blob: 2ddf061c1c4af730885365b07dcb9388d7af98f9 [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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070010#include <linux/aio.h>
Sage Weil124e68e2009-10-06 11:31:08 -070011
12#include "super.h"
13#include "mds_client.h"
14
15/*
16 * Ceph file operations
17 *
18 * Implement basic open/close functionality, and implement
19 * read/write.
20 *
21 * We implement three modes of file I/O:
22 * - buffered uses the generic_file_aio_{read,write} helpers
23 *
24 * - synchronous is used when there is multi-client read/write
25 * sharing, avoids the page cache, and synchronously waits for an
26 * ack from the OSD.
27 *
28 * - direct io takes the variant of the sync path that references
29 * user pages directly.
30 *
31 * fsync() flushes and waits on dirty pages, but just queues metadata
32 * for writeback: since the MDS can recover size and mtime there is no
33 * need to wait for MDS acknowledgement.
34 */
35
36
37/*
38 * Prepare an open request. Preallocate ceph_cap to avoid an
39 * inopportune ENOMEM later.
40 */
41static struct ceph_mds_request *
42prepare_open_request(struct super_block *sb, int flags, int create_mode)
43{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070044 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
45 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -070046 struct ceph_mds_request *req;
47 int want_auth = USE_ANY_MDS;
48 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
49
50 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
51 want_auth = USE_AUTH_MDS;
52
53 req = ceph_mdsc_create_request(mdsc, op, want_auth);
54 if (IS_ERR(req))
55 goto out;
56 req->r_fmode = ceph_flags_to_mode(flags);
57 req->r_args.open.flags = cpu_to_le32(flags);
58 req->r_args.open.mode = cpu_to_le32(create_mode);
Sage Weil124e68e2009-10-06 11:31:08 -070059out:
60 return req;
61}
62
63/*
64 * initialize private struct file data.
65 * if we fail, clean up by dropping fmode reference on the ceph_inode
66 */
67static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
68{
69 struct ceph_file_info *cf;
70 int ret = 0;
71
72 switch (inode->i_mode & S_IFMT) {
73 case S_IFREG:
74 case S_IFDIR:
75 dout("init_file %p %p 0%o (regular)\n", inode, file,
76 inode->i_mode);
77 cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
78 if (cf == NULL) {
79 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
80 return -ENOMEM;
81 }
82 cf->fmode = fmode;
83 cf->next_offset = 2;
84 file->private_data = cf;
85 BUG_ON(inode->i_fop->release != ceph_release);
86 break;
87
88 case S_IFLNK:
89 dout("init_file %p %p 0%o (symlink)\n", inode, file,
90 inode->i_mode);
91 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
92 break;
93
94 default:
95 dout("init_file %p %p 0%o (special)\n", inode, file,
96 inode->i_mode);
97 /*
98 * we need to drop the open ref now, since we don't
99 * have .release set to ceph_release.
100 */
101 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
102 BUG_ON(inode->i_fop->release == ceph_release);
103
104 /* call the proper open fop */
105 ret = inode->i_fop->open(inode, file);
106 }
107 return ret;
108}
109
110/*
Sage Weil124e68e2009-10-06 11:31:08 -0700111 * If we already have the requisite capabilities, we can satisfy
112 * the open request locally (no need to request new caps from the
113 * MDS). We do, however, need to inform the MDS (asynchronously)
114 * if our wanted caps set expands.
115 */
116int ceph_open(struct inode *inode, struct file *file)
117{
118 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700119 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
120 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700121 struct ceph_mds_request *req;
122 struct ceph_file_info *cf = file->private_data;
Sage Weil5f21c962011-07-26 11:30:29 -0700123 struct inode *parent_inode = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700124 int err;
125 int flags, fmode, wanted;
126
127 if (cf) {
128 dout("open file %p is already opened\n", file);
129 return 0;
130 }
131
132 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
133 flags = file->f_flags & ~(O_CREAT|O_EXCL);
134 if (S_ISDIR(inode->i_mode))
135 flags = O_DIRECTORY; /* mds likes to know */
136
137 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
138 ceph_vinop(inode), file, flags, file->f_flags);
139 fmode = ceph_flags_to_mode(flags);
140 wanted = ceph_caps_for_mode(fmode);
141
142 /* snapped files are read-only */
143 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
144 return -EROFS;
145
146 /* trivially open snapdir */
147 if (ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weilbe655592011-11-30 09:47:09 -0800148 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700149 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800150 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700151 return ceph_init_file(inode, file, fmode);
152 }
153
154 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800155 * No need to block if we have caps on the auth MDS (for
156 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700157 * asynchronously.
158 */
Sage Weilbe655592011-11-30 09:47:09 -0800159 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800160 if (__ceph_is_any_real_caps(ci) &&
161 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700162 int mds_wanted = __ceph_caps_mds_wanted(ci);
163 int issued = __ceph_caps_issued(ci, NULL);
164
165 dout("open %p fmode %d want %s issued %s using existing\n",
166 inode, fmode, ceph_cap_string(wanted),
167 ceph_cap_string(issued));
168 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800169 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700170
171 /* adjust wanted? */
172 if ((issued & wanted) != wanted &&
173 (mds_wanted & wanted) != wanted &&
174 ceph_snap(inode) != CEPH_SNAPDIR)
175 ceph_check_caps(ci, 0, NULL);
176
177 return ceph_init_file(inode, file, fmode);
178 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
179 (ci->i_snap_caps & wanted) == wanted) {
180 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800181 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700182 return ceph_init_file(inode, file, fmode);
183 }
Sage Weilbe655592011-11-30 09:47:09 -0800184 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700185
186 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
187 req = prepare_open_request(inode->i_sb, flags, 0);
188 if (IS_ERR(req)) {
189 err = PTR_ERR(req);
190 goto out;
191 }
Sage Weil70b666c2011-05-27 09:24:26 -0700192 req->r_inode = inode;
193 ihold(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700194 req->r_num_caps = 1;
Sage Weil5f21c962011-07-26 11:30:29 -0700195 if (flags & (O_CREAT|O_TRUNC))
196 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
Sage Weil124e68e2009-10-06 11:31:08 -0700197 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -0700198 iput(parent_inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700199 if (!err)
200 err = ceph_init_file(inode, file, req->r_fmode);
201 ceph_mdsc_put_request(req);
202 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
203out:
204 return err;
205}
206
207
208/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700209 * Do a lookup + open with a single request. If we get a non-existent
210 * file or symlink, return 1 so the VFS can retry.
Sage Weil124e68e2009-10-06 11:31:08 -0700211 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700212int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400213 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400214 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700215{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700216 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
217 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700218 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700219 struct dentry *dn;
Sage Weil124e68e2009-10-06 11:31:08 -0700220 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700221
Sage Weil5ef50c32012-07-31 11:27:36 -0700222 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
223 dir, dentry, dentry->d_name.len, dentry->d_name.name,
224 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
225
226 if (dentry->d_name.len > NAME_MAX)
227 return -ENAMETOOLONG;
228
229 err = ceph_init_dentry(dentry);
230 if (err < 0)
231 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700232
233 /* do the open */
234 req = prepare_open_request(dir->i_sb, flags, mode);
235 if (IS_ERR(req))
Al Virod9585272012-06-22 12:39:14 +0400236 return PTR_ERR(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700237 req->r_dentry = dget(dentry);
238 req->r_num_caps = 2;
239 if (flags & O_CREAT) {
240 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
241 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
242 }
243 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700244 err = ceph_mdsc_do_request(mdsc,
245 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
246 req);
Sam Lang79aec9842012-12-19 09:44:23 -1000247 if (err)
248 goto out_err;
249
Sage Weil468640e2011-07-26 11:28:11 -0700250 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil5ef50c32012-07-31 11:27:36 -0700251 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700252 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700253
254 if (d_unhashed(dentry)) {
255 dn = ceph_finish_lookup(req, dentry, err);
256 if (IS_ERR(dn))
257 err = PTR_ERR(dn);
258 } else {
259 /* we were given a hashed negative dentry */
260 dn = NULL;
261 }
Sage Weil468640e2011-07-26 11:28:11 -0700262 if (err)
Sage Weil5ef50c32012-07-31 11:27:36 -0700263 goto out_err;
264 if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) {
265 /* make vfs retry on splice, ENOENT, or symlink */
266 dout("atomic_open finish_no_open on dn %p\n", dn);
267 err = finish_no_open(file, dn);
268 } else {
269 dout("atomic_open finish_open on dn %p\n", dn);
Sam Lang6e8575f2012-12-28 09:56:46 -0800270 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
271 *opened |= FILE_CREATED;
272 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700273 err = finish_open(file, dentry, ceph_open, opened);
274 }
275
276out_err:
Sage Weil124e68e2009-10-06 11:31:08 -0700277 ceph_mdsc_put_request(req);
Sage Weil5ef50c32012-07-31 11:27:36 -0700278 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400279 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700280}
281
282int ceph_release(struct inode *inode, struct file *file)
283{
284 struct ceph_inode_info *ci = ceph_inode(inode);
285 struct ceph_file_info *cf = file->private_data;
286
287 dout("release inode %p file %p\n", inode, file);
288 ceph_put_fmode(ci, cf->fmode);
289 if (cf->last_readdir)
290 ceph_mdsc_put_request(cf->last_readdir);
291 kfree(cf->last_name);
292 kfree(cf->dir_info);
293 dput(cf->dentry);
294 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800295
296 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700297 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700298 return 0;
299}
300
301/*
Sage Weil124e68e2009-10-06 11:31:08 -0700302 * Read a range of bytes striped over one or more objects. Iterate over
303 * objects we stripe over. (That's not atomic, but good enough for now.)
304 *
305 * If we get a short result from the OSD, check against i_size; we need to
306 * only return a short read to the caller if we hit EOF.
307 */
308static int striped_read(struct inode *inode,
309 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800310 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700311 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800312 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700313{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700314 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700315 struct ceph_inode_info *ci = ceph_inode(inode);
316 u64 pos, this_len;
Sage Weilb7495fc2010-11-09 12:43:12 -0800317 int io_align, page_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700318 int left, pages_left;
319 int read;
320 struct page **page_pos;
321 int ret;
322 bool hit_stripe, was_short;
323
324 /*
325 * we may need to do multiple reads. not atomic, unfortunately.
326 */
327 pos = off;
328 left = len;
329 page_pos = pages;
330 pages_left = num_pages;
331 read = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800332 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700333
334more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700335 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800336 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800337 else
338 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700339 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700340 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700341 &ci->i_layout, pos, &this_len,
342 ci->i_truncate_seq,
343 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800344 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700345 if (ret == -ENOENT)
346 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700347 hit_stripe = this_len < left;
348 was_short = ret >= 0 && ret < this_len;
Sage Weil124e68e2009-10-06 11:31:08 -0700349 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read,
350 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
351
352 if (ret > 0) {
Sage Weil773e9b42011-06-07 20:57:14 -0700353 int didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700354
355 if (read < pos - off) {
356 dout(" zero gap %llu to %llu\n", off + read, pos);
Sage Weil773e9b42011-06-07 20:57:14 -0700357 ceph_zero_page_vector_range(page_align + read,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700358 pos - off - read, pages);
Sage Weil124e68e2009-10-06 11:31:08 -0700359 }
360 pos += ret;
361 read = pos - off;
362 left -= ret;
363 page_pos += didpages;
364 pages_left -= didpages;
365
366 /* hit stripe? */
367 if (left && hit_stripe)
368 goto more;
369 }
370
371 if (was_short) {
Sage Weilc3cd6282011-06-01 16:08:44 -0700372 /* did we bounce off eof? */
373 if (pos + left > inode->i_size)
374 *checkeof = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700375
Sage Weilc3cd6282011-06-01 16:08:44 -0700376 /* zero trailing bytes (inside i_size) */
377 if (left > 0 && pos < inode->i_size) {
378 if (pos + left > inode->i_size)
379 left = inode->i_size - pos;
380
381 dout("zero tail %d\n", left);
Sage Weil773e9b42011-06-07 20:57:14 -0700382 ceph_zero_page_vector_range(page_align + read, left,
Sage Weilc3cd6282011-06-01 16:08:44 -0700383 pages);
384 read += left;
385 }
Sage Weil124e68e2009-10-06 11:31:08 -0700386 }
387
Sage Weil124e68e2009-10-06 11:31:08 -0700388 if (ret >= 0)
389 ret = read;
390 dout("striped_read returns %d\n", ret);
391 return ret;
392}
393
394/*
395 * Completely synchronous read and write methods. Direct from __user
396 * buffer to osd, or directly to user pages (if O_DIRECT).
397 *
398 * If the read spans object boundary, just do multiple reads.
399 */
400static ssize_t ceph_sync_read(struct file *file, char __user *data,
Sage Weil6a026582010-02-09 14:04:02 -0800401 unsigned len, loff_t *poff, int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700402{
Al Viro496ad9a2013-01-23 17:07:38 -0500403 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700404 struct page **pages;
405 u64 off = *poff;
Henry C Changab226e22010-12-15 20:41:54 -0800406 int num_pages, ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700407
408 dout("sync_read on file %p %llu~%u %s\n", file, off, len,
409 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
410
Henry C Changab226e22010-12-15 20:41:54 -0800411 if (file->f_flags & O_DIRECT) {
412 num_pages = calc_pages_for((unsigned long)data, len);
Henry C Changb6aa5902010-12-15 20:45:41 -0800413 pages = ceph_get_direct_page_vector(data, num_pages, true);
Henry C Changab226e22010-12-15 20:41:54 -0800414 } else {
415 num_pages = calc_pages_for(off, len);
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700416 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Henry C Changab226e22010-12-15 20:41:54 -0800417 }
Sage Weil124e68e2009-10-06 11:31:08 -0700418 if (IS_ERR(pages))
419 return PTR_ERR(pages);
420
Sage Weile98b6fe2010-11-09 12:24:53 -0800421 /*
422 * flush any page cache pages in this range. this
423 * will make concurrent normal and sync io slow,
424 * but it will at least behave sensibly when they are
425 * in sequence.
426 */
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800427 ret = filemap_write_and_wait(inode->i_mapping);
428 if (ret < 0)
429 goto done;
430
Sage Weilb7495fc2010-11-09 12:43:12 -0800431 ret = striped_read(inode, off, len, pages, num_pages, checkeof,
Henry C Changab226e22010-12-15 20:41:54 -0800432 file->f_flags & O_DIRECT,
433 (unsigned long)data & ~PAGE_MASK);
Sage Weil124e68e2009-10-06 11:31:08 -0700434
435 if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700436 ret = ceph_copy_page_vector_to_user(pages, data, off, ret);
Sage Weil124e68e2009-10-06 11:31:08 -0700437 if (ret >= 0)
438 *poff = off + ret;
439
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800440done:
Sage Weil124e68e2009-10-06 11:31:08 -0700441 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800442 ceph_put_page_vector(pages, num_pages, true);
Sage Weil124e68e2009-10-06 11:31:08 -0700443 else
444 ceph_release_page_vector(pages, num_pages);
445 dout("sync_read result %d\n", ret);
446 return ret;
447}
448
449/*
Alex Elder26be8802013-04-15 11:20:42 -0500450 * Write commit request unsafe callback, called to tell us when a
451 * request is unsafe (that is, in flight--has been handed to the
452 * messenger to send to its target osd). It is called again when
453 * we've received a response message indicating the request is
454 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
455 * is completed early (and unsuccessfully) due to a timeout or
456 * interrupt.
457 *
458 * This is used if we requested both an ACK and ONDISK commit reply
459 * from the OSD.
Sage Weil124e68e2009-10-06 11:31:08 -0700460 */
Alex Elder26be8802013-04-15 11:20:42 -0500461static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700462{
463 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
464
Alex Elder26be8802013-04-15 11:20:42 -0500465 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
466 unsafe ? "un" : "");
467 if (unsafe) {
468 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
469 spin_lock(&ci->i_unsafe_lock);
470 list_add_tail(&req->r_unsafe_item,
471 &ci->i_unsafe_writes);
472 spin_unlock(&ci->i_unsafe_lock);
473 } else {
474 spin_lock(&ci->i_unsafe_lock);
475 list_del_init(&req->r_unsafe_item);
476 spin_unlock(&ci->i_unsafe_lock);
477 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
478 }
Sage Weil124e68e2009-10-06 11:31:08 -0700479}
480
481/*
482 * Synchronous write, straight from __user pointer or user pages (if
483 * O_DIRECT).
484 *
485 * If write spans object boundary, just do multiple writes. (For a
486 * correct atomic write, we should e.g. take write locks on all
487 * objects, rollback on failure, etc.)
488 */
489static ssize_t ceph_sync_write(struct file *file, const char __user *data,
Yan, Zheng03d254e2013-04-12 16:11:13 +0800490 size_t left, loff_t pos, loff_t *ppos)
Sage Weil124e68e2009-10-06 11:31:08 -0700491{
Al Viro496ad9a2013-01-23 17:07:38 -0500492 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700493 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700494 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500495 struct ceph_snap_context *snapc;
496 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700497 struct ceph_osd_request *req;
Alex Elderacead002013-03-14 14:09:05 -0500498 int num_ops = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700499 struct page **pages;
500 int num_pages;
Sage Weil124e68e2009-10-06 11:31:08 -0700501 u64 len;
502 int written = 0;
503 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700504 int check_caps = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800505 int page_align, io_align;
Henry C Changab226e22010-12-15 20:41:54 -0800506 unsigned long buf_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700507 int ret;
508 struct timespec mtime = CURRENT_TIME;
Alex Elder43bfe5d2013-04-03 01:28:57 -0500509 bool own_pages = false;
Sage Weil124e68e2009-10-06 11:31:08 -0700510
Al Viro496ad9a2013-01-23 17:07:38 -0500511 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700512 return -EROFS;
513
Yan, Zheng03d254e2013-04-12 16:11:13 +0800514 dout("sync_write on file %p %lld~%u %s\n", file, pos,
Sage Weil124e68e2009-10-06 11:31:08 -0700515 (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
516
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800517 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left);
518 if (ret < 0)
519 return ret;
520
521 ret = invalidate_inode_pages2_range(inode->i_mapping,
522 pos >> PAGE_CACHE_SHIFT,
523 (pos + left) >> PAGE_CACHE_SHIFT);
524 if (ret < 0)
525 dout("invalidate_inode_pages2_range returned %d\n", ret);
526
Sage Weil124e68e2009-10-06 11:31:08 -0700527 flags = CEPH_OSD_FLAG_ORDERSNAP |
528 CEPH_OSD_FLAG_ONDISK |
529 CEPH_OSD_FLAG_WRITE;
530 if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0)
531 flags |= CEPH_OSD_FLAG_ACK;
532 else
Alex Elderacead002013-03-14 14:09:05 -0500533 num_ops++; /* Also include a 'startsync' command. */
Sage Weil124e68e2009-10-06 11:31:08 -0700534
535 /*
536 * we may need to do multiple writes here if we span an object
537 * boundary. this isn't atomic, unfortunately. :(
538 */
539more:
Sage Weild7f124f2011-06-13 16:22:18 -0700540 io_align = pos & ~PAGE_MASK;
541 buf_align = (unsigned long)data & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700542 len = left;
Alex Elder3a42b6c2013-02-16 17:07:32 +0000543
Alex Elderacead002013-03-14 14:09:05 -0500544 snapc = ci->i_snap_realm->cached_context;
545 vino = ceph_vino(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700546 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Alex Elder79528732013-04-03 21:32:51 -0500547 vino, pos, &len, num_ops,
Alex Elderacead002013-03-14 14:09:05 -0500548 CEPH_OSD_OP_WRITE, flags, snapc,
Sage Weil124e68e2009-10-06 11:31:08 -0700549 ci->i_truncate_seq, ci->i_truncate_size,
Alex Elderacead002013-03-14 14:09:05 -0500550 false);
Sage Weil68162822012-09-24 21:01:02 -0700551 if (IS_ERR(req))
552 return PTR_ERR(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700553
Alex Elder153e5162013-03-01 18:00:15 -0600554 /* write from beginning of first page, regardless of io alignment */
555 page_align = file->f_flags & O_DIRECT ? buf_align : io_align;
556 num_pages = calc_pages_for(page_align, len);
Sage Weil124e68e2009-10-06 11:31:08 -0700557 if (file->f_flags & O_DIRECT) {
Henry C Changb6aa5902010-12-15 20:45:41 -0800558 pages = ceph_get_direct_page_vector(data, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700559 if (IS_ERR(pages)) {
560 ret = PTR_ERR(pages);
561 goto out;
562 }
563
564 /*
565 * throw out any page cache pages in this range. this
566 * may block.
567 */
Sage Weil213c99e2010-08-03 10:25:11 -0700568 truncate_inode_pages_range(inode->i_mapping, pos,
Sage Weil5c6a2cd2010-04-22 13:48:59 -0700569 (pos+len) | (PAGE_CACHE_SIZE-1));
Sage Weil124e68e2009-10-06 11:31:08 -0700570 } else {
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700571 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700572 if (IS_ERR(pages)) {
573 ret = PTR_ERR(pages);
574 goto out;
575 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700576 ret = ceph_copy_user_to_page_vector(pages, data, pos, len);
Sage Weil124e68e2009-10-06 11:31:08 -0700577 if (ret < 0) {
578 ceph_release_page_vector(pages, num_pages);
579 goto out;
580 }
581
582 if ((file->f_flags & O_SYNC) == 0) {
583 /* get a second commit callback */
Alex Elder26be8802013-04-15 11:20:42 -0500584 req->r_unsafe_callback = ceph_sync_write_unsafe;
585 req->r_inode = inode;
Alex Elder43bfe5d2013-04-03 01:28:57 -0500586 own_pages = true;
Sage Weil124e68e2009-10-06 11:31:08 -0700587 }
588 }
Alex Elder406e2c92013-04-15 14:50:36 -0500589 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
590 false, own_pages);
Sage Weil124e68e2009-10-06 11:31:08 -0700591
Alex Elder02ee07d2013-03-14 14:09:06 -0500592 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
Alex Elder79528732013-04-03 21:32:51 -0500593 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
Alex Elder02ee07d2013-03-14 14:09:06 -0500594
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700595 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
Alex Elder26be8802013-04-15 11:20:42 -0500596 if (!ret)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700597 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700598
599 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800600 ceph_put_page_vector(pages, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700601 else if (file->f_flags & O_SYNC)
602 ceph_release_page_vector(pages, num_pages);
603
604out:
605 ceph_osdc_put_request(req);
606 if (ret == 0) {
607 pos += len;
608 written += len;
609 left -= len;
Henry C Chang022f3e22013-03-19 09:46:26 +0800610 data += len;
Sage Weil124e68e2009-10-06 11:31:08 -0700611 if (left)
612 goto more;
613
614 ret = written;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800615 *ppos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700616 if (pos > i_size_read(inode))
617 check_caps = ceph_inode_set_size(inode, pos);
618 if (check_caps)
619 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY,
620 NULL);
621 }
622 return ret;
623}
624
625/*
626 * Wrap generic_file_aio_read with checks for cap bits on the inode.
627 * Atomically grab references, so that those bits are not released
628 * back to the MDS mid-read.
629 *
630 * Hmm, the sync read case isn't actually async... should it be?
631 */
632static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
633 unsigned long nr_segs, loff_t pos)
634{
635 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700636 struct ceph_file_info *fi = filp->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700637 loff_t *ppos = &iocb->ki_pos;
638 size_t len = iov->iov_len;
Al Viro496ad9a2013-01-23 17:07:38 -0500639 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -0700640 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700641 void __user *base = iov->iov_base;
Sage Weil124e68e2009-10-06 11:31:08 -0700642 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700643 int want, got = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800644 int checkeof = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700645
646 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
647 inode, ceph_vinop(inode), pos, (unsigned)len, inode);
Sage Weil6a026582010-02-09 14:04:02 -0800648again:
Sage Weil29625072010-05-27 10:40:43 -0700649 if (fi->fmode & CEPH_FILE_MODE_LAZY)
650 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
651 else
652 want = CEPH_CAP_FILE_CACHE;
653 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
Sage Weil124e68e2009-10-06 11:31:08 -0700654 if (ret < 0)
655 goto out;
656 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
657 inode, ceph_vinop(inode), pos, (unsigned)len,
658 ceph_cap_string(got));
659
Sage Weil29625072010-05-27 10:40:43 -0700660 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700661 (iocb->ki_filp->f_flags & O_DIRECT) ||
Sage Weil4918b6d2011-07-26 11:26:07 -0700662 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
663 (fi->flags & CEPH_F_SYNC))
Sage Weil124e68e2009-10-06 11:31:08 -0700664 /* hmm, this isn't really async... */
Sage Weil6a026582010-02-09 14:04:02 -0800665 ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
Sage Weil124e68e2009-10-06 11:31:08 -0700666 else
667 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
668
669out:
670 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
671 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
672 ceph_put_cap_refs(ci, got);
Sage Weil6a026582010-02-09 14:04:02 -0800673
674 if (checkeof && ret >= 0) {
675 int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
676
677 /* hit EOF or hole? */
678 if (statret == 0 && *ppos < inode->i_size) {
Sage Weilc3cd6282011-06-01 16:08:44 -0700679 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size);
Sage Weil6a026582010-02-09 14:04:02 -0800680 read += ret;
681 base += ret;
682 len -= ret;
683 checkeof = 0;
684 goto again;
685 }
686 }
687 if (ret >= 0)
688 ret += read;
689
Sage Weil124e68e2009-10-06 11:31:08 -0700690 return ret;
691}
692
693/*
694 * Take cap references to avoid releasing caps to MDS mid-write.
695 *
696 * If we are synchronous, and write with an old snap context, the OSD
697 * may return EOLDSNAPC. In that case, retry the write.. _after_
698 * dropping our cap refs and allowing the pending snap to logically
699 * complete _before_ this write occurs.
700 *
701 * If we are near ENOSPC, write synchronously.
702 */
703static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov,
704 unsigned long nr_segs, loff_t pos)
705{
706 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700707 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -0500708 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700709 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700710 struct ceph_osd_client *osdc =
711 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800712 ssize_t count, written = 0;
713 int err, want, got;
714 bool hold_mutex;
Sage Weil124e68e2009-10-06 11:31:08 -0700715
716 if (ceph_snap(inode) != CEPH_NOSNAP)
717 return -EROFS;
718
Yan, Zheng03d254e2013-04-12 16:11:13 +0800719 mutex_lock(&inode->i_mutex);
720 hold_mutex = true;
721
722 err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ);
723 if (err)
724 goto out;
725
726 /* We can write back this queue in page reclaim */
727 current->backing_dev_info = file->f_mapping->backing_dev_info;
728
729 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
730 if (err)
731 goto out;
732
733 if (count == 0)
734 goto out;
735
736 err = file_remove_suid(file);
737 if (err)
738 goto out;
739
740 err = file_update_time(file);
741 if (err)
742 goto out;
743
Sage Weil124e68e2009-10-06 11:31:08 -0700744retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800745 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800746 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800747 goto out;
748 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800749
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700750 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800751 inode, ceph_vinop(inode), pos, count, inode->i_size);
Sage Weil7971bd92013-05-01 21:15:58 -0700752 if (fi->fmode & CEPH_FILE_MODE_LAZY)
753 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
754 else
755 want = CEPH_CAP_FILE_BUFFER;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800756 got = 0;
757 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count);
758 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +0800759 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -0700760
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700761 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800762 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -0700763
764 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
765 (iocb->ki_filp->f_flags & O_DIRECT) ||
766 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
767 (fi->flags & CEPH_F_SYNC)) {
Yan, Zheng37505d52013-04-12 16:11:10 +0800768 mutex_unlock(&inode->i_mutex);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800769 written = ceph_sync_write(file, iov->iov_base, count,
770 pos, &iocb->ki_pos);
Sage Weil7971bd92013-05-01 21:15:58 -0700771 } else {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800772 written = generic_file_buffered_write(iocb, iov, nr_segs,
773 pos, &iocb->ki_pos,
774 count, 0);
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800775 mutex_unlock(&inode->i_mutex);
Sage Weil124e68e2009-10-06 11:31:08 -0700776 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800777 hold_mutex = false;
Sage Weild8de9ab2011-07-26 11:27:34 -0700778
Yan, Zheng03d254e2013-04-12 16:11:13 +0800779 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -0700780 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -0800781 spin_lock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700782 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -0800783 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700784 if (dirty)
785 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -0700786 }
Sage Weil7971bd92013-05-01 21:15:58 -0700787
Sage Weil124e68e2009-10-06 11:31:08 -0700788 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Sage Weil7971bd92013-05-01 21:15:58 -0700789 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
790 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -0700791 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -0700792
Yan, Zheng03d254e2013-04-12 16:11:13 +0800793 if (written >= 0 &&
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800794 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
795 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800796 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800797 if (err < 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +0800798 written = err;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800799 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800800
801 if (written == -EOLDSNAPC) {
Sage Weil124e68e2009-10-06 11:31:08 -0700802 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
803 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800804 mutex_lock(&inode->i_mutex);
805 hold_mutex = true;
Sage Weil124e68e2009-10-06 11:31:08 -0700806 goto retry_snap;
807 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800808out:
809 if (hold_mutex)
810 mutex_unlock(&inode->i_mutex);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800811 current->backing_dev_info = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700812
Yan, Zheng03d254e2013-04-12 16:11:13 +0800813 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -0700814}
815
816/*
817 * llseek. be sure to verify file size on SEEK_END.
818 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800819static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -0700820{
821 struct inode *inode = file->f_mapping->host;
822 int ret;
823
824 mutex_lock(&inode->i_mutex);
Yan, Zhengb415bf42013-07-02 12:40:19 +0800825 __ceph_do_pending_vmtruncate(inode);
Sage Weil6a82c472011-12-13 09:19:26 -0800826
Andrew Morton965c8e52012-12-17 15:59:39 -0800827 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Sage Weil124e68e2009-10-06 11:31:08 -0700828 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
829 if (ret < 0) {
830 offset = ret;
831 goto out;
832 }
Josef Bacik06222e42011-07-18 13:21:38 -0400833 }
834
Andrew Morton965c8e52012-12-17 15:59:39 -0800835 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -0400836 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -0700837 offset += inode->i_size;
838 break;
839 case SEEK_CUR:
840 /*
841 * Here we special-case the lseek(fd, 0, SEEK_CUR)
842 * position-querying operation. Avoid rewriting the "same"
843 * f_pos value back to the file because a concurrent read(),
844 * write() or lseek() might have altered it
845 */
846 if (offset == 0) {
847 offset = file->f_pos;
848 goto out;
849 }
850 offset += file->f_pos;
851 break;
Josef Bacik06222e42011-07-18 13:21:38 -0400852 case SEEK_DATA:
853 if (offset >= inode->i_size) {
854 ret = -ENXIO;
855 goto out;
856 }
857 break;
858 case SEEK_HOLE:
859 if (offset >= inode->i_size) {
860 ret = -ENXIO;
861 goto out;
862 }
863 offset = inode->i_size;
864 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700865 }
866
Jie Liu46a1c2c2013-06-25 12:02:13 +0800867 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -0700868
869out:
870 mutex_unlock(&inode->i_mutex);
871 return offset;
872}
873
874const struct file_operations ceph_file_fops = {
875 .open = ceph_open,
876 .release = ceph_release,
877 .llseek = ceph_llseek,
878 .read = do_sync_read,
879 .write = do_sync_write,
880 .aio_read = ceph_aio_read,
881 .aio_write = ceph_aio_write,
882 .mmap = ceph_mmap,
883 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -0700884 .lock = ceph_lock,
885 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -0700886 .splice_read = generic_file_splice_read,
887 .splice_write = generic_file_splice_write,
888 .unlocked_ioctl = ceph_ioctl,
889 .compat_ioctl = ceph_ioctl,
890};
891