1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/spinlock.h>
9 #include <linux/compat.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/pagemap.h>
13 #include <linux/uio.h>
14 #include <linux/blkdev.h>
16 #include <linux/mount.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/falloc.h>
20 #include <linux/swap.h>
21 #include <linux/crc32.h>
22 #include <linux/writeback.h>
23 #include <linux/uaccess.h>
24 #include <linux/dlm.h>
25 #include <linux/dlm_plock.h>
26 #include <linux/delay.h>
27 #include <linux/backing-dev.h>
28 #include <linux/fileattr.h>
46 * gfs2_llseek - seek to a location in a file
49 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
51 * SEEK_END requires the glock for the file because it references the
54 * Returns: The new offset, or errno
57 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
59 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
60 struct gfs2_holder i_gh;
65 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
68 error = generic_file_llseek(file, offset, whence);
69 gfs2_glock_dq_uninit(&i_gh);
74 error = gfs2_seek_data(file, offset);
78 error = gfs2_seek_hole(file, offset);
84 * These don't reference inode->i_size and don't depend on the
85 * block mapping, so we don't need the glock.
87 error = generic_file_llseek(file, offset, whence);
97 * gfs2_readdir - Iterator for a directory
98 * @file: The directory to read from
99 * @ctx: What to feed directory entries to
104 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
106 struct inode *dir = file->f_mapping->host;
107 struct gfs2_inode *dip = GFS2_I(dir);
108 struct gfs2_holder d_gh;
111 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
115 error = gfs2_dir_read(dir, ctx, &file->f_ra);
117 gfs2_glock_dq_uninit(&d_gh);
123 * struct fsflag_gfs2flag
125 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126 * and to GFS2_DIF_JDATA for non-directories.
131 } fsflag_gfs2flag[] = {
132 {FS_SYNC_FL, GFS2_DIF_SYNC},
133 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
134 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
135 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
136 {FS_INDEX_FL, GFS2_DIF_EXHASH},
137 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
138 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
141 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
146 if (S_ISDIR(inode->i_mode))
147 gfsflags &= ~GFS2_DIF_JDATA;
149 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
151 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
152 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
153 fsflags |= fsflag_gfs2flag[i].fsflag;
157 int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
159 struct inode *inode = d_inode(dentry);
160 struct gfs2_inode *ip = GFS2_I(inode);
161 struct gfs2_holder gh;
165 if (d_is_special(dentry))
168 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
169 error = gfs2_glock_nq(&gh);
173 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
175 fileattr_fill_flags(fa, fsflags);
179 gfs2_holder_uninit(&gh);
183 void gfs2_set_inode_flags(struct inode *inode)
185 struct gfs2_inode *ip = GFS2_I(inode);
186 unsigned int flags = inode->i_flags;
188 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
189 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
191 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
192 flags |= S_IMMUTABLE;
193 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
195 if (ip->i_diskflags & GFS2_DIF_NOATIME)
197 if (ip->i_diskflags & GFS2_DIF_SYNC)
199 inode->i_flags = flags;
202 /* Flags that can be set by user space */
203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
204 GFS2_DIF_IMMUTABLE| \
205 GFS2_DIF_APPENDONLY| \
209 GFS2_DIF_INHERIT_JDATA)
212 * do_gfs2_set_flags - set flags on an inode
214 * @reqflags: The flags to set
215 * @mask: Indicates which flags are valid
216 * @fsflags: The FS_* inode flags passed in
219 static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask,
222 struct gfs2_inode *ip = GFS2_I(inode);
223 struct gfs2_sbd *sdp = GFS2_SB(inode);
224 struct buffer_head *bh;
225 struct gfs2_holder gh;
227 u32 new_flags, flags;
229 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
234 flags = ip->i_diskflags;
235 new_flags = (flags & ~mask) | (reqflags & mask);
236 if ((new_flags ^ flags) == 0)
240 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
242 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
244 if (!IS_IMMUTABLE(inode)) {
245 error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
249 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
250 if (new_flags & GFS2_DIF_JDATA)
251 gfs2_log_flush(sdp, ip->i_gl,
252 GFS2_LOG_HEAD_FLUSH_NORMAL |
254 error = filemap_fdatawrite(inode->i_mapping);
257 error = filemap_fdatawait(inode->i_mapping);
260 if (new_flags & GFS2_DIF_JDATA)
261 gfs2_ordered_del_inode(ip);
263 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
266 error = gfs2_meta_inode_buffer(ip, &bh);
269 inode->i_ctime = current_time(inode);
270 gfs2_trans_add_meta(ip->i_gl, bh);
271 ip->i_diskflags = new_flags;
272 gfs2_dinode_out(ip, bh->b_data);
274 gfs2_set_inode_flags(inode);
275 gfs2_set_aops(inode);
279 gfs2_glock_dq_uninit(&gh);
283 int gfs2_fileattr_set(struct user_namespace *mnt_userns,
284 struct dentry *dentry, struct fileattr *fa)
286 struct inode *inode = d_inode(dentry);
287 u32 fsflags = fa->flags, gfsflags = 0;
291 if (d_is_special(dentry))
294 if (fileattr_has_fsx(fa))
297 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
298 if (fsflags & fsflag_gfs2flag[i].fsflag) {
299 fsflags &= ~fsflag_gfs2flag[i].fsflag;
300 gfsflags |= fsflag_gfs2flag[i].gfsflag;
303 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
306 mask = GFS2_FLAGS_USER_SET;
307 if (S_ISDIR(inode->i_mode)) {
308 mask &= ~GFS2_DIF_JDATA;
310 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
311 if (gfsflags & GFS2_DIF_TOPDIR)
313 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
316 return do_gfs2_set_flags(inode, gfsflags, mask, fsflags);
319 static int gfs2_getlabel(struct file *filp, char __user *label)
321 struct inode *inode = file_inode(filp);
322 struct gfs2_sbd *sdp = GFS2_SB(inode);
324 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
330 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
334 return gfs2_fitrim(filp, (void __user *)arg);
335 case FS_IOC_GETFSLABEL:
336 return gfs2_getlabel(filp, (char __user *)arg);
343 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
346 /* Keep this list in sync with gfs2_ioctl */
348 case FS_IOC_GETFSLABEL:
354 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
357 #define gfs2_compat_ioctl NULL
361 * gfs2_size_hint - Give a hint to the size of a write request
362 * @filep: The struct file
363 * @offset: The file offset of the write
364 * @size: The length of the write
366 * When we are about to do a write, this function records the total
367 * write size in order to provide a suitable hint to the lower layers
368 * about how many blocks will be required.
372 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
374 struct inode *inode = file_inode(filep);
375 struct gfs2_sbd *sdp = GFS2_SB(inode);
376 struct gfs2_inode *ip = GFS2_I(inode);
377 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
378 int hint = min_t(size_t, INT_MAX, blks);
380 if (hint > atomic_read(&ip->i_sizehint))
381 atomic_set(&ip->i_sizehint, hint);
385 * gfs2_allocate_page_backing - Allocate blocks for a write fault
386 * @page: The (locked) page to allocate backing for
387 * @length: Size of the allocation
389 * We try to allocate all the blocks required for the page in one go. This
390 * might fail for various reasons, so we keep trying until all the blocks to
391 * back this page are allocated. If some of the blocks are already allocated,
394 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
396 u64 pos = page_offset(page);
399 struct iomap iomap = { };
401 if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
404 if (length < iomap.length)
405 iomap.length = length;
406 length -= iomap.length;
408 } while (length > 0);
414 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
415 * @vmf: The virtual memory fault containing the page to become writable
417 * When the page becomes writable, we need to ensure that we have
418 * blocks allocated on disk to back that page.
421 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
423 struct page *page = vmf->page;
424 struct inode *inode = file_inode(vmf->vma->vm_file);
425 struct gfs2_inode *ip = GFS2_I(inode);
426 struct gfs2_sbd *sdp = GFS2_SB(inode);
427 struct gfs2_alloc_parms ap = { .aflags = 0, };
428 u64 offset = page_offset(page);
429 unsigned int data_blocks, ind_blocks, rblocks;
430 vm_fault_t ret = VM_FAULT_LOCKED;
431 struct gfs2_holder gh;
436 sb_start_pagefault(inode->i_sb);
438 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
439 err = gfs2_glock_nq(&gh);
441 ret = block_page_mkwrite_return(err);
445 /* Check page index against inode size */
446 size = i_size_read(inode);
447 if (offset >= size) {
448 ret = VM_FAULT_SIGBUS;
452 /* Update file times before taking page lock */
453 file_update_time(vmf->vma->vm_file);
455 /* page is wholly or partially inside EOF */
456 if (size - offset < PAGE_SIZE)
457 length = size - offset;
461 gfs2_size_hint(vmf->vma->vm_file, offset, length);
463 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
464 set_bit(GIF_SW_PAGED, &ip->i_flags);
467 * iomap_writepage / iomap_writepages currently don't support inline
468 * files, so always unstuff here.
471 if (!gfs2_is_stuffed(ip) &&
472 !gfs2_write_alloc_required(ip, offset, length)) {
474 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
475 ret = VM_FAULT_NOPAGE;
481 err = gfs2_rindex_update(sdp);
483 ret = block_page_mkwrite_return(err);
487 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
488 ap.target = data_blocks + ind_blocks;
489 err = gfs2_quota_lock_check(ip, &ap);
491 ret = block_page_mkwrite_return(err);
494 err = gfs2_inplace_reserve(ip, &ap);
496 ret = block_page_mkwrite_return(err);
497 goto out_quota_unlock;
500 rblocks = RES_DINODE + ind_blocks;
501 if (gfs2_is_jdata(ip))
502 rblocks += data_blocks ? data_blocks : 1;
503 if (ind_blocks || data_blocks) {
504 rblocks += RES_STATFS + RES_QUOTA;
505 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
507 err = gfs2_trans_begin(sdp, rblocks, 0);
509 ret = block_page_mkwrite_return(err);
513 /* Unstuff, if required, and allocate backing blocks for page */
514 if (gfs2_is_stuffed(ip)) {
515 err = gfs2_unstuff_dinode(ip);
517 ret = block_page_mkwrite_return(err);
523 /* If truncated, we must retry the operation, we may have raced
524 * with the glock demotion code.
526 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
527 ret = VM_FAULT_NOPAGE;
528 goto out_page_locked;
531 err = gfs2_allocate_page_backing(page, length);
533 ret = block_page_mkwrite_return(err);
536 if (ret != VM_FAULT_LOCKED)
541 gfs2_inplace_release(ip);
543 gfs2_quota_unlock(ip);
547 gfs2_holder_uninit(&gh);
548 if (ret == VM_FAULT_LOCKED) {
549 set_page_dirty(page);
550 wait_for_stable_page(page);
552 sb_end_pagefault(inode->i_sb);
556 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
558 struct inode *inode = file_inode(vmf->vma->vm_file);
559 struct gfs2_inode *ip = GFS2_I(inode);
560 struct gfs2_holder gh;
564 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
565 err = gfs2_glock_nq(&gh);
567 ret = block_page_mkwrite_return(err);
570 ret = filemap_fault(vmf);
573 gfs2_holder_uninit(&gh);
577 static const struct vm_operations_struct gfs2_vm_ops = {
579 .map_pages = filemap_map_pages,
580 .page_mkwrite = gfs2_page_mkwrite,
585 * @file: The file to map
586 * @vma: The VMA which described the mapping
588 * There is no need to get a lock here unless we should be updating
589 * atime. We ignore any locking errors since the only consequence is
590 * a missed atime update (which will just be deferred until later).
595 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
597 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
599 if (!(file->f_flags & O_NOATIME) &&
600 !IS_NOATIME(&ip->i_inode)) {
601 struct gfs2_holder i_gh;
604 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
608 /* grab lock to update inode */
609 gfs2_glock_dq_uninit(&i_gh);
612 vma->vm_ops = &gfs2_vm_ops;
618 * gfs2_open_common - This is common to open and atomic_open
619 * @inode: The inode being opened
620 * @file: The file being opened
622 * This maybe called under a glock or not depending upon how it has
623 * been called. We must always be called under a glock for regular
624 * files, however. For other file types, it does not matter whether
625 * we hold the glock or not.
627 * Returns: Error code or 0 for success
630 int gfs2_open_common(struct inode *inode, struct file *file)
632 struct gfs2_file *fp;
635 if (S_ISREG(inode->i_mode)) {
636 ret = generic_file_open(inode, file);
641 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
645 mutex_init(&fp->f_fl_mutex);
647 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
648 file->private_data = fp;
649 if (file->f_mode & FMODE_WRITE) {
650 ret = gfs2_qa_get(GFS2_I(inode));
657 kfree(file->private_data);
658 file->private_data = NULL;
663 * gfs2_open - open a file
664 * @inode: the inode to open
665 * @file: the struct file for this opening
667 * After atomic_open, this function is only used for opening files
668 * which are already cached. We must still get the glock for regular
669 * files to ensure that we have the file size uptodate for the large
670 * file check which is in the common code. That is only an issue for
671 * regular files though.
676 static int gfs2_open(struct inode *inode, struct file *file)
678 struct gfs2_inode *ip = GFS2_I(inode);
679 struct gfs2_holder i_gh;
681 bool need_unlock = false;
683 if (S_ISREG(ip->i_inode.i_mode)) {
684 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
691 error = gfs2_open_common(inode, file);
694 gfs2_glock_dq_uninit(&i_gh);
700 * gfs2_release - called to close a struct file
701 * @inode: the inode the struct file belongs to
702 * @file: the struct file being closed
707 static int gfs2_release(struct inode *inode, struct file *file)
709 struct gfs2_inode *ip = GFS2_I(inode);
711 kfree(file->private_data);
712 file->private_data = NULL;
714 if (gfs2_rs_active(&ip->i_res))
715 gfs2_rs_delete(ip, &inode->i_writecount);
716 if (file->f_mode & FMODE_WRITE)
722 * gfs2_fsync - sync the dirty data for a file (across the cluster)
723 * @file: the file that points to the dentry
724 * @start: the start position in the file to sync
725 * @end: the end position in the file to sync
726 * @datasync: set if we can ignore timestamp changes
728 * We split the data flushing here so that we don't wait for the data
729 * until after we've also sent the metadata to disk. Note that for
730 * data=ordered, we will write & wait for the data at the log flush
731 * stage anyway, so this is unlikely to make much of a difference
732 * except in the data=writeback case.
734 * If the fdatawrite fails due to any reason except -EIO, we will
735 * continue the remainder of the fsync, although we'll still report
736 * the error at the end. This is to match filemap_write_and_wait_range()
742 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
745 struct address_space *mapping = file->f_mapping;
746 struct inode *inode = mapping->host;
747 int sync_state = inode->i_state & I_DIRTY;
748 struct gfs2_inode *ip = GFS2_I(inode);
749 int ret = 0, ret1 = 0;
751 if (mapping->nrpages) {
752 ret1 = filemap_fdatawrite_range(mapping, start, end);
757 if (!gfs2_is_jdata(ip))
758 sync_state &= ~I_DIRTY_PAGES;
760 sync_state &= ~I_DIRTY_SYNC;
763 ret = sync_inode_metadata(inode, 1);
766 if (gfs2_is_jdata(ip))
767 ret = file_write_and_wait(file);
770 gfs2_ail_flush(ip->i_gl, 1);
773 if (mapping->nrpages)
774 ret = file_fdatawait_range(file, start, end);
776 return ret ? ret : ret1;
779 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
780 struct gfs2_holder *gh)
782 struct file *file = iocb->ki_filp;
783 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
784 size_t count = iov_iter_count(to);
788 return 0; /* skip atime */
790 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
791 ret = gfs2_glock_nq(gh);
795 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, 0);
798 gfs2_holder_uninit(gh);
802 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
803 struct gfs2_holder *gh)
805 struct file *file = iocb->ki_filp;
806 struct inode *inode = file->f_mapping->host;
807 struct gfs2_inode *ip = GFS2_I(inode);
808 size_t len = iov_iter_count(from);
809 loff_t offset = iocb->ki_pos;
813 * Deferred lock, even if its a write, since we do no allocation on
814 * this path. All we need to change is the atime, and this lock mode
815 * ensures that other nodes have flushed their buffered read caches
816 * (i.e. their page cache entries for this inode). We do not,
817 * unfortunately, have the option of only flushing a range like the
820 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
821 ret = gfs2_glock_nq(gh);
825 /* Silently fall back to buffered I/O when writing beyond EOF */
826 if (offset + len > i_size_read(&ip->i_inode))
829 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, 0);
835 gfs2_holder_uninit(gh);
839 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
841 struct gfs2_inode *ip;
842 struct gfs2_holder gh;
846 if (iocb->ki_flags & IOCB_DIRECT) {
847 ret = gfs2_file_direct_read(iocb, to, &gh);
848 if (likely(ret != -ENOTBLK))
850 iocb->ki_flags &= ~IOCB_DIRECT;
852 iocb->ki_flags |= IOCB_NOIO;
853 ret = generic_file_read_iter(iocb, to);
854 iocb->ki_flags &= ~IOCB_NOIO;
856 if (!iov_iter_count(to))
862 if (iocb->ki_flags & IOCB_NOWAIT)
865 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
866 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
867 ret = gfs2_glock_nq(&gh);
870 ret = generic_file_read_iter(iocb, to);
875 gfs2_holder_uninit(&gh);
876 return written ? written : ret;
880 * gfs2_file_write_iter - Perform a write to a file
881 * @iocb: The io context
882 * @from: The data to write
884 * We have to do a lock/unlock here to refresh the inode size for
885 * O_APPEND writes, otherwise we can land up writing at the wrong
886 * offset. There is still a race, but provided the app is using its
887 * own file locking, this will make O_APPEND work as expected.
891 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
893 struct file *file = iocb->ki_filp;
894 struct inode *inode = file_inode(file);
895 struct gfs2_inode *ip = GFS2_I(inode);
896 struct gfs2_holder gh;
899 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
901 if (iocb->ki_flags & IOCB_APPEND) {
902 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
905 gfs2_glock_dq_uninit(&gh);
909 ret = generic_write_checks(iocb, from);
913 ret = file_remove_privs(file);
917 ret = file_update_time(file);
921 if (iocb->ki_flags & IOCB_DIRECT) {
922 struct address_space *mapping = file->f_mapping;
923 ssize_t buffered, ret2;
925 ret = gfs2_file_direct_write(iocb, from, &gh);
926 if (ret < 0 || !iov_iter_count(from))
929 iocb->ki_flags |= IOCB_DSYNC;
930 current->backing_dev_info = inode_to_bdi(inode);
931 buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
932 current->backing_dev_info = NULL;
933 if (unlikely(buffered <= 0)) {
940 * We need to ensure that the page cache pages are written to
941 * disk and invalidated to preserve the expected O_DIRECT
942 * semantics. If the writeback or invalidate fails, only report
943 * the direct I/O range as we don't know if the buffered pages
946 iocb->ki_pos += buffered;
947 ret2 = generic_write_sync(iocb, buffered);
948 invalidate_mapping_pages(mapping,
949 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
950 (iocb->ki_pos - 1) >> PAGE_SHIFT);
951 if (!ret || ret2 > 0)
954 current->backing_dev_info = inode_to_bdi(inode);
955 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
956 current->backing_dev_info = NULL;
957 if (likely(ret > 0)) {
959 ret = generic_write_sync(iocb, ret);
968 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
971 struct super_block *sb = inode->i_sb;
972 struct gfs2_inode *ip = GFS2_I(inode);
973 loff_t end = offset + len;
974 struct buffer_head *dibh;
977 error = gfs2_meta_inode_buffer(ip, &dibh);
981 gfs2_trans_add_meta(ip->i_gl, dibh);
983 if (gfs2_is_stuffed(ip)) {
984 error = gfs2_unstuff_dinode(ip);
989 while (offset < end) {
990 struct iomap iomap = { };
992 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
995 offset = iomap.offset + iomap.length;
996 if (!(iomap.flags & IOMAP_F_NEW))
998 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
999 iomap.length >> inode->i_blkbits,
1002 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1012 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1013 * blocks, determine how many bytes can be written.
1014 * @ip: The inode in question.
1015 * @len: Max cap of bytes. What we return in *len must be <= this.
1016 * @data_blocks: Compute and return the number of data blocks needed
1017 * @ind_blocks: Compute and return the number of indirect blocks needed
1018 * @max_blocks: The total blocks available to work with.
1020 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1022 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1023 unsigned int *data_blocks, unsigned int *ind_blocks,
1024 unsigned int max_blocks)
1027 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1028 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1030 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1031 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1035 *data_blocks = max_data;
1036 *ind_blocks = max_blocks - max_data;
1037 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1040 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1044 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1046 struct inode *inode = file_inode(file);
1047 struct gfs2_sbd *sdp = GFS2_SB(inode);
1048 struct gfs2_inode *ip = GFS2_I(inode);
1049 struct gfs2_alloc_parms ap = { .aflags = 0, };
1050 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1051 loff_t bytes, max_bytes, max_blks;
1053 const loff_t pos = offset;
1054 const loff_t count = len;
1055 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1056 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1057 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1059 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1061 offset &= bsize_mask;
1063 len = next - offset;
1064 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1067 bytes &= bsize_mask;
1069 bytes = sdp->sd_sb.sb_bsize;
1071 gfs2_size_hint(file, offset, len);
1073 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1074 ap.min_target = data_blocks + ind_blocks;
1079 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1085 /* We need to determine how many bytes we can actually
1086 * fallocate without exceeding quota or going over the
1087 * end of the fs. We start off optimistically by assuming
1088 * we can write max_bytes */
1089 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1091 /* Since max_bytes is most likely a theoretical max, we
1092 * calculate a more realistic 'bytes' to serve as a good
1093 * starting point for the number of bytes we may be able
1095 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1096 ap.target = data_blocks + ind_blocks;
1098 error = gfs2_quota_lock_check(ip, &ap);
1101 /* ap.allowed tells us how many blocks quota will allow
1102 * us to write. Check if this reduces max_blks */
1103 max_blks = UINT_MAX;
1105 max_blks = ap.allowed;
1107 error = gfs2_inplace_reserve(ip, &ap);
1111 /* check if the selected rgrp limits our max_blks further */
1112 if (ip->i_res.rs_reserved < max_blks)
1113 max_blks = ip->i_res.rs_reserved;
1115 /* Almost done. Calculate bytes that can be written using
1116 * max_blks. We also recompute max_bytes, data_blocks and
1118 calc_max_reserv(ip, &max_bytes, &data_blocks,
1119 &ind_blocks, max_blks);
1121 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1122 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1123 if (gfs2_is_jdata(ip))
1124 rblocks += data_blocks ? data_blocks : 1;
1126 error = gfs2_trans_begin(sdp, rblocks,
1127 PAGE_SIZE >> inode->i_blkbits);
1129 goto out_trans_fail;
1131 error = fallocate_chunk(inode, offset, max_bytes, mode);
1132 gfs2_trans_end(sdp);
1135 goto out_trans_fail;
1138 offset += max_bytes;
1139 gfs2_inplace_release(ip);
1140 gfs2_quota_unlock(ip);
1143 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1144 i_size_write(inode, pos + count);
1145 file_update_time(file);
1146 mark_inode_dirty(inode);
1148 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1149 return vfs_fsync_range(file, pos, pos + count - 1,
1150 (file->f_flags & __O_SYNC) ? 0 : 1);
1154 gfs2_inplace_release(ip);
1156 gfs2_quota_unlock(ip);
1160 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1162 struct inode *inode = file_inode(file);
1163 struct gfs2_sbd *sdp = GFS2_SB(inode);
1164 struct gfs2_inode *ip = GFS2_I(inode);
1165 struct gfs2_holder gh;
1168 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1170 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1171 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1176 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1177 ret = gfs2_glock_nq(&gh);
1181 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1182 (offset + len) > inode->i_size) {
1183 ret = inode_newsize_ok(inode, offset + len);
1188 ret = get_write_access(inode);
1192 if (mode & FALLOC_FL_PUNCH_HOLE) {
1193 ret = __gfs2_punch_hole(file, offset, len);
1195 ret = __gfs2_fallocate(file, mode, offset, len);
1197 gfs2_rs_deltree(&ip->i_res);
1200 put_write_access(inode);
1204 gfs2_holder_uninit(&gh);
1205 inode_unlock(inode);
1209 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1210 struct file *out, loff_t *ppos,
1211 size_t len, unsigned int flags)
1215 gfs2_size_hint(out, *ppos, len);
1217 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1221 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1224 * gfs2_lock - acquire/release a posix lock on a file
1225 * @file: the file pointer
1226 * @cmd: either modify or retrieve lock state, possibly wait
1227 * @fl: type and range of lock
1232 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1234 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1235 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1236 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1238 if (!(fl->fl_flags & FL_POSIX))
1240 if (cmd == F_CANCELLK) {
1243 fl->fl_type = F_UNLCK;
1245 if (unlikely(gfs2_withdrawn(sdp))) {
1246 if (fl->fl_type == F_UNLCK)
1247 locks_lock_file_wait(file, fl);
1251 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1252 else if (fl->fl_type == F_UNLCK)
1253 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1255 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1258 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1260 struct gfs2_file *fp = file->private_data;
1261 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1262 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1263 struct gfs2_glock *gl;
1269 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1270 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1272 mutex_lock(&fp->f_fl_mutex);
1274 if (gfs2_holder_initialized(fl_gh)) {
1275 struct file_lock request;
1276 if (fl_gh->gh_state == state)
1278 locks_init_lock(&request);
1279 request.fl_type = F_UNLCK;
1280 request.fl_flags = FL_FLOCK;
1281 locks_lock_file_wait(file, &request);
1282 gfs2_glock_dq(fl_gh);
1283 gfs2_holder_reinit(state, flags, fl_gh);
1285 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1286 &gfs2_flock_glops, CREATE, &gl);
1289 gfs2_holder_init(gl, state, flags, fl_gh);
1292 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1293 error = gfs2_glock_nq(fl_gh);
1294 if (error != GLR_TRYFAILED)
1296 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1297 fl_gh->gh_error = 0;
1301 gfs2_holder_uninit(fl_gh);
1302 if (error == GLR_TRYFAILED)
1305 error = locks_lock_file_wait(file, fl);
1306 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1310 mutex_unlock(&fp->f_fl_mutex);
1314 static void do_unflock(struct file *file, struct file_lock *fl)
1316 struct gfs2_file *fp = file->private_data;
1317 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1319 mutex_lock(&fp->f_fl_mutex);
1320 locks_lock_file_wait(file, fl);
1321 if (gfs2_holder_initialized(fl_gh)) {
1322 gfs2_glock_dq(fl_gh);
1323 gfs2_holder_uninit(fl_gh);
1325 mutex_unlock(&fp->f_fl_mutex);
1329 * gfs2_flock - acquire/release a flock lock on a file
1330 * @file: the file pointer
1331 * @cmd: either modify or retrieve lock state, possibly wait
1332 * @fl: type and range of lock
1337 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1339 if (!(fl->fl_flags & FL_FLOCK))
1341 if (fl->fl_type & LOCK_MAND)
1344 if (fl->fl_type == F_UNLCK) {
1345 do_unflock(file, fl);
1348 return do_flock(file, cmd, fl);
1352 const struct file_operations gfs2_file_fops = {
1353 .llseek = gfs2_llseek,
1354 .read_iter = gfs2_file_read_iter,
1355 .write_iter = gfs2_file_write_iter,
1356 .iopoll = iomap_dio_iopoll,
1357 .unlocked_ioctl = gfs2_ioctl,
1358 .compat_ioctl = gfs2_compat_ioctl,
1361 .release = gfs2_release,
1362 .fsync = gfs2_fsync,
1364 .flock = gfs2_flock,
1365 .splice_read = generic_file_splice_read,
1366 .splice_write = gfs2_file_splice_write,
1367 .setlease = simple_nosetlease,
1368 .fallocate = gfs2_fallocate,
1371 const struct file_operations gfs2_dir_fops = {
1372 .iterate_shared = gfs2_readdir,
1373 .unlocked_ioctl = gfs2_ioctl,
1374 .compat_ioctl = gfs2_compat_ioctl,
1376 .release = gfs2_release,
1377 .fsync = gfs2_fsync,
1379 .flock = gfs2_flock,
1380 .llseek = default_llseek,
1383 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1385 const struct file_operations gfs2_file_fops_nolock = {
1386 .llseek = gfs2_llseek,
1387 .read_iter = gfs2_file_read_iter,
1388 .write_iter = gfs2_file_write_iter,
1389 .iopoll = iomap_dio_iopoll,
1390 .unlocked_ioctl = gfs2_ioctl,
1391 .compat_ioctl = gfs2_compat_ioctl,
1394 .release = gfs2_release,
1395 .fsync = gfs2_fsync,
1396 .splice_read = generic_file_splice_read,
1397 .splice_write = gfs2_file_splice_write,
1398 .setlease = generic_setlease,
1399 .fallocate = gfs2_fallocate,
1402 const struct file_operations gfs2_dir_fops_nolock = {
1403 .iterate_shared = gfs2_readdir,
1404 .unlocked_ioctl = gfs2_ioctl,
1405 .compat_ioctl = gfs2_compat_ioctl,
1407 .release = gfs2_release,
1408 .fsync = gfs2_fsync,
1409 .llseek = default_llseek,