blob: 9d470397e249e8cd1b2fd763ad27701dfc0777b6 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil124e68e2009-10-06 11:31:08 -07002
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003#include <linux/module.h>
Sage Weil124e68e2009-10-06 11:31:08 -07004#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weil124e68e2009-10-06 11:31:08 -07006#include <linux/file.h>
Sage Weil5ef50c32012-07-31 11:27:36 -07007#include <linux/mount.h>
Sage Weil124e68e2009-10-06 11:31:08 -07008#include <linux/namei.h>
9#include <linux/writeback.h>
Li Wangad7a60d2013-08-15 11:51:44 +080010#include <linux/falloc.h>
Sage Weil124e68e2009-10-06 11:31:08 -070011
12#include "super.h"
13#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040014#include "cache.h"
Sage Weil124e68e2009-10-06 11:31:08 -070015
16/*
17 * Ceph file operations
18 *
19 * Implement basic open/close functionality, and implement
20 * read/write.
21 *
22 * We implement three modes of file I/O:
23 * - buffered uses the generic_file_aio_{read,write} helpers
24 *
25 * - synchronous is used when there is multi-client read/write
26 * sharing, avoids the page cache, and synchronously waits for an
27 * ack from the OSD.
28 *
29 * - direct io takes the variant of the sync path that references
30 * user pages directly.
31 *
32 * fsync() flushes and waits on dirty pages, but just queues metadata
33 * for writeback: since the MDS can recover size and mtime there is no
34 * need to wait for MDS acknowledgement.
35 */
36
Zhu, Caifengb5b98982015-10-08 15:26:15 +080037/*
38 * Calculate the length sum of direct io vectors that can
39 * be combined into one page vector.
40 */
41static size_t dio_get_pagev_size(const struct iov_iter *it)
42{
43 const struct iovec *iov = it->iov;
44 const struct iovec *iovend = iov + it->nr_segs;
45 size_t size;
46
47 size = iov->iov_len - it->iov_offset;
48 /*
49 * An iov can be page vectored when both the current tail
50 * and the next base are page aligned.
51 */
52 while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) &&
53 (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) {
54 size += iov->iov_len;
55 }
56 dout("dio_get_pagevlen len = %zu\n", size);
57 return size;
58}
59
60/*
61 * Allocate a page vector based on (@it, @nbytes).
62 * The return value is the tuple describing a page vector,
63 * that is (@pages, @page_align, @num_pages).
64 */
65static struct page **
66dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes,
67 size_t *page_align, int *num_pages)
68{
69 struct iov_iter tmp_it = *it;
70 size_t align;
71 struct page **pages;
72 int ret = 0, idx, npages;
73
74 align = (unsigned long)(it->iov->iov_base + it->iov_offset) &
75 (PAGE_SIZE - 1);
76 npages = calc_pages_for(align, nbytes);
77 pages = kmalloc(sizeof(*pages) * npages, GFP_KERNEL);
78 if (!pages) {
79 pages = vmalloc(sizeof(*pages) * npages);
80 if (!pages)
81 return ERR_PTR(-ENOMEM);
82 }
83
84 for (idx = 0; idx < npages; ) {
85 size_t start;
86 ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes,
87 npages - idx, &start);
88 if (ret < 0)
89 goto fail;
90
91 iov_iter_advance(&tmp_it, ret);
92 nbytes -= ret;
93 idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE;
94 }
95
96 BUG_ON(nbytes != 0);
97 *num_pages = npages;
98 *page_align = align;
99 dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align);
100 return pages;
101fail:
102 ceph_put_page_vector(pages, idx, false);
103 return ERR_PTR(ret);
104}
Sage Weil124e68e2009-10-06 11:31:08 -0700105
106/*
107 * Prepare an open request. Preallocate ceph_cap to avoid an
108 * inopportune ENOMEM later.
109 */
110static struct ceph_mds_request *
111prepare_open_request(struct super_block *sb, int flags, int create_mode)
112{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700113 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
114 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700115 struct ceph_mds_request *req;
116 int want_auth = USE_ANY_MDS;
117 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
118
119 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
120 want_auth = USE_AUTH_MDS;
121
122 req = ceph_mdsc_create_request(mdsc, op, want_auth);
123 if (IS_ERR(req))
124 goto out;
125 req->r_fmode = ceph_flags_to_mode(flags);
126 req->r_args.open.flags = cpu_to_le32(flags);
127 req->r_args.open.mode = cpu_to_le32(create_mode);
Sage Weil124e68e2009-10-06 11:31:08 -0700128out:
129 return req;
130}
131
132/*
133 * initialize private struct file data.
134 * if we fail, clean up by dropping fmode reference on the ceph_inode
135 */
136static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
137{
138 struct ceph_file_info *cf;
139 int ret = 0;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400140 struct ceph_inode_info *ci = ceph_inode(inode);
141 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
142 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700143
144 switch (inode->i_mode & S_IFMT) {
145 case S_IFREG:
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400146 /* First file open request creates the cookie, we want to keep
147 * this cookie around for the filetime of the inode as not to
148 * have to worry about fscache register / revoke / operation
149 * races.
150 *
151 * Also, if we know the operation is going to invalidate data
152 * (non readonly) just nuke the cache right away.
153 */
154 ceph_fscache_register_inode_cookie(mdsc->fsc, ci);
155 if ((fmode & CEPH_FILE_MODE_WR))
156 ceph_fscache_invalidate(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700157 case S_IFDIR:
158 dout("init_file %p %p 0%o (regular)\n", inode, file,
159 inode->i_mode);
Geliang Tang99ec2692016-03-13 15:26:29 +0800160 cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
Sage Weil124e68e2009-10-06 11:31:08 -0700161 if (cf == NULL) {
162 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
163 return -ENOMEM;
164 }
165 cf->fmode = fmode;
166 cf->next_offset = 2;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800167 cf->readdir_cache_idx = -1;
Sage Weil124e68e2009-10-06 11:31:08 -0700168 file->private_data = cf;
169 BUG_ON(inode->i_fop->release != ceph_release);
170 break;
171
172 case S_IFLNK:
173 dout("init_file %p %p 0%o (symlink)\n", inode, file,
174 inode->i_mode);
175 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
176 break;
177
178 default:
179 dout("init_file %p %p 0%o (special)\n", inode, file,
180 inode->i_mode);
181 /*
182 * we need to drop the open ref now, since we don't
183 * have .release set to ceph_release.
184 */
185 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
186 BUG_ON(inode->i_fop->release == ceph_release);
187
188 /* call the proper open fop */
189 ret = inode->i_fop->open(inode, file);
190 }
191 return ret;
192}
193
194/*
Sage Weil124e68e2009-10-06 11:31:08 -0700195 * If we already have the requisite capabilities, we can satisfy
196 * the open request locally (no need to request new caps from the
197 * MDS). We do, however, need to inform the MDS (asynchronously)
198 * if our wanted caps set expands.
199 */
200int ceph_open(struct inode *inode, struct file *file)
201{
202 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700203 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
204 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700205 struct ceph_mds_request *req;
206 struct ceph_file_info *cf = file->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700207 int err;
208 int flags, fmode, wanted;
209
210 if (cf) {
211 dout("open file %p is already opened\n", file);
212 return 0;
213 }
214
215 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
216 flags = file->f_flags & ~(O_CREAT|O_EXCL);
217 if (S_ISDIR(inode->i_mode))
218 flags = O_DIRECTORY; /* mds likes to know */
219
220 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
221 ceph_vinop(inode), file, flags, file->f_flags);
222 fmode = ceph_flags_to_mode(flags);
223 wanted = ceph_caps_for_mode(fmode);
224
225 /* snapped files are read-only */
226 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
227 return -EROFS;
228
229 /* trivially open snapdir */
230 if (ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weilbe655592011-11-30 09:47:09 -0800231 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700232 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800233 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700234 return ceph_init_file(inode, file, fmode);
235 }
236
237 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800238 * No need to block if we have caps on the auth MDS (for
239 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700240 * asynchronously.
241 */
Sage Weilbe655592011-11-30 09:47:09 -0800242 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800243 if (__ceph_is_any_real_caps(ci) &&
244 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700245 int mds_wanted = __ceph_caps_mds_wanted(ci);
246 int issued = __ceph_caps_issued(ci, NULL);
247
248 dout("open %p fmode %d want %s issued %s using existing\n",
249 inode, fmode, ceph_cap_string(wanted),
250 ceph_cap_string(issued));
251 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800252 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700253
254 /* adjust wanted? */
255 if ((issued & wanted) != wanted &&
256 (mds_wanted & wanted) != wanted &&
257 ceph_snap(inode) != CEPH_SNAPDIR)
258 ceph_check_caps(ci, 0, NULL);
259
260 return ceph_init_file(inode, file, fmode);
261 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
262 (ci->i_snap_caps & wanted) == wanted) {
263 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800264 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700265 return ceph_init_file(inode, file, fmode);
266 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400267
Sage Weilbe655592011-11-30 09:47:09 -0800268 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700269
270 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
271 req = prepare_open_request(inode->i_sb, flags, 0);
272 if (IS_ERR(req)) {
273 err = PTR_ERR(req);
274 goto out;
275 }
Sage Weil70b666c2011-05-27 09:24:26 -0700276 req->r_inode = inode;
277 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400278
Sage Weil124e68e2009-10-06 11:31:08 -0700279 req->r_num_caps = 1;
Jianpeng Mae36d5712015-08-18 10:25:35 +0800280 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700281 if (!err)
282 err = ceph_init_file(inode, file, req->r_fmode);
283 ceph_mdsc_put_request(req);
284 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
285out:
286 return err;
287}
288
289
290/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700291 * Do a lookup + open with a single request. If we get a non-existent
292 * file or symlink, return 1 so the VFS can retry.
Sage Weil124e68e2009-10-06 11:31:08 -0700293 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700294int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400295 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400296 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700297{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700298 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
299 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700300 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700301 struct dentry *dn;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800302 struct ceph_acls_info acls = {};
Yan, Zheng315f2402016-03-07 10:34:50 +0800303 int mask;
Sage Weil124e68e2009-10-06 11:31:08 -0700304 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700305
Al Viroa4555892014-10-21 20:11:25 -0400306 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
307 dir, dentry, dentry,
Sage Weil5ef50c32012-07-31 11:27:36 -0700308 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
309
310 if (dentry->d_name.len > NAME_MAX)
311 return -ENAMETOOLONG;
312
313 err = ceph_init_dentry(dentry);
314 if (err < 0)
315 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700316
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800317 if (flags & O_CREAT) {
318 err = ceph_pre_init_acls(dir, &mode, &acls);
319 if (err < 0)
320 return err;
321 }
322
Sage Weil124e68e2009-10-06 11:31:08 -0700323 /* do the open */
324 req = prepare_open_request(dir->i_sb, flags, mode);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800325 if (IS_ERR(req)) {
326 err = PTR_ERR(req);
327 goto out_acl;
328 }
Sage Weil124e68e2009-10-06 11:31:08 -0700329 req->r_dentry = dget(dentry);
330 req->r_num_caps = 2;
331 if (flags & O_CREAT) {
332 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
333 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800334 if (acls.pagelist) {
335 req->r_pagelist = acls.pagelist;
336 acls.pagelist = NULL;
337 }
Sage Weil124e68e2009-10-06 11:31:08 -0700338 }
Yan, Zheng315f2402016-03-07 10:34:50 +0800339
340 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
341 if (ceph_security_xattr_wanted(dir))
342 mask |= CEPH_CAP_XATTR_SHARED;
343 req->r_args.open.mask = cpu_to_le32(mask);
344
Sage Weil124e68e2009-10-06 11:31:08 -0700345 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700346 err = ceph_mdsc_do_request(mdsc,
347 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
348 req);
Yan, Zhengbf91c312015-01-19 13:23:20 +0800349 err = ceph_handle_snapdir(req, dentry, err);
Sam Lang79aec9842012-12-19 09:44:23 -1000350 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800351 goto out_req;
Sam Lang79aec9842012-12-19 09:44:23 -1000352
Jianpeng Maa43137f2015-08-18 10:23:50 +0800353 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700354 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700355
356 if (d_unhashed(dentry)) {
357 dn = ceph_finish_lookup(req, dentry, err);
358 if (IS_ERR(dn))
359 err = PTR_ERR(dn);
360 } else {
361 /* we were given a hashed negative dentry */
362 dn = NULL;
363 }
Sage Weil468640e2011-07-26 11:28:11 -0700364 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800365 goto out_req;
David Howells2b0143b2015-03-17 22:25:59 +0000366 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
Sage Weil5ef50c32012-07-31 11:27:36 -0700367 /* make vfs retry on splice, ENOENT, or symlink */
368 dout("atomic_open finish_no_open on dn %p\n", dn);
369 err = finish_no_open(file, dn);
370 } else {
371 dout("atomic_open finish_open on dn %p\n", dn);
Sam Lang6e8575f2012-12-28 09:56:46 -0800372 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
David Howells2b0143b2015-03-17 22:25:59 +0000373 ceph_init_inode_acls(d_inode(dentry), &acls);
Sam Lang6e8575f2012-12-28 09:56:46 -0800374 *opened |= FILE_CREATED;
375 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700376 err = finish_open(file, dentry, ceph_open, opened);
377 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800378out_req:
Yan, Zhengab866542014-04-01 20:20:17 +0800379 if (!req->r_err && req->r_target_inode)
380 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
Sage Weil124e68e2009-10-06 11:31:08 -0700381 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800382out_acl:
383 ceph_release_acls_info(&acls);
Sage Weil5ef50c32012-07-31 11:27:36 -0700384 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400385 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700386}
387
388int ceph_release(struct inode *inode, struct file *file)
389{
390 struct ceph_inode_info *ci = ceph_inode(inode);
391 struct ceph_file_info *cf = file->private_data;
392
393 dout("release inode %p file %p\n", inode, file);
394 ceph_put_fmode(ci, cf->fmode);
395 if (cf->last_readdir)
396 ceph_mdsc_put_request(cf->last_readdir);
397 kfree(cf->last_name);
398 kfree(cf->dir_info);
Sage Weil124e68e2009-10-06 11:31:08 -0700399 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800400
401 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700402 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700403 return 0;
404}
405
Yan, Zheng83701242014-11-14 22:36:18 +0800406enum {
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800407 HAVE_RETRIED = 1,
408 CHECK_EOF = 2,
409 READ_INLINE = 3,
Yan, Zheng83701242014-11-14 22:36:18 +0800410};
411
Sage Weil124e68e2009-10-06 11:31:08 -0700412/*
Sage Weil124e68e2009-10-06 11:31:08 -0700413 * Read a range of bytes striped over one or more objects. Iterate over
414 * objects we stripe over. (That's not atomic, but good enough for now.)
415 *
416 * If we get a short result from the OSD, check against i_size; we need to
417 * only return a short read to the caller if we hit EOF.
418 */
419static int striped_read(struct inode *inode,
420 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800421 struct page **pages, int num_pages,
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800422 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700423{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700424 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700425 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpenter688bac42013-07-23 16:48:01 +0300426 u64 pos, this_len, left;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800427 loff_t i_size;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800428 int page_align, pages_left;
429 int read, ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700430 struct page **page_pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700431 bool hit_stripe, was_short;
432
433 /*
434 * we may need to do multiple reads. not atomic, unfortunately.
435 */
436 pos = off;
437 left = len;
438 page_pos = pages;
439 pages_left = num_pages;
440 read = 0;
441
442more:
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800443 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700444 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700445 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700446 &ci->i_layout, pos, &this_len,
447 ci->i_truncate_seq,
448 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800449 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700450 if (ret == -ENOENT)
451 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700452 hit_stripe = this_len < left;
453 was_short = ret >= 0 && ret < this_len;
Dan Carpenter688bac42013-07-23 16:48:01 +0300454 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
Sage Weil124e68e2009-10-06 11:31:08 -0700455 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
456
Yan, Zheng99c88e62015-12-30 11:32:46 +0800457 i_size = i_size_read(inode);
majianpeng02ae66d2013-08-06 16:20:38 +0800458 if (ret >= 0) {
459 int didpages;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800460 if (was_short && (pos + ret < i_size)) {
461 int zlen = min(this_len - ret, i_size - pos - ret);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800462 int zoff = (off & ~PAGE_MASK) + read + ret;
majianpeng02ae66d2013-08-06 16:20:38 +0800463 dout(" zero gap %llu to %llu\n",
Yan, Zheng1487a682015-01-06 15:29:14 +0800464 pos + ret, pos + ret + zlen);
465 ceph_zero_page_vector_range(zoff, zlen, pages);
466 ret += zlen;
Sage Weil124e68e2009-10-06 11:31:08 -0700467 }
majianpeng02ae66d2013-08-06 16:20:38 +0800468
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300469 didpages = (page_align + ret) >> PAGE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700470 pos += ret;
471 read = pos - off;
472 left -= ret;
473 page_pos += didpages;
474 pages_left -= didpages;
475
majianpeng02ae66d2013-08-06 16:20:38 +0800476 /* hit stripe and need continue*/
Yan, Zheng99c88e62015-12-30 11:32:46 +0800477 if (left && hit_stripe && pos < i_size)
Sage Weil124e68e2009-10-06 11:31:08 -0700478 goto more;
479 }
480
majianpengee7289b2013-08-21 15:02:51 +0800481 if (read > 0) {
majianpeng02ae66d2013-08-06 16:20:38 +0800482 ret = read;
Sage Weilc3cd6282011-06-01 16:08:44 -0700483 /* did we bounce off eof? */
Yan, Zheng99c88e62015-12-30 11:32:46 +0800484 if (pos + left > i_size)
Yan, Zheng83701242014-11-14 22:36:18 +0800485 *checkeof = CHECK_EOF;
Sage Weil124e68e2009-10-06 11:31:08 -0700486 }
487
Sage Weil124e68e2009-10-06 11:31:08 -0700488 dout("striped_read returns %d\n", ret);
489 return ret;
490}
491
492/*
493 * Completely synchronous read and write methods. Direct from __user
494 * buffer to osd, or directly to user pages (if O_DIRECT).
495 *
496 * If the read spans object boundary, just do multiple reads.
497 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800498static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
499 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700500{
majianpeng8eb4efb2013-09-26 14:42:17 +0800501 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500502 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700503 struct page **pages;
majianpeng8eb4efb2013-09-26 14:42:17 +0800504 u64 off = iocb->ki_pos;
Henry C Changab226e22010-12-15 20:41:54 -0800505 int num_pages, ret;
Al Viro2b777c92014-04-03 22:31:22 -0400506 size_t len = iov_iter_count(i);
Sage Weil124e68e2009-10-06 11:31:08 -0700507
majianpeng8eb4efb2013-09-26 14:42:17 +0800508 dout("sync_read on file %p %llu~%u %s\n", file, off,
509 (unsigned)len,
Sage Weil124e68e2009-10-06 11:31:08 -0700510 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
Yan, Zhengd0d0db22014-07-21 10:15:48 +0800511
512 if (!len)
513 return 0;
Sage Weile98b6fe2010-11-09 12:24:53 -0800514 /*
515 * flush any page cache pages in this range. this
516 * will make concurrent normal and sync io slow,
517 * but it will at least behave sensibly when they are
518 * in sequence.
519 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800520 ret = filemap_write_and_wait_range(inode->i_mapping, off,
521 off + len);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800522 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800523 return ret;
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800524
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800525 num_pages = calc_pages_for(off, len);
526 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
527 if (IS_ERR(pages))
528 return PTR_ERR(pages);
529 ret = striped_read(inode, off, len, pages,
530 num_pages, checkeof);
531 if (ret > 0) {
532 int l, k = 0;
533 size_t left = ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700534
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800535 while (left) {
536 size_t page_off = off & ~PAGE_MASK;
537 size_t copy = min_t(size_t, left,
538 PAGE_SIZE - page_off);
539 l = copy_page_to_iter(pages[k++], page_off, copy, i);
540 off += l;
541 left -= l;
542 if (l < copy)
majianpeng8eb4efb2013-09-26 14:42:17 +0800543 break;
544 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800545 }
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800546 ceph_release_page_vector(pages, num_pages);
majianpeng8eb4efb2013-09-26 14:42:17 +0800547
548 if (off > iocb->ki_pos) {
549 ret = off - iocb->ki_pos;
550 iocb->ki_pos = off;
551 }
552
Sage Weil124e68e2009-10-06 11:31:08 -0700553 dout("sync_read result %d\n", ret);
554 return ret;
555}
556
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800557struct ceph_aio_request {
558 struct kiocb *iocb;
559 size_t total_len;
560 int write;
561 int error;
562 struct list_head osd_reqs;
563 unsigned num_reqs;
564 atomic_t pending_reqs;
Yan, Zheng5be03892015-12-24 08:44:20 +0800565 struct timespec mtime;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800566 struct ceph_cap_flush *prealloc_cf;
567};
568
Yan, Zheng5be03892015-12-24 08:44:20 +0800569struct ceph_aio_work {
570 struct work_struct work;
571 struct ceph_osd_request *req;
572};
573
574static void ceph_aio_retry_work(struct work_struct *work);
575
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800576static void ceph_aio_complete(struct inode *inode,
577 struct ceph_aio_request *aio_req)
578{
579 struct ceph_inode_info *ci = ceph_inode(inode);
580 int ret;
581
582 if (!atomic_dec_and_test(&aio_req->pending_reqs))
583 return;
584
585 ret = aio_req->error;
586 if (!ret)
587 ret = aio_req->total_len;
588
589 dout("ceph_aio_complete %p rc %d\n", inode, ret);
590
591 if (ret >= 0 && aio_req->write) {
592 int dirty;
593
594 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
595 if (endoff > i_size_read(inode)) {
596 if (ceph_inode_set_size(inode, endoff))
597 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
598 }
599
600 spin_lock(&ci->i_ceph_lock);
601 ci->i_inline_version = CEPH_INLINE_NONE;
602 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
603 &aio_req->prealloc_cf);
604 spin_unlock(&ci->i_ceph_lock);
605 if (dirty)
606 __mark_inode_dirty(inode, dirty);
607
608 }
609
610 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
611 CEPH_CAP_FILE_RD));
612
613 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
614
615 ceph_free_cap_flush(aio_req->prealloc_cf);
616 kfree(aio_req);
617}
618
619static void ceph_aio_complete_req(struct ceph_osd_request *req,
620 struct ceph_msg *msg)
621{
622 int rc = req->r_result;
623 struct inode *inode = req->r_inode;
624 struct ceph_aio_request *aio_req = req->r_priv;
625 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
626 int num_pages = calc_pages_for((u64)osd_data->alignment,
627 osd_data->length);
628
629 dout("ceph_aio_complete_req %p rc %d bytes %llu\n",
630 inode, rc, osd_data->length);
631
632 if (rc == -EOLDSNAPC) {
Yan, Zheng5be03892015-12-24 08:44:20 +0800633 struct ceph_aio_work *aio_work;
634 BUG_ON(!aio_req->write);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800635
Yan, Zheng5be03892015-12-24 08:44:20 +0800636 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
637 if (aio_work) {
638 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
639 aio_work->req = req;
640 queue_work(ceph_inode_to_client(inode)->wb_wq,
641 &aio_work->work);
642 return;
643 }
644 rc = -ENOMEM;
645 } else if (!aio_req->write) {
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800646 if (rc == -ENOENT)
647 rc = 0;
648 if (rc >= 0 && osd_data->length > rc) {
649 int zoff = osd_data->alignment + rc;
650 int zlen = osd_data->length - rc;
651 /*
652 * If read is satisfied by single OSD request,
653 * it can pass EOF. Otherwise read is within
654 * i_size.
655 */
656 if (aio_req->num_reqs == 1) {
657 loff_t i_size = i_size_read(inode);
658 loff_t endoff = aio_req->iocb->ki_pos + rc;
659 if (endoff < i_size)
660 zlen = min_t(size_t, zlen,
661 i_size - endoff);
662 aio_req->total_len = rc + zlen;
663 }
664
665 if (zlen > 0)
666 ceph_zero_page_vector_range(zoff, zlen,
667 osd_data->pages);
668 }
669 }
670
671 ceph_put_page_vector(osd_data->pages, num_pages, false);
672 ceph_osdc_put_request(req);
673
674 if (rc < 0)
675 cmpxchg(&aio_req->error, 0, rc);
676
677 ceph_aio_complete(inode, aio_req);
678 return;
679}
680
Yan, Zheng5be03892015-12-24 08:44:20 +0800681static void ceph_aio_retry_work(struct work_struct *work)
682{
683 struct ceph_aio_work *aio_work =
684 container_of(work, struct ceph_aio_work, work);
685 struct ceph_osd_request *orig_req = aio_work->req;
686 struct ceph_aio_request *aio_req = orig_req->r_priv;
687 struct inode *inode = orig_req->r_inode;
688 struct ceph_inode_info *ci = ceph_inode(inode);
689 struct ceph_snap_context *snapc;
690 struct ceph_osd_request *req;
691 int ret;
692
693 spin_lock(&ci->i_ceph_lock);
694 if (__ceph_have_pending_cap_snap(ci)) {
695 struct ceph_cap_snap *capsnap =
696 list_last_entry(&ci->i_cap_snaps,
697 struct ceph_cap_snap,
698 ci_item);
699 snapc = ceph_get_snap_context(capsnap->context);
700 } else {
701 BUG_ON(!ci->i_head_snapc);
702 snapc = ceph_get_snap_context(ci->i_head_snapc);
703 }
704 spin_unlock(&ci->i_ceph_lock);
705
706 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2,
707 false, GFP_NOFS);
Dan Carpenter1418bf02016-01-26 12:24:44 +0300708 if (!req) {
709 ret = -ENOMEM;
Yan, Zheng5be03892015-12-24 08:44:20 +0800710 req = orig_req;
711 goto out;
712 }
713
714 req->r_flags = CEPH_OSD_FLAG_ORDERSNAP |
715 CEPH_OSD_FLAG_ONDISK |
716 CEPH_OSD_FLAG_WRITE;
717 req->r_base_oloc = orig_req->r_base_oloc;
Ilya Dryomovd30291b2016-04-29 19:54:20 +0200718 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
Yan, Zheng5be03892015-12-24 08:44:20 +0800719
Ilya Dryomov13d1ad12016-04-27 14:15:51 +0200720 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
721 if (ret) {
722 ceph_osdc_put_request(req);
723 req = orig_req;
724 goto out;
725 }
726
Yan, Zheng5be03892015-12-24 08:44:20 +0800727 req->r_ops[0] = orig_req->r_ops[0];
728 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
729
730 ceph_osdc_build_request(req, req->r_ops[0].extent.offset,
731 snapc, CEPH_NOSNAP, &aio_req->mtime);
732
Yan, Zheng5be03892015-12-24 08:44:20 +0800733 ceph_osdc_put_request(orig_req);
734
735 req->r_callback = ceph_aio_complete_req;
736 req->r_inode = inode;
737 req->r_priv = aio_req;
738
739 ret = ceph_osdc_start_request(req->r_osdc, req, false);
740out:
741 if (ret < 0) {
Yan, Zheng5be03892015-12-24 08:44:20 +0800742 req->r_result = ret;
743 ceph_aio_complete_req(req, NULL);
744 }
745
Yan, Zhengdb6aed72016-01-26 23:05:37 +0800746 ceph_put_snap_context(snapc);
Yan, Zheng5be03892015-12-24 08:44:20 +0800747 kfree(aio_work);
748}
749
Sage Weil124e68e2009-10-06 11:31:08 -0700750/*
Alex Elder26be8802013-04-15 11:20:42 -0500751 * Write commit request unsafe callback, called to tell us when a
752 * request is unsafe (that is, in flight--has been handed to the
753 * messenger to send to its target osd). It is called again when
754 * we've received a response message indicating the request is
755 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
756 * is completed early (and unsuccessfully) due to a timeout or
757 * interrupt.
758 *
759 * This is used if we requested both an ACK and ONDISK commit reply
760 * from the OSD.
Sage Weil124e68e2009-10-06 11:31:08 -0700761 */
Alex Elder26be8802013-04-15 11:20:42 -0500762static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700763{
764 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
765
Alex Elder26be8802013-04-15 11:20:42 -0500766 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
767 unsafe ? "un" : "");
768 if (unsafe) {
769 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
770 spin_lock(&ci->i_unsafe_lock);
771 list_add_tail(&req->r_unsafe_item,
772 &ci->i_unsafe_writes);
773 spin_unlock(&ci->i_unsafe_lock);
774 } else {
775 spin_lock(&ci->i_unsafe_lock);
776 list_del_init(&req->r_unsafe_item);
777 spin_unlock(&ci->i_unsafe_lock);
778 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
779 }
Sage Weil124e68e2009-10-06 11:31:08 -0700780}
781
majianpenge8344e62013-09-12 13:54:26 +0800782
majianpenge8344e62013-09-12 13:54:26 +0800783static ssize_t
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800784ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
785 struct ceph_snap_context *snapc,
786 struct ceph_cap_flush **pcf)
Sage Weil124e68e2009-10-06 11:31:08 -0700787{
majianpenge8344e62013-09-12 13:54:26 +0800788 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500789 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700790 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700791 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500792 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700793 struct ceph_osd_request *req;
794 struct page **pages;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800795 struct ceph_aio_request *aio_req = NULL;
796 int num_pages = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700797 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700798 int ret;
Deepa Dinamani8bbd4712016-02-02 22:07:48 -0800799 struct timespec mtime = current_fs_time(inode->i_sb);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800800 size_t count = iov_iter_count(iter);
801 loff_t pos = iocb->ki_pos;
802 bool write = iov_iter_rw(iter) == WRITE;
Sage Weil124e68e2009-10-06 11:31:08 -0700803
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800804 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700805 return -EROFS;
806
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800807 dout("sync_direct_read_write (%s) on file %p %lld~%u\n",
808 (write ? "write" : "read"), file, pos, (unsigned)count);
Sage Weil124e68e2009-10-06 11:31:08 -0700809
majianpenge8344e62013-09-12 13:54:26 +0800810 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800811 if (ret < 0)
812 return ret;
813
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800814 if (write) {
815 ret = invalidate_inode_pages2_range(inode->i_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300816 pos >> PAGE_SHIFT,
817 (pos + count) >> PAGE_SHIFT);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800818 if (ret < 0)
819 dout("invalidate_inode_pages2_range returned %d\n", ret);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800820
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800821 flags = CEPH_OSD_FLAG_ORDERSNAP |
822 CEPH_OSD_FLAG_ONDISK |
823 CEPH_OSD_FLAG_WRITE;
824 } else {
825 flags = CEPH_OSD_FLAG_READ;
826 }
Sage Weil124e68e2009-10-06 11:31:08 -0700827
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800828 while (iov_iter_count(iter) > 0) {
829 u64 size = dio_get_pagev_size(iter);
830 size_t start = 0;
831 ssize_t len;
majianpenge8344e62013-09-12 13:54:26 +0800832
majianpenge8344e62013-09-12 13:54:26 +0800833 vino = ceph_vino(inode);
834 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800835 vino, pos, &size, 0,
836 /*include a 'startsync' command*/
837 write ? 2 : 1,
838 write ? CEPH_OSD_OP_WRITE :
839 CEPH_OSD_OP_READ,
840 flags, snapc,
majianpenge8344e62013-09-12 13:54:26 +0800841 ci->i_truncate_seq,
842 ci->i_truncate_size,
843 false);
844 if (IS_ERR(req)) {
845 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400846 break;
majianpenge8344e62013-09-12 13:54:26 +0800847 }
848
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800849 len = size;
850 pages = dio_get_pages_alloc(iter, len, &start, &num_pages);
Zhu, Caifengb5b98982015-10-08 15:26:15 +0800851 if (IS_ERR(pages)) {
Al Viro64c31312014-04-03 22:58:25 -0400852 ceph_osdc_put_request(req);
Zhu, Caifengb5b98982015-10-08 15:26:15 +0800853 ret = PTR_ERR(pages);
Al Viro64c31312014-04-03 22:58:25 -0400854 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700855 }
856
857 /*
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800858 * To simplify error handling, allow AIO when IO within i_size
859 * or IO can be satisfied by single OSD request.
Sage Weil124e68e2009-10-06 11:31:08 -0700860 */
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800861 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
862 (len == count || pos + count <= i_size_read(inode))) {
863 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
864 if (aio_req) {
865 aio_req->iocb = iocb;
866 aio_req->write = write;
867 INIT_LIST_HEAD(&aio_req->osd_reqs);
868 if (write) {
Yan, Zheng5be03892015-12-24 08:44:20 +0800869 aio_req->mtime = mtime;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800870 swap(aio_req->prealloc_cf, *pcf);
871 }
872 }
873 /* ignore error */
874 }
majianpenge8344e62013-09-12 13:54:26 +0800875
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800876 if (write) {
877 /*
878 * throw out any page cache pages in this range. this
879 * may block.
880 */
881 truncate_inode_pages_range(inode->i_mapping, pos,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300882 (pos+len) | (PAGE_SIZE - 1));
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800883
884 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
885 }
886
887
888 osd_req_op_extent_osd_data_pages(req, 0, pages, len, start,
889 false, false);
890
majianpenge8344e62013-09-12 13:54:26 +0800891 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
892
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800893 if (aio_req) {
894 aio_req->total_len += len;
895 aio_req->num_reqs++;
896 atomic_inc(&aio_req->pending_reqs);
897
898 req->r_callback = ceph_aio_complete_req;
899 req->r_inode = inode;
900 req->r_priv = aio_req;
901 list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs);
902
903 pos += len;
904 iov_iter_advance(iter, len);
905 continue;
906 }
907
908 ret = ceph_osdc_start_request(req->r_osdc, req, false);
majianpenge8344e62013-09-12 13:54:26 +0800909 if (!ret)
910 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
911
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800912 size = i_size_read(inode);
913 if (!write) {
914 if (ret == -ENOENT)
915 ret = 0;
916 if (ret >= 0 && ret < len && pos + ret < size) {
917 int zlen = min_t(size_t, len - ret,
918 size - pos - ret);
919 ceph_zero_page_vector_range(start + ret, zlen,
920 pages);
921 ret += zlen;
922 }
923 if (ret >= 0)
924 len = ret;
925 }
926
majianpenge8344e62013-09-12 13:54:26 +0800927 ceph_put_page_vector(pages, num_pages, false);
928
majianpenge8344e62013-09-12 13:54:26 +0800929 ceph_osdc_put_request(req);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800930 if (ret < 0)
majianpenge8344e62013-09-12 13:54:26 +0800931 break;
Al Viro64c31312014-04-03 22:58:25 -0400932
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800933 pos += len;
934 iov_iter_advance(iter, len);
935
936 if (!write && pos >= size)
937 break;
938
939 if (write && pos > size) {
940 if (ceph_inode_set_size(inode, pos))
Al Viro64c31312014-04-03 22:58:25 -0400941 ceph_check_caps(ceph_inode(inode),
942 CHECK_CAPS_AUTHONLY,
943 NULL);
944 }
majianpenge8344e62013-09-12 13:54:26 +0800945 }
946
Yan, Zhengc8fe9b12015-12-23 21:23:38 +0800947 if (aio_req) {
948 if (aio_req->num_reqs == 0) {
949 kfree(aio_req);
950 return ret;
951 }
952
953 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
954 CEPH_CAP_FILE_RD);
955
956 while (!list_empty(&aio_req->osd_reqs)) {
957 req = list_first_entry(&aio_req->osd_reqs,
958 struct ceph_osd_request,
959 r_unsafe_item);
960 list_del_init(&req->r_unsafe_item);
961 if (ret >= 0)
962 ret = ceph_osdc_start_request(req->r_osdc,
963 req, false);
964 if (ret < 0) {
965 req->r_result = ret;
966 ceph_aio_complete_req(req, NULL);
967 }
968 }
969 return -EIOCBQUEUED;
970 }
971
972 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
973 ret = pos - iocb->ki_pos;
majianpenge8344e62013-09-12 13:54:26 +0800974 iocb->ki_pos = pos;
majianpenge8344e62013-09-12 13:54:26 +0800975 }
976 return ret;
977}
978
majianpenge8344e62013-09-12 13:54:26 +0800979/*
980 * Synchronous write, straight from __user pointer or user pages.
981 *
982 * If write spans object boundary, just do multiple writes. (For a
983 * correct atomic write, we should e.g. take write locks on all
984 * objects, rollback on failure, etc.)
985 */
Yan, Zheng06fee302014-07-28 14:33:46 +0800986static ssize_t
Yan, Zheng5dda377c2015-04-30 14:40:54 +0800987ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
988 struct ceph_snap_context *snapc)
majianpenge8344e62013-09-12 13:54:26 +0800989{
990 struct file *file = iocb->ki_filp;
991 struct inode *inode = file_inode(file);
992 struct ceph_inode_info *ci = ceph_inode(inode);
993 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
majianpenge8344e62013-09-12 13:54:26 +0800994 struct ceph_vino vino;
995 struct ceph_osd_request *req;
996 struct page **pages;
997 u64 len;
998 int num_pages;
999 int written = 0;
1000 int flags;
1001 int check_caps = 0;
1002 int ret;
Deepa Dinamani8bbd4712016-02-02 22:07:48 -08001003 struct timespec mtime = current_fs_time(inode->i_sb);
Al Viro4908b822014-04-03 23:09:01 -04001004 size_t count = iov_iter_count(from);
majianpenge8344e62013-09-12 13:54:26 +08001005
1006 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1007 return -EROFS;
1008
1009 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
1010
1011 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
1012 if (ret < 0)
1013 return ret;
1014
1015 ret = invalidate_inode_pages2_range(inode->i_mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001016 pos >> PAGE_SHIFT,
1017 (pos + count) >> PAGE_SHIFT);
majianpenge8344e62013-09-12 13:54:26 +08001018 if (ret < 0)
1019 dout("invalidate_inode_pages2_range returned %d\n", ret);
1020
1021 flags = CEPH_OSD_FLAG_ORDERSNAP |
1022 CEPH_OSD_FLAG_ONDISK |
1023 CEPH_OSD_FLAG_WRITE |
1024 CEPH_OSD_FLAG_ACK;
1025
Al Viro4908b822014-04-03 23:09:01 -04001026 while ((len = iov_iter_count(from)) > 0) {
majianpenge8344e62013-09-12 13:54:26 +08001027 size_t left;
1028 int n;
1029
majianpenge8344e62013-09-12 13:54:26 +08001030 vino = ceph_vino(inode);
1031 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Yan, Zheng715e4cd2014-11-13 14:40:37 +08001032 vino, pos, &len, 0, 1,
majianpenge8344e62013-09-12 13:54:26 +08001033 CEPH_OSD_OP_WRITE, flags, snapc,
1034 ci->i_truncate_seq,
1035 ci->i_truncate_size,
1036 false);
1037 if (IS_ERR(req)) {
1038 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -04001039 break;
majianpenge8344e62013-09-12 13:54:26 +08001040 }
1041
1042 /*
1043 * write from beginning of first page,
1044 * regardless of io alignment
1045 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001046 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
majianpenge8344e62013-09-12 13:54:26 +08001047
Yan, Zheng687265e2015-06-13 17:27:05 +08001048 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
Sage Weil124e68e2009-10-06 11:31:08 -07001049 if (IS_ERR(pages)) {
1050 ret = PTR_ERR(pages);
1051 goto out;
1052 }
majianpenge8344e62013-09-12 13:54:26 +08001053
1054 left = len;
1055 for (n = 0; n < num_pages; n++) {
Ilya Dryomov125d7252014-01-28 19:16:18 +02001056 size_t plen = min_t(size_t, left, PAGE_SIZE);
Al Viro4908b822014-04-03 23:09:01 -04001057 ret = copy_page_from_iter(pages[n], 0, plen, from);
majianpenge8344e62013-09-12 13:54:26 +08001058 if (ret != plen) {
1059 ret = -EFAULT;
1060 break;
1061 }
1062 left -= ret;
majianpenge8344e62013-09-12 13:54:26 +08001063 }
1064
Sage Weil124e68e2009-10-06 11:31:08 -07001065 if (ret < 0) {
1066 ceph_release_page_vector(pages, num_pages);
1067 goto out;
1068 }
1069
majianpenge8344e62013-09-12 13:54:26 +08001070 /* get a second commit callback */
1071 req->r_unsafe_callback = ceph_sync_write_unsafe;
1072 req->r_inode = inode;
Sage Weil124e68e2009-10-06 11:31:08 -07001073
majianpenge8344e62013-09-12 13:54:26 +08001074 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1075 false, true);
Alex Elder02ee07d2013-03-14 14:09:06 -05001076
majianpenge8344e62013-09-12 13:54:26 +08001077 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
1078 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
Sage Weil124e68e2009-10-06 11:31:08 -07001079
majianpenge8344e62013-09-12 13:54:26 +08001080 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1081 if (!ret)
1082 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -07001083
1084out:
majianpenge8344e62013-09-12 13:54:26 +08001085 ceph_osdc_put_request(req);
1086 if (ret == 0) {
1087 pos += len;
1088 written += len;
Sage Weil124e68e2009-10-06 11:31:08 -07001089
majianpenge8344e62013-09-12 13:54:26 +08001090 if (pos > i_size_read(inode)) {
1091 check_caps = ceph_inode_set_size(inode, pos);
1092 if (check_caps)
1093 ceph_check_caps(ceph_inode(inode),
1094 CHECK_CAPS_AUTHONLY,
1095 NULL);
1096 }
1097 } else
1098 break;
1099 }
1100
1101 if (ret != -EOLDSNAPC && written > 0) {
Sage Weil124e68e2009-10-06 11:31:08 -07001102 ret = written;
majianpenge8344e62013-09-12 13:54:26 +08001103 iocb->ki_pos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -07001104 }
1105 return ret;
1106}
1107
1108/*
1109 * Wrap generic_file_aio_read with checks for cap bits on the inode.
1110 * Atomically grab references, so that those bits are not released
1111 * back to the MDS mid-read.
1112 *
1113 * Hmm, the sync read case isn't actually async... should it be?
1114 */
Al Viro36444242014-04-02 20:28:01 -04001115static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
Sage Weil124e68e2009-10-06 11:31:08 -07001116{
1117 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -07001118 struct ceph_file_info *fi = filp->private_data;
Christoph Hellwig66ee59a2015-02-11 19:56:46 +01001119 size_t len = iov_iter_count(to);
Al Viro496ad9a2013-01-23 17:07:38 -05001120 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -07001121 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001122 struct page *pinned_page = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -07001123 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -07001124 int want, got = 0;
Yan, Zheng83701242014-11-14 22:36:18 +08001125 int retry_op = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -07001126
Sage Weil6a026582010-02-09 14:04:02 -08001127again:
majianpeng8eb4efb2013-09-26 14:42:17 +08001128 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1129 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1130
Sage Weil29625072010-05-27 10:40:43 -07001131 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1132 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1133 else
1134 want = CEPH_CAP_FILE_CACHE;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001135 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
Sage Weil124e68e2009-10-06 11:31:08 -07001136 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +08001137 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -07001138
Sage Weil29625072010-05-27 10:40:43 -07001139 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -04001140 (iocb->ki_flags & IOCB_DIRECT) ||
majianpeng8eb4efb2013-09-26 14:42:17 +08001141 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -07001142
majianpeng8eb4efb2013-09-26 14:42:17 +08001143 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1144 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1145 ceph_cap_string(got));
1146
Yan, Zheng83701242014-11-14 22:36:18 +08001147 if (ci->i_inline_version == CEPH_INLINE_NONE) {
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001148 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1149 ret = ceph_direct_read_write(iocb, to,
1150 NULL, NULL);
1151 if (ret >= 0 && ret < len)
1152 retry_op = CHECK_EOF;
1153 } else {
1154 ret = ceph_sync_read(iocb, to, &retry_op);
1155 }
Yan, Zheng83701242014-11-14 22:36:18 +08001156 } else {
1157 retry_op = READ_INLINE;
1158 }
majianpeng8eb4efb2013-09-26 14:42:17 +08001159 } else {
majianpeng8eb4efb2013-09-26 14:42:17 +08001160 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
Al Viro36444242014-04-02 20:28:01 -04001161 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
majianpeng8eb4efb2013-09-26 14:42:17 +08001162 ceph_cap_string(got));
1163
Al Viro36444242014-04-02 20:28:01 -04001164 ret = generic_file_read_iter(iocb, to);
majianpeng8eb4efb2013-09-26 14:42:17 +08001165 }
Sage Weil124e68e2009-10-06 11:31:08 -07001166 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1167 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001168 if (pinned_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001169 put_page(pinned_page);
Yan, Zheng3738daa2014-11-14 22:10:07 +08001170 pinned_page = NULL;
1171 }
Sage Weil124e68e2009-10-06 11:31:08 -07001172 ceph_put_cap_refs(ci, got);
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001173 if (retry_op > HAVE_RETRIED && ret >= 0) {
Yan, Zheng83701242014-11-14 22:36:18 +08001174 int statret;
1175 struct page *page = NULL;
1176 loff_t i_size;
1177 if (retry_op == READ_INLINE) {
Yan, Zheng687265e2015-06-13 17:27:05 +08001178 page = __page_cache_alloc(GFP_KERNEL);
Yan, Zheng83701242014-11-14 22:36:18 +08001179 if (!page)
1180 return -ENOMEM;
1181 }
Sage Weil6a026582010-02-09 14:04:02 -08001182
Yan, Zheng83701242014-11-14 22:36:18 +08001183 statret = __ceph_do_getattr(inode, page,
1184 CEPH_STAT_CAP_INLINE_DATA, !!page);
1185 if (statret < 0) {
1186 __free_page(page);
1187 if (statret == -ENODATA) {
1188 BUG_ON(retry_op != READ_INLINE);
1189 goto again;
1190 }
1191 return statret;
1192 }
1193
1194 i_size = i_size_read(inode);
1195 if (retry_op == READ_INLINE) {
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001196 BUG_ON(ret > 0 || read > 0);
1197 if (iocb->ki_pos < i_size &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001198 iocb->ki_pos < PAGE_SIZE) {
Yan, Zheng83701242014-11-14 22:36:18 +08001199 loff_t end = min_t(loff_t, i_size,
1200 iocb->ki_pos + len);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001201 end = min_t(loff_t, end, PAGE_SIZE);
Yan, Zheng83701242014-11-14 22:36:18 +08001202 if (statret < end)
1203 zero_user_segment(page, statret, end);
1204 ret = copy_page_to_iter(page,
1205 iocb->ki_pos & ~PAGE_MASK,
1206 end - iocb->ki_pos, to);
1207 iocb->ki_pos += ret;
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001208 read += ret;
1209 }
1210 if (iocb->ki_pos < i_size && read < len) {
1211 size_t zlen = min_t(size_t, len - read,
1212 i_size - iocb->ki_pos);
1213 ret = iov_iter_zero(zlen, to);
1214 iocb->ki_pos += ret;
1215 read += ret;
Yan, Zheng83701242014-11-14 22:36:18 +08001216 }
1217 __free_pages(page, 0);
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001218 return read;
Yan, Zheng83701242014-11-14 22:36:18 +08001219 }
Sage Weil6a026582010-02-09 14:04:02 -08001220
1221 /* hit EOF or hole? */
Yan, Zheng83701242014-11-14 22:36:18 +08001222 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
Yan, Zhengfcc02d22015-01-10 11:43:12 +08001223 ret < len) {
majianpeng8eb4efb2013-09-26 14:42:17 +08001224 dout("sync_read hit hole, ppos %lld < size %lld"
Yan, Zheng99c88e62015-12-30 11:32:46 +08001225 ", reading more\n", iocb->ki_pos, i_size);
majianpeng8eb4efb2013-09-26 14:42:17 +08001226
Sage Weil6a026582010-02-09 14:04:02 -08001227 read += ret;
Sage Weil6a026582010-02-09 14:04:02 -08001228 len -= ret;
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001229 retry_op = HAVE_RETRIED;
Sage Weil6a026582010-02-09 14:04:02 -08001230 goto again;
1231 }
1232 }
majianpeng8eb4efb2013-09-26 14:42:17 +08001233
Sage Weil6a026582010-02-09 14:04:02 -08001234 if (ret >= 0)
1235 ret += read;
1236
Sage Weil124e68e2009-10-06 11:31:08 -07001237 return ret;
1238}
1239
1240/*
1241 * Take cap references to avoid releasing caps to MDS mid-write.
1242 *
1243 * If we are synchronous, and write with an old snap context, the OSD
1244 * may return EOLDSNAPC. In that case, retry the write.. _after_
1245 * dropping our cap refs and allowing the pending snap to logically
1246 * complete _before_ this write occurs.
1247 *
1248 * If we are near ENOSPC, write synchronously.
1249 */
Al Viro4908b822014-04-03 23:09:01 -04001250static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
Sage Weil124e68e2009-10-06 11:31:08 -07001251{
1252 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -07001253 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -05001254 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -07001255 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001256 struct ceph_osd_client *osdc =
1257 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001258 struct ceph_cap_flush *prealloc_cf;
Al Viro3309dd02015-04-09 12:55:47 -04001259 ssize_t count, written = 0;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001260 int err, want, got;
Al Viro3309dd02015-04-09 12:55:47 -04001261 loff_t pos;
Sage Weil124e68e2009-10-06 11:31:08 -07001262
1263 if (ceph_snap(inode) != CEPH_NOSNAP)
1264 return -EROFS;
1265
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001266 prealloc_cf = ceph_alloc_cap_flush();
1267 if (!prealloc_cf)
1268 return -ENOMEM;
1269
Al Viro59551022016-01-22 15:40:57 -05001270 inode_lock(inode);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001271
Yan, Zheng03d254e2013-04-12 16:11:13 +08001272 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01001273 current->backing_dev_info = inode_to_bdi(inode);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001274
Yan, Zheng55b0b312015-09-07 11:35:01 +08001275 if (iocb->ki_flags & IOCB_APPEND) {
1276 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1277 if (err < 0)
1278 goto out;
1279 }
1280
Al Viro3309dd02015-04-09 12:55:47 -04001281 err = generic_write_checks(iocb, from);
1282 if (err <= 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +08001283 goto out;
1284
Al Viro3309dd02015-04-09 12:55:47 -04001285 pos = iocb->ki_pos;
1286 count = iov_iter_count(from);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02001287 err = file_remove_privs(file);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001288 if (err)
1289 goto out;
1290
1291 err = file_update_time(file);
1292 if (err)
1293 goto out;
1294
Yan, Zheng28127bd2014-11-14 22:38:29 +08001295 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1296 err = ceph_uninline_data(file, NULL);
1297 if (err < 0)
1298 goto out;
1299 }
1300
Sage Weil124e68e2009-10-06 11:31:08 -07001301retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001302 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +08001303 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001304 goto out;
1305 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001306
Randy Dunlapac7f29b2013-04-19 14:20:07 -07001307 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng99c88e62015-12-30 11:32:46 +08001308 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
Sage Weil7971bd92013-05-01 21:15:58 -07001309 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1310 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1311 else
1312 want = CEPH_CAP_FILE_BUFFER;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001313 got = 0;
Yan, Zheng3738daa2014-11-14 22:10:07 +08001314 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1315 &got, NULL);
Yan, Zheng03d254e2013-04-12 16:11:13 +08001316 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +08001317 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -07001318
Randy Dunlapac7f29b2013-04-19 14:20:07 -07001319 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +08001320 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -07001321
1322 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Al Viro2ba48ce2015-04-09 13:52:01 -04001323 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001324 struct ceph_snap_context *snapc;
Al Viro4908b822014-04-03 23:09:01 -04001325 struct iov_iter data;
Al Viro59551022016-01-22 15:40:57 -05001326 inode_unlock(inode);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001327
1328 spin_lock(&ci->i_ceph_lock);
1329 if (__ceph_have_pending_cap_snap(ci)) {
1330 struct ceph_cap_snap *capsnap =
1331 list_last_entry(&ci->i_cap_snaps,
1332 struct ceph_cap_snap,
1333 ci_item);
1334 snapc = ceph_get_snap_context(capsnap->context);
1335 } else {
1336 BUG_ON(!ci->i_head_snapc);
1337 snapc = ceph_get_snap_context(ci->i_head_snapc);
1338 }
1339 spin_unlock(&ci->i_ceph_lock);
1340
Al Viro4908b822014-04-03 23:09:01 -04001341 /* we might need to revert back to that point */
1342 data = *from;
Al Viro2ba48ce2015-04-09 13:52:01 -04001343 if (iocb->ki_flags & IOCB_DIRECT)
Yan, Zhengc8fe9b12015-12-23 21:23:38 +08001344 written = ceph_direct_read_write(iocb, &data, snapc,
1345 &prealloc_cf);
majianpenge8344e62013-09-12 13:54:26 +08001346 else
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001347 written = ceph_sync_write(iocb, &data, pos, snapc);
majianpeng0e5dd452013-08-08 15:32:19 +08001348 if (written == -EOLDSNAPC) {
1349 dout("aio_write %p %llx.%llx %llu~%u"
1350 "got EOLDSNAPC, retrying\n",
1351 inode, ceph_vinop(inode),
Al Viro4908b822014-04-03 23:09:01 -04001352 pos, (unsigned)count);
Al Viro59551022016-01-22 15:40:57 -05001353 inode_lock(inode);
majianpeng0e5dd452013-08-08 15:32:19 +08001354 goto retry_snap;
1355 }
Al Viro4908b822014-04-03 23:09:01 -04001356 if (written > 0)
1357 iov_iter_advance(from, written);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08001358 ceph_put_snap_context(snapc);
Sage Weil7971bd92013-05-01 21:15:58 -07001359 } else {
Yan, Zheng99c88e62015-12-30 11:32:46 +08001360 loff_t old_size = i_size_read(inode);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001361 /*
1362 * No need to acquire the i_truncate_mutex. Because
1363 * the MDS revokes Fwb caps before sending truncate
1364 * message to us. We can't get Fwb cap while there
1365 * are pending vmtruncate. So write and vmtruncate
1366 * can not run at the same time
1367 */
Al Viro4908b822014-04-03 23:09:01 -04001368 written = generic_perform_write(file, from, pos);
Al Viroaec605f42014-02-11 22:28:43 -05001369 if (likely(written >= 0))
1370 iocb->ki_pos = pos + written;
Yan, Zheng99c88e62015-12-30 11:32:46 +08001371 if (i_size_read(inode) > old_size)
Yunchuan Wen32d3e142013-12-26 06:29:27 -08001372 ceph_fscache_update_objectsize(inode);
Al Viro59551022016-01-22 15:40:57 -05001373 inode_unlock(inode);
Sage Weil124e68e2009-10-06 11:31:08 -07001374 }
Sage Weild8de9ab2011-07-26 11:27:34 -07001375
Yan, Zheng03d254e2013-04-12 16:11:13 +08001376 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -07001377 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -08001378 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001379 ci->i_inline_version = CEPH_INLINE_NONE;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001380 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1381 &prealloc_cf);
Sage Weilbe655592011-11-30 09:47:09 -08001382 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -07001383 if (dirty)
1384 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -07001385 }
Sage Weil7971bd92013-05-01 21:15:58 -07001386
Sage Weil124e68e2009-10-06 11:31:08 -07001387 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Al Viro4908b822014-04-03 23:09:01 -04001388 inode, ceph_vinop(inode), pos, (unsigned)count,
Sage Weil7971bd92013-05-01 21:15:58 -07001389 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -07001390 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -07001391
Yan, Zheng03d254e2013-04-12 16:11:13 +08001392 if (written >= 0 &&
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001393 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
1394 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Yan, Zheng03d254e2013-04-12 16:11:13 +08001395 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001396 if (err < 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +08001397 written = err;
Yan, Zheng6070e0c2013-03-01 10:55:39 +08001398 }
Yan, Zheng03d254e2013-04-12 16:11:13 +08001399
Sage Weil2f75e9e2013-08-09 09:57:58 -07001400 goto out_unlocked;
Sage Weil124e68e2009-10-06 11:31:08 -07001401
Sage Weil2f75e9e2013-08-09 09:57:58 -07001402out:
Al Viro59551022016-01-22 15:40:57 -05001403 inode_unlock(inode);
Sage Weil2f75e9e2013-08-09 09:57:58 -07001404out_unlocked:
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001405 ceph_free_cap_flush(prealloc_cf);
Sage Weil2f75e9e2013-08-09 09:57:58 -07001406 current->backing_dev_info = NULL;
Yan, Zheng03d254e2013-04-12 16:11:13 +08001407 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -07001408}
1409
1410/*
1411 * llseek. be sure to verify file size on SEEK_END.
1412 */
Andrew Morton965c8e52012-12-17 15:59:39 -08001413static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -07001414{
1415 struct inode *inode = file->f_mapping->host;
Yan, Zheng99c88e62015-12-30 11:32:46 +08001416 loff_t i_size;
Sage Weil124e68e2009-10-06 11:31:08 -07001417 int ret;
1418
Al Viro59551022016-01-22 15:40:57 -05001419 inode_lock(inode);
Sage Weil6a82c472011-12-13 09:19:26 -08001420
Andrew Morton965c8e52012-12-17 15:59:39 -08001421 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Yan, Zheng508b32d2014-09-16 21:46:17 +08001422 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
Sage Weil124e68e2009-10-06 11:31:08 -07001423 if (ret < 0) {
1424 offset = ret;
1425 goto out;
1426 }
Josef Bacik06222e42011-07-18 13:21:38 -04001427 }
1428
Yan, Zheng99c88e62015-12-30 11:32:46 +08001429 i_size = i_size_read(inode);
Andrew Morton965c8e52012-12-17 15:59:39 -08001430 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -04001431 case SEEK_END:
Yan, Zheng99c88e62015-12-30 11:32:46 +08001432 offset += i_size;
Sage Weil124e68e2009-10-06 11:31:08 -07001433 break;
1434 case SEEK_CUR:
1435 /*
1436 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1437 * position-querying operation. Avoid rewriting the "same"
1438 * f_pos value back to the file because a concurrent read(),
1439 * write() or lseek() might have altered it
1440 */
1441 if (offset == 0) {
1442 offset = file->f_pos;
1443 goto out;
1444 }
1445 offset += file->f_pos;
1446 break;
Josef Bacik06222e42011-07-18 13:21:38 -04001447 case SEEK_DATA:
Yan, Zheng99c88e62015-12-30 11:32:46 +08001448 if (offset >= i_size) {
Josef Bacik06222e42011-07-18 13:21:38 -04001449 ret = -ENXIO;
1450 goto out;
1451 }
1452 break;
1453 case SEEK_HOLE:
Yan, Zheng99c88e62015-12-30 11:32:46 +08001454 if (offset >= i_size) {
Josef Bacik06222e42011-07-18 13:21:38 -04001455 ret = -ENXIO;
1456 goto out;
1457 }
Yan, Zheng99c88e62015-12-30 11:32:46 +08001458 offset = i_size;
Josef Bacik06222e42011-07-18 13:21:38 -04001459 break;
Sage Weil124e68e2009-10-06 11:31:08 -07001460 }
1461
Jie Liu46a1c2c2013-06-25 12:02:13 +08001462 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -07001463
1464out:
Al Viro59551022016-01-22 15:40:57 -05001465 inode_unlock(inode);
Sage Weil124e68e2009-10-06 11:31:08 -07001466 return offset;
1467}
1468
Li Wangad7a60d2013-08-15 11:51:44 +08001469static inline void ceph_zero_partial_page(
1470 struct inode *inode, loff_t offset, unsigned size)
1471{
1472 struct page *page;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001473 pgoff_t index = offset >> PAGE_SHIFT;
Li Wangad7a60d2013-08-15 11:51:44 +08001474
1475 page = find_lock_page(inode->i_mapping, index);
1476 if (page) {
1477 wait_on_page_writeback(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001478 zero_user(page, offset & (PAGE_SIZE - 1), size);
Li Wangad7a60d2013-08-15 11:51:44 +08001479 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001480 put_page(page);
Li Wangad7a60d2013-08-15 11:51:44 +08001481 }
1482}
1483
1484static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1485 loff_t length)
1486{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001487 loff_t nearly = round_up(offset, PAGE_SIZE);
Li Wangad7a60d2013-08-15 11:51:44 +08001488 if (offset < nearly) {
1489 loff_t size = nearly - offset;
1490 if (length < size)
1491 size = length;
1492 ceph_zero_partial_page(inode, offset, size);
1493 offset += size;
1494 length -= size;
1495 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001496 if (length >= PAGE_SIZE) {
1497 loff_t size = round_down(length, PAGE_SIZE);
Li Wangad7a60d2013-08-15 11:51:44 +08001498 truncate_pagecache_range(inode, offset, offset + size - 1);
1499 offset += size;
1500 length -= size;
1501 }
1502 if (length)
1503 ceph_zero_partial_page(inode, offset, length);
1504}
1505
1506static int ceph_zero_partial_object(struct inode *inode,
1507 loff_t offset, loff_t *length)
1508{
1509 struct ceph_inode_info *ci = ceph_inode(inode);
1510 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1511 struct ceph_osd_request *req;
1512 int ret = 0;
1513 loff_t zero = 0;
1514 int op;
1515
1516 if (!length) {
1517 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1518 length = &zero;
1519 } else {
1520 op = CEPH_OSD_OP_ZERO;
1521 }
1522
1523 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1524 ceph_vino(inode),
1525 offset, length,
Yan, Zheng715e4cd2014-11-13 14:40:37 +08001526 0, 1, op,
Li Wangad7a60d2013-08-15 11:51:44 +08001527 CEPH_OSD_FLAG_WRITE |
1528 CEPH_OSD_FLAG_ONDISK,
1529 NULL, 0, 0, false);
1530 if (IS_ERR(req)) {
1531 ret = PTR_ERR(req);
1532 goto out;
1533 }
1534
1535 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1536 &inode->i_mtime);
1537
1538 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1539 if (!ret) {
1540 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1541 if (ret == -ENOENT)
1542 ret = 0;
1543 }
1544 ceph_osdc_put_request(req);
1545
1546out:
1547 return ret;
1548}
1549
1550static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1551{
1552 int ret = 0;
1553 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilb314a902013-08-27 12:15:16 -07001554 s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1555 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1556 s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1557 u64 object_set_size = object_size * stripe_count;
1558 u64 nearly, t;
Li Wangad7a60d2013-08-15 11:51:44 +08001559
Sage Weilb314a902013-08-27 12:15:16 -07001560 /* round offset up to next period boundary */
1561 nearly = offset + object_set_size - 1;
1562 t = nearly;
1563 nearly -= do_div(t, object_set_size);
1564
Li Wangad7a60d2013-08-15 11:51:44 +08001565 while (length && offset < nearly) {
1566 loff_t size = length;
1567 ret = ceph_zero_partial_object(inode, offset, &size);
1568 if (ret < 0)
1569 return ret;
1570 offset += size;
1571 length -= size;
1572 }
1573 while (length >= object_set_size) {
1574 int i;
1575 loff_t pos = offset;
1576 for (i = 0; i < stripe_count; ++i) {
1577 ret = ceph_zero_partial_object(inode, pos, NULL);
1578 if (ret < 0)
1579 return ret;
1580 pos += stripe_unit;
1581 }
1582 offset += object_set_size;
1583 length -= object_set_size;
1584 }
1585 while (length) {
1586 loff_t size = length;
1587 ret = ceph_zero_partial_object(inode, offset, &size);
1588 if (ret < 0)
1589 return ret;
1590 offset += size;
1591 length -= size;
1592 }
1593 return ret;
1594}
1595
1596static long ceph_fallocate(struct file *file, int mode,
1597 loff_t offset, loff_t length)
1598{
1599 struct ceph_file_info *fi = file->private_data;
Libo Chenaa8b60e2013-12-11 13:49:11 +08001600 struct inode *inode = file_inode(file);
Li Wangad7a60d2013-08-15 11:51:44 +08001601 struct ceph_inode_info *ci = ceph_inode(inode);
1602 struct ceph_osd_client *osdc =
1603 &ceph_inode_to_client(inode)->client->osdc;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001604 struct ceph_cap_flush *prealloc_cf;
Li Wangad7a60d2013-08-15 11:51:44 +08001605 int want, got = 0;
1606 int dirty;
1607 int ret = 0;
1608 loff_t endoff = 0;
1609 loff_t size;
1610
Yan, Zheng494d77b2014-06-26 15:25:17 +08001611 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1612 return -EOPNOTSUPP;
1613
Li Wangad7a60d2013-08-15 11:51:44 +08001614 if (!S_ISREG(inode->i_mode))
1615 return -EOPNOTSUPP;
1616
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001617 prealloc_cf = ceph_alloc_cap_flush();
1618 if (!prealloc_cf)
1619 return -ENOMEM;
1620
Al Viro59551022016-01-22 15:40:57 -05001621 inode_lock(inode);
Li Wangad7a60d2013-08-15 11:51:44 +08001622
1623 if (ceph_snap(inode) != CEPH_NOSNAP) {
1624 ret = -EROFS;
1625 goto unlock;
1626 }
1627
1628 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1629 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1630 ret = -ENOSPC;
1631 goto unlock;
1632 }
1633
Yan, Zheng28127bd2014-11-14 22:38:29 +08001634 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1635 ret = ceph_uninline_data(file, NULL);
1636 if (ret < 0)
1637 goto unlock;
1638 }
1639
Li Wangad7a60d2013-08-15 11:51:44 +08001640 size = i_size_read(inode);
1641 if (!(mode & FALLOC_FL_KEEP_SIZE))
1642 endoff = offset + length;
1643
1644 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1645 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1646 else
1647 want = CEPH_CAP_FILE_BUFFER;
1648
Yan, Zheng3738daa2014-11-14 22:10:07 +08001649 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
Li Wangad7a60d2013-08-15 11:51:44 +08001650 if (ret < 0)
1651 goto unlock;
1652
1653 if (mode & FALLOC_FL_PUNCH_HOLE) {
1654 if (offset < size)
1655 ceph_zero_pagecache_range(inode, offset, length);
1656 ret = ceph_zero_objects(inode, offset, length);
1657 } else if (endoff > size) {
1658 truncate_pagecache_range(inode, size, -1);
1659 if (ceph_inode_set_size(inode, endoff))
1660 ceph_check_caps(ceph_inode(inode),
1661 CHECK_CAPS_AUTHONLY, NULL);
1662 }
1663
1664 if (!ret) {
1665 spin_lock(&ci->i_ceph_lock);
Yan, Zheng28127bd2014-11-14 22:38:29 +08001666 ci->i_inline_version = CEPH_INLINE_NONE;
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001667 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1668 &prealloc_cf);
Li Wangad7a60d2013-08-15 11:51:44 +08001669 spin_unlock(&ci->i_ceph_lock);
1670 if (dirty)
1671 __mark_inode_dirty(inode, dirty);
1672 }
1673
1674 ceph_put_cap_refs(ci, got);
1675unlock:
Al Viro59551022016-01-22 15:40:57 -05001676 inode_unlock(inode);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001677 ceph_free_cap_flush(prealloc_cf);
Li Wangad7a60d2013-08-15 11:51:44 +08001678 return ret;
1679}
1680
Sage Weil124e68e2009-10-06 11:31:08 -07001681const struct file_operations ceph_file_fops = {
1682 .open = ceph_open,
1683 .release = ceph_release,
1684 .llseek = ceph_llseek,
Al Viro36444242014-04-02 20:28:01 -04001685 .read_iter = ceph_read_iter,
Al Viro4908b822014-04-03 23:09:01 -04001686 .write_iter = ceph_write_iter,
Sage Weil124e68e2009-10-06 11:31:08 -07001687 .mmap = ceph_mmap,
1688 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -07001689 .lock = ceph_lock,
1690 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -07001691 .splice_read = generic_file_splice_read,
Al Viro3551dd72014-04-05 04:40:12 -04001692 .splice_write = iter_file_splice_write,
Sage Weil124e68e2009-10-06 11:31:08 -07001693 .unlocked_ioctl = ceph_ioctl,
1694 .compat_ioctl = ceph_ioctl,
Li Wangad7a60d2013-08-15 11:51:44 +08001695 .fallocate = ceph_fallocate,
Sage Weil124e68e2009-10-06 11:31:08 -07001696};
1697