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 v.2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/pagemap.h>
16 #include <linux/uio.h>
17 #include <linux/blkdev.h>
19 #include <linux/smp_lock.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/ext2_fs.h>
23 #include <linux/crc32.h>
24 #include <linux/iflags.h>
25 #include <asm/uaccess.h>
28 #include "lm_interface.h"
46 /* "bad" is for NFS support */
47 struct filldir_bad_entry {
49 unsigned int fbe_length;
51 struct gfs2_inum fbe_inum;
52 unsigned int fbe_type;
56 struct gfs2_sbd *fdb_sbd;
58 struct filldir_bad_entry *fdb_entry;
59 unsigned int fdb_entry_num;
60 unsigned int fdb_entry_off;
63 unsigned int fdb_name_size;
64 unsigned int fdb_name_off;
67 /* For regular, non-NFS */
69 struct gfs2_sbd *fdr_sbd;
72 filldir_t fdr_filldir;
77 * Most fields left uninitialised to catch anybody who tries to
78 * use them. f_flags set to prevent file_accessed() from touching
79 * any other part of this. Its use is purely as a flag so that we
80 * know (in readpage()) whether or not do to locking.
82 struct file gfs2_internal_file_sentinal = {
83 .f_flags = O_NOATIME|O_RDONLY,
86 static int gfs2_read_actor(read_descriptor_t *desc, struct page *page,
87 unsigned long offset, unsigned long size)
90 unsigned long count = desc->count;
96 memcpy(desc->arg.buf, kaddr + offset, size);
99 desc->count = count - size;
100 desc->written += size;
101 desc->arg.buf += size;
105 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
106 char *buf, loff_t *pos, unsigned size)
108 struct inode *inode = &ip->i_inode;
109 read_descriptor_t desc;
114 do_generic_mapping_read(inode->i_mapping, ra_state,
115 &gfs2_internal_file_sentinal, pos, &desc,
117 return desc.written ? desc.written : desc.error;
121 * gfs2_llseek - seek to a location in a file
123 * @offset: the offset
124 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
126 * SEEK_END requires the glock for the file because it references the
129 * Returns: The new offset, or errno
132 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
134 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
135 struct gfs2_holder i_gh;
139 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
142 error = remote_llseek(file, offset, origin);
143 gfs2_glock_dq_uninit(&i_gh);
146 error = remote_llseek(file, offset, origin);
152 * filldir_reg_func - Report a directory entry to the caller of gfs2_dir_read()
153 * @opaque: opaque data used by the function
154 * @name: the name of the directory entry
155 * @length: the length of the name
156 * @offset: the entry's offset in the directory
157 * @inum: the inode number the entry points to
158 * @type: the type of inode the entry points to
160 * Returns: 0 on success, 1 if buffer full
163 static int filldir_reg_func(void *opaque, const char *name, unsigned int length,
164 uint64_t offset, struct gfs2_inum *inum,
167 struct filldir_reg *fdr = (struct filldir_reg *)opaque;
168 struct gfs2_sbd *sdp = fdr->fdr_sbd;
171 error = fdr->fdr_filldir(fdr->fdr_opaque, name, length, offset,
172 inum->no_addr, type);
176 if (fdr->fdr_prefetch && !(length == 1 && *name == '.')) {
177 gfs2_glock_prefetch_num(sdp,
178 inum->no_addr, &gfs2_inode_glops,
179 LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY);
180 gfs2_glock_prefetch_num(sdp,
181 inum->no_addr, &gfs2_iopen_glops,
182 LM_ST_SHARED, LM_FLAG_TRY);
189 * readdir_reg - Read directory entries from a directory
190 * @file: The directory to read from
191 * @dirent: Buffer for dirents
192 * @filldir: Function used to do the copying
197 static int readdir_reg(struct file *file, void *dirent, filldir_t filldir)
199 struct inode *dir = file->f_mapping->host;
200 struct gfs2_inode *dip = GFS2_I(dir);
201 struct filldir_reg fdr;
202 struct gfs2_holder d_gh;
203 uint64_t offset = file->f_pos;
206 fdr.fdr_sbd = GFS2_SB(dir);
207 fdr.fdr_prefetch = 1;
208 fdr.fdr_filldir = filldir;
209 fdr.fdr_opaque = dirent;
211 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh);
212 error = gfs2_glock_nq_atime(&d_gh);
214 gfs2_holder_uninit(&d_gh);
218 error = gfs2_dir_read(dir, &offset, &fdr, filldir_reg_func);
220 gfs2_glock_dq_uninit(&d_gh);
222 file->f_pos = offset;
228 * filldir_bad_func - Report a directory entry to the caller of gfs2_dir_read()
229 * @opaque: opaque data used by the function
230 * @name: the name of the directory entry
231 * @length: the length of the name
232 * @offset: the entry's offset in the directory
233 * @inum: the inode number the entry points to
234 * @type: the type of inode the entry points to
236 * For supporting NFS.
238 * Returns: 0 on success, 1 if buffer full
241 static int filldir_bad_func(void *opaque, const char *name, unsigned int length,
242 uint64_t offset, struct gfs2_inum *inum,
245 struct filldir_bad *fdb = (struct filldir_bad *)opaque;
246 struct gfs2_sbd *sdp = fdb->fdb_sbd;
247 struct filldir_bad_entry *fbe;
249 if (fdb->fdb_entry_off == fdb->fdb_entry_num ||
250 fdb->fdb_name_off + length > fdb->fdb_name_size)
253 fbe = &fdb->fdb_entry[fdb->fdb_entry_off];
254 fbe->fbe_name = fdb->fdb_name + fdb->fdb_name_off;
255 memcpy(fbe->fbe_name, name, length);
256 fbe->fbe_length = length;
257 fbe->fbe_offset = offset;
258 fbe->fbe_inum = *inum;
259 fbe->fbe_type = type;
261 fdb->fdb_entry_off++;
262 fdb->fdb_name_off += length;
264 if (!(length == 1 && *name == '.')) {
265 gfs2_glock_prefetch_num(sdp,
266 inum->no_addr, &gfs2_inode_glops,
267 LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY);
268 gfs2_glock_prefetch_num(sdp,
269 inum->no_addr, &gfs2_iopen_glops,
270 LM_ST_SHARED, LM_FLAG_TRY);
277 * readdir_bad - Read directory entries from a directory
278 * @file: The directory to read from
279 * @dirent: Buffer for dirents
280 * @filldir: Function used to do the copying
282 * For supporting NFS.
287 static int readdir_bad(struct file *file, void *dirent, filldir_t filldir)
289 struct inode *dir = file->f_mapping->host;
290 struct gfs2_inode *dip = GFS2_I(dir);
291 struct gfs2_sbd *sdp = GFS2_SB(dir);
292 struct filldir_reg fdr;
293 unsigned int entries, size;
294 struct filldir_bad *fdb;
295 struct gfs2_holder d_gh;
296 uint64_t offset = file->f_pos;
298 struct filldir_bad_entry *fbe;
301 entries = gfs2_tune_get(sdp, gt_entries_per_readdir);
302 size = sizeof(struct filldir_bad) +
303 entries * (sizeof(struct filldir_bad_entry) + GFS2_FAST_NAME_SIZE);
305 fdb = kzalloc(size, GFP_KERNEL);
310 fdb->fdb_entry = (struct filldir_bad_entry *)(fdb + 1);
311 fdb->fdb_entry_num = entries;
312 fdb->fdb_name = ((char *)fdb) + sizeof(struct filldir_bad) +
313 entries * sizeof(struct filldir_bad_entry);
314 fdb->fdb_name_size = entries * GFS2_FAST_NAME_SIZE;
316 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh);
317 error = gfs2_glock_nq_atime(&d_gh);
319 gfs2_holder_uninit(&d_gh);
323 error = gfs2_dir_read(dir, &offset, fdb, filldir_bad_func);
325 gfs2_glock_dq_uninit(&d_gh);
328 fdr.fdr_prefetch = 0;
329 fdr.fdr_filldir = filldir;
330 fdr.fdr_opaque = dirent;
332 for (x = 0; x < fdb->fdb_entry_off; x++) {
333 fbe = &fdb->fdb_entry[x];
335 error = filldir_reg_func(&fdr,
336 fbe->fbe_name, fbe->fbe_length,
338 &fbe->fbe_inum, fbe->fbe_type);
340 file->f_pos = fbe->fbe_offset;
346 file->f_pos = offset;
355 * gfs2_readdir - Read directory entries from a directory
356 * @file: The directory to read from
357 * @dirent: Buffer for dirents
358 * @filldir: Function used to do the copying
363 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
367 if (strcmp(current->comm, "nfsd") != 0)
368 error = readdir_reg(file, dirent, filldir);
370 error = readdir_bad(file, dirent, filldir);
375 static const u32 iflags_to_gfs2[32] = {
376 [iflag_Sync] = GFS2_DIF_SYNC,
377 [iflag_Immutable] = GFS2_DIF_IMMUTABLE,
378 [iflag_Append] = GFS2_DIF_APPENDONLY,
379 [iflag_NoAtime] = GFS2_DIF_NOATIME,
380 [iflag_Index] = GFS2_DIF_EXHASH,
381 [iflag_JournalData] = GFS2_DIF_JDATA,
382 [iflag_DirectIO] = GFS2_DIF_DIRECTIO,
385 static const u32 gfs2_to_iflags[32] = {
386 [gfs2fl_Sync] = IFLAG_SYNC,
387 [gfs2fl_Immutable] = IFLAG_IMMUTABLE,
388 [gfs2fl_AppendOnly] = IFLAG_APPEND,
389 [gfs2fl_NoAtime] = IFLAG_NOATIME,
390 [gfs2fl_ExHash] = IFLAG_INDEX,
391 [gfs2fl_Jdata] = IFLAG_JOURNAL_DATA,
392 [gfs2fl_Directio] = IFLAG_DIRECTIO,
393 [gfs2fl_InheritDirectio] = IFLAG_DIRECTIO,
394 [gfs2fl_InheritJdata] = IFLAG_JOURNAL_DATA,
397 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
399 struct inode *inode = filp->f_dentry->d_inode;
400 struct gfs2_inode *ip = GFS2_I(inode);
401 struct gfs2_holder gh;
405 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
406 error = gfs2_glock_nq_m_atime(1, &gh);
410 iflags = iflags_cvt(gfs2_to_iflags, ip->i_di.di_flags);
411 if (put_user(iflags, ptr))
414 gfs2_glock_dq_m(1, &gh);
415 gfs2_holder_uninit(&gh);
419 /* Flags that can be set by user space */
420 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
422 GFS2_DIF_IMMUTABLE| \
423 GFS2_DIF_APPENDONLY| \
427 GFS2_DIF_INHERIT_DIRECTIO| \
428 GFS2_DIF_INHERIT_JDATA)
431 * gfs2_set_flags - set flags on an inode
433 * @flags: The flags to set
434 * @mask: Indicates which flags are valid
437 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
439 struct inode *inode = filp->f_dentry->d_inode;
440 struct gfs2_inode *ip = GFS2_I(inode);
441 struct gfs2_sbd *sdp = GFS2_SB(inode);
442 struct buffer_head *bh;
443 struct gfs2_holder gh;
445 u32 new_flags, flags;
447 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
451 flags = ip->i_di.di_flags;
452 new_flags = (flags & ~mask) | (reqflags & mask);
453 if ((new_flags ^ flags) == 0)
456 if (S_ISDIR(inode->i_mode)) {
457 if ((new_flags ^ flags) & GFS2_DIF_JDATA)
458 new_flags ^= (GFS2_DIF_JDATA|GFS2_DIF_INHERIT_JDATA);
459 if ((new_flags ^ flags) & GFS2_DIF_DIRECTIO)
460 new_flags ^= (GFS2_DIF_DIRECTIO|GFS2_DIF_INHERIT_DIRECTIO);
464 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
468 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
470 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
472 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
473 !capable(CAP_LINUX_IMMUTABLE))
475 if (!IS_IMMUTABLE(inode)) {
476 error = permission(inode, MAY_WRITE, NULL);
481 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
484 error = gfs2_meta_inode_buffer(ip, &bh);
487 gfs2_trans_add_bh(ip->i_gl, bh, 1);
488 ip->i_di.di_flags = new_flags;
489 gfs2_dinode_out(&ip->i_di, bh->b_data);
494 gfs2_glock_dq_uninit(&gh);
498 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
500 u32 iflags, gfsflags;
501 if (get_user(iflags, ptr))
503 gfsflags = iflags_cvt(iflags_to_gfs2, iflags);
504 return do_gfs2_set_flags(filp, gfsflags, ~0);
507 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
511 return gfs2_get_flags(filp, (u32 __user *)arg);
513 return gfs2_set_flags(filp, (u32 __user *)arg);
521 * @file: The file to map
522 * @vma: The VMA which described the mapping
524 * Returns: 0 or error code
527 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
529 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
530 struct gfs2_holder i_gh;
533 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
534 error = gfs2_glock_nq_atime(&i_gh);
536 gfs2_holder_uninit(&i_gh);
540 /* This is VM_MAYWRITE instead of VM_WRITE because a call
541 to mprotect() can turn on VM_WRITE later. */
543 if ((vma->vm_flags & (VM_MAYSHARE | VM_MAYWRITE)) ==
544 (VM_MAYSHARE | VM_MAYWRITE))
545 vma->vm_ops = &gfs2_vm_ops_sharewrite;
547 vma->vm_ops = &gfs2_vm_ops_private;
549 gfs2_glock_dq_uninit(&i_gh);
555 * gfs2_open - open a file
556 * @inode: the inode to open
557 * @file: the struct file for this opening
562 static int gfs2_open(struct inode *inode, struct file *file)
564 struct gfs2_inode *ip = GFS2_I(inode);
565 struct gfs2_holder i_gh;
566 struct gfs2_file *fp;
569 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
573 mutex_init(&fp->f_fl_mutex);
575 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
576 file->private_data = fp;
578 if (S_ISREG(ip->i_di.di_mode)) {
579 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
584 if (!(file->f_flags & O_LARGEFILE) &&
585 ip->i_di.di_size > MAX_NON_LFS) {
590 /* Listen to the Direct I/O flag */
592 if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO)
593 file->f_flags |= O_DIRECT;
595 gfs2_glock_dq_uninit(&i_gh);
601 gfs2_glock_dq_uninit(&i_gh);
603 file->private_data = NULL;
609 * gfs2_close - called to close a struct file
610 * @inode: the inode the struct file belongs to
611 * @file: the struct file being closed
616 static int gfs2_close(struct inode *inode, struct file *file)
618 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
619 struct gfs2_file *fp;
621 fp = file->private_data;
622 file->private_data = NULL;
624 if (gfs2_assert_warn(sdp, fp))
633 * gfs2_fsync - sync the dirty data for a file (across the cluster)
634 * @file: the file that points to the dentry (we ignore this)
635 * @dentry: the dentry that points to the inode to sync
640 static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync)
642 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
644 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
650 * gfs2_lock - acquire/release a posix lock on a file
651 * @file: the file pointer
652 * @cmd: either modify or retrieve lock state, possibly wait
653 * @fl: type and range of lock
658 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
660 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
661 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
662 struct lm_lockname name =
663 { .ln_number = ip->i_num.no_addr,
664 .ln_type = LM_TYPE_PLOCK };
666 if (!(fl->fl_flags & FL_POSIX))
668 if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
671 if (sdp->sd_args.ar_localflocks) {
673 struct file_lock tmp;
675 ret = posix_test_lock(file, fl, &tmp);
676 fl->fl_type = F_UNLCK;
678 memcpy(fl, &tmp, sizeof(struct file_lock));
681 return posix_lock_file_wait(file, fl);
686 return gfs2_lm_plock_get(sdp, &name, file, fl);
687 else if (fl->fl_type == F_UNLCK)
688 return gfs2_lm_punlock(sdp, &name, file, fl);
690 return gfs2_lm_plock(sdp, &name, file, cmd, fl);
693 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
695 struct gfs2_file *fp = file->private_data;
696 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
697 struct gfs2_inode *ip = GFS2_I(file->f_dentry->d_inode);
698 struct gfs2_glock *gl;
703 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
704 flags = ((IS_SETLKW(cmd)) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
706 mutex_lock(&fp->f_fl_mutex);
710 if (fl_gh->gh_state == state)
713 flock_lock_file_wait(file,
714 &(struct file_lock){.fl_type = F_UNLCK});
715 gfs2_glock_dq_uninit(fl_gh);
717 error = gfs2_glock_get(GFS2_SB(&ip->i_inode),
718 ip->i_num.no_addr, &gfs2_flock_glops,
724 gfs2_holder_init(gl, state, flags, fl_gh);
727 error = gfs2_glock_nq(fl_gh);
729 gfs2_holder_uninit(fl_gh);
730 if (error == GLR_TRYFAILED)
733 error = flock_lock_file_wait(file, fl);
734 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
738 mutex_unlock(&fp->f_fl_mutex);
742 static void do_unflock(struct file *file, struct file_lock *fl)
744 struct gfs2_file *fp = file->private_data;
745 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
747 mutex_lock(&fp->f_fl_mutex);
748 flock_lock_file_wait(file, fl);
750 gfs2_glock_dq_uninit(fl_gh);
751 mutex_unlock(&fp->f_fl_mutex);
755 * gfs2_flock - acquire/release a flock lock on a file
756 * @file: the file pointer
757 * @cmd: either modify or retrieve lock state, possibly wait
758 * @fl: type and range of lock
763 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
765 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
766 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
768 if (!(fl->fl_flags & FL_FLOCK))
770 if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
773 if (sdp->sd_args.ar_localflocks)
774 return flock_lock_file_wait(file, fl);
776 if (fl->fl_type == F_UNLCK) {
777 do_unflock(file, fl);
780 return do_flock(file, cmd, fl);
783 const struct file_operations gfs2_file_fops = {
784 .llseek = gfs2_llseek,
785 .read = generic_file_read,
786 .readv = generic_file_readv,
787 .aio_read = generic_file_aio_read,
788 .write = generic_file_write,
789 .writev = generic_file_writev,
790 .aio_write = generic_file_aio_write,
791 .unlocked_ioctl = gfs2_ioctl,
794 .release = gfs2_close,
797 .sendfile = generic_file_sendfile,
799 .splice_read = generic_file_splice_read,
800 .splice_write = generic_file_splice_write,
803 const struct file_operations gfs2_dir_fops = {
804 .readdir = gfs2_readdir,
805 .unlocked_ioctl = gfs2_ioctl,
807 .release = gfs2_close,