blob: ca3f630db90f22a0cefe4bdb7355033704b4154c [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil124e68e2009-10-06 11:31:08 -07002
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003#include <linux/module.h>
Sage Weil124e68e2009-10-06 11:31:08 -07004#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weil124e68e2009-10-06 11:31:08 -07006#include <linux/file.h>
Sage Weil5ef50c32012-07-31 11:27:36 -07007#include <linux/mount.h>
Sage Weil124e68e2009-10-06 11:31:08 -07008#include <linux/namei.h>
9#include <linux/writeback.h>
Li Wangad7a60d2013-08-15 11:51:44 +080010#include <linux/falloc.h>
Sage Weil124e68e2009-10-06 11:31:08 -070011
12#include "super.h"
13#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040014#include "cache.h"
Sage Weil124e68e2009-10-06 11:31:08 -070015
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
Zhu, Caifengb5b98982015-10-08 15:26:15 +080037/*
38 * Calculate the length sum of direct io vectors that can
39 * be combined into one page vector.
40 */
41static size_t dio_get_pagev_size(const struct iov_iter *it)
42{
43 const struct iovec *iov = it->iov;
44 const struct iovec *iovend = iov + it->nr_segs;
45 size_t size;
46
47 size = iov->iov_len - it->iov_offset;
48 /*
49 * An iov can be page vectored when both the current tail
50 * and the next base are page aligned.
51 */
52 while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) &&
53 (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) {
54 size += iov->iov_len;
55 }
56 dout("dio_get_pagevlen len = %zu\n", size);
57 return size;
58}
59
60/*
61 * Allocate a page vector based on (@it, @nbytes).
62 * The return value is the tuple describing a page vector,
63 * that is (@pages, @page_align, @num_pages).
64 */
65static struct page **
66dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes,
67 size_t *page_align, int *num_pages)
68{
69 struct iov_iter tmp_it = *it;
70 size_t align;
71 struct page **pages;
72 int ret = 0, idx, npages;
73
74 align = (unsigned long)(it->iov->iov_base + it->iov_offset) &
75 (PAGE_SIZE - 1);
76 npages = calc_pages_for(align, nbytes);
77 pages = kmalloc(sizeof(*pages) * npages, GFP_KERNEL);
78 if (!pages) {
79 pages = vmalloc(sizeof(*pages) * npages);
80 if (!pages)
81 return ERR_PTR(-ENOMEM);
82 }
83
84 for (idx = 0; idx < npages; ) {
85 size_t start;
86 ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes,
87 npages - idx, &start);
88 if (ret < 0)
89 goto fail;
90
91 iov_iter_advance(&tmp_it, ret);
92 nbytes -= ret;
93 idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE;
94 }
95
96 BUG_ON(nbytes != 0);
97 *num_pages = npages;
98 *page_align = align;
99 dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align);
100 return pages;
101fail:
102 ceph_put_page_vector(pages, idx, false);
103 return ERR_PTR(ret);
104}
Sage Weil124e68e2009-10-06 11:31:08 -0700105
106/*
107 * Prepare an open request. Preallocate ceph_cap to avoid an
108 * inopportune ENOMEM later.
109 */
110static struct ceph_mds_request *
111prepare_open_request(struct super_block *sb, int flags, int create_mode)
112{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700113 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
114 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700115 struct ceph_mds_request *req;
116 int want_auth = USE_ANY_MDS;
117 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
118
119 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
120 want_auth = USE_AUTH_MDS;
121
122 req = ceph_mdsc_create_request(mdsc, op, want_auth);
123 if (IS_ERR(req))
124 goto out;
125 req->r_fmode = ceph_flags_to_mode(flags);
126 req->r_args.open.flags = cpu_to_le32(flags);
127 req->r_args.open.mode = cpu_to_le32(create_mode);
Sage Weil124e68e2009-10-06 11:31:08 -0700128out:
129 return req;
130}
131
132/*
133 * initialize private struct file data.
134 * if we fail, clean up by dropping fmode reference on the ceph_inode
135 */
136static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
137{
138 struct ceph_file_info *cf;
139 int ret = 0;
140
141 switch (inode->i_mode & S_IFMT) {
142 case S_IFREG:
Yan, Zheng46b59b22016-05-18 15:25:03 +0800143 ceph_fscache_register_inode_cookie(inode);
144 ceph_fscache_file_set_cookie(inode, file);
Sage Weil124e68e2009-10-06 11:31:08 -0700145 case S_IFDIR:
146 dout("init_file %p %p 0%o (regular)\n", inode, file,
147 inode->i_mode);
Geliang Tang99ec2692016-03-13 15:26:29 +0800148 cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
Sage Weil124e68e2009-10-06 11:31:08 -0700149 if (cf == NULL) {
150 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
151 return -ENOMEM;
152 }
153 cf->fmode = fmode;
154 cf->next_offset = 2;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800155 cf->readdir_cache_idx = -1;
Sage Weil124e68e2009-10-06 11:31:08 -0700156 file->private_data = cf;
157 BUG_ON(inode->i_fop->release != ceph_release);
158 break;
159
160 case S_IFLNK:
161 dout("init_file %p %p 0%o (symlink)\n", inode, file,
162 inode->i_mode);
163 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
164 break;
165
166 default:
167 dout("init_file %p %p 0%o (special)\n", inode, file,
168 inode->i_mode);
169 /*
170 * we need to drop the open ref now, since we don't
171 * have .release set to ceph_release.
172 */
173 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
174 BUG_ON(inode->i_fop->release == ceph_release);
175
176 /* call the proper open fop */
177 ret = inode->i_fop->open(inode, file);
178 }
179 return ret;
180}
181
182/*
Yan, Zheng77310322016-04-08 15:27:16 +0800183 * try renew caps after session gets killed.
184 */
185int ceph_renew_caps(struct inode *inode)
186{
187 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
188 struct ceph_inode_info *ci = ceph_inode(inode);
189 struct ceph_mds_request *req;
190 int err, flags, wanted;
191
192 spin_lock(&ci->i_ceph_lock);
193 wanted = __ceph_caps_file_wanted(ci);
194 if (__ceph_is_any_real_caps(ci) &&
195 (!(wanted & CEPH_CAP_ANY_WR) == 0 || ci->i_auth_cap)) {
196 int issued = __ceph_caps_issued(ci, NULL);
197 spin_unlock(&ci->i_ceph_lock);
198 dout("renew caps %p want %s issued %s updating mds_wanted\n",
199 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
200 ceph_check_caps(ci, 0, NULL);
201 return 0;
202 }
203 spin_unlock(&ci->i_ceph_lock);
204
205 flags = 0;
206 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
207 flags = O_RDWR;
208 else if (wanted & CEPH_CAP_FILE_RD)
209 flags = O_RDONLY;
210 else if (wanted & CEPH_CAP_FILE_WR)
211 flags = O_WRONLY;
212#ifdef O_LAZY
213 if (wanted & CEPH_CAP_FILE_LAZYIO)
214 flags |= O_LAZY;
215#endif
216
217 req = prepare_open_request(inode->i_sb, flags, 0);
218 if (IS_ERR(req)) {
219 err = PTR_ERR(req);
220 goto out;
221 }
222
223 req->r_inode = inode;
224 ihold(inode);
225 req->r_num_caps = 1;
226 req->r_fmode = -1;
227
228 err = ceph_mdsc_do_request(mdsc, NULL, req);
229 ceph_mdsc_put_request(req);
230out:
231 dout("renew caps %p open result=%d\n", inode, err);
232 return err < 0 ? err : 0;
233}
234
235/*
Sage Weil124e68e2009-10-06 11:31:08 -0700236 * If we already have the requisite capabilities, we can satisfy
237 * the open request locally (no need to request new caps from the
238 * MDS). We do, however, need to inform the MDS (asynchronously)
239 * if our wanted caps set expands.
240 */
241int ceph_open(struct inode *inode, struct file *file)
242{
243 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700244 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
245 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700246 struct ceph_mds_request *req;
247 struct ceph_file_info *cf = file->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700248 int err;
249 int flags, fmode, wanted;
250
251 if (cf) {
252 dout("open file %p is already opened\n", file);
253 return 0;
254 }
255
256 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
257 flags = file->f_flags & ~(O_CREAT|O_EXCL);
258 if (S_ISDIR(inode->i_mode))
259 flags = O_DIRECTORY; /* mds likes to know */
260
261 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
262 ceph_vinop(inode), file, flags, file->f_flags);
263 fmode = ceph_flags_to_mode(flags);
264 wanted = ceph_caps_for_mode(fmode);
265
266 /* snapped files are read-only */
267 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
268 return -EROFS;
269
270 /* trivially open snapdir */
271 if (ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weilbe655592011-11-30 09:47:09 -0800272 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700273 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800274 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700275 return ceph_init_file(inode, file, fmode);
276 }
277
278 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800279 * No need to block if we have caps on the auth MDS (for
280 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700281 * asynchronously.
282 */
Sage Weilbe655592011-11-30 09:47:09 -0800283 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800284 if (__ceph_is_any_real_caps(ci) &&
285 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700286 int mds_wanted = __ceph_caps_mds_wanted(ci);
287 int issued = __ceph_caps_issued(ci, NULL);
288
289 dout("open %p fmode %d want %s issued %s using existing\n",
290 inode, fmode, ceph_cap_string(wanted),
291 ceph_cap_string(issued));
292 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800293 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700294
295 /* adjust wanted? */
296 if ((issued & wanted) != wanted &&
297 (mds_wanted & wanted) != wanted &&
298 ceph_snap(inode) != CEPH_SNAPDIR)
299 ceph_check_caps(ci, 0, NULL);
300
301 return ceph_init_file(inode, file, fmode);
302 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
303 (ci->i_snap_caps & wanted) == wanted) {
304 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800305 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700306 return ceph_init_file(inode, file, fmode);
307 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400308
Sage Weilbe655592011-11-30 09:47:09 -0800309 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700310
311 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
312 req = prepare_open_request(inode->i_sb, flags, 0);
313 if (IS_ERR(req)) {
314 err = PTR_ERR(req);
315 goto out;
316 }
Sage Weil70b666c2011-05-27 09:24:26 -0700317 req->r_inode = inode;
318 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400319
Sage Weil124e68e2009-10-06 11:31:08 -0700320 req->r_num_caps = 1;
Jianpeng Mae36d5712015-08-18 10:25:35 +0800321 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700322 if (!err)
323 err = ceph_init_file(inode, file, req->r_fmode);
324 ceph_mdsc_put_request(req);
325 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
326out:
327 return err;
328}
329
330
331/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700332 * Do a lookup + open with a single request. If we get a non-existent
333 * file or symlink, return 1 so the VFS can retry.
Sage Weil124e68e2009-10-06 11:31:08 -0700334 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700335int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400336 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400337 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700338{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700339 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
340 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700341 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700342 struct dentry *dn;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800343 struct ceph_acls_info acls = {};
Yan, Zheng315f2402016-03-07 10:34:50 +0800344 int mask;
Sage Weil124e68e2009-10-06 11:31:08 -0700345 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700346
Al Viroa4555892014-10-21 20:11:25 -0400347 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
348 dir, dentry, dentry,
Sage Weil5ef50c32012-07-31 11:27:36 -0700349 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
350
351 if (dentry->d_name.len > NAME_MAX)
352 return -ENAMETOOLONG;
353
354 err = ceph_init_dentry(dentry);
355 if (err < 0)
356 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700357
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800358 if (flags & O_CREAT) {
359 err = ceph_pre_init_acls(dir, &mode, &acls);
360 if (err < 0)
361 return err;
362 }
363
Sage Weil124e68e2009-10-06 11:31:08 -0700364 /* do the open */
365 req = prepare_open_request(dir->i_sb, flags, mode);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800366 if (IS_ERR(req)) {
367 err = PTR_ERR(req);
368 goto out_acl;
369 }
Sage Weil124e68e2009-10-06 11:31:08 -0700370 req->r_dentry = dget(dentry);
371 req->r_num_caps = 2;
372 if (flags & O_CREAT) {
373 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
374 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800375 if (acls.pagelist) {
376 req->r_pagelist = acls.pagelist;
377 acls.pagelist = NULL;
378 }
Sage Weil124e68e2009-10-06 11:31:08 -0700379 }
Yan, Zheng315f2402016-03-07 10:34:50 +0800380
381 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
382 if (ceph_security_xattr_wanted(dir))
383 mask |= CEPH_CAP_XATTR_SHARED;
384 req->r_args.open.mask = cpu_to_le32(mask);
385
Sage Weil124e68e2009-10-06 11:31:08 -0700386 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700387 err = ceph_mdsc_do_request(mdsc,
388 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
389 req);
Yan, Zhengbf91c312015-01-19 13:23:20 +0800390 err = ceph_handle_snapdir(req, dentry, err);
Sam Lang79aec9842012-12-19 09:44:23 -1000391 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800392 goto out_req;
Sam Lang79aec9842012-12-19 09:44:23 -1000393
Jianpeng Maa43137f2015-08-18 10:23:50 +0800394 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700395 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700396
Al Viro00699ad2016-07-05 09:44:53 -0400397 if (d_in_lookup(dentry)) {
Sage Weil5ef50c32012-07-31 11:27:36 -0700398 dn = ceph_finish_lookup(req, dentry, err);
399 if (IS_ERR(dn))
400 err = PTR_ERR(dn);
401 } else {
402 /* we were given a hashed negative dentry */
403 dn = NULL;
404 }
Sage Weil468640e2011-07-26 11:28:11 -0700405 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800406 goto out_req;
David Howells2b0143b2015-03-17 22:25:59 +0000407 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
Sage Weil5ef50c32012-07-31 11:27:36 -0700408 /* make vfs retry on splice, ENOENT, or symlink */
409 dout("atomic_open finish_no_open on dn %p\n", dn);
410 err = finish_no_open(file, dn);
411 } else {
412 dout("atomic_open finish_open on dn %p\n", dn);
Sam Lang6e8575f2012-12-28 09:56:46 -0800413 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
David Howells2b0143b2015-03-17 22:25:59 +0000414 ceph_init_inode_acls(d_inode(dentry), &acls);
Sam Lang6e8575f2012-12-28 09:56:46 -0800415 *opened |= FILE_CREATED;
416 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700417 err = finish_open(file, dentry, ceph_open, opened);
418 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800419out_req:
Yan, Zhengab866542014-04-01 20:20:17 +0800420 if (!req->r_err && req->r_target_inode)
421 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
Sage Weil124e68e2009-10-06 11:31:08 -0700422 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800423out_acl:
424 ceph_release_acls_info(&acls);
Sage Weil5ef50c32012-07-31 11:27:36 -0700425 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400426 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700427}
428
429int ceph_release(struct inode *inode, struct file *file)
430{
431 struct ceph_inode_info *ci = ceph_inode(inode);
432 struct ceph_file_info *cf = file->private_data;
433
434 dout("release inode %p file %p\n", inode, file);
435 ceph_put_fmode(ci, cf->fmode);
436 if (cf->last_readdir)
437 ceph_mdsc_put_request(cf->last_readdir);
438 kfree(cf->last_name);
439 kfree(cf->dir_info);
Sage Weil124e68e2009-10-06 11:31:08 -0700440 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800441
442 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700443 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700444 return 0;
445}
446
Yan, Zheng83701242014-11-14 22:36:18 +0800447enum {
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800448 HAVE_RETRIED = 1,
449 CHECK_EOF = 2,
450 READ_INLINE = 3,
Yan, Zheng83701242014-11-14 22:36:18 +0800451};
452
Sage Weil124e68e2009-10-06 11:31:08 -0700453/*
Sage Weil124e68e2009-10-06 11:31:08 -0700454 * Read a range of bytes striped over one or more objects. Iterate over
455 * objects we stripe over. (That's not atomic, but good enough for now.)
456 *
457 * If we get a short result from the OSD, check against i_size; we need to
458 * only return a short read to the caller if we hit EOF.
459 */
460static int striped_read(struct inode *inode,
461 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800462 struct page **pages, int num_pages,
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800463 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700464{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700465 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700466 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpenter688bac42013-07-23 16:48:01 +0300467 u64 pos, this_len, left;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800468 loff_t i_size;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800469 int page_align, pages_left;
470 int read, ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700471 struct page **page_pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700472 bool hit_stripe, was_short;
473
474 /*
475 * we may need to do multiple reads. not atomic, unfortunately.
476 */
477 pos = off;
478 left = len;
479 page_pos = pages;
480 pages_left = num_pages;
481 read = 0;
482
483more:
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800484 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700485 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700486 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700487 &ci->i_layout, pos, &this_len,
488 ci->i_truncate_seq,
489 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800490 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700491 if (ret == -ENOENT)
492 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700493 hit_stripe = this_len < left;
494 was_short = ret >= 0 && ret < this_len;
Dan Carpenter688bac42013-07-23 16:48:01 +0300495 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
Sage Weil124e68e2009-10-06 11:31:08 -0700496 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
497
Yan, Zheng99c88e62015-12-30 11:32:46 +0800498 i_size = i_size_read(inode);
majianpeng02ae66d2013-08-06 16:20:38 +0800499 if (ret >= 0) {
500 int didpages;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800501 if (was_short && (pos + ret < i_size)) {
502 int zlen = min(this_len - ret, i_size - pos - ret);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800503 int zoff = (off & ~PAGE_MASK) + read + ret;
majianpeng02ae66d2013-08-06 16:20:38 +0800504 dout(" zero gap %llu to %llu\n",
Yan, Zheng1487a682015-01-06 15:29:14 +0800505 pos + ret, pos + ret + zlen);
506 ceph_zero_page_vector_range(zoff, zlen, pages);
507 ret += zlen;
Sage Weil124e68e2009-10-06 11:31:08 -0700508 }
majianpeng02ae66d2013-08-06 16:20:38 +0800509
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300510 didpages = (page_align + ret) >> PAGE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700511 pos += ret;
512 read = pos - off;
513 left -= ret;
514 page_pos += didpages;
515 pages_left -= didpages;
516
majianpeng02ae66d2013-08-06 16:20:38 +0800517 /* hit stripe and need continue*/
Yan, Zheng99c88e62015-12-30 11:32:46 +0800518 if (left && hit_stripe && pos < i_size)
Sage Weil124e68e2009-10-06 11:31:08 -0700519 goto more;
520 }
521
majianpengee7289b2013-08-21 15:02:51 +0800522 if (read > 0) {
majianpeng02ae66d2013-08-06 16:20:38 +0800523 ret = read;
Sage Weilc3cd6282011-06-01 16:08:44 -0700524 /* did we bounce off eof? */
Yan, Zheng99c88e62015-12-30 11:32:46 +0800525 if (pos + left > i_size)
Yan, Zheng83701242014-11-14 22:36:18 +0800526 *checkeof = CHECK_EOF;
Sage Weil124e68e2009-10-06 11:31:08 -0700527 }
528
Sage Weil124e68e2009-10-06 11:31:08 -0700529 dout("striped_read returns %d\n", ret);
530 return ret;
531}
532
533/*
534 * Completely synchronous read and write methods. Direct from __user
535 * buffer to osd, or directly to user pages (if O_DIRECT).
536 *
537 * If the read spans object boundary, just do multiple reads.
538 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800539static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
540 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700541{
majianpeng8eb4efb2013-09-26 14:42:17 +0800542 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500543 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700544 struct page **pages;
majianpeng8eb4efb2013-09-26 14:42:17 +0800545 u64 off = iocb->ki_pos;
Henry C Changab226e22010-12-15 20:41:54 -0800546 int num_pages, ret;
Al Viro2b777c92014-04-03 22:31:22 -0400547 size_t len = iov_iter_count(i);
Sage Weil124e68e2009-10-06 11:31:08 -0700548
majianpeng8eb4efb2013-09-26 14:42:17 +0800549 dout("sync_read on file %p %llu~%u %s\n", file, off,
550 (unsigned)len,
Sage Weil124e68e2009-10-06 11:31:08 -0700551 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
Yan, Zhengd0d0db22014-07-21 10:15:48 +0800552
553 if (!len)
554 return 0;
Sage Weile98b6fe2010-11-09 12:24:53 -0800555 /*
556 * flush any page cache pages in this range. this
557 * will make concurrent normal and sync io slow,
558 * but it will at least behave sensibly when they are
559 * in sequence.
560 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800561 ret = filemap_write_and_wait_range(inode->i_mapping, off,
562 off + len);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800563 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800564 return ret;
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800565
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800566 num_pages = calc_pages_for(off, len);
567 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
568 if (IS_ERR(pages))
569 return PTR_ERR(pages);
570 ret = striped_read(inode, off, len, pages,
571 num_pages, checkeof);
572 if (ret > 0) {
573 int l, k = 0;
574 size_t left = ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700575
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800576 while (left) {
577 size_t page_off = off & ~PAGE_MASK;
578 size_t copy = min_t(size_t, left,
579 PAGE_SIZE - page_off);
580 l = copy_page_to_iter(pages[k++], page_off, copy, i);
581 off += l;
582 left -= l;
583 if (l < copy)
majianpeng8eb4efb2013-09-26 14:42:17 +0800584 break;
585 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800586 }
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800587 ceph_release_page_vector(pages, num_pages);
majianpeng8eb4efb2013-09-26 14:42:17 +0800588
589 if (off > iocb->ki_pos) {
590 ret = off - iocb->ki_pos;
591 iocb->ki_pos = off;
592 }
593
Sage Weil124e68e2009-10-06 11:31:08 -0700594 dout("sync_read result %d\n", ret);
595 return ret;
596}
597
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800598struct ceph_aio_request {
599 struct kiocb *iocb;
600 size_t total_len;
601 int write;
602 int error;
603 struct list_head osd_reqs;
604 unsigned num_reqs;
605 atomic_t pending_reqs;
Yan, Zheng5be03892015-12-24 08:44:20 +0800606 struct timespec mtime;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800607 struct ceph_cap_flush *prealloc_cf;
608};
609
Yan, Zheng5be03892015-12-24 08:44:20 +0800610struct ceph_aio_work {
611 struct work_struct work;
612 struct ceph_osd_request *req;
613};
614
615static void ceph_aio_retry_work(struct work_struct *work);
616
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800617static void ceph_aio_complete(struct inode *inode,
618 struct ceph_aio_request *aio_req)
619{
620 struct ceph_inode_info *ci = ceph_inode(inode);
621 int ret;
622
623 if (!atomic_dec_and_test(&aio_req->pending_reqs))
624 return;
625
626 ret = aio_req->error;
627 if (!ret)
628 ret = aio_req->total_len;
629
630 dout("ceph_aio_complete %p rc %d\n", inode, ret);
631
632 if (ret >= 0 && aio_req->write) {
633 int dirty;
634
635 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
636 if (endoff > i_size_read(inode)) {
637 if (ceph_inode_set_size(inode, endoff))
638 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
639 }
640
641 spin_lock(&ci->i_ceph_lock);
642 ci->i_inline_version = CEPH_INLINE_NONE;
643 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
644 &aio_req->prealloc_cf);
645 spin_unlock(&ci->i_ceph_lock);
646 if (dirty)
647 __mark_inode_dirty(inode, dirty);
648
649 }
650
651 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
652 CEPH_CAP_FILE_RD));
653
654 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
655
656 ceph_free_cap_flush(aio_req->prealloc_cf);
657 kfree(aio_req);
658}
659
Ilya Dryomov85e084f2016-04-28 16:07:24 +0200660static void ceph_aio_complete_req(struct ceph_osd_request *req)
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800661{
662 int rc = req->r_result;
663 struct inode *inode = req->r_inode;
664 struct ceph_aio_request *aio_req = req->r_priv;
665 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
666 int num_pages = calc_pages_for((u64)osd_data->alignment,
667 osd_data->length);
668
669 dout("ceph_aio_complete_req %p rc %d bytes %llu\n",
670 inode, rc, osd_data->length);
671
672 if (rc == -EOLDSNAPC) {
Yan, Zheng5be03892015-12-24 08:44:20 +0800673 struct ceph_aio_work *aio_work;
674 BUG_ON(!aio_req->write);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800675
Yan, Zheng5be03892015-12-24 08:44:20 +0800676 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
677 if (aio_work) {
678 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
679 aio_work->req = req;
680 queue_work(ceph_inode_to_client(inode)->wb_wq,
681 &aio_work->work);
682 return;
683 }
684 rc = -ENOMEM;
685 } else if (!aio_req->write) {
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800686 if (rc == -ENOENT)
687 rc = 0;
688 if (rc >= 0 && osd_data->length > rc) {
689 int zoff = osd_data->alignment + rc;
690 int zlen = osd_data->length - rc;
691 /*
692 * If read is satisfied by single OSD request,
693 * it can pass EOF. Otherwise read is within
694 * i_size.
695 */
696 if (aio_req->num_reqs == 1) {
697 loff_t i_size = i_size_read(inode);
698 loff_t endoff = aio_req->iocb->ki_pos + rc;
699 if (endoff < i_size)
700 zlen = min_t(size_t, zlen,
701 i_size - endoff);
702 aio_req->total_len = rc + zlen;
703 }
704
705 if (zlen > 0)
706 ceph_zero_page_vector_range(zoff, zlen,
707 osd_data->pages);
708 }
709 }
710
Yan, Zhenga22bd5f2016-05-26 10:30:13 +0800711 ceph_put_page_vector(osd_data->pages, num_pages, !aio_req->write);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800712 ceph_osdc_put_request(req);
713
714 if (rc < 0)
715 cmpxchg(&aio_req->error, 0, rc);
716
717 ceph_aio_complete(inode, aio_req);
718 return;
719}
720
Yan, Zheng5be03892015-12-24 08:44:20 +0800721static void ceph_aio_retry_work(struct work_struct *work)
722{
723 struct ceph_aio_work *aio_work =
724 container_of(work, struct ceph_aio_work, work);
725 struct ceph_osd_request *orig_req = aio_work->req;
726 struct ceph_aio_request *aio_req = orig_req->r_priv;
727 struct inode *inode = orig_req->r_inode;
728 struct ceph_inode_info *ci = ceph_inode(inode);
729 struct ceph_snap_context *snapc;
730 struct ceph_osd_request *req;
731 int ret;
732
733 spin_lock(&ci->i_ceph_lock);
734 if (__ceph_have_pending_cap_snap(ci)) {
735 struct ceph_cap_snap *capsnap =
736 list_last_entry(&ci->i_cap_snaps,
737 struct ceph_cap_snap,
738 ci_item);
739 snapc = ceph_get_snap_context(capsnap->context);
740 } else {
741 BUG_ON(!ci->i_head_snapc);
742 snapc = ceph_get_snap_context(ci->i_head_snapc);
743 }
744 spin_unlock(&ci->i_ceph_lock);
745
746 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2,
747 false, GFP_NOFS);
Dan Carpenter1418bf02016-01-26 12:24:44 +0300748 if (!req) {
749 ret = -ENOMEM;
Yan, Zheng5be03892015-12-24 08:44:20 +0800750 req = orig_req;
751 goto out;
752 }
753
754 req->r_flags = CEPH_OSD_FLAG_ORDERSNAP |
755 CEPH_OSD_FLAG_ONDISK |
756 CEPH_OSD_FLAG_WRITE;
Ilya Dryomov63244fa2016-04-28 16:07:23 +0200757 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
Ilya Dryomovd30291b2016-04-29 19:54:20 +0200758 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
Yan, Zheng5be03892015-12-24 08:44:20 +0800759
Ilya Dryomov13d1ad12016-04-27 14:15:51 +0200760 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
761 if (ret) {
762 ceph_osdc_put_request(req);
763 req = orig_req;
764 goto out;
765 }
Yan, Zheng5be03892015-12-24 08:44:20 +0800766
767 req->r_ops[0] = orig_req->r_ops[0];
768 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
769
Ilya Dryomovbb873b52016-05-26 00:29:52 +0200770 req->r_mtime = aio_req->mtime;
771 req->r_data_offset = req->r_ops[0].extent.offset;
Yan, Zheng5be03892015-12-24 08:44:20 +0800772
Yan, Zheng5be03892015-12-24 08:44:20 +0800773 ceph_osdc_put_request(orig_req);
774
775 req->r_callback = ceph_aio_complete_req;
776 req->r_inode = inode;
777 req->r_priv = aio_req;
778
779 ret = ceph_osdc_start_request(req->r_osdc, req, false);
780out:
781 if (ret < 0) {
Yan, Zheng5be03892015-12-24 08:44:20 +0800782 req->r_result = ret;
Ilya Dryomov85e084f2016-04-28 16:07:24 +0200783 ceph_aio_complete_req(req);
Yan, Zheng5be03892015-12-24 08:44:20 +0800784 }
785
Yan, Zhengdb6aed72016-01-26 23:05:37 +0800786 ceph_put_snap_context(snapc);
Yan, Zheng5be03892015-12-24 08:44:20 +0800787 kfree(aio_work);
788}
789
Sage Weil124e68e2009-10-06 11:31:08 -0700790/*
Alex Elder26be8802013-04-15 11:20:42 -0500791 * Write commit request unsafe callback, called to tell us when a
792 * request is unsafe (that is, in flight--has been handed to the
793 * messenger to send to its target osd). It is called again when
794 * we've received a response message indicating the request is
795 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
796 * is completed early (and unsuccessfully) due to a timeout or
797 * interrupt.
798 *
799 * This is used if we requested both an ACK and ONDISK commit reply
800 * from the OSD.
Sage Weil124e68e2009-10-06 11:31:08 -0700801 */
Alex Elder26be8802013-04-15 11:20:42 -0500802static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700803{
804 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
805
Alex Elder26be8802013-04-15 11:20:42 -0500806 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
807 unsafe ? "un" : "");
808 if (unsafe) {
809 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
810 spin_lock(&ci->i_unsafe_lock);
811 list_add_tail(&req->r_unsafe_item,
812 &ci->i_unsafe_writes);
813 spin_unlock(&ci->i_unsafe_lock);
Ilya Dryomovfe5da052016-04-28 16:07:24 +0200814
815 complete_all(&req->r_completion);
Alex Elder26be8802013-04-15 11:20:42 -0500816 } else {
817 spin_lock(&ci->i_unsafe_lock);
818 list_del_init(&req->r_unsafe_item);
819 spin_unlock(&ci->i_unsafe_lock);
820 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
821 }
Sage Weil124e68e2009-10-06 11:31:08 -0700822}
823
Yan, Zheng9a5530c2016-06-15 16:29:18 +0800824/*
825 * Wait on any unsafe replies for the given inode. First wait on the
826 * newest request, and make that the upper bound. Then, if there are
827 * more requests, keep waiting on the oldest as long as it is still older
828 * than the original request.
829 */
830void ceph_sync_write_wait(struct inode *inode)
831{
832 struct ceph_inode_info *ci = ceph_inode(inode);
833 struct list_head *head = &ci->i_unsafe_writes;
834 struct ceph_osd_request *req;
835 u64 last_tid;
836
837 if (!S_ISREG(inode->i_mode))
838 return;
839
840 spin_lock(&ci->i_unsafe_lock);
841 if (list_empty(head))
842 goto out;
843
844 /* set upper bound as _last_ entry in chain */
845
846 req = list_last_entry(head, struct ceph_osd_request,
847 r_unsafe_item);
848 last_tid = req->r_tid;
849
850 do {
851 ceph_osdc_get_request(req);
852 spin_unlock(&ci->i_unsafe_lock);
853
854 dout("sync_write_wait on tid %llu (until %llu)\n",
855 req->r_tid, last_tid);
856 wait_for_completion(&req->r_safe_completion);
857 ceph_osdc_put_request(req);
858
859 spin_lock(&ci->i_unsafe_lock);
860 /*
861 * from here on look at first entry in chain, since we
862 * only want to wait for anything older than last_tid
863 */
864 if (list_empty(head))
865 break;
866 req = list_first_entry(head, struct ceph_osd_request,
867 r_unsafe_item);
868 } while (req->r_tid < last_tid);
869out:
870 spin_unlock(&ci->i_unsafe_lock);
871}
majianpenge8344e62013-09-12 13:54:26 +0800872
majianpenge8344e62013-09-12 13:54:26 +0800873static ssize_t
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800874ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
875 struct ceph_snap_context *snapc,
876 struct ceph_cap_flush **pcf)
Sage Weil124e68e2009-10-06 11:31:08 -0700877{
majianpenge8344e62013-09-12 13:54:26 +0800878 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500879 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700880 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700881 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500882 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700883 struct ceph_osd_request *req;
884 struct page **pages;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800885 struct ceph_aio_request *aio_req = NULL;
886 int num_pages = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700887 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700888 int ret;
Deepa Dinamanic2050a42016-09-14 07:48:06 -0700889 struct timespec mtime = current_time(inode);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800890 size_t count = iov_iter_count(iter);
891 loff_t pos = iocb->ki_pos;
892 bool write = iov_iter_rw(iter) == WRITE;
Sage Weil124e68e2009-10-06 11:31:08 -0700893
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800894 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700895 return -EROFS;
896
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800897 dout("sync_direct_read_write (%s) on file %p %lld~%u\n",
898 (write ? "write" : "read"), file, pos, (unsigned)count);
Sage Weil124e68e2009-10-06 11:31:08 -0700899
majianpenge8344e62013-09-12 13:54:26 +0800900 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800901 if (ret < 0)
902 return ret;
903
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800904 if (write) {
NeilBrown5d7eb1a2016-09-01 22:26:23 +0800905 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300906 pos >> PAGE_SHIFT,
907 (pos + count) >> PAGE_SHIFT);
NeilBrown5d7eb1a2016-09-01 22:26:23 +0800908 if (ret2 < 0)
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800909 dout("invalidate_inode_pages2_range returned %d\n", ret);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800910
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800911 flags = CEPH_OSD_FLAG_ORDERSNAP |
912 CEPH_OSD_FLAG_ONDISK |
913 CEPH_OSD_FLAG_WRITE;
914 } else {
915 flags = CEPH_OSD_FLAG_READ;
916 }
Sage Weil124e68e2009-10-06 11:31:08 -0700917
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800918 while (iov_iter_count(iter) > 0) {
919 u64 size = dio_get_pagev_size(iter);
920 size_t start = 0;
921 ssize_t len;
majianpenge8344e62013-09-12 13:54:26 +0800922
majianpenge8344e62013-09-12 13:54:26 +0800923 vino = ceph_vino(inode);
924 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800925 vino, pos, &size, 0,
926 /*include a 'startsync' command*/
927 write ? 2 : 1,
928 write ? CEPH_OSD_OP_WRITE :
929 CEPH_OSD_OP_READ,
930 flags, snapc,
majianpenge8344e62013-09-12 13:54:26 +0800931 ci->i_truncate_seq,
932 ci->i_truncate_size,
933 false);
934 if (IS_ERR(req)) {
935 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400936 break;
majianpenge8344e62013-09-12 13:54:26 +0800937 }
938
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800939 len = size;
940 pages = dio_get_pages_alloc(iter, len, &start, &num_pages);
Zhu, Caifengb5b98982015-10-08 15:26:15 +0800941 if (IS_ERR(pages)) {
Al Viro64c31312014-04-03 22:58:25 -0400942 ceph_osdc_put_request(req);
Zhu, Caifengb5b98982015-10-08 15:26:15 +0800943 ret = PTR_ERR(pages);
Al Viro64c31312014-04-03 22:58:25 -0400944 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700945 }
946
947 /*
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800948 * To simplify error handling, allow AIO when IO within i_size
949 * or IO can be satisfied by single OSD request.
Sage Weil124e68e2009-10-06 11:31:08 -0700950 */
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800951 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
952 (len == count || pos + count <= i_size_read(inode))) {
953 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
954 if (aio_req) {
955 aio_req->iocb = iocb;
956 aio_req->write = write;
957 INIT_LIST_HEAD(&aio_req->osd_reqs);
958 if (write) {
Yan, Zheng5be03892015-12-24 08:44:20 +0800959 aio_req->mtime = mtime;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800960 swap(aio_req->prealloc_cf, *pcf);
961 }
962 }
963 /* ignore error */
964 }
majianpenge8344e62013-09-12 13:54:26 +0800965
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800966 if (write) {
967 /*
968 * throw out any page cache pages in this range. this
969 * may block.
970 */
971 truncate_inode_pages_range(inode->i_mapping, pos,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300972 (pos+len) | (PAGE_SIZE - 1));
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800973
974 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
Ilya Dryomovbb873b52016-05-26 00:29:52 +0200975 req->r_mtime = mtime;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800976 }
977
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800978 osd_req_op_extent_osd_data_pages(req, 0, pages, len, start,
979 false, false);
980
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800981 if (aio_req) {
982 aio_req->total_len += len;
983 aio_req->num_reqs++;
984 atomic_inc(&aio_req->pending_reqs);
985
986 req->r_callback = ceph_aio_complete_req;
987 req->r_inode = inode;
988 req->r_priv = aio_req;
989 list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs);
990
991 pos += len;
992 iov_iter_advance(iter, len);
993 continue;
994 }
995
996 ret = ceph_osdc_start_request(req->r_osdc, req, false);
majianpenge8344e62013-09-12 13:54:26 +0800997 if (!ret)
998 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
999
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001000 size = i_size_read(inode);
1001 if (!write) {
1002 if (ret == -ENOENT)
1003 ret = 0;
1004 if (ret >= 0 && ret < len && pos + ret < size) {
1005 int zlen = min_t(size_t, len - ret,
1006 size - pos - ret);
1007 ceph_zero_page_vector_range(start + ret, zlen,
1008 pages);
1009 ret += zlen;
1010 }
1011 if (ret >= 0)
1012 len = ret;
1013 }
1014
Yan, Zhenga22bd5f2016-05-26 10:30:13 +08001015 ceph_put_page_vector(pages, num_pages, !write);
majianpenge8344e62013-09-12 13:54:26 +08001016
majianpenge8344e62013-09-12 13:54:26 +08001017 ceph_osdc_put_request(req);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001018 if (ret < 0)
majianpenge8344e62013-09-12 13:54:26 +08001019 break;
Al Viro64c31312014-04-03 22:58:25 -04001020
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001021 pos += len;
1022 iov_iter_advance(iter, len);
1023
1024 if (!write && pos >= size)
1025 break;
1026
1027 if (write && pos > size) {
1028 if (ceph_inode_set_size(inode, pos))
Al Viro64c31312014-04-03 22:58:25 -04001029 ceph_check_caps(ceph_inode(inode),
1030 CHECK_CAPS_AUTHONLY,
1031 NULL);
1032 }
majianpenge8344e62013-09-12 13:54:26 +08001033 }
1034
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001035 if (aio_req) {
Yan, Zhengfc8c3892016-06-14 11:13:59 +08001036 LIST_HEAD(osd_reqs);
1037
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001038 if (aio_req->num_reqs == 0) {
1039 kfree(aio_req);
1040 return ret;
1041 }
1042
1043 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1044 CEPH_CAP_FILE_RD);
1045
Yan, Zhengfc8c3892016-06-14 11:13:59 +08001046 list_splice(&aio_req->osd_reqs, &osd_reqs);
1047 while (!list_empty(&osd_reqs)) {
1048 req = list_first_entry(&osd_reqs,
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001049 struct ceph_osd_request,
1050 r_unsafe_item);
1051 list_del_init(&req->r_unsafe_item);
1052 if (ret >= 0)
1053 ret = ceph_osdc_start_request(req->r_osdc,
1054 req, false);
1055 if (ret < 0) {
1056 req->r_result = ret;
Ilya Dryomov85e084f2016-04-28 16:07:24 +02001057 ceph_aio_complete_req(req);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001058 }
1059 }
1060 return -EIOCBQUEUED;
1061 }
1062
1063 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1064 ret = pos - iocb->ki_pos;
majianpenge8344e62013-09-12 13:54:26 +08001065 iocb->ki_pos = pos;
majianpenge8344e62013-09-12 13:54:26 +08001066 }
1067 return ret;
1068}
1069
majianpenge8344e62013-09-12 13:54:26 +08001070/*
1071 * Synchronous write, straight from __user pointer or user pages.
1072 *
1073 * If write spans object boundary, just do multiple writes. (For a
1074 * correct atomic write, we should e.g. take write locks on all
1075 * objects, rollback on failure, etc.)
1076 */
Yan, Zheng06fee302014-07-28 14:33:46 +08001077static ssize_t
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001078ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1079 struct ceph_snap_context *snapc)
majianpenge8344e62013-09-12 13:54:26 +08001080{
1081 struct file *file = iocb->ki_filp;
1082 struct inode *inode = file_inode(file);
1083 struct ceph_inode_info *ci = ceph_inode(inode);
1084 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
majianpenge8344e62013-09-12 13:54:26 +08001085 struct ceph_vino vino;
1086 struct ceph_osd_request *req;
1087 struct page **pages;
1088 u64 len;
1089 int num_pages;
1090 int written = 0;
1091 int flags;
1092 int check_caps = 0;
1093 int ret;
Deepa Dinamanic2050a42016-09-14 07:48:06 -07001094 struct timespec mtime = current_time(inode);
Al Viro4908b822014-04-03 23:09:01 -04001095 size_t count = iov_iter_count(from);
majianpenge8344e62013-09-12 13:54:26 +08001096
1097 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1098 return -EROFS;
1099
1100 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
1101
1102 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
1103 if (ret < 0)
1104 return ret;
1105
1106 ret = invalidate_inode_pages2_range(inode->i_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001107 pos >> PAGE_SHIFT,
1108 (pos + count) >> PAGE_SHIFT);
majianpenge8344e62013-09-12 13:54:26 +08001109 if (ret < 0)
1110 dout("invalidate_inode_pages2_range returned %d\n", ret);
1111
1112 flags = CEPH_OSD_FLAG_ORDERSNAP |
1113 CEPH_OSD_FLAG_ONDISK |
1114 CEPH_OSD_FLAG_WRITE |
1115 CEPH_OSD_FLAG_ACK;
1116
Al Viro4908b822014-04-03 23:09:01 -04001117 while ((len = iov_iter_count(from)) > 0) {
majianpenge8344e62013-09-12 13:54:26 +08001118 size_t left;
1119 int n;
1120
majianpenge8344e62013-09-12 13:54:26 +08001121 vino = ceph_vino(inode);
1122 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Yan, Zheng715e4cd2014-11-13 14:40:37 +08001123 vino, pos, &len, 0, 1,
majianpenge8344e62013-09-12 13:54:26 +08001124 CEPH_OSD_OP_WRITE, flags, snapc,
1125 ci->i_truncate_seq,
1126 ci->i_truncate_size,
1127 false);
1128 if (IS_ERR(req)) {
1129 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -04001130 break;
majianpenge8344e62013-09-12 13:54:26 +08001131 }
1132
1133 /*
1134 * write from beginning of first page,
1135 * regardless of io alignment
1136 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001137 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
majianpenge8344e62013-09-12 13:54:26 +08001138
Yan, Zheng687265e2015-06-13 17:27:05 +08001139 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
Sage Weil124e68e2009-10-06 11:31:08 -07001140 if (IS_ERR(pages)) {
1141 ret = PTR_ERR(pages);
1142 goto out;
1143 }
majianpenge8344e62013-09-12 13:54:26 +08001144
1145 left = len;
1146 for (n = 0; n < num_pages; n++) {
Ilya Dryomov125d7252014-01-28 19:16:18 +02001147 size_t plen = min_t(size_t, left, PAGE_SIZE);
Al Viro4908b822014-04-03 23:09:01 -04001148 ret = copy_page_from_iter(pages[n], 0, plen, from);
majianpenge8344e62013-09-12 13:54:26 +08001149 if (ret != plen) {
1150 ret = -EFAULT;
1151 break;
1152 }
1153 left -= ret;
majianpenge8344e62013-09-12 13:54:26 +08001154 }
1155
Sage Weil124e68e2009-10-06 11:31:08 -07001156 if (ret < 0) {
1157 ceph_release_page_vector(pages, num_pages);
1158 goto out;
1159 }
1160
majianpenge8344e62013-09-12 13:54:26 +08001161 /* get a second commit callback */
1162 req->r_unsafe_callback = ceph_sync_write_unsafe;
1163 req->r_inode = inode;
Sage Weil124e68e2009-10-06 11:31:08 -07001164
majianpenge8344e62013-09-12 13:54:26 +08001165 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1166 false, true);
Alex Elder02ee07d2013-03-14 14:09:06 -05001167
Ilya Dryomovbb873b52016-05-26 00:29:52 +02001168 req->r_mtime = mtime;
majianpenge8344e62013-09-12 13:54:26 +08001169 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1170 if (!ret)
1171 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -07001172
1173out:
majianpenge8344e62013-09-12 13:54:26 +08001174 ceph_osdc_put_request(req);
1175 if (ret == 0) {
1176 pos += len;
1177 written += len;
Sage Weil124e68e2009-10-06 11:31:08 -07001178
majianpenge8344e62013-09-12 13:54:26 +08001179 if (pos > i_size_read(inode)) {
1180 check_caps = ceph_inode_set_size(inode, pos);
1181 if (check_caps)
1182 ceph_check_caps(ceph_inode(inode),
1183 CHECK_CAPS_AUTHONLY,
1184 NULL);
1185 }
1186 } else
1187 break;
1188 }
1189
1190 if (ret != -EOLDSNAPC && written > 0) {
Sage Weil124e68e2009-10-06 11:31:08 -07001191 ret = written;
majianpenge8344e62013-09-12 13:54:26 +08001192 iocb->ki_pos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -07001193 }
1194 return ret;
1195}
1196
1197/*
1198 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1199 * Atomically grab references, so that those bits are not released
1200 * back to the MDS mid-read.
1201 *
1202 * Hmm, the sync read case isn't actually async... should it be?
1203 */
Al Viro36444242014-04-02 20:28:01 -04001204static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
Sage Weil124e68e2009-10-06 11:31:08 -07001205{
1206 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -07001207 struct ceph_file_info *fi = filp->private_data;
Christoph Hellwig66ee59a2015-02-11 19:56:46 +01001208 size_t len = iov_iter_count(to);
Al Viro496ad9a2013-01-23 17:07:38 -05001209 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -07001210 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001211 struct page *pinned_page = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -07001212 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -07001213 int want, got = 0;
Yan, Zheng83701242014-11-14 22:36:18 +08001214 int retry_op = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -07001215
Sage Weil6a026582010-02-09 14:04:02 -08001216again:
majianpeng8eb4efb2013-09-26 14:42:17 +08001217 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1218 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1219
Sage Weil29625072010-05-27 10:40:43 -07001220 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1221 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1222 else
1223 want = CEPH_CAP_FILE_CACHE;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001224 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
Sage Weil124e68e2009-10-06 11:31:08 -07001225 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +08001226 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -07001227
Sage Weil29625072010-05-27 10:40:43 -07001228 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -04001229 (iocb->ki_flags & IOCB_DIRECT) ||
majianpeng8eb4efb2013-09-26 14:42:17 +08001230 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -07001231
majianpeng8eb4efb2013-09-26 14:42:17 +08001232 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1233 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1234 ceph_cap_string(got));
1235
Yan, Zheng83701242014-11-14 22:36:18 +08001236 if (ci->i_inline_version == CEPH_INLINE_NONE) {
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001237 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1238 ret = ceph_direct_read_write(iocb, to,
1239 NULL, NULL);
1240 if (ret >= 0 && ret < len)
1241 retry_op = CHECK_EOF;
1242 } else {
1243 ret = ceph_sync_read(iocb, to, &retry_op);
1244 }
Yan, Zheng83701242014-11-14 22:36:18 +08001245 } else {
1246 retry_op = READ_INLINE;
1247 }
majianpeng8eb4efb2013-09-26 14:42:17 +08001248 } else {
majianpeng8eb4efb2013-09-26 14:42:17 +08001249 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
Al Viro36444242014-04-02 20:28:01 -04001250 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
majianpeng8eb4efb2013-09-26 14:42:17 +08001251 ceph_cap_string(got));
Yan, Zhenge99b0ea2016-10-25 10:51:55 +08001252 current->journal_info = filp;
Al Viro36444242014-04-02 20:28:01 -04001253 ret = generic_file_read_iter(iocb, to);
Yan, Zhenge99b0ea2016-10-25 10:51:55 +08001254 current->journal_info = NULL;
majianpeng8eb4efb2013-09-26 14:42:17 +08001255 }
Sage Weil124e68e2009-10-06 11:31:08 -07001256 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1257 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001258 if (pinned_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001259 put_page(pinned_page);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001260 pinned_page = NULL;
1261 }
Sage Weil124e68e2009-10-06 11:31:08 -07001262 ceph_put_cap_refs(ci, got);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001263 if (retry_op > HAVE_RETRIED && ret >= 0) {
Yan, Zheng83701242014-11-14 22:36:18 +08001264 int statret;
1265 struct page *page = NULL;
1266 loff_t i_size;
1267 if (retry_op == READ_INLINE) {
Yan, Zheng687265e2015-06-13 17:27:05 +08001268 page = __page_cache_alloc(GFP_KERNEL);
Yan, Zheng83701242014-11-14 22:36:18 +08001269 if (!page)
1270 return -ENOMEM;
1271 }
Sage Weil6a026582010-02-09 14:04:02 -08001272
Yan, Zheng83701242014-11-14 22:36:18 +08001273 statret = __ceph_do_getattr(inode, page,
1274 CEPH_STAT_CAP_INLINE_DATA, !!page);
1275 if (statret < 0) {
Nikolay Borisov0d7718f62016-10-10 15:38:18 +03001276 if (page)
1277 __free_page(page);
Yan, Zheng83701242014-11-14 22:36:18 +08001278 if (statret == -ENODATA) {
1279 BUG_ON(retry_op != READ_INLINE);
1280 goto again;
1281 }
1282 return statret;
1283 }
1284
1285 i_size = i_size_read(inode);
1286 if (retry_op == READ_INLINE) {
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001287 BUG_ON(ret > 0 || read > 0);
1288 if (iocb->ki_pos < i_size &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001289 iocb->ki_pos < PAGE_SIZE) {
Yan, Zheng83701242014-11-14 22:36:18 +08001290 loff_t end = min_t(loff_t, i_size,
1291 iocb->ki_pos + len);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001292 end = min_t(loff_t, end, PAGE_SIZE);
Yan, Zheng83701242014-11-14 22:36:18 +08001293 if (statret < end)
1294 zero_user_segment(page, statret, end);
1295 ret = copy_page_to_iter(page,
1296 iocb->ki_pos & ~PAGE_MASK,
1297 end - iocb->ki_pos, to);
1298 iocb->ki_pos += ret;
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001299 read += ret;
1300 }
1301 if (iocb->ki_pos < i_size && read < len) {
1302 size_t zlen = min_t(size_t, len - read,
1303 i_size - iocb->ki_pos);
1304 ret = iov_iter_zero(zlen, to);
1305 iocb->ki_pos += ret;
1306 read += ret;
Yan, Zheng83701242014-11-14 22:36:18 +08001307 }
1308 __free_pages(page, 0);
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001309 return read;
Yan, Zheng83701242014-11-14 22:36:18 +08001310 }
Sage Weil6a026582010-02-09 14:04:02 -08001311
1312 /* hit EOF or hole? */
Yan, Zheng83701242014-11-14 22:36:18 +08001313 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001314 ret < len) {
majianpeng8eb4efb2013-09-26 14:42:17 +08001315 dout("sync_read hit hole, ppos %lld < size %lld"
Yan, Zheng99c88e62015-12-30 11:32:46 +08001316 ", reading more\n", iocb->ki_pos, i_size);
majianpeng8eb4efb2013-09-26 14:42:17 +08001317
Sage Weil6a026582010-02-09 14:04:02 -08001318 read += ret;
Sage Weil6a026582010-02-09 14:04:02 -08001319 len -= ret;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001320 retry_op = HAVE_RETRIED;
Sage Weil6a026582010-02-09 14:04:02 -08001321 goto again;
1322 }
1323 }
majianpeng8eb4efb2013-09-26 14:42:17 +08001324
Sage Weil6a026582010-02-09 14:04:02 -08001325 if (ret >= 0)
1326 ret += read;
1327
Sage Weil124e68e2009-10-06 11:31:08 -07001328 return ret;
1329}
1330
1331/*
1332 * Take cap references to avoid releasing caps to MDS mid-write.
1333 *
1334 * If we are synchronous, and write with an old snap context, the OSD
1335 * may return EOLDSNAPC. In that case, retry the write.. _after_
1336 * dropping our cap refs and allowing the pending snap to logically
1337 * complete _before_ this write occurs.
1338 *
1339 * If we are near ENOSPC, write synchronously.
1340 */
Al Viro4908b822014-04-03 23:09:01 -04001341static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
Sage Weil124e68e2009-10-06 11:31:08 -07001342{
1343 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -07001344 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -05001345 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -07001346 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001347 struct ceph_osd_client *osdc =
1348 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001349 struct ceph_cap_flush *prealloc_cf;
Al Viro3309dd02015-04-09 12:55:47 -04001350 ssize_t count, written = 0;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001351 int err, want, got;
Al Viro3309dd02015-04-09 12:55:47 -04001352 loff_t pos;
Sage Weil124e68e2009-10-06 11:31:08 -07001353
1354 if (ceph_snap(inode) != CEPH_NOSNAP)
1355 return -EROFS;
1356
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001357 prealloc_cf = ceph_alloc_cap_flush();
1358 if (!prealloc_cf)
1359 return -ENOMEM;
1360
Al Viro59551022016-01-22 15:40:57 -05001361 inode_lock(inode);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001362
Yan, Zheng03d254e2013-04-12 16:11:13 +08001363 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01001364 current->backing_dev_info = inode_to_bdi(inode);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001365
Yan, Zheng55b0b312015-09-07 11:35:01 +08001366 if (iocb->ki_flags & IOCB_APPEND) {
1367 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1368 if (err < 0)
1369 goto out;
1370 }
1371
Al Viro3309dd02015-04-09 12:55:47 -04001372 err = generic_write_checks(iocb, from);
1373 if (err <= 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +08001374 goto out;
1375
Al Viro3309dd02015-04-09 12:55:47 -04001376 pos = iocb->ki_pos;
1377 count = iov_iter_count(from);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001378 err = file_remove_privs(file);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001379 if (err)
1380 goto out;
1381
1382 err = file_update_time(file);
1383 if (err)
1384 goto out;
1385
Yan, Zheng28127bd2014-11-14 22:38:29 +08001386 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1387 err = ceph_uninline_data(file, NULL);
1388 if (err < 0)
1389 goto out;
1390 }
1391
Sage Weil124e68e2009-10-06 11:31:08 -07001392retry_snap:
Ilya Dryomovb7ec35b2016-04-28 16:07:25 +02001393 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +08001394 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001395 goto out;
1396 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001397
Randy Dunlapac7f29b2013-04-19 14:20:07 -07001398 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng99c88e62015-12-30 11:32:46 +08001399 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
Sage Weil7971bd92013-05-01 21:15:58 -07001400 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1401 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1402 else
1403 want = CEPH_CAP_FILE_BUFFER;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001404 got = 0;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001405 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1406 &got, NULL);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001407 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +08001408 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -07001409
Randy Dunlapac7f29b2013-04-19 14:20:07 -07001410 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +08001411 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -07001412
1413 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -04001414 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001415 struct ceph_snap_context *snapc;
Al Viro4908b822014-04-03 23:09:01 -04001416 struct iov_iter data;
Al Viro59551022016-01-22 15:40:57 -05001417 inode_unlock(inode);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001418
1419 spin_lock(&ci->i_ceph_lock);
1420 if (__ceph_have_pending_cap_snap(ci)) {
1421 struct ceph_cap_snap *capsnap =
1422 list_last_entry(&ci->i_cap_snaps,
1423 struct ceph_cap_snap,
1424 ci_item);
1425 snapc = ceph_get_snap_context(capsnap->context);
1426 } else {
1427 BUG_ON(!ci->i_head_snapc);
1428 snapc = ceph_get_snap_context(ci->i_head_snapc);
1429 }
1430 spin_unlock(&ci->i_ceph_lock);
1431
Al Viro4908b822014-04-03 23:09:01 -04001432 /* we might need to revert back to that point */
1433 data = *from;
Al Viro2ba48ce2015-04-09 13:52:01 -04001434 if (iocb->ki_flags & IOCB_DIRECT)
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001435 written = ceph_direct_read_write(iocb, &data, snapc,
1436 &prealloc_cf);
majianpenge8344e62013-09-12 13:54:26 +08001437 else
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001438 written = ceph_sync_write(iocb, &data, pos, snapc);
majianpeng0e5dd452013-08-08 15:32:19 +08001439 if (written == -EOLDSNAPC) {
1440 dout("aio_write %p %llx.%llx %llu~%u"
1441 "got EOLDSNAPC, retrying\n",
1442 inode, ceph_vinop(inode),
Al Viro4908b822014-04-03 23:09:01 -04001443 pos, (unsigned)count);
Al Viro59551022016-01-22 15:40:57 -05001444 inode_lock(inode);
majianpeng0e5dd452013-08-08 15:32:19 +08001445 goto retry_snap;
1446 }
Al Viro4908b822014-04-03 23:09:01 -04001447 if (written > 0)
1448 iov_iter_advance(from, written);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001449 ceph_put_snap_context(snapc);
Sage Weil7971bd92013-05-01 21:15:58 -07001450 } else {
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001451 /*
1452 * No need to acquire the i_truncate_mutex. Because
1453 * the MDS revokes Fwb caps before sending truncate
1454 * message to us. We can't get Fwb cap while there
1455 * are pending vmtruncate. So write and vmtruncate
1456 * can not run at the same time
1457 */
Al Viro4908b822014-04-03 23:09:01 -04001458 written = generic_perform_write(file, from, pos);
Al Viroaec605f42014-02-11 22:28:43 -05001459 if (likely(written >= 0))
1460 iocb->ki_pos = pos + written;
Al Viro59551022016-01-22 15:40:57 -05001461 inode_unlock(inode);
Sage Weil124e68e2009-10-06 11:31:08 -07001462 }
Sage Weild8de9ab2011-07-26 11:27:34 -07001463
Yan, Zheng03d254e2013-04-12 16:11:13 +08001464 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -07001465 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -08001466 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001467 ci->i_inline_version = CEPH_INLINE_NONE;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001468 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1469 &prealloc_cf);
Sage Weilbe655592011-11-30 09:47:09 -08001470 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -07001471 if (dirty)
1472 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -07001473 }
Sage Weil7971bd92013-05-01 21:15:58 -07001474
Sage Weil124e68e2009-10-06 11:31:08 -07001475 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Al Viro4908b822014-04-03 23:09:01 -04001476 inode, ceph_vinop(inode), pos, (unsigned)count,
Sage Weil7971bd92013-05-01 21:15:58 -07001477 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -07001478 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -07001479
Christoph Hellwig6aa657c2016-04-07 08:52:02 -07001480 if (written >= 0) {
Ilya Dryomovb7ec35b2016-04-28 16:07:25 +02001481 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL))
Christoph Hellwig6aa657c2016-04-07 08:52:02 -07001482 iocb->ki_flags |= IOCB_DSYNC;
1483
1484 written = generic_write_sync(iocb, written);
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001485 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001486
Sage Weil2f75e9e2013-08-09 09:57:58 -07001487 goto out_unlocked;
Sage Weil124e68e2009-10-06 11:31:08 -07001488
Sage Weil2f75e9e2013-08-09 09:57:58 -07001489out:
Al Viro59551022016-01-22 15:40:57 -05001490 inode_unlock(inode);
Sage Weil2f75e9e2013-08-09 09:57:58 -07001491out_unlocked:
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001492 ceph_free_cap_flush(prealloc_cf);
Sage Weil2f75e9e2013-08-09 09:57:58 -07001493 current->backing_dev_info = NULL;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001494 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -07001495}
1496
1497/*
1498 * llseek. be sure to verify file size on SEEK_END.
1499 */
Andrew Morton965c8e52012-12-17 15:59:39 -08001500static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -07001501{
1502 struct inode *inode = file->f_mapping->host;
Yan, Zheng99c88e62015-12-30 11:32:46 +08001503 loff_t i_size;
Phil Turnbull955818c2016-07-21 13:43:09 -04001504 loff_t ret;
Sage Weil124e68e2009-10-06 11:31:08 -07001505
Al Viro59551022016-01-22 15:40:57 -05001506 inode_lock(inode);
Sage Weil6a82c472011-12-13 09:19:26 -08001507
Andrew Morton965c8e52012-12-17 15:59:39 -08001508 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Yan, Zheng508b32d2014-09-16 21:46:17 +08001509 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
Phil Turnbull955818c2016-07-21 13:43:09 -04001510 if (ret < 0)
Sage Weil124e68e2009-10-06 11:31:08 -07001511 goto out;
Josef Bacik06222e42011-07-18 13:21:38 -04001512 }
1513
Yan, Zheng99c88e62015-12-30 11:32:46 +08001514 i_size = i_size_read(inode);
Andrew Morton965c8e52012-12-17 15:59:39 -08001515 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -04001516 case SEEK_END:
Yan, Zheng99c88e62015-12-30 11:32:46 +08001517 offset += i_size;
Sage Weil124e68e2009-10-06 11:31:08 -07001518 break;
1519 case SEEK_CUR:
1520 /*
1521 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1522 * position-querying operation. Avoid rewriting the "same"
1523 * f_pos value back to the file because a concurrent read(),
1524 * write() or lseek() might have altered it
1525 */
1526 if (offset == 0) {
Phil Turnbull955818c2016-07-21 13:43:09 -04001527 ret = file->f_pos;
Sage Weil124e68e2009-10-06 11:31:08 -07001528 goto out;
1529 }
1530 offset += file->f_pos;
1531 break;
Josef Bacik06222e42011-07-18 13:21:38 -04001532 case SEEK_DATA:
Yan, Zheng99c88e62015-12-30 11:32:46 +08001533 if (offset >= i_size) {
Josef Bacik06222e42011-07-18 13:21:38 -04001534 ret = -ENXIO;
1535 goto out;
1536 }
1537 break;
1538 case SEEK_HOLE:
Yan, Zheng99c88e62015-12-30 11:32:46 +08001539 if (offset >= i_size) {
Josef Bacik06222e42011-07-18 13:21:38 -04001540 ret = -ENXIO;
1541 goto out;
1542 }
Yan, Zheng99c88e62015-12-30 11:32:46 +08001543 offset = i_size;
Josef Bacik06222e42011-07-18 13:21:38 -04001544 break;
Sage Weil124e68e2009-10-06 11:31:08 -07001545 }
1546
Phil Turnbull955818c2016-07-21 13:43:09 -04001547 ret = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -07001548
1549out:
Al Viro59551022016-01-22 15:40:57 -05001550 inode_unlock(inode);
Phil Turnbull955818c2016-07-21 13:43:09 -04001551 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -07001552}
1553
Li Wangad7a60d2013-08-15 11:51:44 +08001554static inline void ceph_zero_partial_page(
1555 struct inode *inode, loff_t offset, unsigned size)
1556{
1557 struct page *page;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001558 pgoff_t index = offset >> PAGE_SHIFT;
Li Wangad7a60d2013-08-15 11:51:44 +08001559
1560 page = find_lock_page(inode->i_mapping, index);
1561 if (page) {
1562 wait_on_page_writeback(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001563 zero_user(page, offset & (PAGE_SIZE - 1), size);
Li Wangad7a60d2013-08-15 11:51:44 +08001564 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001565 put_page(page);
Li Wangad7a60d2013-08-15 11:51:44 +08001566 }
1567}
1568
1569static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1570 loff_t length)
1571{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001572 loff_t nearly = round_up(offset, PAGE_SIZE);
Li Wangad7a60d2013-08-15 11:51:44 +08001573 if (offset < nearly) {
1574 loff_t size = nearly - offset;
1575 if (length < size)
1576 size = length;
1577 ceph_zero_partial_page(inode, offset, size);
1578 offset += size;
1579 length -= size;
1580 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001581 if (length >= PAGE_SIZE) {
1582 loff_t size = round_down(length, PAGE_SIZE);
Li Wangad7a60d2013-08-15 11:51:44 +08001583 truncate_pagecache_range(inode, offset, offset + size - 1);
1584 offset += size;
1585 length -= size;
1586 }
1587 if (length)
1588 ceph_zero_partial_page(inode, offset, length);
1589}
1590
1591static int ceph_zero_partial_object(struct inode *inode,
1592 loff_t offset, loff_t *length)
1593{
1594 struct ceph_inode_info *ci = ceph_inode(inode);
1595 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1596 struct ceph_osd_request *req;
1597 int ret = 0;
1598 loff_t zero = 0;
1599 int op;
1600
1601 if (!length) {
1602 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1603 length = &zero;
1604 } else {
1605 op = CEPH_OSD_OP_ZERO;
1606 }
1607
1608 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1609 ceph_vino(inode),
1610 offset, length,
Yan, Zheng715e4cd2014-11-13 14:40:37 +08001611 0, 1, op,
Li Wangad7a60d2013-08-15 11:51:44 +08001612 CEPH_OSD_FLAG_WRITE |
1613 CEPH_OSD_FLAG_ONDISK,
1614 NULL, 0, 0, false);
1615 if (IS_ERR(req)) {
1616 ret = PTR_ERR(req);
1617 goto out;
1618 }
1619
Ilya Dryomovbb873b52016-05-26 00:29:52 +02001620 req->r_mtime = inode->i_mtime;
Li Wangad7a60d2013-08-15 11:51:44 +08001621 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1622 if (!ret) {
1623 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1624 if (ret == -ENOENT)
1625 ret = 0;
1626 }
1627 ceph_osdc_put_request(req);
1628
1629out:
1630 return ret;
1631}
1632
1633static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1634{
1635 int ret = 0;
1636 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng76271512016-02-03 21:24:49 +08001637 s32 stripe_unit = ci->i_layout.stripe_unit;
1638 s32 stripe_count = ci->i_layout.stripe_count;
1639 s32 object_size = ci->i_layout.object_size;
Sage Weilb314a902013-08-27 12:15:16 -07001640 u64 object_set_size = object_size * stripe_count;
1641 u64 nearly, t;
Li Wangad7a60d2013-08-15 11:51:44 +08001642
Sage Weilb314a902013-08-27 12:15:16 -07001643 /* round offset up to next period boundary */
1644 nearly = offset + object_set_size - 1;
1645 t = nearly;
1646 nearly -= do_div(t, object_set_size);
1647
Li Wangad7a60d2013-08-15 11:51:44 +08001648 while (length && offset < nearly) {
1649 loff_t size = length;
1650 ret = ceph_zero_partial_object(inode, offset, &size);
1651 if (ret < 0)
1652 return ret;
1653 offset += size;
1654 length -= size;
1655 }
1656 while (length >= object_set_size) {
1657 int i;
1658 loff_t pos = offset;
1659 for (i = 0; i < stripe_count; ++i) {
1660 ret = ceph_zero_partial_object(inode, pos, NULL);
1661 if (ret < 0)
1662 return ret;
1663 pos += stripe_unit;
1664 }
1665 offset += object_set_size;
1666 length -= object_set_size;
1667 }
1668 while (length) {
1669 loff_t size = length;
1670 ret = ceph_zero_partial_object(inode, offset, &size);
1671 if (ret < 0)
1672 return ret;
1673 offset += size;
1674 length -= size;
1675 }
1676 return ret;
1677}
1678
1679static long ceph_fallocate(struct file *file, int mode,
1680 loff_t offset, loff_t length)
1681{
1682 struct ceph_file_info *fi = file->private_data;
Libo Chenaa8b60e2013-12-11 13:49:11 +08001683 struct inode *inode = file_inode(file);
Li Wangad7a60d2013-08-15 11:51:44 +08001684 struct ceph_inode_info *ci = ceph_inode(inode);
1685 struct ceph_osd_client *osdc =
1686 &ceph_inode_to_client(inode)->client->osdc;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001687 struct ceph_cap_flush *prealloc_cf;
Li Wangad7a60d2013-08-15 11:51:44 +08001688 int want, got = 0;
1689 int dirty;
1690 int ret = 0;
1691 loff_t endoff = 0;
1692 loff_t size;
1693
Yan, Zheng494d77b2014-06-26 15:25:17 +08001694 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1695 return -EOPNOTSUPP;
1696
Li Wangad7a60d2013-08-15 11:51:44 +08001697 if (!S_ISREG(inode->i_mode))
1698 return -EOPNOTSUPP;
1699
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001700 prealloc_cf = ceph_alloc_cap_flush();
1701 if (!prealloc_cf)
1702 return -ENOMEM;
1703
Al Viro59551022016-01-22 15:40:57 -05001704 inode_lock(inode);
Li Wangad7a60d2013-08-15 11:51:44 +08001705
1706 if (ceph_snap(inode) != CEPH_NOSNAP) {
1707 ret = -EROFS;
1708 goto unlock;
1709 }
1710
Ilya Dryomovb7ec35b2016-04-28 16:07:25 +02001711 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) &&
1712 !(mode & FALLOC_FL_PUNCH_HOLE)) {
Li Wangad7a60d2013-08-15 11:51:44 +08001713 ret = -ENOSPC;
1714 goto unlock;
1715 }
1716
Yan, Zheng28127bd2014-11-14 22:38:29 +08001717 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1718 ret = ceph_uninline_data(file, NULL);
1719 if (ret < 0)
1720 goto unlock;
1721 }
1722
Li Wangad7a60d2013-08-15 11:51:44 +08001723 size = i_size_read(inode);
1724 if (!(mode & FALLOC_FL_KEEP_SIZE))
1725 endoff = offset + length;
1726
1727 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1728 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1729 else
1730 want = CEPH_CAP_FILE_BUFFER;
1731
Yan, Zheng3738daa2014-11-14 22:10:07 +08001732 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
Li Wangad7a60d2013-08-15 11:51:44 +08001733 if (ret < 0)
1734 goto unlock;
1735
1736 if (mode & FALLOC_FL_PUNCH_HOLE) {
1737 if (offset < size)
1738 ceph_zero_pagecache_range(inode, offset, length);
1739 ret = ceph_zero_objects(inode, offset, length);
1740 } else if (endoff > size) {
1741 truncate_pagecache_range(inode, size, -1);
1742 if (ceph_inode_set_size(inode, endoff))
1743 ceph_check_caps(ceph_inode(inode),
1744 CHECK_CAPS_AUTHONLY, NULL);
1745 }
1746
1747 if (!ret) {
1748 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001749 ci->i_inline_version = CEPH_INLINE_NONE;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001750 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1751 &prealloc_cf);
Li Wangad7a60d2013-08-15 11:51:44 +08001752 spin_unlock(&ci->i_ceph_lock);
1753 if (dirty)
1754 __mark_inode_dirty(inode, dirty);
1755 }
1756
1757 ceph_put_cap_refs(ci, got);
1758unlock:
Al Viro59551022016-01-22 15:40:57 -05001759 inode_unlock(inode);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001760 ceph_free_cap_flush(prealloc_cf);
Li Wangad7a60d2013-08-15 11:51:44 +08001761 return ret;
1762}
1763
Sage Weil124e68e2009-10-06 11:31:08 -07001764const struct file_operations ceph_file_fops = {
1765 .open = ceph_open,
1766 .release = ceph_release,
1767 .llseek = ceph_llseek,
Al Viro36444242014-04-02 20:28:01 -04001768 .read_iter = ceph_read_iter,
Al Viro4908b822014-04-03 23:09:01 -04001769 .write_iter = ceph_write_iter,
Sage Weil124e68e2009-10-06 11:31:08 -07001770 .mmap = ceph_mmap,
1771 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -07001772 .lock = ceph_lock,
1773 .flock = ceph_flock,
Al Viro3551dd72014-04-05 04:40:12 -04001774 .splice_write = iter_file_splice_write,
Sage Weil124e68e2009-10-06 11:31:08 -07001775 .unlocked_ioctl = ceph_ioctl,
1776 .compat_ioctl = ceph_ioctl,
Li Wangad7a60d2013-08-15 11:51:44 +08001777 .fallocate = ceph_fallocate,
Sage Weil124e68e2009-10-06 11:31:08 -07001778};
1779