xref: /linux/fs/ceph/file.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
1 #include <linux/ceph/ceph_debug.h>
2 
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/falloc.h>
11 
12 #include "super.h"
13 #include "mds_client.h"
14 #include "cache.h"
15 
16 /*
17  * Ceph file operations
18  *
19  * Implement basic open/close functionality, and implement
20  * read/write.
21  *
22  * We implement three modes of file I/O:
23  *  - buffered uses the generic_file_aio_{read,write} helpers
24  *
25  *  - synchronous is used when there is multi-client read/write
26  *    sharing, avoids the page cache, and synchronously waits for an
27  *    ack from the OSD.
28  *
29  *  - direct io takes the variant of the sync path that references
30  *    user pages directly.
31  *
32  * fsync() flushes and waits on dirty pages, but just queues metadata
33  * for writeback: since the MDS can recover size and mtime there is no
34  * need to wait for MDS acknowledgement.
35  */
36 
37 
38 /*
39  * Prepare an open request.  Preallocate ceph_cap to avoid an
40  * inopportune ENOMEM later.
41  */
42 static struct ceph_mds_request *
43 prepare_open_request(struct super_block *sb, int flags, int create_mode)
44 {
45 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
46 	struct ceph_mds_client *mdsc = fsc->mdsc;
47 	struct ceph_mds_request *req;
48 	int want_auth = USE_ANY_MDS;
49 	int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
50 
51 	if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
52 		want_auth = USE_AUTH_MDS;
53 
54 	req = ceph_mdsc_create_request(mdsc, op, want_auth);
55 	if (IS_ERR(req))
56 		goto out;
57 	req->r_fmode = ceph_flags_to_mode(flags);
58 	req->r_args.open.flags = cpu_to_le32(flags);
59 	req->r_args.open.mode = cpu_to_le32(create_mode);
60 out:
61 	return req;
62 }
63 
64 /*
65  * initialize private struct file data.
66  * if we fail, clean up by dropping fmode reference on the ceph_inode
67  */
68 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
69 {
70 	struct ceph_file_info *cf;
71 	int ret = 0;
72 	struct ceph_inode_info *ci = ceph_inode(inode);
73 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
74 	struct ceph_mds_client *mdsc = fsc->mdsc;
75 
76 	switch (inode->i_mode & S_IFMT) {
77 	case S_IFREG:
78 		/* First file open request creates the cookie, we want to keep
79 		 * this cookie around for the filetime of the inode as not to
80 		 * have to worry about fscache register / revoke / operation
81 		 * races.
82 		 *
83 		 * Also, if we know the operation is going to invalidate data
84 		 * (non readonly) just nuke the cache right away.
85 		 */
86 		ceph_fscache_register_inode_cookie(mdsc->fsc, ci);
87 		if ((fmode & CEPH_FILE_MODE_WR))
88 			ceph_fscache_invalidate(inode);
89 	case S_IFDIR:
90 		dout("init_file %p %p 0%o (regular)\n", inode, file,
91 		     inode->i_mode);
92 		cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
93 		if (cf == NULL) {
94 			ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
95 			return -ENOMEM;
96 		}
97 		cf->fmode = fmode;
98 		cf->next_offset = 2;
99 		file->private_data = cf;
100 		BUG_ON(inode->i_fop->release != ceph_release);
101 		break;
102 
103 	case S_IFLNK:
104 		dout("init_file %p %p 0%o (symlink)\n", inode, file,
105 		     inode->i_mode);
106 		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
107 		break;
108 
109 	default:
110 		dout("init_file %p %p 0%o (special)\n", inode, file,
111 		     inode->i_mode);
112 		/*
113 		 * we need to drop the open ref now, since we don't
114 		 * have .release set to ceph_release.
115 		 */
116 		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
117 		BUG_ON(inode->i_fop->release == ceph_release);
118 
119 		/* call the proper open fop */
120 		ret = inode->i_fop->open(inode, file);
121 	}
122 	return ret;
123 }
124 
125 /*
126  * If we already have the requisite capabilities, we can satisfy
127  * the open request locally (no need to request new caps from the
128  * MDS).  We do, however, need to inform the MDS (asynchronously)
129  * if our wanted caps set expands.
130  */
131 int ceph_open(struct inode *inode, struct file *file)
132 {
133 	struct ceph_inode_info *ci = ceph_inode(inode);
134 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
135 	struct ceph_mds_client *mdsc = fsc->mdsc;
136 	struct ceph_mds_request *req;
137 	struct ceph_file_info *cf = file->private_data;
138 	struct inode *parent_inode = NULL;
139 	int err;
140 	int flags, fmode, wanted;
141 
142 	if (cf) {
143 		dout("open file %p is already opened\n", file);
144 		return 0;
145 	}
146 
147 	/* filter out O_CREAT|O_EXCL; vfs did that already.  yuck. */
148 	flags = file->f_flags & ~(O_CREAT|O_EXCL);
149 	if (S_ISDIR(inode->i_mode))
150 		flags = O_DIRECTORY;  /* mds likes to know */
151 
152 	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
153 	     ceph_vinop(inode), file, flags, file->f_flags);
154 	fmode = ceph_flags_to_mode(flags);
155 	wanted = ceph_caps_for_mode(fmode);
156 
157 	/* snapped files are read-only */
158 	if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
159 		return -EROFS;
160 
161 	/* trivially open snapdir */
162 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
163 		spin_lock(&ci->i_ceph_lock);
164 		__ceph_get_fmode(ci, fmode);
165 		spin_unlock(&ci->i_ceph_lock);
166 		return ceph_init_file(inode, file, fmode);
167 	}
168 
169 	/*
170 	 * No need to block if we have caps on the auth MDS (for
171 	 * write) or any MDS (for read).  Update wanted set
172 	 * asynchronously.
173 	 */
174 	spin_lock(&ci->i_ceph_lock);
175 	if (__ceph_is_any_real_caps(ci) &&
176 	    (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
177 		int mds_wanted = __ceph_caps_mds_wanted(ci);
178 		int issued = __ceph_caps_issued(ci, NULL);
179 
180 		dout("open %p fmode %d want %s issued %s using existing\n",
181 		     inode, fmode, ceph_cap_string(wanted),
182 		     ceph_cap_string(issued));
183 		__ceph_get_fmode(ci, fmode);
184 		spin_unlock(&ci->i_ceph_lock);
185 
186 		/* adjust wanted? */
187 		if ((issued & wanted) != wanted &&
188 		    (mds_wanted & wanted) != wanted &&
189 		    ceph_snap(inode) != CEPH_SNAPDIR)
190 			ceph_check_caps(ci, 0, NULL);
191 
192 		return ceph_init_file(inode, file, fmode);
193 	} else if (ceph_snap(inode) != CEPH_NOSNAP &&
194 		   (ci->i_snap_caps & wanted) == wanted) {
195 		__ceph_get_fmode(ci, fmode);
196 		spin_unlock(&ci->i_ceph_lock);
197 		return ceph_init_file(inode, file, fmode);
198 	}
199 
200 	spin_unlock(&ci->i_ceph_lock);
201 
202 	dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
203 	req = prepare_open_request(inode->i_sb, flags, 0);
204 	if (IS_ERR(req)) {
205 		err = PTR_ERR(req);
206 		goto out;
207 	}
208 	req->r_inode = inode;
209 	ihold(inode);
210 
211 	req->r_num_caps = 1;
212 	if (flags & O_CREAT)
213 		parent_inode = ceph_get_dentry_parent_inode(file->f_path.dentry);
214 	err = ceph_mdsc_do_request(mdsc, parent_inode, req);
215 	iput(parent_inode);
216 	if (!err)
217 		err = ceph_init_file(inode, file, req->r_fmode);
218 	ceph_mdsc_put_request(req);
219 	dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
220 out:
221 	return err;
222 }
223 
224 
225 /*
226  * Do a lookup + open with a single request.  If we get a non-existent
227  * file or symlink, return 1 so the VFS can retry.
228  */
229 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
230 		     struct file *file, unsigned flags, umode_t mode,
231 		     int *opened)
232 {
233 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
234 	struct ceph_mds_client *mdsc = fsc->mdsc;
235 	struct ceph_mds_request *req;
236 	struct dentry *dn;
237 	struct ceph_acls_info acls = {};
238 	int err;
239 
240 	dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
241 	     dir, dentry, dentry,
242 	     d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
243 
244 	if (dentry->d_name.len > NAME_MAX)
245 		return -ENAMETOOLONG;
246 
247 	err = ceph_init_dentry(dentry);
248 	if (err < 0)
249 		return err;
250 
251 	if (flags & O_CREAT) {
252 		err = ceph_pre_init_acls(dir, &mode, &acls);
253 		if (err < 0)
254 			return err;
255 	}
256 
257 	/* do the open */
258 	req = prepare_open_request(dir->i_sb, flags, mode);
259 	if (IS_ERR(req)) {
260 		err = PTR_ERR(req);
261 		goto out_acl;
262 	}
263 	req->r_dentry = dget(dentry);
264 	req->r_num_caps = 2;
265 	if (flags & O_CREAT) {
266 		req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
267 		req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
268 		if (acls.pagelist) {
269 			req->r_pagelist = acls.pagelist;
270 			acls.pagelist = NULL;
271 		}
272 	}
273 	req->r_locked_dir = dir;           /* caller holds dir->i_mutex */
274 	err = ceph_mdsc_do_request(mdsc,
275 				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
276 				   req);
277 	err = ceph_handle_snapdir(req, dentry, err);
278 	if (err)
279 		goto out_req;
280 
281 	if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
282 		err = ceph_handle_notrace_create(dir, dentry);
283 
284 	if (d_unhashed(dentry)) {
285 		dn = ceph_finish_lookup(req, dentry, err);
286 		if (IS_ERR(dn))
287 			err = PTR_ERR(dn);
288 	} else {
289 		/* we were given a hashed negative dentry */
290 		dn = NULL;
291 	}
292 	if (err)
293 		goto out_req;
294 	if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
295 		/* make vfs retry on splice, ENOENT, or symlink */
296 		dout("atomic_open finish_no_open on dn %p\n", dn);
297 		err = finish_no_open(file, dn);
298 	} else {
299 		dout("atomic_open finish_open on dn %p\n", dn);
300 		if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
301 			ceph_init_inode_acls(d_inode(dentry), &acls);
302 			*opened |= FILE_CREATED;
303 		}
304 		err = finish_open(file, dentry, ceph_open, opened);
305 	}
306 out_req:
307 	if (!req->r_err && req->r_target_inode)
308 		ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
309 	ceph_mdsc_put_request(req);
310 out_acl:
311 	ceph_release_acls_info(&acls);
312 	dout("atomic_open result=%d\n", err);
313 	return err;
314 }
315 
316 int ceph_release(struct inode *inode, struct file *file)
317 {
318 	struct ceph_inode_info *ci = ceph_inode(inode);
319 	struct ceph_file_info *cf = file->private_data;
320 
321 	dout("release inode %p file %p\n", inode, file);
322 	ceph_put_fmode(ci, cf->fmode);
323 	if (cf->last_readdir)
324 		ceph_mdsc_put_request(cf->last_readdir);
325 	kfree(cf->last_name);
326 	kfree(cf->dir_info);
327 	dput(cf->dentry);
328 	kmem_cache_free(ceph_file_cachep, cf);
329 
330 	/* wake up anyone waiting for caps on this inode */
331 	wake_up_all(&ci->i_cap_wq);
332 	return 0;
333 }
334 
335 enum {
336 	CHECK_EOF = 1,
337 	READ_INLINE = 2,
338 };
339 
340 /*
341  * Read a range of bytes striped over one or more objects.  Iterate over
342  * objects we stripe over.  (That's not atomic, but good enough for now.)
343  *
344  * If we get a short result from the OSD, check against i_size; we need to
345  * only return a short read to the caller if we hit EOF.
346  */
347 static int striped_read(struct inode *inode,
348 			u64 off, u64 len,
349 			struct page **pages, int num_pages,
350 			int *checkeof, bool o_direct,
351 			unsigned long buf_align)
352 {
353 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
354 	struct ceph_inode_info *ci = ceph_inode(inode);
355 	u64 pos, this_len, left;
356 	int io_align, page_align;
357 	int pages_left;
358 	int read;
359 	struct page **page_pos;
360 	int ret;
361 	bool hit_stripe, was_short;
362 
363 	/*
364 	 * we may need to do multiple reads.  not atomic, unfortunately.
365 	 */
366 	pos = off;
367 	left = len;
368 	page_pos = pages;
369 	pages_left = num_pages;
370 	read = 0;
371 	io_align = off & ~PAGE_MASK;
372 
373 more:
374 	if (o_direct)
375 		page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
376 	else
377 		page_align = pos & ~PAGE_MASK;
378 	this_len = left;
379 	ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
380 				  &ci->i_layout, pos, &this_len,
381 				  ci->i_truncate_seq,
382 				  ci->i_truncate_size,
383 				  page_pos, pages_left, page_align);
384 	if (ret == -ENOENT)
385 		ret = 0;
386 	hit_stripe = this_len < left;
387 	was_short = ret >= 0 && ret < this_len;
388 	dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
389 	     ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
390 
391 	if (ret >= 0) {
392 		int didpages;
393 		if (was_short && (pos + ret < inode->i_size)) {
394 			int zlen = min(this_len - ret,
395 				       inode->i_size - pos - ret);
396 			int zoff = (o_direct ? buf_align : io_align) +
397 				    read + ret;
398 			dout(" zero gap %llu to %llu\n",
399 				pos + ret, pos + ret + zlen);
400 			ceph_zero_page_vector_range(zoff, zlen, pages);
401 			ret += zlen;
402 		}
403 
404 		didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
405 		pos += ret;
406 		read = pos - off;
407 		left -= ret;
408 		page_pos += didpages;
409 		pages_left -= didpages;
410 
411 		/* hit stripe and need continue*/
412 		if (left && hit_stripe && pos < inode->i_size)
413 			goto more;
414 	}
415 
416 	if (read > 0) {
417 		ret = read;
418 		/* did we bounce off eof? */
419 		if (pos + left > inode->i_size)
420 			*checkeof = CHECK_EOF;
421 	}
422 
423 	dout("striped_read returns %d\n", ret);
424 	return ret;
425 }
426 
427 /*
428  * Completely synchronous read and write methods.  Direct from __user
429  * buffer to osd, or directly to user pages (if O_DIRECT).
430  *
431  * If the read spans object boundary, just do multiple reads.
432  */
433 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
434 				int *checkeof)
435 {
436 	struct file *file = iocb->ki_filp;
437 	struct inode *inode = file_inode(file);
438 	struct page **pages;
439 	u64 off = iocb->ki_pos;
440 	int num_pages, ret;
441 	size_t len = iov_iter_count(i);
442 
443 	dout("sync_read on file %p %llu~%u %s\n", file, off,
444 	     (unsigned)len,
445 	     (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
446 
447 	if (!len)
448 		return 0;
449 	/*
450 	 * flush any page cache pages in this range.  this
451 	 * will make concurrent normal and sync io slow,
452 	 * but it will at least behave sensibly when they are
453 	 * in sequence.
454 	 */
455 	ret = filemap_write_and_wait_range(inode->i_mapping, off,
456 						off + len);
457 	if (ret < 0)
458 		return ret;
459 
460 	if (iocb->ki_flags & IOCB_DIRECT) {
461 		while (iov_iter_count(i)) {
462 			size_t start;
463 			ssize_t n;
464 
465 			n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start);
466 			if (n < 0)
467 				return n;
468 
469 			num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
470 
471 			ret = striped_read(inode, off, n,
472 					   pages, num_pages, checkeof,
473 					   1, start);
474 
475 			ceph_put_page_vector(pages, num_pages, true);
476 
477 			if (ret <= 0)
478 				break;
479 			off += ret;
480 			iov_iter_advance(i, ret);
481 			if (ret < n)
482 				break;
483 		}
484 	} else {
485 		num_pages = calc_pages_for(off, len);
486 		pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
487 		if (IS_ERR(pages))
488 			return PTR_ERR(pages);
489 		ret = striped_read(inode, off, len, pages,
490 					num_pages, checkeof, 0, 0);
491 		if (ret > 0) {
492 			int l, k = 0;
493 			size_t left = ret;
494 
495 			while (left) {
496 				size_t page_off = off & ~PAGE_MASK;
497 				size_t copy = min_t(size_t,
498 						    PAGE_SIZE - page_off, left);
499 				l = copy_page_to_iter(pages[k++], page_off,
500 						      copy, i);
501 				off += l;
502 				left -= l;
503 				if (l < copy)
504 					break;
505 			}
506 		}
507 		ceph_release_page_vector(pages, num_pages);
508 	}
509 
510 	if (off > iocb->ki_pos) {
511 		ret = off - iocb->ki_pos;
512 		iocb->ki_pos = off;
513 	}
514 
515 	dout("sync_read result %d\n", ret);
516 	return ret;
517 }
518 
519 /*
520  * Write commit request unsafe callback, called to tell us when a
521  * request is unsafe (that is, in flight--has been handed to the
522  * messenger to send to its target osd).  It is called again when
523  * we've received a response message indicating the request is
524  * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
525  * is completed early (and unsuccessfully) due to a timeout or
526  * interrupt.
527  *
528  * This is used if we requested both an ACK and ONDISK commit reply
529  * from the OSD.
530  */
531 static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
532 {
533 	struct ceph_inode_info *ci = ceph_inode(req->r_inode);
534 
535 	dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
536 		unsafe ? "un" : "");
537 	if (unsafe) {
538 		ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
539 		spin_lock(&ci->i_unsafe_lock);
540 		list_add_tail(&req->r_unsafe_item,
541 			      &ci->i_unsafe_writes);
542 		spin_unlock(&ci->i_unsafe_lock);
543 	} else {
544 		spin_lock(&ci->i_unsafe_lock);
545 		list_del_init(&req->r_unsafe_item);
546 		spin_unlock(&ci->i_unsafe_lock);
547 		ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
548 	}
549 }
550 
551 
552 /*
553  * Synchronous write, straight from __user pointer or user pages.
554  *
555  * If write spans object boundary, just do multiple writes.  (For a
556  * correct atomic write, we should e.g. take write locks on all
557  * objects, rollback on failure, etc.)
558  */
559 static ssize_t
560 ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
561 {
562 	struct file *file = iocb->ki_filp;
563 	struct inode *inode = file_inode(file);
564 	struct ceph_inode_info *ci = ceph_inode(inode);
565 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
566 	struct ceph_snap_context *snapc;
567 	struct ceph_vino vino;
568 	struct ceph_osd_request *req;
569 	struct page **pages;
570 	int num_pages;
571 	int written = 0;
572 	int flags;
573 	int check_caps = 0;
574 	int ret;
575 	struct timespec mtime = CURRENT_TIME;
576 	size_t count = iov_iter_count(from);
577 
578 	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
579 		return -EROFS;
580 
581 	dout("sync_direct_write on file %p %lld~%u\n", file, pos,
582 	     (unsigned)count);
583 
584 	ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
585 	if (ret < 0)
586 		return ret;
587 
588 	ret = invalidate_inode_pages2_range(inode->i_mapping,
589 					    pos >> PAGE_CACHE_SHIFT,
590 					    (pos + count) >> PAGE_CACHE_SHIFT);
591 	if (ret < 0)
592 		dout("invalidate_inode_pages2_range returned %d\n", ret);
593 
594 	flags = CEPH_OSD_FLAG_ORDERSNAP |
595 		CEPH_OSD_FLAG_ONDISK |
596 		CEPH_OSD_FLAG_WRITE;
597 
598 	while (iov_iter_count(from) > 0) {
599 		u64 len = iov_iter_single_seg_count(from);
600 		size_t start;
601 		ssize_t n;
602 
603 		snapc = ci->i_snap_realm->cached_context;
604 		vino = ceph_vino(inode);
605 		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
606 					    vino, pos, &len, 0,
607 					    2,/*include a 'startsync' command*/
608 					    CEPH_OSD_OP_WRITE, flags, snapc,
609 					    ci->i_truncate_seq,
610 					    ci->i_truncate_size,
611 					    false);
612 		if (IS_ERR(req)) {
613 			ret = PTR_ERR(req);
614 			break;
615 		}
616 
617 		osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
618 
619 		n = iov_iter_get_pages_alloc(from, &pages, len, &start);
620 		if (unlikely(n < 0)) {
621 			ret = n;
622 			ceph_osdc_put_request(req);
623 			break;
624 		}
625 
626 		num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
627 		/*
628 		 * throw out any page cache pages in this range. this
629 		 * may block.
630 		 */
631 		truncate_inode_pages_range(inode->i_mapping, pos,
632 				   (pos+n) | (PAGE_CACHE_SIZE-1));
633 		osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
634 						false, false);
635 
636 		/* BUG_ON(vino.snap != CEPH_NOSNAP); */
637 		ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
638 
639 		ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
640 		if (!ret)
641 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
642 
643 		ceph_put_page_vector(pages, num_pages, false);
644 
645 		ceph_osdc_put_request(req);
646 		if (ret)
647 			break;
648 		pos += n;
649 		written += n;
650 		iov_iter_advance(from, n);
651 
652 		if (pos > i_size_read(inode)) {
653 			check_caps = ceph_inode_set_size(inode, pos);
654 			if (check_caps)
655 				ceph_check_caps(ceph_inode(inode),
656 						CHECK_CAPS_AUTHONLY,
657 						NULL);
658 		}
659 	}
660 
661 	if (ret != -EOLDSNAPC && written > 0) {
662 		iocb->ki_pos = pos;
663 		ret = written;
664 	}
665 	return ret;
666 }
667 
668 
669 /*
670  * Synchronous write, straight from __user pointer or user pages.
671  *
672  * If write spans object boundary, just do multiple writes.  (For a
673  * correct atomic write, we should e.g. take write locks on all
674  * objects, rollback on failure, etc.)
675  */
676 static ssize_t
677 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
678 {
679 	struct file *file = iocb->ki_filp;
680 	struct inode *inode = file_inode(file);
681 	struct ceph_inode_info *ci = ceph_inode(inode);
682 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
683 	struct ceph_snap_context *snapc;
684 	struct ceph_vino vino;
685 	struct ceph_osd_request *req;
686 	struct page **pages;
687 	u64 len;
688 	int num_pages;
689 	int written = 0;
690 	int flags;
691 	int check_caps = 0;
692 	int ret;
693 	struct timespec mtime = CURRENT_TIME;
694 	size_t count = iov_iter_count(from);
695 
696 	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
697 		return -EROFS;
698 
699 	dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
700 
701 	ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
702 	if (ret < 0)
703 		return ret;
704 
705 	ret = invalidate_inode_pages2_range(inode->i_mapping,
706 					    pos >> PAGE_CACHE_SHIFT,
707 					    (pos + count) >> PAGE_CACHE_SHIFT);
708 	if (ret < 0)
709 		dout("invalidate_inode_pages2_range returned %d\n", ret);
710 
711 	flags = CEPH_OSD_FLAG_ORDERSNAP |
712 		CEPH_OSD_FLAG_ONDISK |
713 		CEPH_OSD_FLAG_WRITE |
714 		CEPH_OSD_FLAG_ACK;
715 
716 	while ((len = iov_iter_count(from)) > 0) {
717 		size_t left;
718 		int n;
719 
720 		snapc = ci->i_snap_realm->cached_context;
721 		vino = ceph_vino(inode);
722 		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
723 					    vino, pos, &len, 0, 1,
724 					    CEPH_OSD_OP_WRITE, flags, snapc,
725 					    ci->i_truncate_seq,
726 					    ci->i_truncate_size,
727 					    false);
728 		if (IS_ERR(req)) {
729 			ret = PTR_ERR(req);
730 			break;
731 		}
732 
733 		/*
734 		 * write from beginning of first page,
735 		 * regardless of io alignment
736 		 */
737 		num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
738 
739 		pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
740 		if (IS_ERR(pages)) {
741 			ret = PTR_ERR(pages);
742 			goto out;
743 		}
744 
745 		left = len;
746 		for (n = 0; n < num_pages; n++) {
747 			size_t plen = min_t(size_t, left, PAGE_SIZE);
748 			ret = copy_page_from_iter(pages[n], 0, plen, from);
749 			if (ret != plen) {
750 				ret = -EFAULT;
751 				break;
752 			}
753 			left -= ret;
754 		}
755 
756 		if (ret < 0) {
757 			ceph_release_page_vector(pages, num_pages);
758 			goto out;
759 		}
760 
761 		/* get a second commit callback */
762 		req->r_unsafe_callback = ceph_sync_write_unsafe;
763 		req->r_inode = inode;
764 
765 		osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
766 						false, true);
767 
768 		/* BUG_ON(vino.snap != CEPH_NOSNAP); */
769 		ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
770 
771 		ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
772 		if (!ret)
773 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
774 
775 out:
776 		ceph_osdc_put_request(req);
777 		if (ret == 0) {
778 			pos += len;
779 			written += len;
780 
781 			if (pos > i_size_read(inode)) {
782 				check_caps = ceph_inode_set_size(inode, pos);
783 				if (check_caps)
784 					ceph_check_caps(ceph_inode(inode),
785 							CHECK_CAPS_AUTHONLY,
786 							NULL);
787 			}
788 		} else
789 			break;
790 	}
791 
792 	if (ret != -EOLDSNAPC && written > 0) {
793 		ret = written;
794 		iocb->ki_pos = pos;
795 	}
796 	return ret;
797 }
798 
799 /*
800  * Wrap generic_file_aio_read with checks for cap bits on the inode.
801  * Atomically grab references, so that those bits are not released
802  * back to the MDS mid-read.
803  *
804  * Hmm, the sync read case isn't actually async... should it be?
805  */
806 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
807 {
808 	struct file *filp = iocb->ki_filp;
809 	struct ceph_file_info *fi = filp->private_data;
810 	size_t len = iov_iter_count(to);
811 	struct inode *inode = file_inode(filp);
812 	struct ceph_inode_info *ci = ceph_inode(inode);
813 	struct page *pinned_page = NULL;
814 	ssize_t ret;
815 	int want, got = 0;
816 	int retry_op = 0, read = 0;
817 
818 again:
819 	dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
820 	     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
821 
822 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
823 		want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
824 	else
825 		want = CEPH_CAP_FILE_CACHE;
826 	ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
827 	if (ret < 0)
828 		return ret;
829 
830 	if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
831 	    (iocb->ki_flags & IOCB_DIRECT) ||
832 	    (fi->flags & CEPH_F_SYNC)) {
833 
834 		dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
835 		     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
836 		     ceph_cap_string(got));
837 
838 		if (ci->i_inline_version == CEPH_INLINE_NONE) {
839 			/* hmm, this isn't really async... */
840 			ret = ceph_sync_read(iocb, to, &retry_op);
841 		} else {
842 			retry_op = READ_INLINE;
843 		}
844 	} else {
845 		dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
846 		     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
847 		     ceph_cap_string(got));
848 
849 		ret = generic_file_read_iter(iocb, to);
850 	}
851 	dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
852 	     inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
853 	if (pinned_page) {
854 		page_cache_release(pinned_page);
855 		pinned_page = NULL;
856 	}
857 	ceph_put_cap_refs(ci, got);
858 	if (retry_op && ret >= 0) {
859 		int statret;
860 		struct page *page = NULL;
861 		loff_t i_size;
862 		if (retry_op == READ_INLINE) {
863 			page = __page_cache_alloc(GFP_NOFS);
864 			if (!page)
865 				return -ENOMEM;
866 		}
867 
868 		statret = __ceph_do_getattr(inode, page,
869 					    CEPH_STAT_CAP_INLINE_DATA, !!page);
870 		if (statret < 0) {
871 			 __free_page(page);
872 			if (statret == -ENODATA) {
873 				BUG_ON(retry_op != READ_INLINE);
874 				goto again;
875 			}
876 			return statret;
877 		}
878 
879 		i_size = i_size_read(inode);
880 		if (retry_op == READ_INLINE) {
881 			BUG_ON(ret > 0 || read > 0);
882 			if (iocb->ki_pos < i_size &&
883 			    iocb->ki_pos < PAGE_CACHE_SIZE) {
884 				loff_t end = min_t(loff_t, i_size,
885 						   iocb->ki_pos + len);
886 				end = min_t(loff_t, end, PAGE_CACHE_SIZE);
887 				if (statret < end)
888 					zero_user_segment(page, statret, end);
889 				ret = copy_page_to_iter(page,
890 						iocb->ki_pos & ~PAGE_MASK,
891 						end - iocb->ki_pos, to);
892 				iocb->ki_pos += ret;
893 				read += ret;
894 			}
895 			if (iocb->ki_pos < i_size && read < len) {
896 				size_t zlen = min_t(size_t, len - read,
897 						    i_size - iocb->ki_pos);
898 				ret = iov_iter_zero(zlen, to);
899 				iocb->ki_pos += ret;
900 				read += ret;
901 			}
902 			__free_pages(page, 0);
903 			return read;
904 		}
905 
906 		/* hit EOF or hole? */
907 		if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
908 		    ret < len) {
909 			dout("sync_read hit hole, ppos %lld < size %lld"
910 			     ", reading more\n", iocb->ki_pos,
911 			     inode->i_size);
912 
913 			read += ret;
914 			len -= ret;
915 			retry_op = 0;
916 			goto again;
917 		}
918 	}
919 
920 	if (ret >= 0)
921 		ret += read;
922 
923 	return ret;
924 }
925 
926 /*
927  * Take cap references to avoid releasing caps to MDS mid-write.
928  *
929  * If we are synchronous, and write with an old snap context, the OSD
930  * may return EOLDSNAPC.  In that case, retry the write.. _after_
931  * dropping our cap refs and allowing the pending snap to logically
932  * complete _before_ this write occurs.
933  *
934  * If we are near ENOSPC, write synchronously.
935  */
936 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
937 {
938 	struct file *file = iocb->ki_filp;
939 	struct ceph_file_info *fi = file->private_data;
940 	struct inode *inode = file_inode(file);
941 	struct ceph_inode_info *ci = ceph_inode(inode);
942 	struct ceph_osd_client *osdc =
943 		&ceph_sb_to_client(inode->i_sb)->client->osdc;
944 	ssize_t count, written = 0;
945 	int err, want, got;
946 	loff_t pos;
947 
948 	if (ceph_snap(inode) != CEPH_NOSNAP)
949 		return -EROFS;
950 
951 	mutex_lock(&inode->i_mutex);
952 
953 	/* We can write back this queue in page reclaim */
954 	current->backing_dev_info = inode_to_bdi(inode);
955 
956 	err = generic_write_checks(iocb, from);
957 	if (err <= 0)
958 		goto out;
959 
960 	pos = iocb->ki_pos;
961 	count = iov_iter_count(from);
962 	err = file_remove_suid(file);
963 	if (err)
964 		goto out;
965 
966 	err = file_update_time(file);
967 	if (err)
968 		goto out;
969 
970 	if (ci->i_inline_version != CEPH_INLINE_NONE) {
971 		err = ceph_uninline_data(file, NULL);
972 		if (err < 0)
973 			goto out;
974 	}
975 
976 retry_snap:
977 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
978 		err = -ENOSPC;
979 		goto out;
980 	}
981 
982 	dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
983 	     inode, ceph_vinop(inode), pos, count, inode->i_size);
984 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
985 		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
986 	else
987 		want = CEPH_CAP_FILE_BUFFER;
988 	got = 0;
989 	err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
990 			    &got, NULL);
991 	if (err < 0)
992 		goto out;
993 
994 	dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
995 	     inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
996 
997 	if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
998 	    (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
999 		struct iov_iter data;
1000 		mutex_unlock(&inode->i_mutex);
1001 		/* we might need to revert back to that point */
1002 		data = *from;
1003 		if (iocb->ki_flags & IOCB_DIRECT)
1004 			written = ceph_sync_direct_write(iocb, &data, pos);
1005 		else
1006 			written = ceph_sync_write(iocb, &data, pos);
1007 		if (written == -EOLDSNAPC) {
1008 			dout("aio_write %p %llx.%llx %llu~%u"
1009 				"got EOLDSNAPC, retrying\n",
1010 				inode, ceph_vinop(inode),
1011 				pos, (unsigned)count);
1012 			mutex_lock(&inode->i_mutex);
1013 			goto retry_snap;
1014 		}
1015 		if (written > 0)
1016 			iov_iter_advance(from, written);
1017 	} else {
1018 		loff_t old_size = inode->i_size;
1019 		/*
1020 		 * No need to acquire the i_truncate_mutex. Because
1021 		 * the MDS revokes Fwb caps before sending truncate
1022 		 * message to us. We can't get Fwb cap while there
1023 		 * are pending vmtruncate. So write and vmtruncate
1024 		 * can not run at the same time
1025 		 */
1026 		written = generic_perform_write(file, from, pos);
1027 		if (likely(written >= 0))
1028 			iocb->ki_pos = pos + written;
1029 		if (inode->i_size > old_size)
1030 			ceph_fscache_update_objectsize(inode);
1031 		mutex_unlock(&inode->i_mutex);
1032 	}
1033 
1034 	if (written >= 0) {
1035 		int dirty;
1036 		spin_lock(&ci->i_ceph_lock);
1037 		ci->i_inline_version = CEPH_INLINE_NONE;
1038 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1039 		spin_unlock(&ci->i_ceph_lock);
1040 		if (dirty)
1041 			__mark_inode_dirty(inode, dirty);
1042 	}
1043 
1044 	dout("aio_write %p %llx.%llx %llu~%u  dropping cap refs on %s\n",
1045 	     inode, ceph_vinop(inode), pos, (unsigned)count,
1046 	     ceph_cap_string(got));
1047 	ceph_put_cap_refs(ci, got);
1048 
1049 	if (written >= 0 &&
1050 	    ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
1051 	     ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
1052 		err = vfs_fsync_range(file, pos, pos + written - 1, 1);
1053 		if (err < 0)
1054 			written = err;
1055 	}
1056 
1057 	goto out_unlocked;
1058 
1059 out:
1060 	mutex_unlock(&inode->i_mutex);
1061 out_unlocked:
1062 	current->backing_dev_info = NULL;
1063 	return written ? written : err;
1064 }
1065 
1066 /*
1067  * llseek.  be sure to verify file size on SEEK_END.
1068  */
1069 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1070 {
1071 	struct inode *inode = file->f_mapping->host;
1072 	int ret;
1073 
1074 	mutex_lock(&inode->i_mutex);
1075 
1076 	if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1077 		ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1078 		if (ret < 0) {
1079 			offset = ret;
1080 			goto out;
1081 		}
1082 	}
1083 
1084 	switch (whence) {
1085 	case SEEK_END:
1086 		offset += inode->i_size;
1087 		break;
1088 	case SEEK_CUR:
1089 		/*
1090 		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1091 		 * position-querying operation.  Avoid rewriting the "same"
1092 		 * f_pos value back to the file because a concurrent read(),
1093 		 * write() or lseek() might have altered it
1094 		 */
1095 		if (offset == 0) {
1096 			offset = file->f_pos;
1097 			goto out;
1098 		}
1099 		offset += file->f_pos;
1100 		break;
1101 	case SEEK_DATA:
1102 		if (offset >= inode->i_size) {
1103 			ret = -ENXIO;
1104 			goto out;
1105 		}
1106 		break;
1107 	case SEEK_HOLE:
1108 		if (offset >= inode->i_size) {
1109 			ret = -ENXIO;
1110 			goto out;
1111 		}
1112 		offset = inode->i_size;
1113 		break;
1114 	}
1115 
1116 	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1117 
1118 out:
1119 	mutex_unlock(&inode->i_mutex);
1120 	return offset;
1121 }
1122 
1123 static inline void ceph_zero_partial_page(
1124 	struct inode *inode, loff_t offset, unsigned size)
1125 {
1126 	struct page *page;
1127 	pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1128 
1129 	page = find_lock_page(inode->i_mapping, index);
1130 	if (page) {
1131 		wait_on_page_writeback(page);
1132 		zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1133 		unlock_page(page);
1134 		page_cache_release(page);
1135 	}
1136 }
1137 
1138 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1139 				      loff_t length)
1140 {
1141 	loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1142 	if (offset < nearly) {
1143 		loff_t size = nearly - offset;
1144 		if (length < size)
1145 			size = length;
1146 		ceph_zero_partial_page(inode, offset, size);
1147 		offset += size;
1148 		length -= size;
1149 	}
1150 	if (length >= PAGE_CACHE_SIZE) {
1151 		loff_t size = round_down(length, PAGE_CACHE_SIZE);
1152 		truncate_pagecache_range(inode, offset, offset + size - 1);
1153 		offset += size;
1154 		length -= size;
1155 	}
1156 	if (length)
1157 		ceph_zero_partial_page(inode, offset, length);
1158 }
1159 
1160 static int ceph_zero_partial_object(struct inode *inode,
1161 				    loff_t offset, loff_t *length)
1162 {
1163 	struct ceph_inode_info *ci = ceph_inode(inode);
1164 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1165 	struct ceph_osd_request *req;
1166 	int ret = 0;
1167 	loff_t zero = 0;
1168 	int op;
1169 
1170 	if (!length) {
1171 		op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1172 		length = &zero;
1173 	} else {
1174 		op = CEPH_OSD_OP_ZERO;
1175 	}
1176 
1177 	req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1178 					ceph_vino(inode),
1179 					offset, length,
1180 					0, 1, op,
1181 					CEPH_OSD_FLAG_WRITE |
1182 					CEPH_OSD_FLAG_ONDISK,
1183 					NULL, 0, 0, false);
1184 	if (IS_ERR(req)) {
1185 		ret = PTR_ERR(req);
1186 		goto out;
1187 	}
1188 
1189 	ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1190 				&inode->i_mtime);
1191 
1192 	ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1193 	if (!ret) {
1194 		ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1195 		if (ret == -ENOENT)
1196 			ret = 0;
1197 	}
1198 	ceph_osdc_put_request(req);
1199 
1200 out:
1201 	return ret;
1202 }
1203 
1204 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1205 {
1206 	int ret = 0;
1207 	struct ceph_inode_info *ci = ceph_inode(inode);
1208 	s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1209 	s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1210 	s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1211 	u64 object_set_size = object_size * stripe_count;
1212 	u64 nearly, t;
1213 
1214 	/* round offset up to next period boundary */
1215 	nearly = offset + object_set_size - 1;
1216 	t = nearly;
1217 	nearly -= do_div(t, object_set_size);
1218 
1219 	while (length && offset < nearly) {
1220 		loff_t size = length;
1221 		ret = ceph_zero_partial_object(inode, offset, &size);
1222 		if (ret < 0)
1223 			return ret;
1224 		offset += size;
1225 		length -= size;
1226 	}
1227 	while (length >= object_set_size) {
1228 		int i;
1229 		loff_t pos = offset;
1230 		for (i = 0; i < stripe_count; ++i) {
1231 			ret = ceph_zero_partial_object(inode, pos, NULL);
1232 			if (ret < 0)
1233 				return ret;
1234 			pos += stripe_unit;
1235 		}
1236 		offset += object_set_size;
1237 		length -= object_set_size;
1238 	}
1239 	while (length) {
1240 		loff_t size = length;
1241 		ret = ceph_zero_partial_object(inode, offset, &size);
1242 		if (ret < 0)
1243 			return ret;
1244 		offset += size;
1245 		length -= size;
1246 	}
1247 	return ret;
1248 }
1249 
1250 static long ceph_fallocate(struct file *file, int mode,
1251 				loff_t offset, loff_t length)
1252 {
1253 	struct ceph_file_info *fi = file->private_data;
1254 	struct inode *inode = file_inode(file);
1255 	struct ceph_inode_info *ci = ceph_inode(inode);
1256 	struct ceph_osd_client *osdc =
1257 		&ceph_inode_to_client(inode)->client->osdc;
1258 	int want, got = 0;
1259 	int dirty;
1260 	int ret = 0;
1261 	loff_t endoff = 0;
1262 	loff_t size;
1263 
1264 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1265 		return -EOPNOTSUPP;
1266 
1267 	if (!S_ISREG(inode->i_mode))
1268 		return -EOPNOTSUPP;
1269 
1270 	mutex_lock(&inode->i_mutex);
1271 
1272 	if (ceph_snap(inode) != CEPH_NOSNAP) {
1273 		ret = -EROFS;
1274 		goto unlock;
1275 	}
1276 
1277 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1278 		!(mode & FALLOC_FL_PUNCH_HOLE)) {
1279 		ret = -ENOSPC;
1280 		goto unlock;
1281 	}
1282 
1283 	if (ci->i_inline_version != CEPH_INLINE_NONE) {
1284 		ret = ceph_uninline_data(file, NULL);
1285 		if (ret < 0)
1286 			goto unlock;
1287 	}
1288 
1289 	size = i_size_read(inode);
1290 	if (!(mode & FALLOC_FL_KEEP_SIZE))
1291 		endoff = offset + length;
1292 
1293 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
1294 		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1295 	else
1296 		want = CEPH_CAP_FILE_BUFFER;
1297 
1298 	ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1299 	if (ret < 0)
1300 		goto unlock;
1301 
1302 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1303 		if (offset < size)
1304 			ceph_zero_pagecache_range(inode, offset, length);
1305 		ret = ceph_zero_objects(inode, offset, length);
1306 	} else if (endoff > size) {
1307 		truncate_pagecache_range(inode, size, -1);
1308 		if (ceph_inode_set_size(inode, endoff))
1309 			ceph_check_caps(ceph_inode(inode),
1310 				CHECK_CAPS_AUTHONLY, NULL);
1311 	}
1312 
1313 	if (!ret) {
1314 		spin_lock(&ci->i_ceph_lock);
1315 		ci->i_inline_version = CEPH_INLINE_NONE;
1316 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1317 		spin_unlock(&ci->i_ceph_lock);
1318 		if (dirty)
1319 			__mark_inode_dirty(inode, dirty);
1320 	}
1321 
1322 	ceph_put_cap_refs(ci, got);
1323 unlock:
1324 	mutex_unlock(&inode->i_mutex);
1325 	return ret;
1326 }
1327 
1328 const struct file_operations ceph_file_fops = {
1329 	.open = ceph_open,
1330 	.release = ceph_release,
1331 	.llseek = ceph_llseek,
1332 	.read_iter = ceph_read_iter,
1333 	.write_iter = ceph_write_iter,
1334 	.mmap = ceph_mmap,
1335 	.fsync = ceph_fsync,
1336 	.lock = ceph_lock,
1337 	.flock = ceph_flock,
1338 	.splice_read = generic_file_splice_read,
1339 	.splice_write = iter_file_splice_write,
1340 	.unlocked_ioctl = ceph_ioctl,
1341 	.compat_ioctl	= ceph_ioctl,
1342 	.fallocate	= ceph_fallocate,
1343 };
1344 
1345