]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
124e68e7 | 2 | |
3d14c5d2 | 3 | #include <linux/module.h> |
124e68e7 | 4 | #include <linux/sched.h> |
5a0e3ad6 | 5 | #include <linux/slab.h> |
124e68e7 | 6 | #include <linux/file.h> |
5ef50c3b | 7 | #include <linux/mount.h> |
124e68e7 SW |
8 | #include <linux/namei.h> |
9 | #include <linux/writeback.h> | |
ad7a60de | 10 | #include <linux/falloc.h> |
124e68e7 SW |
11 | |
12 | #include "super.h" | |
13 | #include "mds_client.h" | |
99ccbd22 | 14 | #include "cache.h" |
124e68e7 SW |
15 | |
16 | /* | |
17 | * Ceph file operations | |
18 | * | |
19 | * Implement basic open/close functionality, and implement | |
20 | * read/write. | |
21 | * | |
22 | * We implement three modes of file I/O: | |
23 | * - buffered uses the generic_file_aio_{read,write} helpers | |
24 | * | |
25 | * - synchronous is used when there is multi-client read/write | |
26 | * sharing, avoids the page cache, and synchronously waits for an | |
27 | * ack from the OSD. | |
28 | * | |
29 | * - direct io takes the variant of the sync path that references | |
30 | * user pages directly. | |
31 | * | |
32 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
33 | * for writeback: since the MDS can recover size and mtime there is no | |
34 | * need to wait for MDS acknowledgement. | |
35 | */ | |
36 | ||
37 | ||
38 | /* | |
39 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
40 | * inopportune ENOMEM later. | |
41 | */ | |
42 | static struct ceph_mds_request * | |
43 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
44 | { | |
3d14c5d2 YS |
45 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
46 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
47 | struct ceph_mds_request *req; |
48 | int want_auth = USE_ANY_MDS; | |
49 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
50 | ||
51 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
52 | want_auth = USE_AUTH_MDS; | |
53 | ||
54 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
55 | if (IS_ERR(req)) | |
56 | goto out; | |
57 | req->r_fmode = ceph_flags_to_mode(flags); | |
58 | req->r_args.open.flags = cpu_to_le32(flags); | |
59 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
124e68e7 SW |
60 | out: |
61 | return req; | |
62 | } | |
63 | ||
64 | /* | |
65 | * initialize private struct file data. | |
66 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
67 | */ | |
68 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
69 | { | |
70 | struct ceph_file_info *cf; | |
71 | int ret = 0; | |
99ccbd22 MT |
72 | struct ceph_inode_info *ci = ceph_inode(inode); |
73 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); | |
74 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
75 | |
76 | switch (inode->i_mode & S_IFMT) { | |
77 | case S_IFREG: | |
99ccbd22 MT |
78 | /* First file open request creates the cookie, we want to keep |
79 | * this cookie around for the filetime of the inode as not to | |
80 | * have to worry about fscache register / revoke / operation | |
81 | * races. | |
82 | * | |
83 | * Also, if we know the operation is going to invalidate data | |
84 | * (non readonly) just nuke the cache right away. | |
85 | */ | |
86 | ceph_fscache_register_inode_cookie(mdsc->fsc, ci); | |
87 | if ((fmode & CEPH_FILE_MODE_WR)) | |
88 | ceph_fscache_invalidate(inode); | |
124e68e7 SW |
89 | case S_IFDIR: |
90 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
91 | inode->i_mode); | |
92 | cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO); | |
93 | if (cf == NULL) { | |
94 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
95 | return -ENOMEM; | |
96 | } | |
97 | cf->fmode = fmode; | |
98 | cf->next_offset = 2; | |
99 | file->private_data = cf; | |
100 | BUG_ON(inode->i_fop->release != ceph_release); | |
101 | break; | |
102 | ||
103 | case S_IFLNK: | |
104 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
105 | inode->i_mode); | |
106 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
107 | break; | |
108 | ||
109 | default: | |
110 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
111 | inode->i_mode); | |
112 | /* | |
113 | * we need to drop the open ref now, since we don't | |
114 | * have .release set to ceph_release. | |
115 | */ | |
116 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
117 | BUG_ON(inode->i_fop->release == ceph_release); | |
118 | ||
119 | /* call the proper open fop */ | |
120 | ret = inode->i_fop->open(inode, file); | |
121 | } | |
122 | return ret; | |
123 | } | |
124 | ||
125 | /* | |
124e68e7 SW |
126 | * If we already have the requisite capabilities, we can satisfy |
127 | * the open request locally (no need to request new caps from the | |
128 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
129 | * if our wanted caps set expands. | |
130 | */ | |
131 | int ceph_open(struct inode *inode, struct file *file) | |
132 | { | |
133 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
134 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
135 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
136 | struct ceph_mds_request *req; |
137 | struct ceph_file_info *cf = file->private_data; | |
5f21c96d | 138 | struct inode *parent_inode = NULL; |
124e68e7 SW |
139 | int err; |
140 | int flags, fmode, wanted; | |
141 | ||
142 | if (cf) { | |
143 | dout("open file %p is already opened\n", file); | |
144 | return 0; | |
145 | } | |
146 | ||
147 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
148 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
149 | if (S_ISDIR(inode->i_mode)) | |
150 | flags = O_DIRECTORY; /* mds likes to know */ | |
151 | ||
152 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
153 | ceph_vinop(inode), file, flags, file->f_flags); | |
154 | fmode = ceph_flags_to_mode(flags); | |
155 | wanted = ceph_caps_for_mode(fmode); | |
156 | ||
157 | /* snapped files are read-only */ | |
158 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
159 | return -EROFS; | |
160 | ||
161 | /* trivially open snapdir */ | |
162 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 163 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 164 | __ceph_get_fmode(ci, fmode); |
be655596 | 165 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
166 | return ceph_init_file(inode, file, fmode); |
167 | } | |
168 | ||
169 | /* | |
7421ab80 SW |
170 | * No need to block if we have caps on the auth MDS (for |
171 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
172 | * asynchronously. |
173 | */ | |
be655596 | 174 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
175 | if (__ceph_is_any_real_caps(ci) && |
176 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
177 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
178 | int issued = __ceph_caps_issued(ci, NULL); | |
179 | ||
180 | dout("open %p fmode %d want %s issued %s using existing\n", | |
181 | inode, fmode, ceph_cap_string(wanted), | |
182 | ceph_cap_string(issued)); | |
183 | __ceph_get_fmode(ci, fmode); | |
be655596 | 184 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
185 | |
186 | /* adjust wanted? */ | |
187 | if ((issued & wanted) != wanted && | |
188 | (mds_wanted & wanted) != wanted && | |
189 | ceph_snap(inode) != CEPH_SNAPDIR) | |
190 | ceph_check_caps(ci, 0, NULL); | |
191 | ||
192 | return ceph_init_file(inode, file, fmode); | |
193 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
194 | (ci->i_snap_caps & wanted) == wanted) { | |
195 | __ceph_get_fmode(ci, fmode); | |
be655596 | 196 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
197 | return ceph_init_file(inode, file, fmode); |
198 | } | |
99ccbd22 | 199 | |
be655596 | 200 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
201 | |
202 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
203 | req = prepare_open_request(inode->i_sb, flags, 0); | |
204 | if (IS_ERR(req)) { | |
205 | err = PTR_ERR(req); | |
206 | goto out; | |
207 | } | |
70b666c3 SW |
208 | req->r_inode = inode; |
209 | ihold(inode); | |
99ccbd22 | 210 | |
124e68e7 | 211 | req->r_num_caps = 1; |
752c8bdc | 212 | if (flags & O_CREAT) |
b583043e | 213 | parent_inode = ceph_get_dentry_parent_inode(file->f_path.dentry); |
124e68e7 | 214 | err = ceph_mdsc_do_request(mdsc, parent_inode, req); |
5f21c96d | 215 | iput(parent_inode); |
124e68e7 SW |
216 | if (!err) |
217 | err = ceph_init_file(inode, file, req->r_fmode); | |
218 | ceph_mdsc_put_request(req); | |
219 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
220 | out: | |
221 | return err; | |
222 | } | |
223 | ||
224 | ||
225 | /* | |
5ef50c3b SW |
226 | * Do a lookup + open with a single request. If we get a non-existent |
227 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 228 | */ |
5ef50c3b | 229 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 230 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 231 | int *opened) |
124e68e7 | 232 | { |
3d14c5d2 YS |
233 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
234 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 235 | struct ceph_mds_request *req; |
5ef50c3b | 236 | struct dentry *dn; |
b1ee94aa | 237 | struct ceph_acls_info acls = {}; |
124e68e7 | 238 | int err; |
124e68e7 | 239 | |
a455589f AV |
240 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
241 | dir, dentry, dentry, | |
5ef50c3b SW |
242 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
243 | ||
244 | if (dentry->d_name.len > NAME_MAX) | |
245 | return -ENAMETOOLONG; | |
246 | ||
247 | err = ceph_init_dentry(dentry); | |
248 | if (err < 0) | |
249 | return err; | |
124e68e7 | 250 | |
b1ee94aa YZ |
251 | if (flags & O_CREAT) { |
252 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
253 | if (err < 0) | |
254 | return err; | |
255 | } | |
256 | ||
124e68e7 SW |
257 | /* do the open */ |
258 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
259 | if (IS_ERR(req)) { |
260 | err = PTR_ERR(req); | |
261 | goto out_acl; | |
262 | } | |
124e68e7 SW |
263 | req->r_dentry = dget(dentry); |
264 | req->r_num_caps = 2; | |
265 | if (flags & O_CREAT) { | |
266 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
267 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
268 | if (acls.pagelist) { |
269 | req->r_pagelist = acls.pagelist; | |
270 | acls.pagelist = NULL; | |
271 | } | |
124e68e7 SW |
272 | } |
273 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ | |
acda7657 SW |
274 | err = ceph_mdsc_do_request(mdsc, |
275 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
276 | req); | |
bf91c315 | 277 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 278 | if (err) |
b1ee94aa | 279 | goto out_req; |
79aec984 | 280 | |
5ef50c3b | 281 | if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 282 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 283 | |
5ef50c3b SW |
284 | if (d_unhashed(dentry)) { |
285 | dn = ceph_finish_lookup(req, dentry, err); | |
286 | if (IS_ERR(dn)) | |
287 | err = PTR_ERR(dn); | |
288 | } else { | |
289 | /* we were given a hashed negative dentry */ | |
290 | dn = NULL; | |
291 | } | |
292 | if (err) | |
b1ee94aa | 293 | goto out_req; |
2b0143b5 | 294 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
295 | /* make vfs retry on splice, ENOENT, or symlink */ |
296 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
297 | err = finish_no_open(file, dn); | |
298 | } else { | |
299 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 300 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 301 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
302 | *opened |= FILE_CREATED; |
303 | } | |
5ef50c3b SW |
304 | err = finish_open(file, dentry, ceph_open, opened); |
305 | } | |
b1ee94aa | 306 | out_req: |
ab866549 YZ |
307 | if (!req->r_err && req->r_target_inode) |
308 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 309 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
310 | out_acl: |
311 | ceph_release_acls_info(&acls); | |
5ef50c3b | 312 | dout("atomic_open result=%d\n", err); |
d9585277 | 313 | return err; |
124e68e7 SW |
314 | } |
315 | ||
316 | int ceph_release(struct inode *inode, struct file *file) | |
317 | { | |
318 | struct ceph_inode_info *ci = ceph_inode(inode); | |
319 | struct ceph_file_info *cf = file->private_data; | |
320 | ||
321 | dout("release inode %p file %p\n", inode, file); | |
322 | ceph_put_fmode(ci, cf->fmode); | |
323 | if (cf->last_readdir) | |
324 | ceph_mdsc_put_request(cf->last_readdir); | |
325 | kfree(cf->last_name); | |
326 | kfree(cf->dir_info); | |
327 | dput(cf->dentry); | |
328 | kmem_cache_free(ceph_file_cachep, cf); | |
195d3ce2 SW |
329 | |
330 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 331 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
332 | return 0; |
333 | } | |
334 | ||
83701246 YZ |
335 | enum { |
336 | CHECK_EOF = 1, | |
337 | READ_INLINE = 2, | |
338 | }; | |
339 | ||
124e68e7 SW |
340 | /* |
341 | * Read a range of bytes striped over one or more objects. Iterate over | |
342 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
343 | * | |
344 | * If we get a short result from the OSD, check against i_size; we need to | |
345 | * only return a short read to the caller if we hit EOF. | |
346 | */ | |
347 | static int striped_read(struct inode *inode, | |
348 | u64 off, u64 len, | |
6a026589 | 349 | struct page **pages, int num_pages, |
c3cd6283 | 350 | int *checkeof, bool o_direct, |
ab226e21 | 351 | unsigned long buf_align) |
124e68e7 | 352 | { |
3d14c5d2 | 353 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 354 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 355 | u64 pos, this_len, left; |
b7495fc2 | 356 | int io_align, page_align; |
688bac46 | 357 | int pages_left; |
124e68e7 SW |
358 | int read; |
359 | struct page **page_pos; | |
360 | int ret; | |
361 | bool hit_stripe, was_short; | |
362 | ||
363 | /* | |
364 | * we may need to do multiple reads. not atomic, unfortunately. | |
365 | */ | |
366 | pos = off; | |
367 | left = len; | |
368 | page_pos = pages; | |
369 | pages_left = num_pages; | |
370 | read = 0; | |
b7495fc2 | 371 | io_align = off & ~PAGE_MASK; |
124e68e7 SW |
372 | |
373 | more: | |
c3cd6283 | 374 | if (o_direct) |
ab226e21 | 375 | page_align = (pos - io_align + buf_align) & ~PAGE_MASK; |
b7495fc2 SW |
376 | else |
377 | page_align = pos & ~PAGE_MASK; | |
124e68e7 | 378 | this_len = left; |
3d14c5d2 | 379 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
380 | &ci->i_layout, pos, &this_len, |
381 | ci->i_truncate_seq, | |
382 | ci->i_truncate_size, | |
b7495fc2 | 383 | page_pos, pages_left, page_align); |
124e68e7 SW |
384 | if (ret == -ENOENT) |
385 | ret = 0; | |
0e98728f SW |
386 | hit_stripe = this_len < left; |
387 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 388 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
389 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
390 | ||
02ae66d8 | 391 | if (ret >= 0) { |
392 | int didpages; | |
393 | if (was_short && (pos + ret < inode->i_size)) { | |
1487a688 YZ |
394 | int zlen = min(this_len - ret, |
395 | inode->i_size - pos - ret); | |
396 | int zoff = (o_direct ? buf_align : io_align) + | |
397 | read + ret; | |
02ae66d8 | 398 | dout(" zero gap %llu to %llu\n", |
1487a688 YZ |
399 | pos + ret, pos + ret + zlen); |
400 | ceph_zero_page_vector_range(zoff, zlen, pages); | |
401 | ret += zlen; | |
124e68e7 | 402 | } |
02ae66d8 | 403 | |
404 | didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; | |
124e68e7 SW |
405 | pos += ret; |
406 | read = pos - off; | |
407 | left -= ret; | |
408 | page_pos += didpages; | |
409 | pages_left -= didpages; | |
410 | ||
02ae66d8 | 411 | /* hit stripe and need continue*/ |
412 | if (left && hit_stripe && pos < inode->i_size) | |
124e68e7 SW |
413 | goto more; |
414 | } | |
415 | ||
ee7289bf | 416 | if (read > 0) { |
02ae66d8 | 417 | ret = read; |
c3cd6283 SW |
418 | /* did we bounce off eof? */ |
419 | if (pos + left > inode->i_size) | |
83701246 | 420 | *checkeof = CHECK_EOF; |
124e68e7 SW |
421 | } |
422 | ||
124e68e7 SW |
423 | dout("striped_read returns %d\n", ret); |
424 | return ret; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Completely synchronous read and write methods. Direct from __user | |
429 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
430 | * | |
431 | * If the read spans object boundary, just do multiple reads. | |
432 | */ | |
8eb4efb0 | 433 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
434 | int *checkeof) | |
124e68e7 | 435 | { |
8eb4efb0 | 436 | struct file *file = iocb->ki_filp; |
496ad9aa | 437 | struct inode *inode = file_inode(file); |
124e68e7 | 438 | struct page **pages; |
8eb4efb0 | 439 | u64 off = iocb->ki_pos; |
ab226e21 | 440 | int num_pages, ret; |
2b777c9d | 441 | size_t len = iov_iter_count(i); |
124e68e7 | 442 | |
8eb4efb0 | 443 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
444 | (unsigned)len, | |
124e68e7 | 445 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
446 | |
447 | if (!len) | |
448 | return 0; | |
e98b6fed SW |
449 | /* |
450 | * flush any page cache pages in this range. this | |
451 | * will make concurrent normal and sync io slow, | |
452 | * but it will at least behave sensibly when they are | |
453 | * in sequence. | |
454 | */ | |
8eb4efb0 | 455 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
456 | off + len); | |
29065a51 | 457 | if (ret < 0) |
8eb4efb0 | 458 | return ret; |
29065a51 | 459 | |
2ba48ce5 | 460 | if (iocb->ki_flags & IOCB_DIRECT) { |
8eb4efb0 | 461 | while (iov_iter_count(i)) { |
2b777c9d AV |
462 | size_t start; |
463 | ssize_t n; | |
8eb4efb0 | 464 | |
2b777c9d AV |
465 | n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start); |
466 | if (n < 0) | |
467 | return n; | |
8eb4efb0 | 468 | |
2b777c9d AV |
469 | num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE; |
470 | ||
471 | ret = striped_read(inode, off, n, | |
8eb4efb0 | 472 | pages, num_pages, checkeof, |
2b777c9d AV |
473 | 1, start); |
474 | ||
8eb4efb0 | 475 | ceph_put_page_vector(pages, num_pages, true); |
476 | ||
477 | if (ret <= 0) | |
478 | break; | |
479 | off += ret; | |
480 | iov_iter_advance(i, ret); | |
2b777c9d | 481 | if (ret < n) |
8eb4efb0 | 482 | break; |
483 | } | |
484 | } else { | |
485 | num_pages = calc_pages_for(off, len); | |
486 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); | |
487 | if (IS_ERR(pages)) | |
488 | return PTR_ERR(pages); | |
489 | ret = striped_read(inode, off, len, pages, | |
490 | num_pages, checkeof, 0, 0); | |
491 | if (ret > 0) { | |
492 | int l, k = 0; | |
2b777c9d | 493 | size_t left = ret; |
8eb4efb0 | 494 | |
495 | while (left) { | |
5aaa432a YZ |
496 | size_t page_off = off & ~PAGE_MASK; |
497 | size_t copy = min_t(size_t, | |
498 | PAGE_SIZE - page_off, left); | |
499 | l = copy_page_to_iter(pages[k++], page_off, | |
500 | copy, i); | |
2b777c9d AV |
501 | off += l; |
502 | left -= l; | |
503 | if (l < copy) | |
8eb4efb0 | 504 | break; |
505 | } | |
506 | } | |
507 | ceph_release_page_vector(pages, num_pages); | |
508 | } | |
124e68e7 | 509 | |
8eb4efb0 | 510 | if (off > iocb->ki_pos) { |
511 | ret = off - iocb->ki_pos; | |
512 | iocb->ki_pos = off; | |
513 | } | |
124e68e7 | 514 | |
124e68e7 SW |
515 | dout("sync_read result %d\n", ret); |
516 | return ret; | |
517 | } | |
518 | ||
519 | /* | |
26be8808 AE |
520 | * Write commit request unsafe callback, called to tell us when a |
521 | * request is unsafe (that is, in flight--has been handed to the | |
522 | * messenger to send to its target osd). It is called again when | |
523 | * we've received a response message indicating the request is | |
524 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
525 | * is completed early (and unsuccessfully) due to a timeout or | |
526 | * interrupt. | |
527 | * | |
528 | * This is used if we requested both an ACK and ONDISK commit reply | |
529 | * from the OSD. | |
124e68e7 | 530 | */ |
26be8808 | 531 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
532 | { |
533 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
534 | ||
26be8808 AE |
535 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
536 | unsafe ? "un" : ""); | |
537 | if (unsafe) { | |
538 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
539 | spin_lock(&ci->i_unsafe_lock); | |
540 | list_add_tail(&req->r_unsafe_item, | |
541 | &ci->i_unsafe_writes); | |
542 | spin_unlock(&ci->i_unsafe_lock); | |
543 | } else { | |
544 | spin_lock(&ci->i_unsafe_lock); | |
545 | list_del_init(&req->r_unsafe_item); | |
546 | spin_unlock(&ci->i_unsafe_lock); | |
547 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
548 | } | |
124e68e7 SW |
549 | } |
550 | ||
e8344e66 | 551 | |
124e68e7 | 552 | /* |
e8344e66 | 553 | * Synchronous write, straight from __user pointer or user pages. |
124e68e7 SW |
554 | * |
555 | * If write spans object boundary, just do multiple writes. (For a | |
556 | * correct atomic write, we should e.g. take write locks on all | |
557 | * objects, rollback on failure, etc.) | |
558 | */ | |
e8344e66 | 559 | static ssize_t |
5dda377c YZ |
560 | ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
561 | struct ceph_snap_context *snapc) | |
124e68e7 | 562 | { |
e8344e66 | 563 | struct file *file = iocb->ki_filp; |
496ad9aa | 564 | struct inode *inode = file_inode(file); |
124e68e7 | 565 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 566 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 567 | struct ceph_vino vino; |
124e68e7 SW |
568 | struct ceph_osd_request *req; |
569 | struct page **pages; | |
570 | int num_pages; | |
124e68e7 SW |
571 | int written = 0; |
572 | int flags; | |
124e68e7 SW |
573 | int check_caps = 0; |
574 | int ret; | |
575 | struct timespec mtime = CURRENT_TIME; | |
4908b822 | 576 | size_t count = iov_iter_count(from); |
124e68e7 | 577 | |
496ad9aa | 578 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
579 | return -EROFS; |
580 | ||
e8344e66 | 581 | dout("sync_direct_write on file %p %lld~%u\n", file, pos, |
582 | (unsigned)count); | |
124e68e7 | 583 | |
e8344e66 | 584 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
585 | if (ret < 0) |
586 | return ret; | |
587 | ||
588 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
589 | pos >> PAGE_CACHE_SHIFT, | |
e8344e66 | 590 | (pos + count) >> PAGE_CACHE_SHIFT); |
29065a51 YS |
591 | if (ret < 0) |
592 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
593 | ||
124e68e7 SW |
594 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
595 | CEPH_OSD_FLAG_ONDISK | | |
596 | CEPH_OSD_FLAG_WRITE; | |
124e68e7 | 597 | |
4908b822 AV |
598 | while (iov_iter_count(from) > 0) { |
599 | u64 len = iov_iter_single_seg_count(from); | |
64c31311 AV |
600 | size_t start; |
601 | ssize_t n; | |
e8344e66 | 602 | |
e8344e66 | 603 | vino = ceph_vino(inode); |
604 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 605 | vino, pos, &len, 0, |
e8344e66 | 606 | 2,/*include a 'startsync' command*/ |
607 | CEPH_OSD_OP_WRITE, flags, snapc, | |
608 | ci->i_truncate_seq, | |
609 | ci->i_truncate_size, | |
610 | false); | |
611 | if (IS_ERR(req)) { | |
612 | ret = PTR_ERR(req); | |
eab87235 | 613 | break; |
e8344e66 | 614 | } |
124e68e7 | 615 | |
144cba14 | 616 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); |
715e4cd4 | 617 | |
4908b822 | 618 | n = iov_iter_get_pages_alloc(from, &pages, len, &start); |
64c31311 AV |
619 | if (unlikely(n < 0)) { |
620 | ret = n; | |
621 | ceph_osdc_put_request(req); | |
622 | break; | |
124e68e7 SW |
623 | } |
624 | ||
64c31311 | 625 | num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE; |
124e68e7 SW |
626 | /* |
627 | * throw out any page cache pages in this range. this | |
628 | * may block. | |
629 | */ | |
213c99ee | 630 | truncate_inode_pages_range(inode->i_mapping, pos, |
64c31311 AV |
631 | (pos+n) | (PAGE_CACHE_SIZE-1)); |
632 | osd_req_op_extent_osd_data_pages(req, 0, pages, n, start, | |
e8344e66 | 633 | false, false); |
634 | ||
635 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ | |
636 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
637 | ||
638 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
639 | if (!ret) | |
640 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
641 | ||
642 | ceph_put_page_vector(pages, num_pages, false); | |
643 | ||
e8344e66 | 644 | ceph_osdc_put_request(req); |
64c31311 | 645 | if (ret) |
e8344e66 | 646 | break; |
64c31311 AV |
647 | pos += n; |
648 | written += n; | |
4908b822 | 649 | iov_iter_advance(from, n); |
64c31311 AV |
650 | |
651 | if (pos > i_size_read(inode)) { | |
652 | check_caps = ceph_inode_set_size(inode, pos); | |
653 | if (check_caps) | |
654 | ceph_check_caps(ceph_inode(inode), | |
655 | CHECK_CAPS_AUTHONLY, | |
656 | NULL); | |
657 | } | |
e8344e66 | 658 | } |
659 | ||
660 | if (ret != -EOLDSNAPC && written > 0) { | |
661 | iocb->ki_pos = pos; | |
662 | ret = written; | |
663 | } | |
664 | return ret; | |
665 | } | |
666 | ||
667 | ||
668 | /* | |
669 | * Synchronous write, straight from __user pointer or user pages. | |
670 | * | |
671 | * If write spans object boundary, just do multiple writes. (For a | |
672 | * correct atomic write, we should e.g. take write locks on all | |
673 | * objects, rollback on failure, etc.) | |
674 | */ | |
06fee30f | 675 | static ssize_t |
5dda377c YZ |
676 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
677 | struct ceph_snap_context *snapc) | |
e8344e66 | 678 | { |
679 | struct file *file = iocb->ki_filp; | |
680 | struct inode *inode = file_inode(file); | |
681 | struct ceph_inode_info *ci = ceph_inode(inode); | |
682 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 683 | struct ceph_vino vino; |
684 | struct ceph_osd_request *req; | |
685 | struct page **pages; | |
686 | u64 len; | |
687 | int num_pages; | |
688 | int written = 0; | |
689 | int flags; | |
690 | int check_caps = 0; | |
691 | int ret; | |
692 | struct timespec mtime = CURRENT_TIME; | |
4908b822 | 693 | size_t count = iov_iter_count(from); |
e8344e66 | 694 | |
695 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
696 | return -EROFS; | |
697 | ||
698 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
699 | ||
700 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
701 | if (ret < 0) | |
702 | return ret; | |
703 | ||
704 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
705 | pos >> PAGE_CACHE_SHIFT, | |
706 | (pos + count) >> PAGE_CACHE_SHIFT); | |
707 | if (ret < 0) | |
708 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
709 | ||
710 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
711 | CEPH_OSD_FLAG_ONDISK | | |
712 | CEPH_OSD_FLAG_WRITE | | |
713 | CEPH_OSD_FLAG_ACK; | |
714 | ||
4908b822 | 715 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 716 | size_t left; |
717 | int n; | |
718 | ||
e8344e66 | 719 | vino = ceph_vino(inode); |
720 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 721 | vino, pos, &len, 0, 1, |
e8344e66 | 722 | CEPH_OSD_OP_WRITE, flags, snapc, |
723 | ci->i_truncate_seq, | |
724 | ci->i_truncate_size, | |
725 | false); | |
726 | if (IS_ERR(req)) { | |
727 | ret = PTR_ERR(req); | |
eab87235 | 728 | break; |
e8344e66 | 729 | } |
730 | ||
731 | /* | |
732 | * write from beginning of first page, | |
733 | * regardless of io alignment | |
734 | */ | |
735 | num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
736 | ||
34d23762 | 737 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
124e68e7 SW |
738 | if (IS_ERR(pages)) { |
739 | ret = PTR_ERR(pages); | |
740 | goto out; | |
741 | } | |
e8344e66 | 742 | |
743 | left = len; | |
744 | for (n = 0; n < num_pages; n++) { | |
125d725c | 745 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 746 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 747 | if (ret != plen) { |
748 | ret = -EFAULT; | |
749 | break; | |
750 | } | |
751 | left -= ret; | |
e8344e66 | 752 | } |
753 | ||
124e68e7 SW |
754 | if (ret < 0) { |
755 | ceph_release_page_vector(pages, num_pages); | |
756 | goto out; | |
757 | } | |
758 | ||
e8344e66 | 759 | /* get a second commit callback */ |
760 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
761 | req->r_inode = inode; | |
124e68e7 | 762 | |
e8344e66 | 763 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
764 | false, true); | |
02ee07d3 | 765 | |
e8344e66 | 766 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
767 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
124e68e7 | 768 | |
e8344e66 | 769 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
770 | if (!ret) | |
771 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
772 | |
773 | out: | |
e8344e66 | 774 | ceph_osdc_put_request(req); |
775 | if (ret == 0) { | |
776 | pos += len; | |
777 | written += len; | |
778 | ||
779 | if (pos > i_size_read(inode)) { | |
780 | check_caps = ceph_inode_set_size(inode, pos); | |
781 | if (check_caps) | |
782 | ceph_check_caps(ceph_inode(inode), | |
783 | CHECK_CAPS_AUTHONLY, | |
784 | NULL); | |
785 | } | |
786 | } else | |
787 | break; | |
788 | } | |
124e68e7 | 789 | |
e8344e66 | 790 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 791 | ret = written; |
e8344e66 | 792 | iocb->ki_pos = pos; |
124e68e7 SW |
793 | } |
794 | return ret; | |
795 | } | |
796 | ||
797 | /* | |
798 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
799 | * Atomically grab references, so that those bits are not released | |
800 | * back to the MDS mid-read. | |
801 | * | |
802 | * Hmm, the sync read case isn't actually async... should it be? | |
803 | */ | |
3644424d | 804 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
805 | { |
806 | struct file *filp = iocb->ki_filp; | |
2962507c | 807 | struct ceph_file_info *fi = filp->private_data; |
66ee59af | 808 | size_t len = iov_iter_count(to); |
496ad9aa | 809 | struct inode *inode = file_inode(filp); |
124e68e7 | 810 | struct ceph_inode_info *ci = ceph_inode(inode); |
3738daa6 | 811 | struct page *pinned_page = NULL; |
124e68e7 | 812 | ssize_t ret; |
2962507c | 813 | int want, got = 0; |
83701246 | 814 | int retry_op = 0, read = 0; |
124e68e7 | 815 | |
6a026589 | 816 | again: |
8eb4efb0 | 817 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
818 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
819 | ||
2962507c SW |
820 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
821 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
822 | else | |
823 | want = CEPH_CAP_FILE_CACHE; | |
3738daa6 | 824 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
124e68e7 | 825 | if (ret < 0) |
8eb4efb0 | 826 | return ret; |
124e68e7 | 827 | |
2962507c | 828 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
2ba48ce5 | 829 | (iocb->ki_flags & IOCB_DIRECT) || |
8eb4efb0 | 830 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 831 | |
832 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
833 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
834 | ceph_cap_string(got)); | |
835 | ||
83701246 YZ |
836 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
837 | /* hmm, this isn't really async... */ | |
838 | ret = ceph_sync_read(iocb, to, &retry_op); | |
839 | } else { | |
840 | retry_op = READ_INLINE; | |
841 | } | |
8eb4efb0 | 842 | } else { |
8eb4efb0 | 843 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 844 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 845 | ceph_cap_string(got)); |
124e68e7 | 846 | |
3644424d | 847 | ret = generic_file_read_iter(iocb, to); |
8eb4efb0 | 848 | } |
124e68e7 SW |
849 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
850 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 YZ |
851 | if (pinned_page) { |
852 | page_cache_release(pinned_page); | |
853 | pinned_page = NULL; | |
854 | } | |
124e68e7 | 855 | ceph_put_cap_refs(ci, got); |
83701246 YZ |
856 | if (retry_op && ret >= 0) { |
857 | int statret; | |
858 | struct page *page = NULL; | |
859 | loff_t i_size; | |
860 | if (retry_op == READ_INLINE) { | |
861 | page = __page_cache_alloc(GFP_NOFS); | |
862 | if (!page) | |
863 | return -ENOMEM; | |
864 | } | |
6a026589 | 865 | |
83701246 YZ |
866 | statret = __ceph_do_getattr(inode, page, |
867 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
868 | if (statret < 0) { | |
869 | __free_page(page); | |
870 | if (statret == -ENODATA) { | |
871 | BUG_ON(retry_op != READ_INLINE); | |
872 | goto again; | |
873 | } | |
874 | return statret; | |
875 | } | |
6a026589 | 876 | |
83701246 YZ |
877 | i_size = i_size_read(inode); |
878 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
879 | BUG_ON(ret > 0 || read > 0); |
880 | if (iocb->ki_pos < i_size && | |
881 | iocb->ki_pos < PAGE_CACHE_SIZE) { | |
83701246 YZ |
882 | loff_t end = min_t(loff_t, i_size, |
883 | iocb->ki_pos + len); | |
fcc02d2a | 884 | end = min_t(loff_t, end, PAGE_CACHE_SIZE); |
83701246 YZ |
885 | if (statret < end) |
886 | zero_user_segment(page, statret, end); | |
887 | ret = copy_page_to_iter(page, | |
888 | iocb->ki_pos & ~PAGE_MASK, | |
889 | end - iocb->ki_pos, to); | |
890 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
891 | read += ret; |
892 | } | |
893 | if (iocb->ki_pos < i_size && read < len) { | |
894 | size_t zlen = min_t(size_t, len - read, | |
895 | i_size - iocb->ki_pos); | |
896 | ret = iov_iter_zero(zlen, to); | |
897 | iocb->ki_pos += ret; | |
898 | read += ret; | |
83701246 YZ |
899 | } |
900 | __free_pages(page, 0); | |
fcc02d2a | 901 | return read; |
83701246 | 902 | } |
6a026589 SW |
903 | |
904 | /* hit EOF or hole? */ | |
83701246 | 905 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 906 | ret < len) { |
8eb4efb0 | 907 | dout("sync_read hit hole, ppos %lld < size %lld" |
908 | ", reading more\n", iocb->ki_pos, | |
909 | inode->i_size); | |
910 | ||
6a026589 | 911 | read += ret; |
6a026589 | 912 | len -= ret; |
83701246 | 913 | retry_op = 0; |
6a026589 SW |
914 | goto again; |
915 | } | |
916 | } | |
8eb4efb0 | 917 | |
6a026589 SW |
918 | if (ret >= 0) |
919 | ret += read; | |
920 | ||
124e68e7 SW |
921 | return ret; |
922 | } | |
923 | ||
924 | /* | |
925 | * Take cap references to avoid releasing caps to MDS mid-write. | |
926 | * | |
927 | * If we are synchronous, and write with an old snap context, the OSD | |
928 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
929 | * dropping our cap refs and allowing the pending snap to logically | |
930 | * complete _before_ this write occurs. | |
931 | * | |
932 | * If we are near ENOSPC, write synchronously. | |
933 | */ | |
4908b822 | 934 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
935 | { |
936 | struct file *file = iocb->ki_filp; | |
33caad32 | 937 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 938 | struct inode *inode = file_inode(file); |
124e68e7 | 939 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
940 | struct ceph_osd_client *osdc = |
941 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
3309dd04 | 942 | ssize_t count, written = 0; |
03d254ed | 943 | int err, want, got; |
3309dd04 | 944 | loff_t pos; |
124e68e7 SW |
945 | |
946 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
947 | return -EROFS; | |
948 | ||
03d254ed | 949 | mutex_lock(&inode->i_mutex); |
03d254ed | 950 | |
03d254ed | 951 | /* We can write back this queue in page reclaim */ |
de1414a6 | 952 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 953 | |
3309dd04 AV |
954 | err = generic_write_checks(iocb, from); |
955 | if (err <= 0) | |
03d254ed YZ |
956 | goto out; |
957 | ||
3309dd04 AV |
958 | pos = iocb->ki_pos; |
959 | count = iov_iter_count(from); | |
03d254ed YZ |
960 | err = file_remove_suid(file); |
961 | if (err) | |
962 | goto out; | |
963 | ||
964 | err = file_update_time(file); | |
965 | if (err) | |
966 | goto out; | |
967 | ||
28127bdd YZ |
968 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
969 | err = ceph_uninline_data(file, NULL); | |
970 | if (err < 0) | |
971 | goto out; | |
972 | } | |
973 | ||
124e68e7 | 974 | retry_snap: |
6070e0c1 | 975 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 976 | err = -ENOSPC; |
6070e0c1 YZ |
977 | goto out; |
978 | } | |
03d254ed | 979 | |
ac7f29bf | 980 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
03d254ed | 981 | inode, ceph_vinop(inode), pos, count, inode->i_size); |
7971bd92 SW |
982 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
983 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
984 | else | |
985 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 986 | got = 0; |
3738daa6 YZ |
987 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
988 | &got, NULL); | |
03d254ed | 989 | if (err < 0) |
37505d57 | 990 | goto out; |
124e68e7 | 991 | |
ac7f29bf | 992 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 993 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
994 | |
995 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
2ba48ce5 | 996 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
5dda377c | 997 | struct ceph_snap_context *snapc; |
4908b822 | 998 | struct iov_iter data; |
37505d57 | 999 | mutex_unlock(&inode->i_mutex); |
5dda377c YZ |
1000 | |
1001 | spin_lock(&ci->i_ceph_lock); | |
1002 | if (__ceph_have_pending_cap_snap(ci)) { | |
1003 | struct ceph_cap_snap *capsnap = | |
1004 | list_last_entry(&ci->i_cap_snaps, | |
1005 | struct ceph_cap_snap, | |
1006 | ci_item); | |
1007 | snapc = ceph_get_snap_context(capsnap->context); | |
1008 | } else { | |
1009 | BUG_ON(!ci->i_head_snapc); | |
1010 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1011 | } | |
1012 | spin_unlock(&ci->i_ceph_lock); | |
1013 | ||
4908b822 AV |
1014 | /* we might need to revert back to that point */ |
1015 | data = *from; | |
2ba48ce5 | 1016 | if (iocb->ki_flags & IOCB_DIRECT) |
5dda377c YZ |
1017 | written = ceph_sync_direct_write(iocb, &data, pos, |
1018 | snapc); | |
e8344e66 | 1019 | else |
5dda377c | 1020 | written = ceph_sync_write(iocb, &data, pos, snapc); |
0e5dd45c | 1021 | if (written == -EOLDSNAPC) { |
1022 | dout("aio_write %p %llx.%llx %llu~%u" | |
1023 | "got EOLDSNAPC, retrying\n", | |
1024 | inode, ceph_vinop(inode), | |
4908b822 | 1025 | pos, (unsigned)count); |
0e5dd45c | 1026 | mutex_lock(&inode->i_mutex); |
0e5dd45c | 1027 | goto retry_snap; |
1028 | } | |
4908b822 AV |
1029 | if (written > 0) |
1030 | iov_iter_advance(from, written); | |
5dda377c | 1031 | ceph_put_snap_context(snapc); |
7971bd92 | 1032 | } else { |
32d3e148 | 1033 | loff_t old_size = inode->i_size; |
b0d7c223 YZ |
1034 | /* |
1035 | * No need to acquire the i_truncate_mutex. Because | |
1036 | * the MDS revokes Fwb caps before sending truncate | |
1037 | * message to us. We can't get Fwb cap while there | |
1038 | * are pending vmtruncate. So write and vmtruncate | |
1039 | * can not run at the same time | |
1040 | */ | |
4908b822 | 1041 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1042 | if (likely(written >= 0)) |
1043 | iocb->ki_pos = pos + written; | |
32d3e148 YW |
1044 | if (inode->i_size > old_size) |
1045 | ceph_fscache_update_objectsize(inode); | |
6070e0c1 | 1046 | mutex_unlock(&inode->i_mutex); |
7971bd92 | 1047 | } |
d8de9ab6 | 1048 | |
03d254ed | 1049 | if (written >= 0) { |
fca65b4a | 1050 | int dirty; |
be655596 | 1051 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1052 | ci->i_inline_version = CEPH_INLINE_NONE; |
fca65b4a | 1053 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
be655596 | 1054 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1055 | if (dirty) |
1056 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 1057 | } |
7971bd92 | 1058 | |
124e68e7 | 1059 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1060 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1061 | ceph_cap_string(got)); |
124e68e7 | 1062 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1063 | |
03d254ed | 1064 | if (written >= 0 && |
6070e0c1 YZ |
1065 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
1066 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 1067 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 1068 | if (err < 0) |
03d254ed | 1069 | written = err; |
6070e0c1 | 1070 | } |
03d254ed | 1071 | |
2f75e9e1 SW |
1072 | goto out_unlocked; |
1073 | ||
03d254ed | 1074 | out: |
2f75e9e1 SW |
1075 | mutex_unlock(&inode->i_mutex); |
1076 | out_unlocked: | |
03d254ed | 1077 | current->backing_dev_info = NULL; |
03d254ed | 1078 | return written ? written : err; |
124e68e7 SW |
1079 | } |
1080 | ||
1081 | /* | |
1082 | * llseek. be sure to verify file size on SEEK_END. | |
1083 | */ | |
965c8e59 | 1084 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1085 | { |
1086 | struct inode *inode = file->f_mapping->host; | |
1087 | int ret; | |
1088 | ||
1089 | mutex_lock(&inode->i_mutex); | |
6a82c47a | 1090 | |
965c8e59 | 1091 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1092 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
124e68e7 SW |
1093 | if (ret < 0) { |
1094 | offset = ret; | |
1095 | goto out; | |
1096 | } | |
06222e49 JB |
1097 | } |
1098 | ||
965c8e59 | 1099 | switch (whence) { |
06222e49 | 1100 | case SEEK_END: |
124e68e7 SW |
1101 | offset += inode->i_size; |
1102 | break; | |
1103 | case SEEK_CUR: | |
1104 | /* | |
1105 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1106 | * position-querying operation. Avoid rewriting the "same" | |
1107 | * f_pos value back to the file because a concurrent read(), | |
1108 | * write() or lseek() might have altered it | |
1109 | */ | |
1110 | if (offset == 0) { | |
1111 | offset = file->f_pos; | |
1112 | goto out; | |
1113 | } | |
1114 | offset += file->f_pos; | |
1115 | break; | |
06222e49 JB |
1116 | case SEEK_DATA: |
1117 | if (offset >= inode->i_size) { | |
1118 | ret = -ENXIO; | |
1119 | goto out; | |
1120 | } | |
1121 | break; | |
1122 | case SEEK_HOLE: | |
1123 | if (offset >= inode->i_size) { | |
1124 | ret = -ENXIO; | |
1125 | goto out; | |
1126 | } | |
1127 | offset = inode->i_size; | |
1128 | break; | |
124e68e7 SW |
1129 | } |
1130 | ||
46a1c2c7 | 1131 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1132 | |
1133 | out: | |
1134 | mutex_unlock(&inode->i_mutex); | |
1135 | return offset; | |
1136 | } | |
1137 | ||
ad7a60de LW |
1138 | static inline void ceph_zero_partial_page( |
1139 | struct inode *inode, loff_t offset, unsigned size) | |
1140 | { | |
1141 | struct page *page; | |
1142 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
1143 | ||
1144 | page = find_lock_page(inode->i_mapping, index); | |
1145 | if (page) { | |
1146 | wait_on_page_writeback(page); | |
1147 | zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); | |
1148 | unlock_page(page); | |
1149 | page_cache_release(page); | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1154 | loff_t length) | |
1155 | { | |
1156 | loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); | |
1157 | if (offset < nearly) { | |
1158 | loff_t size = nearly - offset; | |
1159 | if (length < size) | |
1160 | size = length; | |
1161 | ceph_zero_partial_page(inode, offset, size); | |
1162 | offset += size; | |
1163 | length -= size; | |
1164 | } | |
1165 | if (length >= PAGE_CACHE_SIZE) { | |
1166 | loff_t size = round_down(length, PAGE_CACHE_SIZE); | |
1167 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
1168 | offset += size; | |
1169 | length -= size; | |
1170 | } | |
1171 | if (length) | |
1172 | ceph_zero_partial_page(inode, offset, length); | |
1173 | } | |
1174 | ||
1175 | static int ceph_zero_partial_object(struct inode *inode, | |
1176 | loff_t offset, loff_t *length) | |
1177 | { | |
1178 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1179 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1180 | struct ceph_osd_request *req; | |
1181 | int ret = 0; | |
1182 | loff_t zero = 0; | |
1183 | int op; | |
1184 | ||
1185 | if (!length) { | |
1186 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1187 | length = &zero; | |
1188 | } else { | |
1189 | op = CEPH_OSD_OP_ZERO; | |
1190 | } | |
1191 | ||
1192 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1193 | ceph_vino(inode), | |
1194 | offset, length, | |
715e4cd4 | 1195 | 0, 1, op, |
ad7a60de LW |
1196 | CEPH_OSD_FLAG_WRITE | |
1197 | CEPH_OSD_FLAG_ONDISK, | |
1198 | NULL, 0, 0, false); | |
1199 | if (IS_ERR(req)) { | |
1200 | ret = PTR_ERR(req); | |
1201 | goto out; | |
1202 | } | |
1203 | ||
1204 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
1205 | &inode->i_mtime); | |
1206 | ||
1207 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1208 | if (!ret) { | |
1209 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1210 | if (ret == -ENOENT) | |
1211 | ret = 0; | |
1212 | } | |
1213 | ceph_osdc_put_request(req); | |
1214 | ||
1215 | out: | |
1216 | return ret; | |
1217 | } | |
1218 | ||
1219 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1220 | { | |
1221 | int ret = 0; | |
1222 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
1223 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
1224 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
1225 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
1226 | u64 object_set_size = object_size * stripe_count; | |
1227 | u64 nearly, t; | |
1228 | ||
1229 | /* round offset up to next period boundary */ | |
1230 | nearly = offset + object_set_size - 1; | |
1231 | t = nearly; | |
1232 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1233 | |
ad7a60de LW |
1234 | while (length && offset < nearly) { |
1235 | loff_t size = length; | |
1236 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1237 | if (ret < 0) | |
1238 | return ret; | |
1239 | offset += size; | |
1240 | length -= size; | |
1241 | } | |
1242 | while (length >= object_set_size) { | |
1243 | int i; | |
1244 | loff_t pos = offset; | |
1245 | for (i = 0; i < stripe_count; ++i) { | |
1246 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1247 | if (ret < 0) | |
1248 | return ret; | |
1249 | pos += stripe_unit; | |
1250 | } | |
1251 | offset += object_set_size; | |
1252 | length -= object_set_size; | |
1253 | } | |
1254 | while (length) { | |
1255 | loff_t size = length; | |
1256 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1257 | if (ret < 0) | |
1258 | return ret; | |
1259 | offset += size; | |
1260 | length -= size; | |
1261 | } | |
1262 | return ret; | |
1263 | } | |
1264 | ||
1265 | static long ceph_fallocate(struct file *file, int mode, | |
1266 | loff_t offset, loff_t length) | |
1267 | { | |
1268 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1269 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1270 | struct ceph_inode_info *ci = ceph_inode(inode); |
1271 | struct ceph_osd_client *osdc = | |
1272 | &ceph_inode_to_client(inode)->client->osdc; | |
1273 | int want, got = 0; | |
1274 | int dirty; | |
1275 | int ret = 0; | |
1276 | loff_t endoff = 0; | |
1277 | loff_t size; | |
1278 | ||
494d77bf YZ |
1279 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1280 | return -EOPNOTSUPP; | |
1281 | ||
ad7a60de LW |
1282 | if (!S_ISREG(inode->i_mode)) |
1283 | return -EOPNOTSUPP; | |
1284 | ||
ad7a60de LW |
1285 | mutex_lock(&inode->i_mutex); |
1286 | ||
1287 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1288 | ret = -EROFS; | |
1289 | goto unlock; | |
1290 | } | |
1291 | ||
1292 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1293 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1294 | ret = -ENOSPC; | |
1295 | goto unlock; | |
1296 | } | |
1297 | ||
28127bdd YZ |
1298 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1299 | ret = ceph_uninline_data(file, NULL); | |
1300 | if (ret < 0) | |
1301 | goto unlock; | |
1302 | } | |
1303 | ||
ad7a60de LW |
1304 | size = i_size_read(inode); |
1305 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1306 | endoff = offset + length; | |
1307 | ||
1308 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1309 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1310 | else | |
1311 | want = CEPH_CAP_FILE_BUFFER; | |
1312 | ||
3738daa6 | 1313 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
ad7a60de LW |
1314 | if (ret < 0) |
1315 | goto unlock; | |
1316 | ||
1317 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1318 | if (offset < size) | |
1319 | ceph_zero_pagecache_range(inode, offset, length); | |
1320 | ret = ceph_zero_objects(inode, offset, length); | |
1321 | } else if (endoff > size) { | |
1322 | truncate_pagecache_range(inode, size, -1); | |
1323 | if (ceph_inode_set_size(inode, endoff)) | |
1324 | ceph_check_caps(ceph_inode(inode), | |
1325 | CHECK_CAPS_AUTHONLY, NULL); | |
1326 | } | |
1327 | ||
1328 | if (!ret) { | |
1329 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1330 | ci->i_inline_version = CEPH_INLINE_NONE; |
ad7a60de LW |
1331 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
1332 | spin_unlock(&ci->i_ceph_lock); | |
1333 | if (dirty) | |
1334 | __mark_inode_dirty(inode, dirty); | |
1335 | } | |
1336 | ||
1337 | ceph_put_cap_refs(ci, got); | |
1338 | unlock: | |
1339 | mutex_unlock(&inode->i_mutex); | |
1340 | return ret; | |
1341 | } | |
1342 | ||
124e68e7 SW |
1343 | const struct file_operations ceph_file_fops = { |
1344 | .open = ceph_open, | |
1345 | .release = ceph_release, | |
1346 | .llseek = ceph_llseek, | |
3644424d | 1347 | .read_iter = ceph_read_iter, |
4908b822 | 1348 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1349 | .mmap = ceph_mmap, |
1350 | .fsync = ceph_fsync, | |
40819f6f GF |
1351 | .lock = ceph_lock, |
1352 | .flock = ceph_flock, | |
124e68e7 | 1353 | .splice_read = generic_file_splice_read, |
3551dd79 | 1354 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1355 | .unlocked_ioctl = ceph_ioctl, |
1356 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1357 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1358 | }; |
1359 |