS_IOC_GETFLAGS:
298 return gfs2_get_flags(filp, (u32 __user *)arg);
299 case FS_IOC_SETFLAGS:
300 return gfs2_set_flags(filp, (u32 __user *)arg);
306 * gfs2_allocate_page_backing - Use bmap to allocate blocks
307 * @page: The (locked) page to allocate backing for
309 * We try to allocate all the blocks required for the page in
310 * one go. This might fail for various reasons, so we keep
311 * trying until all the blocks to back this page are allocated.
312 * If some of the blocks are already allocated, thats ok too.
315 static int gfs2_allocate_page_backing(struct page *page)
317 struct inode *inode = page->mapping->host;
318 struct buffer_head bh;
319 unsigned long size = PAGE_CACHE_SIZE;
320 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
325 gfs2_block_map(inode, lblock, &bh, 1);
326 if (!buffer_mapped(&bh))
329 lblock += (bh.b_size >> inode->i_blkbits);
335 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
336 * @vma: The virtual memory area
337 * @page: The page which is about to become writable
339 * When the page becomes writable, we need to ensure that we have
340 * blocks allocated on disk to back that page.
343 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page)
345 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
346 struct gfs2_inode *ip = GFS2_I(inode);
347 struct gfs2_sbd *sdp = GFS2_SB(inode);
348 unsigned long last_index;
349 u64 pos = page->index << (PAGE_CACHE_SIZE - inode->i_blkbits);
350 unsigned int data_blocks, ind_blocks, rblocks;
351 int alloc_required = 0;
352 struct gfs2_holder gh;
353 struct gfs2_alloc *al;
356 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &gh);
357 ret = gfs2_glock_nq_atime(&gh);
361 set_bit(GIF_SW_PAGED, &ip->i_flags);
362 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
363 ret = gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE, &alloc_required);
364 if (ret || !alloc_required)
367 al = gfs2_alloc_get(ip);
371 ret = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
374 ret = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
376 goto out_quota_unlock;
377 al->al_requested = data_blocks + ind_blocks;
378 ret = gfs2_inplace_reserve(ip);
380 goto out_quota_unlock;
382 rblocks = RES_DINODE + ind_blocks;
383 if (gfs2_is_jdata(ip))
384 rblocks += data_blocks ? data_blocks : 1;
385 if (ind_blocks || data_blocks)
386 rblocks += RES_STATFS + RES_QUOTA;
387 ret = gfs2_trans_begin(sdp, rblocks, 0);
393 last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
394 if (page->index > last_index)
395 goto out_unlock_page;
397 if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
398 goto out_unlock_page;
399 if (gfs2_is_stuffed(ip)) {
400 ret = gfs2_unstuff_dinode(ip, page);
402 goto out_unlock_page;
404 ret = gfs2_allocate_page_backing(page);
410 gfs2_inplace_release(ip);
412 gfs2_quota_unlock(ip);
418 gfs2_holder_uninit(&gh);
422 static struct vm_operations_struct gfs2_vm_ops = {
423 .fault = filemap_fault,
424 .page_mkwrite = gfs2_page_mkwrite,
430 * @file: The file to map
431 * @vma: The VMA which described the mapping
433 * Returns: 0 or error code
436 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
438 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
439 struct gfs2_holder i_gh;
442 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
443 error = gfs2_glock_nq_atime(&i_gh);
445 gfs2_holder_uninit(&i_gh);
449 vma->vm_ops = &gfs2_vm_ops;
451 gfs2_glock_dq_uninit(&i_gh);
457 * gfs2_open - open a file
458 * @inode: the inode to open
459 * @file: the struct file for this opening
464 static int gfs2_open(struct inode *inode, struct file *file)
466 struct gfs2_inode *ip = GFS2_I(inode);
467 struct gfs2_holder i_gh;
468 struct gfs2_file *fp;
471 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
475 mutex_init(&fp->f_fl_mutex);
477 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
478 file->private_data = fp;
480 if (S_ISREG(ip->i_inode.i_mode)) {
481 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
486 if (!(file->f_flags & O_LARGEFILE) &&
487 ip->i_di.di_size > MAX_NON_LFS) {
492 /* Listen to the Direct I/O flag */
494 if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO)
495 file->f_flags |= O_DIRECT;
497 gfs2_glock_dq_uninit(&i_gh);
503 gfs2_glock_dq_uninit(&i_gh);
505 file->private_data = NULL;
511 * gfs2_close - called to close a struct file
512 * @inode: the inode the struct file belongs to
513 * @file: the struct file being closed
518 static int gfs2_close(struct inode *inode, struct file *file)
520 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
521 struct gfs2_file *fp;
523 fp = file->private_data;
524 file->private_data = NULL;
526 if (gfs2_assert_warn(sdp, fp))
535 * gfs2_fsync - sync the dirty data for a file (across the cluster)
536 * @file: the file that points to the dentry (we ignore this)
537 * @dentry: the dentry that points to the inode to sync
539 * The VFS will flush "normal" data for us. We only need to worry
540 * about metadata here. For journaled data, we just do a log flush
541 * as we can't avoid it. Otherwise we can just bale out if datasync
542 * is set. For stuffed inodes we must flush the log in order to
543 * ensure that all data is on disk.
545 * The call to write_inode_now() is there to write back metadata and
546 * the inode itself. It does also try and write the data, but thats
547 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
553 static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
555 struct inode *inode = dentry->d_inode;
556 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
559 if (gfs2_is_jdata(GFS2_I(inode))) {
560 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
564 if (sync_state != 0) {
566 ret = write_inode_now(inode, 0);
568 if (gfs2_is_stuffed(GFS2_I(inode)))
569 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
576 * gfs2_setlease - acquire/release a file lease
577 * @file: the file pointer
584 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
586 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
589 * We don't currently have a way to enforce a lease across the whole
590 * cluster; until we do, disable leases (by just returning -EINVAL),
591 * unless the administrator has requested purely local locking.
593 if (!sdp->sd_args.ar_localflocks)
595 return generic_setlease(file, arg, fl);
598 static int gfs2_lm_plock_get(struct gfs2_sbd *sdp, struct lm_lockname *name,
599 struct file *file, struct file_lock *fl)
602 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
603 error = sdp->sd_lockstruct.ls_ops->lm_plock_get(
604 sdp->sd_lockstruct.ls_lockspace, name, file, fl);
608 static int gfs2_lm_plock(struct gfs2_sbd *sdp, struct lm_lockname *name,
609 struct file *file, int cmd, struct file_lock *fl)
612 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
613 error = sdp->sd_lockstruct.ls_ops->lm_plock(
614 sdp->sd_lockstruct.ls_lockspace, name, file, cmd, fl);
618 static int gfs2_lm_punlock(struct gfs2_sbd *sdp, struct lm_lockname *name,
619 struct file *file, struct file_lock *fl)
622 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
623 error = sdp->sd_lockstruct.ls_ops->lm_punlock(
624 sdp->sd_lockstruct.ls_lockspace, name, file, fl);
629 * gfs2_lock - acquire/release a posix lock on a file
630 * @file: the file pointer
631 * @cmd: either modify or retrieve lock state, possibly wait
632 * @fl: type and range of lock
637 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
639 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
640 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
641 struct lm_lockname name =
642 { .ln_number = ip->i_no_addr,
643 .ln_type = LM_TYPE_PLOCK };
645 if (!(fl->fl_flags & FL_POSIX))
647 if (__mandatory_lock(&ip->i_inode))
650 if (cmd == F_CANCELLK) {
653 fl->fl_type = F_UNLCK;
656 return gfs2_lm_plock_get(sdp, &name, file, fl);
657 else if (fl->fl_type == F_UNLCK)
658 return gfs2_lm_punlock(sdp, &name, file, fl);
660 return gfs2_lm_plock(sdp, &name, file, cmd, fl);
663 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
665 struct gfs2_file *fp = file->private_data;
666 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
667 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
668 struct gfs2_glock *gl;
673 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
674 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE
677 mutex_lock(&fp->f_fl_mutex);
681 if (fl_gh->gh_state == state)
683 flock_lock_file_wait(file,
684 &(struct file_lock){.fl_type = F_UNLCK});
685 gfs2_glock_dq_wait(fl_gh);
686 gfs2_holder_reinit(state, flags, fl_gh);
688 error = gfs2_glock_get(GFS2_SB(&ip->i_inode),
689 ip->i_no_addr, &gfs2_flock_glops,
693 gfs2_holder_init(gl, state, flags, fl_gh);
696 error = gfs2_glock_nq(fl_gh);
698 gfs2_holder_uninit(fl_gh);
699 if (error == GLR_TRYFAILED)
702 error = flock_lock_file_wait(file, fl);
703 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
707 mutex_unlock(&fp->f_fl_mutex);
711 static void do_unflock(struct file *file, struct file_lock *fl)
713 struct gfs2_file *fp = file->private_data;
714 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
716 mutex_lock(&fp->f_fl_mutex);
717 flock_lock_file_wait(file, fl);
719 gfs2_glock_dq_uninit(fl_gh);
720 mutex_unlock(&fp->f_fl_mutex);
724 * gfs2_flock - acquire/release a flock lock on a file
725 * @file: the file pointer
726 * @cmd: either modify or retrieve lock state, possibly wait
727 * @fl: type and range of lock
732 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
734 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
736 if (!(fl->fl_flags & FL_FLOCK))
738 if (__mandatory_lock(&ip->i_inode))
741 if (fl->fl_type == F_UNLCK) {
742 do_unflock(file, fl);
745 return do_flock(file, cmd, fl);
749 const struct file_operations gfs2_file_fops = {
750 .llseek = gfs2_llseek,
751 .read = do_sync_read,
752 .aio_read = generic_file_aio_read,
753 .write = do_sync_write,
754 .aio_write = generic_file_aio_write,
755 .unlocked_ioctl = gfs2_ioctl,
758 .release = gfs2_close,
762 .splice_read = generic_file_splice_read,
763 .splice_write = generic_file_splice_write,
764 .setlease = gfs2_setlease,
767 const struct file_operations gfs2_dir_fops = {
768 .readdir = gfs2_readdir,
769 .unlocked_ioctl = gfs2_ioctl,
771 .release = gfs2_close,
777 const struct file_operations gfs2_file_fops_nolock = {
778 .llseek = gfs2_llseek,
779 .read = do_sync_read,
780 .aio_read = generic_file_aio_read,
781 .write = do_sync_write,
782 .aio_write = generic_file_aio_write,
783 .unlocked_ioctl = gfs2_ioctl,
786 .release = gfs2_close,
788 .splice_read = generic_file_splice_read,
789 .splice_write = generic_file_splice_write,
790 .setlease = gfs2_setlease,
793 const struct file_operations gfs2_dir_fops_nolock = {
794 .readdir = gfs2_readdir,
795 .unlocked_ioctl = gfs2_ioctl,
797 .release = gfs2_close,