blob: dd47026c60648c1e93bb9924b2edc34763dee300 [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>
10
11#include "super.h"
12#include "mds_client.h"
13
14/*
15 * Ceph file operations
16 *
17 * Implement basic open/close functionality, and implement
18 * read/write.
19 *
20 * We implement three modes of file I/O:
21 * - buffered uses the generic_file_aio_{read,write} helpers
22 *
23 * - synchronous is used when there is multi-client read/write
24 * sharing, avoids the page cache, and synchronously waits for an
25 * ack from the OSD.
26 *
27 * - direct io takes the variant of the sync path that references
28 * user pages directly.
29 *
30 * fsync() flushes and waits on dirty pages, but just queues metadata
31 * for writeback: since the MDS can recover size and mtime there is no
32 * need to wait for MDS acknowledgement.
33 */
34
35
36/*
37 * Prepare an open request. Preallocate ceph_cap to avoid an
38 * inopportune ENOMEM later.
39 */
40static struct ceph_mds_request *
41prepare_open_request(struct super_block *sb, int flags, int create_mode)
42{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070043 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
44 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -070045 struct ceph_mds_request *req;
46 int want_auth = USE_ANY_MDS;
47 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
48
49 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
50 want_auth = USE_AUTH_MDS;
51
52 req = ceph_mdsc_create_request(mdsc, op, want_auth);
53 if (IS_ERR(req))
54 goto out;
55 req->r_fmode = ceph_flags_to_mode(flags);
56 req->r_args.open.flags = cpu_to_le32(flags);
57 req->r_args.open.mode = cpu_to_le32(create_mode);
Sage Weil124e68e2009-10-06 11:31:08 -070058out:
59 return req;
60}
61
62/*
63 * initialize private struct file data.
64 * if we fail, clean up by dropping fmode reference on the ceph_inode
65 */
66static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
67{
68 struct ceph_file_info *cf;
69 int ret = 0;
70
71 switch (inode->i_mode & S_IFMT) {
72 case S_IFREG:
73 case S_IFDIR:
74 dout("init_file %p %p 0%o (regular)\n", inode, file,
75 inode->i_mode);
76 cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
77 if (cf == NULL) {
78 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
79 return -ENOMEM;
80 }
81 cf->fmode = fmode;
82 cf->next_offset = 2;
83 file->private_data = cf;
84 BUG_ON(inode->i_fop->release != ceph_release);
85 break;
86
87 case S_IFLNK:
88 dout("init_file %p %p 0%o (symlink)\n", inode, file,
89 inode->i_mode);
90 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
91 break;
92
93 default:
94 dout("init_file %p %p 0%o (special)\n", inode, file,
95 inode->i_mode);
96 /*
97 * we need to drop the open ref now, since we don't
98 * have .release set to ceph_release.
99 */
100 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
101 BUG_ON(inode->i_fop->release == ceph_release);
102
103 /* call the proper open fop */
104 ret = inode->i_fop->open(inode, file);
105 }
106 return ret;
107}
108
109/*
Sage Weil124e68e2009-10-06 11:31:08 -0700110 * If we already have the requisite capabilities, we can satisfy
111 * the open request locally (no need to request new caps from the
112 * MDS). We do, however, need to inform the MDS (asynchronously)
113 * if our wanted caps set expands.
114 */
115int ceph_open(struct inode *inode, struct file *file)
116{
117 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700118 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
119 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700120 struct ceph_mds_request *req;
121 struct ceph_file_info *cf = file->private_data;
Sage Weil5f21c962011-07-26 11:30:29 -0700122 struct inode *parent_inode = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700123 int err;
124 int flags, fmode, wanted;
125
126 if (cf) {
127 dout("open file %p is already opened\n", file);
128 return 0;
129 }
130
131 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
132 flags = file->f_flags & ~(O_CREAT|O_EXCL);
133 if (S_ISDIR(inode->i_mode))
134 flags = O_DIRECTORY; /* mds likes to know */
135
136 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
137 ceph_vinop(inode), file, flags, file->f_flags);
138 fmode = ceph_flags_to_mode(flags);
139 wanted = ceph_caps_for_mode(fmode);
140
141 /* snapped files are read-only */
142 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
143 return -EROFS;
144
145 /* trivially open snapdir */
146 if (ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weilbe655592011-11-30 09:47:09 -0800147 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700148 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800149 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700150 return ceph_init_file(inode, file, fmode);
151 }
152
153 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800154 * No need to block if we have caps on the auth MDS (for
155 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700156 * asynchronously.
157 */
Sage Weilbe655592011-11-30 09:47:09 -0800158 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800159 if (__ceph_is_any_real_caps(ci) &&
160 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700161 int mds_wanted = __ceph_caps_mds_wanted(ci);
162 int issued = __ceph_caps_issued(ci, NULL);
163
164 dout("open %p fmode %d want %s issued %s using existing\n",
165 inode, fmode, ceph_cap_string(wanted),
166 ceph_cap_string(issued));
167 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800168 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700169
170 /* adjust wanted? */
171 if ((issued & wanted) != wanted &&
172 (mds_wanted & wanted) != wanted &&
173 ceph_snap(inode) != CEPH_SNAPDIR)
174 ceph_check_caps(ci, 0, NULL);
175
176 return ceph_init_file(inode, file, fmode);
177 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
178 (ci->i_snap_caps & wanted) == wanted) {
179 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800180 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700181 return ceph_init_file(inode, file, fmode);
182 }
Sage Weilbe655592011-11-30 09:47:09 -0800183 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700184
185 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
186 req = prepare_open_request(inode->i_sb, flags, 0);
187 if (IS_ERR(req)) {
188 err = PTR_ERR(req);
189 goto out;
190 }
Sage Weil70b666c2011-05-27 09:24:26 -0700191 req->r_inode = inode;
192 ihold(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700193 req->r_num_caps = 1;
Sage Weil5f21c962011-07-26 11:30:29 -0700194 if (flags & (O_CREAT|O_TRUNC))
195 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
Sage Weil124e68e2009-10-06 11:31:08 -0700196 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -0700197 iput(parent_inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700198 if (!err)
199 err = ceph_init_file(inode, file, req->r_fmode);
200 ceph_mdsc_put_request(req);
201 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
202out:
203 return err;
204}
205
206
207/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700208 * Do a lookup + open with a single request. If we get a non-existent
209 * file or symlink, return 1 so the VFS can retry.
Sage Weil124e68e2009-10-06 11:31:08 -0700210 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700211int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400212 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400213 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700214{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700215 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
216 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700217 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700218 struct dentry *dn;
Sage Weil124e68e2009-10-06 11:31:08 -0700219 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700220
Sage Weil5ef50c32012-07-31 11:27:36 -0700221 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
222 dir, dentry, dentry->d_name.len, dentry->d_name.name,
223 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
224
225 if (dentry->d_name.len > NAME_MAX)
226 return -ENAMETOOLONG;
227
228 err = ceph_init_dentry(dentry);
229 if (err < 0)
230 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700231
232 /* do the open */
233 req = prepare_open_request(dir->i_sb, flags, mode);
234 if (IS_ERR(req))
Al Virod9585272012-06-22 12:39:14 +0400235 return PTR_ERR(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700236 req->r_dentry = dget(dentry);
237 req->r_num_caps = 2;
238 if (flags & O_CREAT) {
239 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
240 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
241 }
242 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700243 err = ceph_mdsc_do_request(mdsc,
244 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
245 req);
Sam Lang79aec9842012-12-19 09:44:23 -1000246 if (err)
247 goto out_err;
248
Sage Weil468640e2011-07-26 11:28:11 -0700249 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil5ef50c32012-07-31 11:27:36 -0700250 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700251 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700252
253 if (d_unhashed(dentry)) {
254 dn = ceph_finish_lookup(req, dentry, err);
255 if (IS_ERR(dn))
256 err = PTR_ERR(dn);
257 } else {
258 /* we were given a hashed negative dentry */
259 dn = NULL;
260 }
Sage Weil468640e2011-07-26 11:28:11 -0700261 if (err)
Sage Weil5ef50c32012-07-31 11:27:36 -0700262 goto out_err;
263 if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) {
264 /* make vfs retry on splice, ENOENT, or symlink */
265 dout("atomic_open finish_no_open on dn %p\n", dn);
266 err = finish_no_open(file, dn);
267 } else {
268 dout("atomic_open finish_open on dn %p\n", dn);
Sam Lang6e8575f2012-12-28 09:56:46 -0800269 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
270 *opened |= FILE_CREATED;
271 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700272 err = finish_open(file, dentry, ceph_open, opened);
273 }
274
275out_err:
Sage Weil124e68e2009-10-06 11:31:08 -0700276 ceph_mdsc_put_request(req);
Sage Weil5ef50c32012-07-31 11:27:36 -0700277 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400278 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700279}
280
281int ceph_release(struct inode *inode, struct file *file)
282{
283 struct ceph_inode_info *ci = ceph_inode(inode);
284 struct ceph_file_info *cf = file->private_data;
285
286 dout("release inode %p file %p\n", inode, file);
287 ceph_put_fmode(ci, cf->fmode);
288 if (cf->last_readdir)
289 ceph_mdsc_put_request(cf->last_readdir);
290 kfree(cf->last_name);
291 kfree(cf->dir_info);
292 dput(cf->dentry);
293 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800294
295 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700296 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700297 return 0;
298}
299
300/*
Sage Weil124e68e2009-10-06 11:31:08 -0700301 * Read a range of bytes striped over one or more objects. Iterate over
302 * objects we stripe over. (That's not atomic, but good enough for now.)
303 *
304 * If we get a short result from the OSD, check against i_size; we need to
305 * only return a short read to the caller if we hit EOF.
306 */
307static int striped_read(struct inode *inode,
308 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800309 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700310 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800311 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700312{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700313 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700314 struct ceph_inode_info *ci = ceph_inode(inode);
315 u64 pos, this_len;
Sage Weilb7495fc2010-11-09 12:43:12 -0800316 int io_align, page_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700317 int left, pages_left;
318 int read;
319 struct page **page_pos;
320 int ret;
321 bool hit_stripe, was_short;
322
323 /*
324 * we may need to do multiple reads. not atomic, unfortunately.
325 */
326 pos = off;
327 left = len;
328 page_pos = pages;
329 pages_left = num_pages;
330 read = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800331 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700332
333more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700334 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800335 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800336 else
337 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700338 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700339 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700340 &ci->i_layout, pos, &this_len,
341 ci->i_truncate_seq,
342 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800343 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700344 if (ret == -ENOENT)
345 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700346 hit_stripe = this_len < left;
347 was_short = ret >= 0 && ret < this_len;
Sage Weil124e68e2009-10-06 11:31:08 -0700348 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read,
349 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
350
351 if (ret > 0) {
Sage Weil773e9b42011-06-07 20:57:14 -0700352 int didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700353
354 if (read < pos - off) {
355 dout(" zero gap %llu to %llu\n", off + read, pos);
Sage Weil773e9b42011-06-07 20:57:14 -0700356 ceph_zero_page_vector_range(page_align + read,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700357 pos - off - read, pages);
Sage Weil124e68e2009-10-06 11:31:08 -0700358 }
359 pos += ret;
360 read = pos - off;
361 left -= ret;
362 page_pos += didpages;
363 pages_left -= didpages;
364
365 /* hit stripe? */
366 if (left && hit_stripe)
367 goto more;
368 }
369
370 if (was_short) {
Sage Weilc3cd6282011-06-01 16:08:44 -0700371 /* did we bounce off eof? */
372 if (pos + left > inode->i_size)
373 *checkeof = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700374
Sage Weilc3cd6282011-06-01 16:08:44 -0700375 /* zero trailing bytes (inside i_size) */
376 if (left > 0 && pos < inode->i_size) {
377 if (pos + left > inode->i_size)
378 left = inode->i_size - pos;
379
380 dout("zero tail %d\n", left);
Sage Weil773e9b42011-06-07 20:57:14 -0700381 ceph_zero_page_vector_range(page_align + read, left,
Sage Weilc3cd6282011-06-01 16:08:44 -0700382 pages);
383 read += left;
384 }
Sage Weil124e68e2009-10-06 11:31:08 -0700385 }
386
Sage Weil124e68e2009-10-06 11:31:08 -0700387 if (ret >= 0)
388 ret = read;
389 dout("striped_read returns %d\n", ret);
390 return ret;
391}
392
393/*
394 * Completely synchronous read and write methods. Direct from __user
395 * buffer to osd, or directly to user pages (if O_DIRECT).
396 *
397 * If the read spans object boundary, just do multiple reads.
398 */
399static ssize_t ceph_sync_read(struct file *file, char __user *data,
Sage Weil6a026582010-02-09 14:04:02 -0800400 unsigned len, loff_t *poff, int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700401{
Al Viro496ad9a2013-01-23 17:07:38 -0500402 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700403 struct page **pages;
404 u64 off = *poff;
Henry C Changab226e22010-12-15 20:41:54 -0800405 int num_pages, ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700406
407 dout("sync_read on file %p %llu~%u %s\n", file, off, len,
408 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
409
Henry C Changab226e22010-12-15 20:41:54 -0800410 if (file->f_flags & O_DIRECT) {
411 num_pages = calc_pages_for((unsigned long)data, len);
Henry C Changb6aa5902010-12-15 20:45:41 -0800412 pages = ceph_get_direct_page_vector(data, num_pages, true);
Henry C Changab226e22010-12-15 20:41:54 -0800413 } else {
414 num_pages = calc_pages_for(off, len);
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700415 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Henry C Changab226e22010-12-15 20:41:54 -0800416 }
Sage Weil124e68e2009-10-06 11:31:08 -0700417 if (IS_ERR(pages))
418 return PTR_ERR(pages);
419
Sage Weile98b6fe2010-11-09 12:24:53 -0800420 /*
421 * flush any page cache pages in this range. this
422 * will make concurrent normal and sync io slow,
423 * but it will at least behave sensibly when they are
424 * in sequence.
425 */
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800426 ret = filemap_write_and_wait(inode->i_mapping);
427 if (ret < 0)
428 goto done;
429
Sage Weilb7495fc2010-11-09 12:43:12 -0800430 ret = striped_read(inode, off, len, pages, num_pages, checkeof,
Henry C Changab226e22010-12-15 20:41:54 -0800431 file->f_flags & O_DIRECT,
432 (unsigned long)data & ~PAGE_MASK);
Sage Weil124e68e2009-10-06 11:31:08 -0700433
434 if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700435 ret = ceph_copy_page_vector_to_user(pages, data, off, ret);
Sage Weil124e68e2009-10-06 11:31:08 -0700436 if (ret >= 0)
437 *poff = off + ret;
438
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800439done:
Sage Weil124e68e2009-10-06 11:31:08 -0700440 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800441 ceph_put_page_vector(pages, num_pages, true);
Sage Weil124e68e2009-10-06 11:31:08 -0700442 else
443 ceph_release_page_vector(pages, num_pages);
444 dout("sync_read result %d\n", ret);
445 return ret;
446}
447
448/*
449 * Write commit callback, called if we requested both an ACK and
450 * ONDISK commit reply from the OSD.
451 */
452static void sync_write_commit(struct ceph_osd_request *req,
453 struct ceph_msg *msg)
454{
455 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
456
457 dout("sync_write_commit %p tid %llu\n", req, req->r_tid);
458 spin_lock(&ci->i_unsafe_lock);
459 list_del_init(&req->r_unsafe_item);
460 spin_unlock(&ci->i_unsafe_lock);
461 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
462}
463
464/*
465 * Synchronous write, straight from __user pointer or user pages (if
466 * O_DIRECT).
467 *
468 * If write spans object boundary, just do multiple writes. (For a
469 * correct atomic write, we should e.g. take write locks on all
470 * objects, rollback on failure, etc.)
471 */
472static ssize_t ceph_sync_write(struct file *file, const char __user *data,
473 size_t left, loff_t *offset)
474{
Al Viro496ad9a2013-01-23 17:07:38 -0500475 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700476 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700477 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700478 struct ceph_osd_request *req;
479 struct page **pages;
480 int num_pages;
481 long long unsigned pos;
482 u64 len;
483 int written = 0;
484 int flags;
485 int do_sync = 0;
486 int check_caps = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800487 int page_align, io_align;
Henry C Changab226e22010-12-15 20:41:54 -0800488 unsigned long buf_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700489 int ret;
490 struct timespec mtime = CURRENT_TIME;
491
Al Viro496ad9a2013-01-23 17:07:38 -0500492 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700493 return -EROFS;
494
495 dout("sync_write on file %p %lld~%u %s\n", file, *offset,
496 (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
497
498 if (file->f_flags & O_APPEND)
499 pos = i_size_read(inode);
500 else
501 pos = *offset;
502
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800503 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left);
504 if (ret < 0)
505 return ret;
506
507 ret = invalidate_inode_pages2_range(inode->i_mapping,
508 pos >> PAGE_CACHE_SHIFT,
509 (pos + left) >> PAGE_CACHE_SHIFT);
510 if (ret < 0)
511 dout("invalidate_inode_pages2_range returned %d\n", ret);
512
Sage Weil124e68e2009-10-06 11:31:08 -0700513 flags = CEPH_OSD_FLAG_ORDERSNAP |
514 CEPH_OSD_FLAG_ONDISK |
515 CEPH_OSD_FLAG_WRITE;
516 if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0)
517 flags |= CEPH_OSD_FLAG_ACK;
518 else
519 do_sync = 1;
520
521 /*
522 * we may need to do multiple writes here if we span an object
523 * boundary. this isn't atomic, unfortunately. :(
524 */
525more:
Sage Weild7f124f2011-06-13 16:22:18 -0700526 io_align = pos & ~PAGE_MASK;
527 buf_align = (unsigned long)data & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700528 len = left;
Henry C Changab226e22010-12-15 20:41:54 -0800529 if (file->f_flags & O_DIRECT) {
Sage Weilb7495fc2010-11-09 12:43:12 -0800530 /* write from beginning of first page, regardless of
531 io alignment */
Henry C Changab226e22010-12-15 20:41:54 -0800532 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
533 num_pages = calc_pages_for((unsigned long)data, len);
534 } else {
Sage Weilb7495fc2010-11-09 12:43:12 -0800535 page_align = pos & ~PAGE_MASK;
Henry C Changab226e22010-12-15 20:41:54 -0800536 num_pages = calc_pages_for(pos, len);
537 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700538 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Sage Weil124e68e2009-10-06 11:31:08 -0700539 ceph_vino(inode), pos, &len,
540 CEPH_OSD_OP_WRITE, flags,
541 ci->i_snap_realm->cached_context,
542 do_sync,
543 ci->i_truncate_seq, ci->i_truncate_size,
Alex Eldera3bea472013-02-15 11:42:29 -0600544 &mtime, false, page_align);
Sage Weil68162822012-09-24 21:01:02 -0700545 if (IS_ERR(req))
546 return PTR_ERR(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700547
Sage Weil124e68e2009-10-06 11:31:08 -0700548 if (file->f_flags & O_DIRECT) {
Henry C Changb6aa5902010-12-15 20:45:41 -0800549 pages = ceph_get_direct_page_vector(data, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700550 if (IS_ERR(pages)) {
551 ret = PTR_ERR(pages);
552 goto out;
553 }
554
555 /*
556 * throw out any page cache pages in this range. this
557 * may block.
558 */
Sage Weil213c99e2010-08-03 10:25:11 -0700559 truncate_inode_pages_range(inode->i_mapping, pos,
Sage Weil5c6a2cd2010-04-22 13:48:59 -0700560 (pos+len) | (PAGE_CACHE_SIZE-1));
Sage Weil124e68e2009-10-06 11:31:08 -0700561 } else {
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700562 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700563 if (IS_ERR(pages)) {
564 ret = PTR_ERR(pages);
565 goto out;
566 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700567 ret = ceph_copy_user_to_page_vector(pages, data, pos, len);
Sage Weil124e68e2009-10-06 11:31:08 -0700568 if (ret < 0) {
569 ceph_release_page_vector(pages, num_pages);
570 goto out;
571 }
572
573 if ((file->f_flags & O_SYNC) == 0) {
574 /* get a second commit callback */
575 req->r_safe_callback = sync_write_commit;
576 req->r_own_pages = 1;
577 }
578 }
579 req->r_pages = pages;
580 req->r_num_pages = num_pages;
581 req->r_inode = inode;
582
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700583 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700584 if (!ret) {
585 if (req->r_safe_callback) {
586 /*
587 * Add to inode unsafe list only after we
588 * start_request so that a tid has been assigned.
589 */
590 spin_lock(&ci->i_unsafe_lock);
Henry C Chang49bcb932011-03-15 09:18:02 +0000591 list_add_tail(&req->r_unsafe_item,
592 &ci->i_unsafe_writes);
Sage Weil124e68e2009-10-06 11:31:08 -0700593 spin_unlock(&ci->i_unsafe_lock);
594 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
595 }
Henry C Chang78a25562011-03-15 09:18:01 +0000596
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700597 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Henry C Chang78a25562011-03-15 09:18:01 +0000598 if (ret < 0 && req->r_safe_callback) {
599 spin_lock(&ci->i_unsafe_lock);
600 list_del_init(&req->r_unsafe_item);
601 spin_unlock(&ci->i_unsafe_lock);
602 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
603 }
Sage Weil124e68e2009-10-06 11:31:08 -0700604 }
605
606 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800607 ceph_put_page_vector(pages, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700608 else if (file->f_flags & O_SYNC)
609 ceph_release_page_vector(pages, num_pages);
610
611out:
612 ceph_osdc_put_request(req);
613 if (ret == 0) {
614 pos += len;
615 written += len;
616 left -= len;
Sage Weild7f124f2011-06-13 16:22:18 -0700617 data += written;
Sage Weil124e68e2009-10-06 11:31:08 -0700618 if (left)
619 goto more;
620
621 ret = written;
622 *offset = pos;
623 if (pos > i_size_read(inode))
624 check_caps = ceph_inode_set_size(inode, pos);
625 if (check_caps)
626 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY,
627 NULL);
628 }
629 return ret;
630}
631
632/*
633 * Wrap generic_file_aio_read with checks for cap bits on the inode.
634 * Atomically grab references, so that those bits are not released
635 * back to the MDS mid-read.
636 *
637 * Hmm, the sync read case isn't actually async... should it be?
638 */
639static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
640 unsigned long nr_segs, loff_t pos)
641{
642 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700643 struct ceph_file_info *fi = filp->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700644 loff_t *ppos = &iocb->ki_pos;
645 size_t len = iov->iov_len;
Al Viro496ad9a2013-01-23 17:07:38 -0500646 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -0700647 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700648 void __user *base = iov->iov_base;
Sage Weil124e68e2009-10-06 11:31:08 -0700649 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700650 int want, got = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800651 int checkeof = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700652
653 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
654 inode, ceph_vinop(inode), pos, (unsigned)len, inode);
Sage Weil6a026582010-02-09 14:04:02 -0800655again:
Yan, Zheng3f999692013-03-01 10:57:54 +0800656 __ceph_do_pending_vmtruncate(inode, true);
Sage Weil29625072010-05-27 10:40:43 -0700657 if (fi->fmode & CEPH_FILE_MODE_LAZY)
658 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
659 else
660 want = CEPH_CAP_FILE_CACHE;
661 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
Sage Weil124e68e2009-10-06 11:31:08 -0700662 if (ret < 0)
663 goto out;
664 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
665 inode, ceph_vinop(inode), pos, (unsigned)len,
666 ceph_cap_string(got));
667
Sage Weil29625072010-05-27 10:40:43 -0700668 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700669 (iocb->ki_filp->f_flags & O_DIRECT) ||
Sage Weil4918b6d2011-07-26 11:26:07 -0700670 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
671 (fi->flags & CEPH_F_SYNC))
Sage Weil124e68e2009-10-06 11:31:08 -0700672 /* hmm, this isn't really async... */
Sage Weil6a026582010-02-09 14:04:02 -0800673 ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
Sage Weil124e68e2009-10-06 11:31:08 -0700674 else
675 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
676
677out:
678 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
679 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
680 ceph_put_cap_refs(ci, got);
Sage Weil6a026582010-02-09 14:04:02 -0800681
682 if (checkeof && ret >= 0) {
683 int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
684
685 /* hit EOF or hole? */
686 if (statret == 0 && *ppos < inode->i_size) {
Sage Weilc3cd6282011-06-01 16:08:44 -0700687 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 -0800688 read += ret;
689 base += ret;
690 len -= ret;
691 checkeof = 0;
692 goto again;
693 }
694 }
695 if (ret >= 0)
696 ret += read;
697
Sage Weil124e68e2009-10-06 11:31:08 -0700698 return ret;
699}
700
701/*
702 * Take cap references to avoid releasing caps to MDS mid-write.
703 *
704 * If we are synchronous, and write with an old snap context, the OSD
705 * may return EOLDSNAPC. In that case, retry the write.. _after_
706 * dropping our cap refs and allowing the pending snap to logically
707 * complete _before_ this write occurs.
708 *
709 * If we are near ENOSPC, write synchronously.
710 */
711static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov,
712 unsigned long nr_segs, loff_t pos)
713{
714 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700715 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -0500716 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700717 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700718 struct ceph_osd_client *osdc =
719 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Sage Weil124e68e2009-10-06 11:31:08 -0700720 loff_t endoff = pos + iov->iov_len;
Sage Weil7971bd92013-05-01 21:15:58 -0700721 int want, got = 0;
722 int ret, err;
Sage Weil124e68e2009-10-06 11:31:08 -0700723
724 if (ceph_snap(inode) != CEPH_NOSNAP)
725 return -EROFS;
726
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800727 sb_start_write(inode->i_sb);
Sage Weil124e68e2009-10-06 11:31:08 -0700728retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800729 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
730 ret = -ENOSPC;
731 goto out;
732 }
Yan, Zheng3f999692013-03-01 10:57:54 +0800733 __ceph_do_pending_vmtruncate(inode, true);
Sage Weil7971bd92013-05-01 21:15:58 -0700734 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
735 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
736 inode->i_size);
737 if (fi->fmode & CEPH_FILE_MODE_LAZY)
738 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
739 else
740 want = CEPH_CAP_FILE_BUFFER;
741 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
742 if (ret < 0)
743 goto out_put;
Sage Weil124e68e2009-10-06 11:31:08 -0700744
Sage Weil7971bd92013-05-01 21:15:58 -0700745 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
746 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
747 ceph_cap_string(got));
748
749 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
750 (iocb->ki_filp->f_flags & O_DIRECT) ||
751 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
752 (fi->flags & CEPH_F_SYNC)) {
753 ret = ceph_sync_write(file, iov->iov_base, iov->iov_len,
754 &iocb->ki_pos);
755 } else {
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800756 mutex_lock(&inode->i_mutex);
757 ret = __generic_file_aio_write(iocb, iov, nr_segs,
758 &iocb->ki_pos);
759 mutex_unlock(&inode->i_mutex);
Sage Weil124e68e2009-10-06 11:31:08 -0700760 }
Sage Weild8de9ab2011-07-26 11:27:34 -0700761
Sage Weil124e68e2009-10-06 11:31:08 -0700762 if (ret >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -0700763 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -0800764 spin_lock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700765 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -0800766 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700767 if (dirty)
768 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -0700769 }
Sage Weil7971bd92013-05-01 21:15:58 -0700770
771out_put:
Sage Weil124e68e2009-10-06 11:31:08 -0700772 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Sage Weil7971bd92013-05-01 21:15:58 -0700773 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
774 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -0700775 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -0700776
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800777 if (ret >= 0 &&
778 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
779 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
780 err = vfs_fsync_range(file, pos, pos + ret - 1, 1);
781 if (err < 0)
782 ret = err;
783 }
Sage Weild8de9ab2011-07-26 11:27:34 -0700784out:
Sage Weil124e68e2009-10-06 11:31:08 -0700785 if (ret == -EOLDSNAPC) {
786 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
787 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);
788 goto retry_snap;
789 }
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800790 sb_end_write(inode->i_sb);
Sage Weil124e68e2009-10-06 11:31:08 -0700791
792 return ret;
793}
794
795/*
796 * llseek. be sure to verify file size on SEEK_END.
797 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800798static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -0700799{
800 struct inode *inode = file->f_mapping->host;
801 int ret;
802
803 mutex_lock(&inode->i_mutex);
Yan, Zheng3f999692013-03-01 10:57:54 +0800804 __ceph_do_pending_vmtruncate(inode, false);
Sage Weil6a82c472011-12-13 09:19:26 -0800805
Andrew Morton965c8e52012-12-17 15:59:39 -0800806 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Sage Weil124e68e2009-10-06 11:31:08 -0700807 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
808 if (ret < 0) {
809 offset = ret;
810 goto out;
811 }
Josef Bacik06222e42011-07-18 13:21:38 -0400812 }
813
Andrew Morton965c8e52012-12-17 15:59:39 -0800814 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -0400815 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -0700816 offset += inode->i_size;
817 break;
818 case SEEK_CUR:
819 /*
820 * Here we special-case the lseek(fd, 0, SEEK_CUR)
821 * position-querying operation. Avoid rewriting the "same"
822 * f_pos value back to the file because a concurrent read(),
823 * write() or lseek() might have altered it
824 */
825 if (offset == 0) {
826 offset = file->f_pos;
827 goto out;
828 }
829 offset += file->f_pos;
830 break;
Josef Bacik06222e42011-07-18 13:21:38 -0400831 case SEEK_DATA:
832 if (offset >= inode->i_size) {
833 ret = -ENXIO;
834 goto out;
835 }
836 break;
837 case SEEK_HOLE:
838 if (offset >= inode->i_size) {
839 ret = -ENXIO;
840 goto out;
841 }
842 offset = inode->i_size;
843 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700844 }
845
846 if (offset < 0 || offset > inode->i_sb->s_maxbytes) {
847 offset = -EINVAL;
848 goto out;
849 }
850
851 /* Special lock needed here? */
852 if (offset != file->f_pos) {
853 file->f_pos = offset;
854 file->f_version = 0;
855 }
856
857out:
858 mutex_unlock(&inode->i_mutex);
859 return offset;
860}
861
862const struct file_operations ceph_file_fops = {
863 .open = ceph_open,
864 .release = ceph_release,
865 .llseek = ceph_llseek,
866 .read = do_sync_read,
867 .write = do_sync_write,
868 .aio_read = ceph_aio_read,
869 .aio_write = ceph_aio_write,
870 .mmap = ceph_mmap,
871 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -0700872 .lock = ceph_lock,
873 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -0700874 .splice_read = generic_file_splice_read,
875 .splice_write = generic_file_splice_write,
876 .unlocked_ioctl = ceph_ioctl,
877 .compat_ioctl = ceph_ioctl,
878};
879