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/mount.h>
8 #include <linux/namei.h>
9 #include <linux/writeback.h>
10 #include <linux/aio.h>
11 #include <linux/falloc.h>
14 #include "mds_client.h"
18 * Ceph file operations
20 * Implement basic open/close functionality, and implement
23 * We implement three modes of file I/O:
24 * - buffered uses the generic_file_aio_{read,write} helpers
26 * - synchronous is used when there is multi-client read/write
27 * sharing, avoids the page cache, and synchronously waits for an
30 * - direct io takes the variant of the sync path that references
31 * user pages directly.
33 * fsync() flushes and waits on dirty pages, but just queues metadata
34 * for writeback: since the MDS can recover size and mtime there is no
35 * need to wait for MDS acknowledgement.
40 * Prepare an open request. Preallocate ceph_cap to avoid an
41 * inopportune ENOMEM later.
43 static struct ceph_mds_request *
44 prepare_open_request(struct super_block *sb, int flags, int create_mode)
46 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
47 struct ceph_mds_client *mdsc = fsc->mdsc;
48 struct ceph_mds_request *req;
49 int want_auth = USE_ANY_MDS;
50 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
52 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
53 want_auth = USE_AUTH_MDS;
55 req = ceph_mdsc_create_request(mdsc, op, want_auth);
58 req->r_fmode = ceph_flags_to_mode(flags);
59 req->r_args.open.flags = cpu_to_le32(flags);
60 req->r_args.open.mode = cpu_to_le32(create_mode);
66 * initialize private struct file data.
67 * if we fail, clean up by dropping fmode reference on the ceph_inode
69 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
71 struct ceph_file_info *cf;
73 struct ceph_inode_info *ci = ceph_inode(inode);
74 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
75 struct ceph_mds_client *mdsc = fsc->mdsc;
77 switch (inode->i_mode & S_IFMT) {
79 /* First file open request creates the cookie, we want to keep
80 * this cookie around for the filetime of the inode as not to
81 * have to worry about fscache register / revoke / operation
84 * Also, if we know the operation is going to invalidate data
85 * (non readonly) just nuke the cache right away.
87 ceph_fscache_register_inode_cookie(mdsc->fsc, ci);
88 if ((fmode & CEPH_FILE_MODE_WR))
89 ceph_fscache_invalidate(inode);
91 dout("init_file %p %p 0%o (regular)\n", inode, file,
93 cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
95 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
100 file->private_data = cf;
101 BUG_ON(inode->i_fop->release != ceph_release);
105 dout("init_file %p %p 0%o (symlink)\n", inode, file,
107 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
111 dout("init_file %p %p 0%o (special)\n", inode, file,
114 * we need to drop the open ref now, since we don't
115 * have .release set to ceph_release.
117 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
118 BUG_ON(inode->i_fop->release == ceph_release);
120 /* call the proper open fop */
121 ret = inode->i_fop->open(inode, file);
127 * If we already have the requisite capabilities, we can satisfy
128 * the open request locally (no need to request new caps from the
129 * MDS). We do, however, need to inform the MDS (asynchronously)
130 * if our wanted caps set expands.
132 int ceph_open(struct inode *inode, struct file *file)
134 struct ceph_inode_info *ci = ceph_inode(inode);
135 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
136 struct ceph_mds_client *mdsc = fsc->mdsc;
137 struct ceph_mds_request *req;
138 struct ceph_file_info *cf = file->private_data;
139 struct inode *parent_inode = NULL;
141 int flags, fmode, wanted;
144 dout("open file %p is already opened\n", file);
148 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
149 flags = file->f_flags & ~(O_CREAT|O_EXCL);
150 if (S_ISDIR(inode->i_mode))
151 flags = O_DIRECTORY; /* mds likes to know */
153 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
154 ceph_vinop(inode), file, flags, file->f_flags);
155 fmode = ceph_flags_to_mode(flags);
156 wanted = ceph_caps_for_mode(fmode);
158 /* snapped files are read-only */
159 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
162 /* trivially open snapdir */
163 if (ceph_snap(inode) == CEPH_SNAPDIR) {
164 spin_lock(&ci->i_ceph_lock);
165 __ceph_get_fmode(ci, fmode);
166 spin_unlock(&ci->i_ceph_lock);
167 return ceph_init_file(inode, file, fmode);
171 * No need to block if we have caps on the auth MDS (for
172 * write) or any MDS (for read). Update wanted set
175 spin_lock(&ci->i_ceph_lock);
176 if (__ceph_is_any_real_caps(ci) &&
177 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
178 int mds_wanted = __ceph_caps_mds_wanted(ci);
179 int issued = __ceph_caps_issued(ci, NULL);
181 dout("open %p fmode %d want %s issued %s using existing\n",
182 inode, fmode, ceph_cap_string(wanted),
183 ceph_cap_string(issued));
184 __ceph_get_fmode(ci, fmode);
185 spin_unlock(&ci->i_ceph_lock);
188 if ((issued & wanted) != wanted &&
189 (mds_wanted & wanted) != wanted &&
190 ceph_snap(inode) != CEPH_SNAPDIR)
191 ceph_check_caps(ci, 0, NULL);
193 return ceph_init_file(inode, file, fmode);
194 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
195 (ci->i_snap_caps & wanted) == wanted) {
196 __ceph_get_fmode(ci, fmode);
197 spin_unlock(&ci->i_ceph_lock);
198 return ceph_init_file(inode, file, fmode);
201 spin_unlock(&ci->i_ceph_lock);
203 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
204 req = prepare_open_request(inode->i_sb, flags, 0);
209 req->r_inode = inode;
214 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
215 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
218 err = ceph_init_file(inode, file, req->r_fmode);
219 ceph_mdsc_put_request(req);
220 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
227 * Do a lookup + open with a single request. If we get a non-existent
228 * file or symlink, return 1 so the VFS can retry.
230 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
231 struct file *file, unsigned flags, umode_t mode,
234 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
235 struct ceph_mds_client *mdsc = fsc->mdsc;
236 struct ceph_mds_request *req;
240 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
241 dir, dentry, dentry->d_name.len, dentry->d_name.name,
242 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
244 if (dentry->d_name.len > NAME_MAX)
245 return -ENAMETOOLONG;
247 err = ceph_init_dentry(dentry);
252 req = prepare_open_request(dir->i_sb, flags, mode);
255 req->r_dentry = dget(dentry);
257 if (flags & O_CREAT) {
258 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
259 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
261 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
262 err = ceph_mdsc_do_request(mdsc,
263 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
268 err = ceph_handle_snapdir(req, dentry, err);
269 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
270 err = ceph_handle_notrace_create(dir, dentry);
272 if (d_unhashed(dentry)) {
273 dn = ceph_finish_lookup(req, dentry, err);
277 /* we were given a hashed negative dentry */
282 if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) {
283 /* make vfs retry on splice, ENOENT, or symlink */
284 dout("atomic_open finish_no_open on dn %p\n", dn);
285 err = finish_no_open(file, dn);
287 dout("atomic_open finish_open on dn %p\n", dn);
288 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
289 ceph_init_acl(dentry, dentry->d_inode, dir);
290 *opened |= FILE_CREATED;
292 err = finish_open(file, dentry, ceph_open, opened);
295 if (!req->r_err && req->r_target_inode)
296 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
297 ceph_mdsc_put_request(req);
298 dout("atomic_open result=%d\n", err);
302 int ceph_release(struct inode *inode, struct file *file)
304 struct ceph_inode_info *ci = ceph_inode(inode);
305 struct ceph_file_info *cf = file->private_data;
307 dout("release inode %p file %p\n", inode, file);
308 ceph_put_fmode(ci, cf->fmode);
309 if (cf->last_readdir)
310 ceph_mdsc_put_request(cf->last_readdir);
311 kfree(cf->last_name);
314 kmem_cache_free(ceph_file_cachep, cf);
316 /* wake up anyone waiting for caps on this inode */
317 wake_up_all(&ci->i_cap_wq);
322 * Read a range of bytes striped over one or more objects. Iterate over
323 * objects we stripe over. (That's not atomic, but good enough for now.)
325 * If we get a short result from the OSD, check against i_size; we need to
326 * only return a short read to the caller if we hit EOF.
328 static int striped_read(struct inode *inode,
330 struct page **pages, int num_pages,
331 int *checkeof, bool o_direct,
332 unsigned long buf_align)
334 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
335 struct ceph_inode_info *ci = ceph_inode(inode);
336 u64 pos, this_len, left;
337 int io_align, page_align;
340 struct page **page_pos;
342 bool hit_stripe, was_short;
345 * we may need to do multiple reads. not atomic, unfortunately.
350 pages_left = num_pages;
352 io_align = off & ~PAGE_MASK;
356 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
358 page_align = pos & ~PAGE_MASK;
360 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
361 &ci->i_layout, pos, &this_len,
364 page_pos, pages_left, page_align);
367 hit_stripe = this_len < left;
368 was_short = ret >= 0 && ret < this_len;
369 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
370 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
374 if (was_short && (pos + ret < inode->i_size)) {
375 u64 tmp = min(this_len - ret,
376 inode->i_size - pos - ret);
377 dout(" zero gap %llu to %llu\n",
378 pos + ret, pos + ret + tmp);
379 ceph_zero_page_vector_range(page_align + read + ret,
384 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
388 page_pos += didpages;
389 pages_left -= didpages;
391 /* hit stripe and need continue*/
392 if (left && hit_stripe && pos < inode->i_size)
398 /* did we bounce off eof? */
399 if (pos + left > inode->i_size)
403 dout("striped_read returns %d\n", ret);
408 * Completely synchronous read and write methods. Direct from __user
409 * buffer to osd, or directly to user pages (if O_DIRECT).
411 * If the read spans object boundary, just do multiple reads.
413 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
416 struct file *file = iocb->ki_filp;
417 struct inode *inode = file_inode(file);
419 u64 off = iocb->ki_pos;
421 size_t len = iov_iter_count(i);
423 dout("sync_read on file %p %llu~%u %s\n", file, off,
425 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
427 * flush any page cache pages in this range. this
428 * will make concurrent normal and sync io slow,
429 * but it will at least behave sensibly when they are
432 ret = filemap_write_and_wait_range(inode->i_mapping, off,
437 if (file->f_flags & O_DIRECT) {
438 while (iov_iter_count(i)) {
442 n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start);
446 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
448 ret = striped_read(inode, off, n,
449 pages, num_pages, checkeof,
452 ceph_put_page_vector(pages, num_pages, true);
457 iov_iter_advance(i, ret);
462 num_pages = calc_pages_for(off, len);
463 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
465 return PTR_ERR(pages);
466 ret = striped_read(inode, off, len, pages,
467 num_pages, checkeof, 0, 0);
473 int copy = min_t(size_t, PAGE_SIZE, left);
474 l = copy_page_to_iter(pages[k++], 0, copy, i);
481 ceph_release_page_vector(pages, num_pages);
484 if (off > iocb->ki_pos) {
485 ret = off - iocb->ki_pos;
489 dout("sync_read result %d\n", ret);
494 * Write commit request unsafe callback, called to tell us when a
495 * request is unsafe (that is, in flight--has been handed to the
496 * messenger to send to its target osd). It is called again when
497 * we've received a response message indicating the request is
498 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
499 * is completed early (and unsuccessfully) due to a timeout or
502 * This is used if we requested both an ACK and ONDISK commit reply
505 static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
507 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
509 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
512 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
513 spin_lock(&ci->i_unsafe_lock);
514 list_add_tail(&req->r_unsafe_item,
515 &ci->i_unsafe_writes);
516 spin_unlock(&ci->i_unsafe_lock);
518 spin_lock(&ci->i_unsafe_lock);
519 list_del_init(&req->r_unsafe_item);
520 spin_unlock(&ci->i_unsafe_lock);
521 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
527 * Synchronous write, straight from __user pointer or user pages.
529 * If write spans object boundary, just do multiple writes. (For a
530 * correct atomic write, we should e.g. take write locks on all
531 * objects, rollback on failure, etc.)
534 ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from)
536 struct file *file = iocb->ki_filp;
537 struct inode *inode = file_inode(file);
538 struct ceph_inode_info *ci = ceph_inode(inode);
539 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
540 struct ceph_snap_context *snapc;
541 struct ceph_vino vino;
542 struct ceph_osd_request *req;
549 struct timespec mtime = CURRENT_TIME;
550 loff_t pos = iocb->ki_pos;
551 size_t count = iov_iter_count(from);
553 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
556 dout("sync_direct_write on file %p %lld~%u\n", file, pos,
559 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
563 ret = invalidate_inode_pages2_range(inode->i_mapping,
564 pos >> PAGE_CACHE_SHIFT,
565 (pos + count) >> PAGE_CACHE_SHIFT);
567 dout("invalidate_inode_pages2_range returned %d\n", ret);
569 flags = CEPH_OSD_FLAG_ORDERSNAP |
570 CEPH_OSD_FLAG_ONDISK |
573 while (iov_iter_count(from) > 0) {
574 u64 len = iov_iter_single_seg_count(from);
578 snapc = ci->i_snap_realm->cached_context;
579 vino = ceph_vino(inode);
580 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
582 2,/*include a 'startsync' command*/
583 CEPH_OSD_OP_WRITE, flags, snapc,
592 n = iov_iter_get_pages_alloc(from, &pages, len, &start);
593 if (unlikely(n < 0)) {
595 ceph_osdc_put_request(req);
599 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
601 * throw out any page cache pages in this range. this
604 truncate_inode_pages_range(inode->i_mapping, pos,
605 (pos+n) | (PAGE_CACHE_SIZE-1));
606 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
609 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
610 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
612 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
614 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
616 ceph_put_page_vector(pages, num_pages, false);
618 ceph_osdc_put_request(req);
623 iov_iter_advance(from, n);
625 if (pos > i_size_read(inode)) {
626 check_caps = ceph_inode_set_size(inode, pos);
628 ceph_check_caps(ceph_inode(inode),
634 if (ret != -EOLDSNAPC && written > 0) {
643 * Synchronous write, straight from __user pointer or user pages.
645 * If write spans object boundary, just do multiple writes. (For a
646 * correct atomic write, we should e.g. take write locks on all
647 * objects, rollback on failure, etc.)
649 static ssize_t ceph_sync_write(struct kiocb *iocb, struct iov_iter *from)
651 struct file *file = iocb->ki_filp;
652 struct inode *inode = file_inode(file);
653 struct ceph_inode_info *ci = ceph_inode(inode);
654 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
655 struct ceph_snap_context *snapc;
656 struct ceph_vino vino;
657 struct ceph_osd_request *req;
665 struct timespec mtime = CURRENT_TIME;
666 loff_t pos = iocb->ki_pos;
667 size_t count = iov_iter_count(from);
669 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
672 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
674 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
678 ret = invalidate_inode_pages2_range(inode->i_mapping,
679 pos >> PAGE_CACHE_SHIFT,
680 (pos + count) >> PAGE_CACHE_SHIFT);
682 dout("invalidate_inode_pages2_range returned %d\n", ret);
684 flags = CEPH_OSD_FLAG_ORDERSNAP |
685 CEPH_OSD_FLAG_ONDISK |
686 CEPH_OSD_FLAG_WRITE |
689 while ((len = iov_iter_count(from)) > 0) {
693 snapc = ci->i_snap_realm->cached_context;
694 vino = ceph_vino(inode);
695 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
697 CEPH_OSD_OP_WRITE, flags, snapc,
707 * write from beginning of first page,
708 * regardless of io alignment
710 num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
712 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
714 ret = PTR_ERR(pages);
719 for (n = 0; n < num_pages; n++) {
720 size_t plen = min_t(size_t, left, PAGE_SIZE);
721 ret = copy_page_from_iter(pages[n], 0, plen, from);
730 ceph_release_page_vector(pages, num_pages);
734 /* get a second commit callback */
735 req->r_unsafe_callback = ceph_sync_write_unsafe;
736 req->r_inode = inode;
738 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
741 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
742 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
744 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
746 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
749 ceph_osdc_put_request(req);
754 if (pos > i_size_read(inode)) {
755 check_caps = ceph_inode_set_size(inode, pos);
757 ceph_check_caps(ceph_inode(inode),
765 if (ret != -EOLDSNAPC && written > 0) {
773 * Wrap generic_file_aio_read with checks for cap bits on the inode.
774 * Atomically grab references, so that those bits are not released
775 * back to the MDS mid-read.
777 * Hmm, the sync read case isn't actually async... should it be?
779 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
781 struct file *filp = iocb->ki_filp;
782 struct ceph_file_info *fi = filp->private_data;
783 size_t len = iocb->ki_nbytes;
784 struct inode *inode = file_inode(filp);
785 struct ceph_inode_info *ci = ceph_inode(inode);
788 int checkeof = 0, read = 0;
791 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
792 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
794 if (fi->fmode & CEPH_FILE_MODE_LAZY)
795 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
797 want = CEPH_CAP_FILE_CACHE;
798 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
802 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
803 (iocb->ki_filp->f_flags & O_DIRECT) ||
804 (fi->flags & CEPH_F_SYNC)) {
806 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
807 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
808 ceph_cap_string(got));
810 /* hmm, this isn't really async... */
811 ret = ceph_sync_read(iocb, to, &checkeof);
813 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
814 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
815 ceph_cap_string(got));
817 ret = generic_file_read_iter(iocb, to);
819 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
820 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
821 ceph_put_cap_refs(ci, got);
823 if (checkeof && ret >= 0) {
824 int statret = ceph_do_getattr(inode,
827 /* hit EOF or hole? */
828 if (statret == 0 && iocb->ki_pos < inode->i_size &&
830 dout("sync_read hit hole, ppos %lld < size %lld"
831 ", reading more\n", iocb->ki_pos,
834 iov_iter_advance(to, ret);
849 * Take cap references to avoid releasing caps to MDS mid-write.
851 * If we are synchronous, and write with an old snap context, the OSD
852 * may return EOLDSNAPC. In that case, retry the write.. _after_
853 * dropping our cap refs and allowing the pending snap to logically
854 * complete _before_ this write occurs.
856 * If we are near ENOSPC, write synchronously.
858 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
860 struct file *file = iocb->ki_filp;
861 struct ceph_file_info *fi = file->private_data;
862 struct inode *inode = file_inode(file);
863 struct ceph_inode_info *ci = ceph_inode(inode);
864 struct ceph_osd_client *osdc =
865 &ceph_sb_to_client(inode->i_sb)->client->osdc;
866 ssize_t count = iov_iter_count(from), written = 0;
868 loff_t pos = iocb->ki_pos;
870 if (ceph_snap(inode) != CEPH_NOSNAP)
873 mutex_lock(&inode->i_mutex);
875 /* We can write back this queue in page reclaim */
876 current->backing_dev_info = file->f_mapping->backing_dev_info;
878 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
884 iov_iter_truncate(from, count);
886 err = file_remove_suid(file);
890 err = file_update_time(file);
895 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
900 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
901 inode, ceph_vinop(inode), pos, count, inode->i_size);
902 if (fi->fmode & CEPH_FILE_MODE_LAZY)
903 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
905 want = CEPH_CAP_FILE_BUFFER;
907 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count);
911 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
912 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
914 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
915 (file->f_flags & O_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
916 struct iov_iter data;
917 mutex_unlock(&inode->i_mutex);
918 /* we might need to revert back to that point */
920 if (file->f_flags & O_DIRECT)
921 written = ceph_sync_direct_write(iocb, &data);
923 written = ceph_sync_write(iocb, &data);
924 if (written == -EOLDSNAPC) {
925 dout("aio_write %p %llx.%llx %llu~%u"
926 "got EOLDSNAPC, retrying\n",
927 inode, ceph_vinop(inode),
928 pos, (unsigned)count);
929 mutex_lock(&inode->i_mutex);
933 iov_iter_advance(from, written);
935 loff_t old_size = inode->i_size;
937 * No need to acquire the i_truncate_mutex. Because
938 * the MDS revokes Fwb caps before sending truncate
939 * message to us. We can't get Fwb cap while there
940 * are pending vmtruncate. So write and vmtruncate
941 * can not run at the same time
943 written = generic_perform_write(file, from, pos);
944 if (likely(written >= 0))
945 iocb->ki_pos = pos + written;
946 if (inode->i_size > old_size)
947 ceph_fscache_update_objectsize(inode);
948 mutex_unlock(&inode->i_mutex);
953 spin_lock(&ci->i_ceph_lock);
954 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
955 spin_unlock(&ci->i_ceph_lock);
957 __mark_inode_dirty(inode, dirty);
960 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
961 inode, ceph_vinop(inode), pos, (unsigned)count,
962 ceph_cap_string(got));
963 ceph_put_cap_refs(ci, got);
966 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
967 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
968 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
976 mutex_unlock(&inode->i_mutex);
978 current->backing_dev_info = NULL;
979 return written ? written : err;
983 * llseek. be sure to verify file size on SEEK_END.
985 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
987 struct inode *inode = file->f_mapping->host;
990 mutex_lock(&inode->i_mutex);
992 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
993 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
1002 offset += inode->i_size;
1006 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1007 * position-querying operation. Avoid rewriting the "same"
1008 * f_pos value back to the file because a concurrent read(),
1009 * write() or lseek() might have altered it
1012 offset = file->f_pos;
1015 offset += file->f_pos;
1018 if (offset >= inode->i_size) {
1024 if (offset >= inode->i_size) {
1028 offset = inode->i_size;
1032 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1035 mutex_unlock(&inode->i_mutex);
1039 static inline void ceph_zero_partial_page(
1040 struct inode *inode, loff_t offset, unsigned size)
1043 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1045 page = find_lock_page(inode->i_mapping, index);
1047 wait_on_page_writeback(page);
1048 zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1050 page_cache_release(page);
1054 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1057 loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1058 if (offset < nearly) {
1059 loff_t size = nearly - offset;
1062 ceph_zero_partial_page(inode, offset, size);
1066 if (length >= PAGE_CACHE_SIZE) {
1067 loff_t size = round_down(length, PAGE_CACHE_SIZE);
1068 truncate_pagecache_range(inode, offset, offset + size - 1);
1073 ceph_zero_partial_page(inode, offset, length);
1076 static int ceph_zero_partial_object(struct inode *inode,
1077 loff_t offset, loff_t *length)
1079 struct ceph_inode_info *ci = ceph_inode(inode);
1080 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1081 struct ceph_osd_request *req;
1087 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1090 op = CEPH_OSD_OP_ZERO;
1093 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1097 CEPH_OSD_FLAG_WRITE |
1098 CEPH_OSD_FLAG_ONDISK,
1105 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1108 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1110 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1114 ceph_osdc_put_request(req);
1120 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1123 struct ceph_inode_info *ci = ceph_inode(inode);
1124 s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1125 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1126 s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1127 u64 object_set_size = object_size * stripe_count;
1130 /* round offset up to next period boundary */
1131 nearly = offset + object_set_size - 1;
1133 nearly -= do_div(t, object_set_size);
1135 while (length && offset < nearly) {
1136 loff_t size = length;
1137 ret = ceph_zero_partial_object(inode, offset, &size);
1143 while (length >= object_set_size) {
1145 loff_t pos = offset;
1146 for (i = 0; i < stripe_count; ++i) {
1147 ret = ceph_zero_partial_object(inode, pos, NULL);
1152 offset += object_set_size;
1153 length -= object_set_size;
1156 loff_t size = length;
1157 ret = ceph_zero_partial_object(inode, offset, &size);
1166 static long ceph_fallocate(struct file *file, int mode,
1167 loff_t offset, loff_t length)
1169 struct ceph_file_info *fi = file->private_data;
1170 struct inode *inode = file_inode(file);
1171 struct ceph_inode_info *ci = ceph_inode(inode);
1172 struct ceph_osd_client *osdc =
1173 &ceph_inode_to_client(inode)->client->osdc;
1180 if (!S_ISREG(inode->i_mode))
1183 mutex_lock(&inode->i_mutex);
1185 if (ceph_snap(inode) != CEPH_NOSNAP) {
1190 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1191 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1196 size = i_size_read(inode);
1197 if (!(mode & FALLOC_FL_KEEP_SIZE))
1198 endoff = offset + length;
1200 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1201 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1203 want = CEPH_CAP_FILE_BUFFER;
1205 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
1209 if (mode & FALLOC_FL_PUNCH_HOLE) {
1211 ceph_zero_pagecache_range(inode, offset, length);
1212 ret = ceph_zero_objects(inode, offset, length);
1213 } else if (endoff > size) {
1214 truncate_pagecache_range(inode, size, -1);
1215 if (ceph_inode_set_size(inode, endoff))
1216 ceph_check_caps(ceph_inode(inode),
1217 CHECK_CAPS_AUTHONLY, NULL);
1221 spin_lock(&ci->i_ceph_lock);
1222 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1223 spin_unlock(&ci->i_ceph_lock);
1225 __mark_inode_dirty(inode, dirty);
1228 ceph_put_cap_refs(ci, got);
1230 mutex_unlock(&inode->i_mutex);
1234 const struct file_operations ceph_file_fops = {
1236 .release = ceph_release,
1237 .llseek = ceph_llseek,
1238 .read = new_sync_read,
1239 .write = new_sync_write,
1240 .read_iter = ceph_read_iter,
1241 .write_iter = ceph_write_iter,
1243 .fsync = ceph_fsync,
1245 .flock = ceph_flock,
1246 .splice_read = generic_file_splice_read,
1247 .splice_write = iter_file_splice_write,
1248 .unlocked_ioctl = ceph_ioctl,
1249 .compat_ioctl = ceph_ioctl,
1250 .fallocate = ceph_fallocate,