1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
4 #include <linux/sched.h>
5 #include <linux/slab.h>
6 #include <linux/file.h>
7 #include <linux/namei.h>
8 #include <linux/writeback.h>
11 #include "mds_client.h"
14 * Ceph file operations
16 * Implement basic open/close functionality, and implement
19 * We implement three modes of file I/O:
20 * - buffered uses the generic_file_aio_{read,write} helpers
22 * - synchronous is used when there is multi-client read/write
23 * sharing, avoids the page cache, and synchronously waits for an
26 * - direct io takes the variant of the sync path that references
27 * user pages directly.
29 * fsync() flushes and waits on dirty pages, but just queues metadata
30 * for writeback: since the MDS can recover size and mtime there is no
31 * need to wait for MDS acknowledgement.
36 * Prepare an open request. Preallocate ceph_cap to avoid an
37 * inopportune ENOMEM later.
39 static struct ceph_mds_request *
40 prepare_open_request(struct super_block *sb, int flags, int create_mode)
42 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
43 struct ceph_mds_client *mdsc = fsc->mdsc;
44 struct ceph_mds_request *req;
45 int want_auth = USE_ANY_MDS;
46 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
48 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
49 want_auth = USE_AUTH_MDS;
51 req = ceph_mdsc_create_request(mdsc, op, want_auth);
54 req->r_fmode = ceph_flags_to_mode(flags);
55 req->r_args.open.flags = cpu_to_le32(flags);
56 req->r_args.open.mode = cpu_to_le32(create_mode);
57 req->r_args.open.preferred = cpu_to_le32(-1);
63 * initialize private struct file data.
64 * if we fail, clean up by dropping fmode reference on the ceph_inode
66 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
68 struct ceph_file_info *cf;
71 switch (inode->i_mode & S_IFMT) {
74 dout("init_file %p %p 0%o (regular)\n", inode, file,
76 cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
78 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
83 file->private_data = cf;
84 BUG_ON(inode->i_fop->release != ceph_release);
88 dout("init_file %p %p 0%o (symlink)\n", inode, file,
90 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
94 dout("init_file %p %p 0%o (special)\n", inode, file,
97 * we need to drop the open ref now, since we don't
98 * have .release set to ceph_release.
100 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
101 BUG_ON(inode->i_fop->release == ceph_release);
103 /* call the proper open fop */
104 ret = inode->i_fop->open(inode, file);
110 * If the filp already has private_data, that means the file was
111 * already opened by intent during lookup, and we do nothing.
113 * If we already have the requisite capabilities, we can satisfy
114 * the open request locally (no need to request new caps from the
115 * MDS). We do, however, need to inform the MDS (asynchronously)
116 * if our wanted caps set expands.
118 int ceph_open(struct inode *inode, struct file *file)
120 struct ceph_inode_info *ci = ceph_inode(inode);
121 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
122 struct ceph_mds_client *mdsc = fsc->mdsc;
123 struct ceph_mds_request *req;
124 struct ceph_file_info *cf = file->private_data;
125 struct inode *parent_inode = NULL;
127 int flags, fmode, wanted;
130 dout("open file %p is already opened\n", file);
134 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
135 flags = file->f_flags & ~(O_CREAT|O_EXCL);
136 if (S_ISDIR(inode->i_mode))
137 flags = O_DIRECTORY; /* mds likes to know */
139 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
140 ceph_vinop(inode), file, flags, file->f_flags);
141 fmode = ceph_flags_to_mode(flags);
142 wanted = ceph_caps_for_mode(fmode);
144 /* snapped files are read-only */
145 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
148 /* trivially open snapdir */
149 if (ceph_snap(inode) == CEPH_SNAPDIR) {
150 spin_lock(&ci->i_ceph_lock);
151 __ceph_get_fmode(ci, fmode);
152 spin_unlock(&ci->i_ceph_lock);
153 return ceph_init_file(inode, file, fmode);
157 * No need to block if we have caps on the auth MDS (for
158 * write) or any MDS (for read). Update wanted set
161 spin_lock(&ci->i_ceph_lock);
162 if (__ceph_is_any_real_caps(ci) &&
163 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
164 int mds_wanted = __ceph_caps_mds_wanted(ci);
165 int issued = __ceph_caps_issued(ci, NULL);
167 dout("open %p fmode %d want %s issued %s using existing\n",
168 inode, fmode, ceph_cap_string(wanted),
169 ceph_cap_string(issued));
170 __ceph_get_fmode(ci, fmode);
171 spin_unlock(&ci->i_ceph_lock);
174 if ((issued & wanted) != wanted &&
175 (mds_wanted & wanted) != wanted &&
176 ceph_snap(inode) != CEPH_SNAPDIR)
177 ceph_check_caps(ci, 0, NULL);
179 return ceph_init_file(inode, file, fmode);
180 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
181 (ci->i_snap_caps & wanted) == wanted) {
182 __ceph_get_fmode(ci, fmode);
183 spin_unlock(&ci->i_ceph_lock);
184 return ceph_init_file(inode, file, fmode);
186 spin_unlock(&ci->i_ceph_lock);
188 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
189 req = prepare_open_request(inode->i_sb, flags, 0);
194 req->r_inode = inode;
197 if (flags & (O_CREAT|O_TRUNC))
198 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
199 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
202 err = ceph_init_file(inode, file, req->r_fmode);
203 ceph_mdsc_put_request(req);
204 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
211 * Do a lookup + open with a single request.
213 * If this succeeds, but some subsequent check in the vfs
214 * may_open() fails, the struct *file gets cleaned up (i.e.
215 * ceph_release gets called). So fear not!
219 * path_lookup_open -> LOOKUP_OPEN
220 * path_lookup_create -> LOOKUP_OPEN|LOOKUP_CREATE
222 struct dentry *ceph_lookup_open(struct inode *dir, struct dentry *dentry,
223 struct nameidata *nd, int mode,
226 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
227 struct ceph_mds_client *mdsc = fsc->mdsc;
229 struct ceph_mds_request *req;
232 int flags = nd->intent.open.flags;
234 dout("ceph_lookup_open dentry %p '%.*s' flags %d mode 0%o\n",
235 dentry, dentry->d_name.len, dentry->d_name.name, flags, mode);
238 req = prepare_open_request(dir->i_sb, flags, mode);
240 return ERR_CAST(req);
241 req->r_dentry = dget(dentry);
243 if (flags & O_CREAT) {
244 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
245 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
247 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
248 err = ceph_mdsc_do_request(mdsc,
249 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
251 err = ceph_handle_snapdir(req, dentry, err);
254 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
255 err = ceph_handle_notrace_create(dir, dentry);
258 file = lookup_instantiate_filp(nd, req->r_dentry, ceph_open);
262 ret = ceph_finish_lookup(req, dentry, err);
263 ceph_mdsc_put_request(req);
264 dout("ceph_lookup_open result=%p\n", ret);
268 int ceph_release(struct inode *inode, struct file *file)
270 struct ceph_inode_info *ci = ceph_inode(inode);
271 struct ceph_file_info *cf = file->private_data;
273 dout("release inode %p file %p\n", inode, file);
274 ceph_put_fmode(ci, cf->fmode);
275 if (cf->last_readdir)
276 ceph_mdsc_put_request(cf->last_readdir);
277 kfree(cf->last_name);
280 kmem_cache_free(ceph_file_cachep, cf);
282 /* wake up anyone waiting for caps on this inode */
283 wake_up_all(&ci->i_cap_wq);
288 * Read a range of bytes striped over one or more objects. Iterate over
289 * objects we stripe over. (That's not atomic, but good enough for now.)
291 * If we get a short result from the OSD, check against i_size; we need to
292 * only return a short read to the caller if we hit EOF.
294 static int striped_read(struct inode *inode,
296 struct page **pages, int num_pages,
297 int *checkeof, bool o_direct,
298 unsigned long buf_align)
300 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
301 struct ceph_inode_info *ci = ceph_inode(inode);
303 int io_align, page_align;
304 int left, pages_left;
306 struct page **page_pos;
308 bool hit_stripe, was_short;
311 * we may need to do multiple reads. not atomic, unfortunately.
316 pages_left = num_pages;
318 io_align = off & ~PAGE_MASK;
322 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
324 page_align = pos & ~PAGE_MASK;
326 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
327 &ci->i_layout, pos, &this_len,
330 page_pos, pages_left, page_align);
333 hit_stripe = this_len < left;
334 was_short = ret >= 0 && ret < this_len;
335 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read,
336 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
339 int didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
341 if (read < pos - off) {
342 dout(" zero gap %llu to %llu\n", off + read, pos);
343 ceph_zero_page_vector_range(page_align + read,
344 pos - off - read, pages);
349 page_pos += didpages;
350 pages_left -= didpages;
353 if (left && hit_stripe)
358 /* did we bounce off eof? */
359 if (pos + left > inode->i_size)
362 /* zero trailing bytes (inside i_size) */
363 if (left > 0 && pos < inode->i_size) {
364 if (pos + left > inode->i_size)
365 left = inode->i_size - pos;
367 dout("zero tail %d\n", left);
368 ceph_zero_page_vector_range(page_align + read, left,
376 dout("striped_read returns %d\n", ret);
381 * Completely synchronous read and write methods. Direct from __user
382 * buffer to osd, or directly to user pages (if O_DIRECT).
384 * If the read spans object boundary, just do multiple reads.
386 static ssize_t ceph_sync_read(struct file *file, char __user *data,
387 unsigned len, loff_t *poff, int *checkeof)
389 struct inode *inode = file->f_dentry->d_inode;
394 dout("sync_read on file %p %llu~%u %s\n", file, off, len,
395 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
397 if (file->f_flags & O_DIRECT) {
398 num_pages = calc_pages_for((unsigned long)data, len);
399 pages = ceph_get_direct_page_vector(data, num_pages, true);
401 num_pages = calc_pages_for(off, len);
402 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
405 return PTR_ERR(pages);
408 * flush any page cache pages in this range. this
409 * will make concurrent normal and sync io slow,
410 * but it will at least behave sensibly when they are
413 ret = filemap_write_and_wait(inode->i_mapping);
417 ret = striped_read(inode, off, len, pages, num_pages, checkeof,
418 file->f_flags & O_DIRECT,
419 (unsigned long)data & ~PAGE_MASK);
421 if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
422 ret = ceph_copy_page_vector_to_user(pages, data, off, ret);
427 if (file->f_flags & O_DIRECT)
428 ceph_put_page_vector(pages, num_pages, true);
430 ceph_release_page_vector(pages, num_pages);
431 dout("sync_read result %d\n", ret);
436 * Write commit callback, called if we requested both an ACK and
437 * ONDISK commit reply from the OSD.
439 static void sync_write_commit(struct ceph_osd_request *req,
440 struct ceph_msg *msg)
442 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
444 dout("sync_write_commit %p tid %llu\n", req, req->r_tid);
445 spin_lock(&ci->i_unsafe_lock);
446 list_del_init(&req->r_unsafe_item);
447 spin_unlock(&ci->i_unsafe_lock);
448 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
452 * Synchronous write, straight from __user pointer or user pages (if
455 * If write spans object boundary, just do multiple writes. (For a
456 * correct atomic write, we should e.g. take write locks on all
457 * objects, rollback on failure, etc.)
459 static ssize_t ceph_sync_write(struct file *file, const char __user *data,
460 size_t left, loff_t *offset)
462 struct inode *inode = file->f_dentry->d_inode;
463 struct ceph_inode_info *ci = ceph_inode(inode);
464 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
465 struct ceph_osd_request *req;
468 long long unsigned pos;
474 int page_align, io_align;
475 unsigned long buf_align;
477 struct timespec mtime = CURRENT_TIME;
479 if (ceph_snap(file->f_dentry->d_inode) != CEPH_NOSNAP)
482 dout("sync_write on file %p %lld~%u %s\n", file, *offset,
483 (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
485 if (file->f_flags & O_APPEND)
486 pos = i_size_read(inode);
490 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left);
494 ret = invalidate_inode_pages2_range(inode->i_mapping,
495 pos >> PAGE_CACHE_SHIFT,
496 (pos + left) >> PAGE_CACHE_SHIFT);
498 dout("invalidate_inode_pages2_range returned %d\n", ret);
500 flags = CEPH_OSD_FLAG_ORDERSNAP |
501 CEPH_OSD_FLAG_ONDISK |
503 if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0)
504 flags |= CEPH_OSD_FLAG_ACK;
509 * we may need to do multiple writes here if we span an object
510 * boundary. this isn't atomic, unfortunately. :(
513 io_align = pos & ~PAGE_MASK;
514 buf_align = (unsigned long)data & ~PAGE_MASK;
516 if (file->f_flags & O_DIRECT) {
517 /* write from beginning of first page, regardless of
519 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
520 num_pages = calc_pages_for((unsigned long)data, len);
522 page_align = pos & ~PAGE_MASK;
523 num_pages = calc_pages_for(pos, len);
525 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
526 ceph_vino(inode), pos, &len,
527 CEPH_OSD_OP_WRITE, flags,
528 ci->i_snap_realm->cached_context,
530 ci->i_truncate_seq, ci->i_truncate_size,
531 &mtime, false, 2, page_align);
535 if (file->f_flags & O_DIRECT) {
536 pages = ceph_get_direct_page_vector(data, num_pages, false);
538 ret = PTR_ERR(pages);
543 * throw out any page cache pages in this range. this
546 truncate_inode_pages_range(inode->i_mapping, pos,
547 (pos+len) | (PAGE_CACHE_SIZE-1));
549 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
551 ret = PTR_ERR(pages);
554 ret = ceph_copy_user_to_page_vector(pages, data, pos, len);
556 ceph_release_page_vector(pages, num_pages);
560 if ((file->f_flags & O_SYNC) == 0) {
561 /* get a second commit callback */
562 req->r_safe_callback = sync_write_commit;
563 req->r_own_pages = 1;
566 req->r_pages = pages;
567 req->r_num_pages = num_pages;
568 req->r_inode = inode;
570 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
572 if (req->r_safe_callback) {
574 * Add to inode unsafe list only after we
575 * start_request so that a tid has been assigned.
577 spin_lock(&ci->i_unsafe_lock);
578 list_add_tail(&req->r_unsafe_item,
579 &ci->i_unsafe_writes);
580 spin_unlock(&ci->i_unsafe_lock);
581 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
584 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
585 if (ret < 0 && req->r_safe_callback) {
586 spin_lock(&ci->i_unsafe_lock);
587 list_del_init(&req->r_unsafe_item);
588 spin_unlock(&ci->i_unsafe_lock);
589 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
593 if (file->f_flags & O_DIRECT)
594 ceph_put_page_vector(pages, num_pages, false);
595 else if (file->f_flags & O_SYNC)
596 ceph_release_page_vector(pages, num_pages);
599 ceph_osdc_put_request(req);
610 if (pos > i_size_read(inode))
611 check_caps = ceph_inode_set_size(inode, pos);
613 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY,
620 * Wrap generic_file_aio_read with checks for cap bits on the inode.
621 * Atomically grab references, so that those bits are not released
622 * back to the MDS mid-read.
624 * Hmm, the sync read case isn't actually async... should it be?
626 static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
627 unsigned long nr_segs, loff_t pos)
629 struct file *filp = iocb->ki_filp;
630 struct ceph_file_info *fi = filp->private_data;
631 loff_t *ppos = &iocb->ki_pos;
632 size_t len = iov->iov_len;
633 struct inode *inode = filp->f_dentry->d_inode;
634 struct ceph_inode_info *ci = ceph_inode(inode);
635 void __user *base = iov->iov_base;
638 int checkeof = 0, read = 0;
640 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
641 inode, ceph_vinop(inode), pos, (unsigned)len, inode);
643 __ceph_do_pending_vmtruncate(inode);
644 if (fi->fmode & CEPH_FILE_MODE_LAZY)
645 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
647 want = CEPH_CAP_FILE_CACHE;
648 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
651 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
652 inode, ceph_vinop(inode), pos, (unsigned)len,
653 ceph_cap_string(got));
655 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
656 (iocb->ki_filp->f_flags & O_DIRECT) ||
657 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
658 (fi->flags & CEPH_F_SYNC))
659 /* hmm, this isn't really async... */
660 ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
662 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
665 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
666 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
667 ceph_put_cap_refs(ci, got);
669 if (checkeof && ret >= 0) {
670 int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
672 /* hit EOF or hole? */
673 if (statret == 0 && *ppos < inode->i_size) {
674 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size);
689 * Take cap references to avoid releasing caps to MDS mid-write.
691 * If we are synchronous, and write with an old snap context, the OSD
692 * may return EOLDSNAPC. In that case, retry the write.. _after_
693 * dropping our cap refs and allowing the pending snap to logically
694 * complete _before_ this write occurs.
696 * If we are near ENOSPC, write synchronously.
698 static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov,
699 unsigned long nr_segs, loff_t pos)
701 struct file *file = iocb->ki_filp;
702 struct ceph_file_info *fi = file->private_data;
703 struct inode *inode = file->f_dentry->d_inode;
704 struct ceph_inode_info *ci = ceph_inode(inode);
705 struct ceph_osd_client *osdc =
706 &ceph_sb_to_client(inode->i_sb)->client->osdc;
707 loff_t endoff = pos + iov->iov_len;
711 if (ceph_snap(inode) != CEPH_NOSNAP)
715 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
717 __ceph_do_pending_vmtruncate(inode);
718 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
719 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
721 if (fi->fmode & CEPH_FILE_MODE_LAZY)
722 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
724 want = CEPH_CAP_FILE_BUFFER;
725 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
729 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
730 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
731 ceph_cap_string(got));
733 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
734 (iocb->ki_filp->f_flags & O_DIRECT) ||
735 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
736 (fi->flags & CEPH_F_SYNC)) {
737 ret = ceph_sync_write(file, iov->iov_base, iov->iov_len,
741 * buffered write; drop Fw early to avoid slow
742 * revocation if we get stuck on balance_dirty_pages
746 spin_lock(&ci->i_ceph_lock);
747 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
748 spin_unlock(&ci->i_ceph_lock);
749 ceph_put_cap_refs(ci, got);
751 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
752 if ((ret >= 0 || ret == -EIOCBQUEUED) &&
753 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host)
754 || ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
755 err = vfs_fsync_range(file, pos, pos + ret - 1, 1);
761 __mark_inode_dirty(inode, dirty);
767 spin_lock(&ci->i_ceph_lock);
768 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
769 spin_unlock(&ci->i_ceph_lock);
771 __mark_inode_dirty(inode, dirty);
775 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
776 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
777 ceph_cap_string(got));
778 ceph_put_cap_refs(ci, got);
781 if (ret == -EOLDSNAPC) {
782 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
783 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);
791 * llseek. be sure to verify file size on SEEK_END.
793 static loff_t ceph_llseek(struct file *file, loff_t offset, int origin)
795 struct inode *inode = file->f_mapping->host;
798 mutex_lock(&inode->i_mutex);
799 __ceph_do_pending_vmtruncate(inode);
801 if (origin == SEEK_END || origin == SEEK_DATA || origin == SEEK_HOLE) {
802 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
811 offset += inode->i_size;
815 * Here we special-case the lseek(fd, 0, SEEK_CUR)
816 * position-querying operation. Avoid rewriting the "same"
817 * f_pos value back to the file because a concurrent read(),
818 * write() or lseek() might have altered it
821 offset = file->f_pos;
824 offset += file->f_pos;
827 if (offset >= inode->i_size) {
833 if (offset >= inode->i_size) {
837 offset = inode->i_size;
841 if (offset < 0 || offset > inode->i_sb->s_maxbytes) {
846 /* Special lock needed here? */
847 if (offset != file->f_pos) {
848 file->f_pos = offset;
853 mutex_unlock(&inode->i_mutex);
857 const struct file_operations ceph_file_fops = {
859 .release = ceph_release,
860 .llseek = ceph_llseek,
861 .read = do_sync_read,
862 .write = do_sync_write,
863 .aio_read = ceph_aio_read,
864 .aio_write = ceph_aio_write,
869 .splice_read = generic_file_splice_read,
870 .splice_write = generic_file_splice_write,
871 .unlocked_ioctl = ceph_ioctl,
872 .compat_ioctl = ceph_ioctl,