blob: 3b6b522b4b31ed9e2f7193c661894787bfe63627 [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
37
38/*
39 * Prepare an open request. Preallocate ceph_cap to avoid an
40 * inopportune ENOMEM later.
41 */
42static struct ceph_mds_request *
43prepare_open_request(struct super_block *sb, int flags, int create_mode)
44{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070045 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
46 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -070047 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);
Sage Weil124e68e2009-10-06 11:31:08 -070060out:
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 */
68static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
69{
70 struct ceph_file_info *cf;
71 int ret = 0;
Milosz Tanski99ccbd22013-08-21 17:29:54 -040072 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;
Sage Weil124e68e2009-10-06 11:31:08 -070075
76 switch (inode->i_mode & S_IFMT) {
77 case S_IFREG:
Milosz Tanski99ccbd22013-08-21 17:29:54 -040078 /* 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);
Sage Weil124e68e2009-10-06 11:31:08 -070089 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/*
Sage Weil124e68e2009-10-06 11:31:08 -0700126 * 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 */
131int ceph_open(struct inode *inode, struct file *file)
132{
133 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700134 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
135 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700136 struct ceph_mds_request *req;
137 struct ceph_file_info *cf = file->private_data;
Sage Weil5f21c962011-07-26 11:30:29 -0700138 struct inode *parent_inode = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700139 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) {
Sage Weilbe655592011-11-30 09:47:09 -0800163 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700164 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800165 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700166 return ceph_init_file(inode, file, fmode);
167 }
168
169 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800170 * No need to block if we have caps on the auth MDS (for
171 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700172 * asynchronously.
173 */
Sage Weilbe655592011-11-30 09:47:09 -0800174 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800175 if (__ceph_is_any_real_caps(ci) &&
176 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700177 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);
Sage Weilbe655592011-11-30 09:47:09 -0800184 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700185
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);
Sage Weilbe655592011-11-30 09:47:09 -0800196 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700197 return ceph_init_file(inode, file, fmode);
198 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400199
Sage Weilbe655592011-11-30 09:47:09 -0800200 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700201
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 }
Sage Weil70b666c2011-05-27 09:24:26 -0700208 req->r_inode = inode;
209 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400210
Sage Weil124e68e2009-10-06 11:31:08 -0700211 req->r_num_caps = 1;
Sage Weil752c8bd2013-02-05 13:52:29 -0800212 if (flags & O_CREAT)
Al Virob5830432014-10-31 01:22:04 -0400213 parent_inode = ceph_get_dentry_parent_inode(file->f_path.dentry);
Sage Weil124e68e2009-10-06 11:31:08 -0700214 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -0700215 iput(parent_inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700216 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));
220out:
221 return err;
222}
223
224
225/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700226 * 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.
Sage Weil124e68e2009-10-06 11:31:08 -0700228 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700229int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400230 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400231 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700232{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700233 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
234 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700235 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700236 struct dentry *dn;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800237 struct ceph_acls_info acls = {};
Sage Weil124e68e2009-10-06 11:31:08 -0700238 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700239
Al Viroa4555892014-10-21 20:11:25 -0400240 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
241 dir, dentry, dentry,
Sage Weil5ef50c32012-07-31 11:27:36 -0700242 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;
Sage Weil124e68e2009-10-06 11:31:08 -0700250
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800251 if (flags & O_CREAT) {
252 err = ceph_pre_init_acls(dir, &mode, &acls);
253 if (err < 0)
254 return err;
255 }
256
Sage Weil124e68e2009-10-06 11:31:08 -0700257 /* do the open */
258 req = prepare_open_request(dir->i_sb, flags, mode);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800259 if (IS_ERR(req)) {
260 err = PTR_ERR(req);
261 goto out_acl;
262 }
Sage Weil124e68e2009-10-06 11:31:08 -0700263 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;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800268 if (acls.pagelist) {
269 req->r_pagelist = acls.pagelist;
270 acls.pagelist = NULL;
271 }
Sage Weil124e68e2009-10-06 11:31:08 -0700272 }
273 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700274 err = ceph_mdsc_do_request(mdsc,
275 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
276 req);
Yan, Zhengbf91c312015-01-19 13:23:20 +0800277 err = ceph_handle_snapdir(req, dentry, err);
Sam Lang79aec9842012-12-19 09:44:23 -1000278 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800279 goto out_req;
Sam Lang79aec9842012-12-19 09:44:23 -1000280
Sage Weil5ef50c32012-07-31 11:27:36 -0700281 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700282 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700283
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 }
Sage Weil468640e2011-07-26 11:28:11 -0700292 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800293 goto out_req;
David Howells2b0143b2015-03-17 22:25:59 +0000294 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
Sage Weil5ef50c32012-07-31 11:27:36 -0700295 /* 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);
Sam Lang6e8575f2012-12-28 09:56:46 -0800300 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
David Howells2b0143b2015-03-17 22:25:59 +0000301 ceph_init_inode_acls(d_inode(dentry), &acls);
Sam Lang6e8575f2012-12-28 09:56:46 -0800302 *opened |= FILE_CREATED;
303 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700304 err = finish_open(file, dentry, ceph_open, opened);
305 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800306out_req:
Yan, Zhengab866542014-04-01 20:20:17 +0800307 if (!req->r_err && req->r_target_inode)
308 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
Sage Weil124e68e2009-10-06 11:31:08 -0700309 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800310out_acl:
311 ceph_release_acls_info(&acls);
Sage Weil5ef50c32012-07-31 11:27:36 -0700312 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400313 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700314}
315
316int 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);
Sage Weil195d3ce2010-03-01 09:57:54 -0800329
330 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700331 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700332 return 0;
333}
334
Yan, Zheng83701242014-11-14 22:36:18 +0800335enum {
336 CHECK_EOF = 1,
337 READ_INLINE = 2,
338};
339
Sage Weil124e68e2009-10-06 11:31:08 -0700340/*
Sage Weil124e68e2009-10-06 11:31:08 -0700341 * 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 */
347static int striped_read(struct inode *inode,
348 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800349 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700350 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800351 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700352{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700353 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700354 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpenter688bac42013-07-23 16:48:01 +0300355 u64 pos, this_len, left;
Sage Weilb7495fc2010-11-09 12:43:12 -0800356 int io_align, page_align;
Dan Carpenter688bac42013-07-23 16:48:01 +0300357 int pages_left;
Sage Weil124e68e2009-10-06 11:31:08 -0700358 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;
Sage Weilb7495fc2010-11-09 12:43:12 -0800371 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700372
373more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700374 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800375 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800376 else
377 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700378 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700379 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700380 &ci->i_layout, pos, &this_len,
381 ci->i_truncate_seq,
382 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800383 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700384 if (ret == -ENOENT)
385 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700386 hit_stripe = this_len < left;
387 was_short = ret >= 0 && ret < this_len;
Dan Carpenter688bac42013-07-23 16:48:01 +0300388 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
Sage Weil124e68e2009-10-06 11:31:08 -0700389 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
390
majianpeng02ae66d2013-08-06 16:20:38 +0800391 if (ret >= 0) {
392 int didpages;
393 if (was_short && (pos + ret < inode->i_size)) {
Yan, Zheng1487a682015-01-06 15:29:14 +0800394 int zlen = min(this_len - ret,
395 inode->i_size - pos - ret);
396 int zoff = (o_direct ? buf_align : io_align) +
397 read + ret;
majianpeng02ae66d2013-08-06 16:20:38 +0800398 dout(" zero gap %llu to %llu\n",
Yan, Zheng1487a682015-01-06 15:29:14 +0800399 pos + ret, pos + ret + zlen);
400 ceph_zero_page_vector_range(zoff, zlen, pages);
401 ret += zlen;
Sage Weil124e68e2009-10-06 11:31:08 -0700402 }
majianpeng02ae66d2013-08-06 16:20:38 +0800403
404 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700405 pos += ret;
406 read = pos - off;
407 left -= ret;
408 page_pos += didpages;
409 pages_left -= didpages;
410
majianpeng02ae66d2013-08-06 16:20:38 +0800411 /* hit stripe and need continue*/
412 if (left && hit_stripe && pos < inode->i_size)
Sage Weil124e68e2009-10-06 11:31:08 -0700413 goto more;
414 }
415
majianpengee7289b2013-08-21 15:02:51 +0800416 if (read > 0) {
majianpeng02ae66d2013-08-06 16:20:38 +0800417 ret = read;
Sage Weilc3cd6282011-06-01 16:08:44 -0700418 /* did we bounce off eof? */
419 if (pos + left > inode->i_size)
Yan, Zheng83701242014-11-14 22:36:18 +0800420 *checkeof = CHECK_EOF;
Sage Weil124e68e2009-10-06 11:31:08 -0700421 }
422
Sage Weil124e68e2009-10-06 11:31:08 -0700423 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 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800433static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
434 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700435{
majianpeng8eb4efb2013-09-26 14:42:17 +0800436 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500437 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700438 struct page **pages;
majianpeng8eb4efb2013-09-26 14:42:17 +0800439 u64 off = iocb->ki_pos;
Henry C Changab226e22010-12-15 20:41:54 -0800440 int num_pages, ret;
Al Viro2b777c92014-04-03 22:31:22 -0400441 size_t len = iov_iter_count(i);
Sage Weil124e68e2009-10-06 11:31:08 -0700442
majianpeng8eb4efb2013-09-26 14:42:17 +0800443 dout("sync_read on file %p %llu~%u %s\n", file, off,
444 (unsigned)len,
Sage Weil124e68e2009-10-06 11:31:08 -0700445 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
Yan, Zhengd0d0db22014-07-21 10:15:48 +0800446
447 if (!len)
448 return 0;
Sage Weile98b6fe2010-11-09 12:24:53 -0800449 /*
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 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800455 ret = filemap_write_and_wait_range(inode->i_mapping, off,
456 off + len);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800457 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800458 return ret;
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800459
Al Viro2ba48ce2015-04-09 13:52:01 -0400460 if (iocb->ki_flags & IOCB_DIRECT) {
majianpeng8eb4efb2013-09-26 14:42:17 +0800461 while (iov_iter_count(i)) {
Al Viro2b777c92014-04-03 22:31:22 -0400462 size_t start;
463 ssize_t n;
Sage Weil124e68e2009-10-06 11:31:08 -0700464
Al Viro2b777c92014-04-03 22:31:22 -0400465 n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start);
466 if (n < 0)
467 return n;
Sage Weil124e68e2009-10-06 11:31:08 -0700468
Al Viro2b777c92014-04-03 22:31:22 -0400469 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
470
471 ret = striped_read(inode, off, n,
majianpeng8eb4efb2013-09-26 14:42:17 +0800472 pages, num_pages, checkeof,
Al Viro2b777c92014-04-03 22:31:22 -0400473 1, start);
474
majianpeng8eb4efb2013-09-26 14:42:17 +0800475 ceph_put_page_vector(pages, num_pages, true);
476
477 if (ret <= 0)
478 break;
479 off += ret;
480 iov_iter_advance(i, ret);
Al Viro2b777c92014-04-03 22:31:22 -0400481 if (ret < n)
majianpeng8eb4efb2013-09-26 14:42:17 +0800482 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;
Al Viro2b777c92014-04-03 22:31:22 -0400493 size_t left = ret;
majianpeng8eb4efb2013-09-26 14:42:17 +0800494
495 while (left) {
Yan, Zheng5aaa4322014-07-02 10:13:00 +0800496 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);
Al Viro2b777c92014-04-03 22:31:22 -0400501 off += l;
502 left -= l;
503 if (l < copy)
majianpeng8eb4efb2013-09-26 14:42:17 +0800504 break;
505 }
506 }
Sage Weil124e68e2009-10-06 11:31:08 -0700507 ceph_release_page_vector(pages, num_pages);
majianpeng8eb4efb2013-09-26 14:42:17 +0800508 }
509
510 if (off > iocb->ki_pos) {
511 ret = off - iocb->ki_pos;
512 iocb->ki_pos = off;
513 }
514
Sage Weil124e68e2009-10-06 11:31:08 -0700515 dout("sync_read result %d\n", ret);
516 return ret;
517}
518
519/*
Alex Elder26be8802013-04-15 11:20:42 -0500520 * 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.
Sage Weil124e68e2009-10-06 11:31:08 -0700530 */
Alex Elder26be8802013-04-15 11:20:42 -0500531static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700532{
533 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
534
Alex Elder26be8802013-04-15 11:20:42 -0500535 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 }
Sage Weil124e68e2009-10-06 11:31:08 -0700549}
550
majianpenge8344e62013-09-12 13:54:26 +0800551
Sage Weil124e68e2009-10-06 11:31:08 -0700552/*
majianpenge8344e62013-09-12 13:54:26 +0800553 * Synchronous write, straight from __user pointer or user pages.
Sage Weil124e68e2009-10-06 11:31:08 -0700554 *
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 */
majianpenge8344e62013-09-12 13:54:26 +0800559static ssize_t
Yan, Zheng06fee302014-07-28 14:33:46 +0800560ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Sage Weil124e68e2009-10-06 11:31:08 -0700561{
majianpenge8344e62013-09-12 13:54:26 +0800562 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500563 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700564 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700565 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500566 struct ceph_snap_context *snapc;
567 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700568 struct ceph_osd_request *req;
569 struct page **pages;
570 int num_pages;
Sage Weil124e68e2009-10-06 11:31:08 -0700571 int written = 0;
572 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700573 int check_caps = 0;
574 int ret;
575 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400576 size_t count = iov_iter_count(from);
Sage Weil124e68e2009-10-06 11:31:08 -0700577
Al Viro496ad9a2013-01-23 17:07:38 -0500578 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700579 return -EROFS;
580
majianpenge8344e62013-09-12 13:54:26 +0800581 dout("sync_direct_write on file %p %lld~%u\n", file, pos,
582 (unsigned)count);
Sage Weil124e68e2009-10-06 11:31:08 -0700583
majianpenge8344e62013-09-12 13:54:26 +0800584 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800585 if (ret < 0)
586 return ret;
587
588 ret = invalidate_inode_pages2_range(inode->i_mapping,
589 pos >> PAGE_CACHE_SHIFT,
majianpenge8344e62013-09-12 13:54:26 +0800590 (pos + count) >> PAGE_CACHE_SHIFT);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800591 if (ret < 0)
592 dout("invalidate_inode_pages2_range returned %d\n", ret);
593
Sage Weil124e68e2009-10-06 11:31:08 -0700594 flags = CEPH_OSD_FLAG_ORDERSNAP |
595 CEPH_OSD_FLAG_ONDISK |
596 CEPH_OSD_FLAG_WRITE;
Sage Weil124e68e2009-10-06 11:31:08 -0700597
Al Viro4908b822014-04-03 23:09:01 -0400598 while (iov_iter_count(from) > 0) {
599 u64 len = iov_iter_single_seg_count(from);
Al Viro64c31312014-04-03 22:58:25 -0400600 size_t start;
601 ssize_t n;
majianpenge8344e62013-09-12 13:54:26 +0800602
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,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800606 vino, pos, &len, 0,
majianpenge8344e62013-09-12 13:54:26 +0800607 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);
Al Viroeab87232014-04-03 22:44:19 -0400614 break;
majianpenge8344e62013-09-12 13:54:26 +0800615 }
616
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800617 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
618
Al Viro4908b822014-04-03 23:09:01 -0400619 n = iov_iter_get_pages_alloc(from, &pages, len, &start);
Al Viro64c31312014-04-03 22:58:25 -0400620 if (unlikely(n < 0)) {
621 ret = n;
622 ceph_osdc_put_request(req);
623 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700624 }
625
Al Viro64c31312014-04-03 22:58:25 -0400626 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
Sage Weil124e68e2009-10-06 11:31:08 -0700627 /*
628 * throw out any page cache pages in this range. this
629 * may block.
630 */
Sage Weil213c99e2010-08-03 10:25:11 -0700631 truncate_inode_pages_range(inode->i_mapping, pos,
Al Viro64c31312014-04-03 22:58:25 -0400632 (pos+n) | (PAGE_CACHE_SIZE-1));
633 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
majianpenge8344e62013-09-12 13:54:26 +0800634 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
majianpenge8344e62013-09-12 13:54:26 +0800645 ceph_osdc_put_request(req);
Al Viro64c31312014-04-03 22:58:25 -0400646 if (ret)
majianpenge8344e62013-09-12 13:54:26 +0800647 break;
Al Viro64c31312014-04-03 22:58:25 -0400648 pos += n;
649 written += n;
Al Viro4908b822014-04-03 23:09:01 -0400650 iov_iter_advance(from, n);
Al Viro64c31312014-04-03 22:58:25 -0400651
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 }
majianpenge8344e62013-09-12 13:54:26 +0800659 }
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 */
Yan, Zheng06fee302014-07-28 14:33:46 +0800676static ssize_t
677ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
majianpenge8344e62013-09-12 13:54:26 +0800678{
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;
Al Viro4908b822014-04-03 23:09:01 -0400694 size_t count = iov_iter_count(from);
majianpenge8344e62013-09-12 13:54:26 +0800695
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
Al Viro4908b822014-04-03 23:09:01 -0400716 while ((len = iov_iter_count(from)) > 0) {
majianpenge8344e62013-09-12 13:54:26 +0800717 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,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800723 vino, pos, &len, 0, 1,
majianpenge8344e62013-09-12 13:54:26 +0800724 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);
Al Viroeab87232014-04-03 22:44:19 -0400730 break;
majianpenge8344e62013-09-12 13:54:26 +0800731 }
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
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700739 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700740 if (IS_ERR(pages)) {
741 ret = PTR_ERR(pages);
742 goto out;
743 }
majianpenge8344e62013-09-12 13:54:26 +0800744
745 left = len;
746 for (n = 0; n < num_pages; n++) {
Ilya Dryomov125d7252014-01-28 19:16:18 +0200747 size_t plen = min_t(size_t, left, PAGE_SIZE);
Al Viro4908b822014-04-03 23:09:01 -0400748 ret = copy_page_from_iter(pages[n], 0, plen, from);
majianpenge8344e62013-09-12 13:54:26 +0800749 if (ret != plen) {
750 ret = -EFAULT;
751 break;
752 }
753 left -= ret;
majianpenge8344e62013-09-12 13:54:26 +0800754 }
755
Sage Weil124e68e2009-10-06 11:31:08 -0700756 if (ret < 0) {
757 ceph_release_page_vector(pages, num_pages);
758 goto out;
759 }
760
majianpenge8344e62013-09-12 13:54:26 +0800761 /* get a second commit callback */
762 req->r_unsafe_callback = ceph_sync_write_unsafe;
763 req->r_inode = inode;
Sage Weil124e68e2009-10-06 11:31:08 -0700764
majianpenge8344e62013-09-12 13:54:26 +0800765 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
766 false, true);
Alex Elder02ee07d2013-03-14 14:09:06 -0500767
majianpenge8344e62013-09-12 13:54:26 +0800768 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
769 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
Sage Weil124e68e2009-10-06 11:31:08 -0700770
majianpenge8344e62013-09-12 13:54:26 +0800771 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
772 if (!ret)
773 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700774
775out:
majianpenge8344e62013-09-12 13:54:26 +0800776 ceph_osdc_put_request(req);
777 if (ret == 0) {
778 pos += len;
779 written += len;
Sage Weil124e68e2009-10-06 11:31:08 -0700780
majianpenge8344e62013-09-12 13:54:26 +0800781 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) {
Sage Weil124e68e2009-10-06 11:31:08 -0700793 ret = written;
majianpenge8344e62013-09-12 13:54:26 +0800794 iocb->ki_pos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700795 }
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 */
Al Viro36444242014-04-02 20:28:01 -0400806static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
Sage Weil124e68e2009-10-06 11:31:08 -0700807{
808 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700809 struct ceph_file_info *fi = filp->private_data;
Christoph Hellwig66ee59a2015-02-11 19:56:46 +0100810 size_t len = iov_iter_count(to);
Al Viro496ad9a2013-01-23 17:07:38 -0500811 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -0700812 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +0800813 struct page *pinned_page = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700814 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700815 int want, got = 0;
Yan, Zheng83701242014-11-14 22:36:18 +0800816 int retry_op = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700817
Sage Weil6a026582010-02-09 14:04:02 -0800818again:
majianpeng8eb4efb2013-09-26 14:42:17 +0800819 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
Sage Weil29625072010-05-27 10:40:43 -0700822 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;
Yan, Zheng3738daa2014-11-14 22:10:07 +0800826 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
Sage Weil124e68e2009-10-06 11:31:08 -0700827 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800828 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700829
Sage Weil29625072010-05-27 10:40:43 -0700830 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -0400831 (iocb->ki_flags & IOCB_DIRECT) ||
majianpeng8eb4efb2013-09-26 14:42:17 +0800832 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700833
majianpeng8eb4efb2013-09-26 14:42:17 +0800834 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
Yan, Zheng83701242014-11-14 22:36:18 +0800838 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 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800844 } else {
majianpeng8eb4efb2013-09-26 14:42:17 +0800845 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
Al Viro36444242014-04-02 20:28:01 -0400846 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
majianpeng8eb4efb2013-09-26 14:42:17 +0800847 ceph_cap_string(got));
848
Al Viro36444242014-04-02 20:28:01 -0400849 ret = generic_file_read_iter(iocb, to);
majianpeng8eb4efb2013-09-26 14:42:17 +0800850 }
Sage Weil124e68e2009-10-06 11:31:08 -0700851 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
852 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
Yan, Zheng3738daa2014-11-14 22:10:07 +0800853 if (pinned_page) {
854 page_cache_release(pinned_page);
855 pinned_page = NULL;
856 }
Sage Weil124e68e2009-10-06 11:31:08 -0700857 ceph_put_cap_refs(ci, got);
Yan, Zheng83701242014-11-14 22:36:18 +0800858 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 }
Sage Weil6a026582010-02-09 14:04:02 -0800867
Yan, Zheng83701242014-11-14 22:36:18 +0800868 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) {
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800881 BUG_ON(ret > 0 || read > 0);
882 if (iocb->ki_pos < i_size &&
883 iocb->ki_pos < PAGE_CACHE_SIZE) {
Yan, Zheng83701242014-11-14 22:36:18 +0800884 loff_t end = min_t(loff_t, i_size,
885 iocb->ki_pos + len);
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800886 end = min_t(loff_t, end, PAGE_CACHE_SIZE);
Yan, Zheng83701242014-11-14 22:36:18 +0800887 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;
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800893 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;
Yan, Zheng83701242014-11-14 22:36:18 +0800901 }
902 __free_pages(page, 0);
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800903 return read;
Yan, Zheng83701242014-11-14 22:36:18 +0800904 }
Sage Weil6a026582010-02-09 14:04:02 -0800905
906 /* hit EOF or hole? */
Yan, Zheng83701242014-11-14 22:36:18 +0800907 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
Yan, Zhengfcc02d22015-01-10 11:43:12 +0800908 ret < len) {
majianpeng8eb4efb2013-09-26 14:42:17 +0800909 dout("sync_read hit hole, ppos %lld < size %lld"
910 ", reading more\n", iocb->ki_pos,
911 inode->i_size);
912
Sage Weil6a026582010-02-09 14:04:02 -0800913 read += ret;
Sage Weil6a026582010-02-09 14:04:02 -0800914 len -= ret;
Yan, Zheng83701242014-11-14 22:36:18 +0800915 retry_op = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800916 goto again;
917 }
918 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800919
Sage Weil6a026582010-02-09 14:04:02 -0800920 if (ret >= 0)
921 ret += read;
922
Sage Weil124e68e2009-10-06 11:31:08 -0700923 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 */
Al Viro4908b822014-04-03 23:09:01 -0400936static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
Sage Weil124e68e2009-10-06 11:31:08 -0700937{
938 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700939 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -0500940 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700941 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700942 struct ceph_osd_client *osdc =
943 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Al Viro3309dd02015-04-09 12:55:47 -0400944 ssize_t count, written = 0;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800945 int err, want, got;
Al Viro3309dd02015-04-09 12:55:47 -0400946 loff_t pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700947
948 if (ceph_snap(inode) != CEPH_NOSNAP)
949 return -EROFS;
950
Yan, Zheng03d254e2013-04-12 16:11:13 +0800951 mutex_lock(&inode->i_mutex);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800952
Yan, Zheng03d254e2013-04-12 16:11:13 +0800953 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100954 current->backing_dev_info = inode_to_bdi(inode);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800955
Al Viro3309dd02015-04-09 12:55:47 -0400956 err = generic_write_checks(iocb, from);
957 if (err <= 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +0800958 goto out;
959
Al Viro3309dd02015-04-09 12:55:47 -0400960 pos = iocb->ki_pos;
961 count = iov_iter_count(from);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800962 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
Yan, Zheng28127bd2014-11-14 22:38:29 +0800970 if (ci->i_inline_version != CEPH_INLINE_NONE) {
971 err = ceph_uninline_data(file, NULL);
972 if (err < 0)
973 goto out;
974 }
975
Sage Weil124e68e2009-10-06 11:31:08 -0700976retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800977 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800978 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800979 goto out;
980 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800981
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700982 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800983 inode, ceph_vinop(inode), pos, count, inode->i_size);
Sage Weil7971bd92013-05-01 21:15:58 -0700984 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;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800988 got = 0;
Yan, Zheng3738daa2014-11-14 22:10:07 +0800989 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
990 &got, NULL);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800991 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +0800992 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -0700993
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700994 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800995 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -0700996
997 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -0400998 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
Al Viro4908b822014-04-03 23:09:01 -0400999 struct iov_iter data;
Yan, Zheng37505d52013-04-12 16:11:10 +08001000 mutex_unlock(&inode->i_mutex);
Al Viro4908b822014-04-03 23:09:01 -04001001 /* we might need to revert back to that point */
1002 data = *from;
Al Viro2ba48ce2015-04-09 13:52:01 -04001003 if (iocb->ki_flags & IOCB_DIRECT)
Yan, Zheng06fee302014-07-28 14:33:46 +08001004 written = ceph_sync_direct_write(iocb, &data, pos);
majianpenge8344e62013-09-12 13:54:26 +08001005 else
Yan, Zheng06fee302014-07-28 14:33:46 +08001006 written = ceph_sync_write(iocb, &data, pos);
majianpeng0e5dd452013-08-08 15:32:19 +08001007 if (written == -EOLDSNAPC) {
1008 dout("aio_write %p %llx.%llx %llu~%u"
1009 "got EOLDSNAPC, retrying\n",
1010 inode, ceph_vinop(inode),
Al Viro4908b822014-04-03 23:09:01 -04001011 pos, (unsigned)count);
majianpeng0e5dd452013-08-08 15:32:19 +08001012 mutex_lock(&inode->i_mutex);
majianpeng0e5dd452013-08-08 15:32:19 +08001013 goto retry_snap;
1014 }
Al Viro4908b822014-04-03 23:09:01 -04001015 if (written > 0)
1016 iov_iter_advance(from, written);
Sage Weil7971bd92013-05-01 21:15:58 -07001017 } else {
Yunchuan Wen32d3e142013-12-26 06:29:27 -08001018 loff_t old_size = inode->i_size;
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001019 /*
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 */
Al Viro4908b822014-04-03 23:09:01 -04001026 written = generic_perform_write(file, from, pos);
Al Viroaec605f42014-02-11 22:28:43 -05001027 if (likely(written >= 0))
1028 iocb->ki_pos = pos + written;
Yunchuan Wen32d3e142013-12-26 06:29:27 -08001029 if (inode->i_size > old_size)
1030 ceph_fscache_update_objectsize(inode);
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001031 mutex_unlock(&inode->i_mutex);
Sage Weil124e68e2009-10-06 11:31:08 -07001032 }
Sage Weild8de9ab2011-07-26 11:27:34 -07001033
Yan, Zheng03d254e2013-04-12 16:11:13 +08001034 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -07001035 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -08001036 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001037 ci->i_inline_version = CEPH_INLINE_NONE;
Sage Weilfca65b42011-05-04 11:33:47 -07001038 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -08001039 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -07001040 if (dirty)
1041 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -07001042 }
Sage Weil7971bd92013-05-01 21:15:58 -07001043
Sage Weil124e68e2009-10-06 11:31:08 -07001044 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Al Viro4908b822014-04-03 23:09:01 -04001045 inode, ceph_vinop(inode), pos, (unsigned)count,
Sage Weil7971bd92013-05-01 21:15:58 -07001046 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -07001047 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -07001048
Yan, Zheng03d254e2013-04-12 16:11:13 +08001049 if (written >= 0 &&
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001050 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
1051 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Yan, Zheng03d254e2013-04-12 16:11:13 +08001052 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001053 if (err < 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +08001054 written = err;
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001055 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001056
Sage Weil2f75e9e2013-08-09 09:57:58 -07001057 goto out_unlocked;
Sage Weil124e68e2009-10-06 11:31:08 -07001058
Sage Weil2f75e9e2013-08-09 09:57:58 -07001059out:
1060 mutex_unlock(&inode->i_mutex);
1061out_unlocked:
1062 current->backing_dev_info = NULL;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001063 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -07001064}
1065
1066/*
1067 * llseek. be sure to verify file size on SEEK_END.
1068 */
Andrew Morton965c8e52012-12-17 15:59:39 -08001069static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -07001070{
1071 struct inode *inode = file->f_mapping->host;
1072 int ret;
1073
1074 mutex_lock(&inode->i_mutex);
Sage Weil6a82c472011-12-13 09:19:26 -08001075
Andrew Morton965c8e52012-12-17 15:59:39 -08001076 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Yan, Zheng508b32d2014-09-16 21:46:17 +08001077 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
Sage Weil124e68e2009-10-06 11:31:08 -07001078 if (ret < 0) {
1079 offset = ret;
1080 goto out;
1081 }
Josef Bacik06222e42011-07-18 13:21:38 -04001082 }
1083
Andrew Morton965c8e52012-12-17 15:59:39 -08001084 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -04001085 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -07001086 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;
Josef Bacik06222e42011-07-18 13:21:38 -04001101 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;
Sage Weil124e68e2009-10-06 11:31:08 -07001114 }
1115
Jie Liu46a1c2c2013-06-25 12:02:13 +08001116 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -07001117
1118out:
1119 mutex_unlock(&inode->i_mutex);
1120 return offset;
1121}
1122
Li Wangad7a60d2013-08-15 11:51:44 +08001123static 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
1138static 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
1160static 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,
Yan, Zheng715e4cd2014-11-13 14:40:37 +08001180 0, 1, op,
Li Wangad7a60d2013-08-15 11:51:44 +08001181 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
1200out:
1201 return ret;
1202}
1203
1204static 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);
Sage Weilb314a902013-08-27 12:15:16 -07001208 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;
Li Wangad7a60d2013-08-15 11:51:44 +08001213
Sage Weilb314a902013-08-27 12:15:16 -07001214 /* 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
Li Wangad7a60d2013-08-15 11:51:44 +08001219 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
1250static 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;
Libo Chenaa8b60e2013-12-11 13:49:11 +08001254 struct inode *inode = file_inode(file);
Li Wangad7a60d2013-08-15 11:51:44 +08001255 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
Yan, Zheng494d77b2014-06-26 15:25:17 +08001264 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1265 return -EOPNOTSUPP;
1266
Li Wangad7a60d2013-08-15 11:51:44 +08001267 if (!S_ISREG(inode->i_mode))
1268 return -EOPNOTSUPP;
1269
Li Wangad7a60d2013-08-15 11:51:44 +08001270 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
Yan, Zheng28127bd2014-11-14 22:38:29 +08001283 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1284 ret = ceph_uninline_data(file, NULL);
1285 if (ret < 0)
1286 goto unlock;
1287 }
1288
Li Wangad7a60d2013-08-15 11:51:44 +08001289 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
Yan, Zheng3738daa2014-11-14 22:10:07 +08001298 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
Li Wangad7a60d2013-08-15 11:51:44 +08001299 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);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001315 ci->i_inline_version = CEPH_INLINE_NONE;
Li Wangad7a60d2013-08-15 11:51:44 +08001316 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);
1323unlock:
1324 mutex_unlock(&inode->i_mutex);
1325 return ret;
1326}
1327
Sage Weil124e68e2009-10-06 11:31:08 -07001328const struct file_operations ceph_file_fops = {
1329 .open = ceph_open,
1330 .release = ceph_release,
1331 .llseek = ceph_llseek,
Al Viro36444242014-04-02 20:28:01 -04001332 .read_iter = ceph_read_iter,
Al Viro4908b822014-04-03 23:09:01 -04001333 .write_iter = ceph_write_iter,
Sage Weil124e68e2009-10-06 11:31:08 -07001334 .mmap = ceph_mmap,
1335 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -07001336 .lock = ceph_lock,
1337 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -07001338 .splice_read = generic_file_splice_read,
Al Viro3551dd72014-04-05 04:40:12 -04001339 .splice_write = iter_file_splice_write,
Sage Weil124e68e2009-10-06 11:31:08 -07001340 .unlocked_ioctl = ceph_ioctl,
1341 .compat_ioctl = ceph_ioctl,
Li Wangad7a60d2013-08-15 11:51:44 +08001342 .fallocate = ceph_fallocate,
Sage Weil124e68e2009-10-06 11:31:08 -07001343};
1344