blob: dad8d44bf90e85a89ac0115efd9992fa297c4678 [file] [log] [blame]
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -04001// SPDX-License-Identifier: GPL-2.0
2/*
3 * NVMe Over Fabrics Target File I/O commands implementation.
4 * Copyright (c) 2017-2018 Western Digital Corporation or its
5 * affiliates.
6 */
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8#include <linux/uio.h>
9#include <linux/falloc.h>
10#include <linux/file.h>
11#include "nvmet.h"
12
13#define NVMET_MAX_MPOOL_BVEC 16
14#define NVMET_MIN_MPOOL_OBJ 16
15
16void nvmet_file_ns_disable(struct nvmet_ns *ns)
17{
18 if (ns->file) {
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -040019 if (ns->buffered_io)
20 flush_workqueue(buffered_io_wq);
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040021 mempool_destroy(ns->bvec_pool);
22 ns->bvec_pool = NULL;
23 kmem_cache_destroy(ns->bvec_cache);
24 ns->bvec_cache = NULL;
25 fput(ns->file);
26 ns->file = NULL;
27 }
28}
29
30int nvmet_file_ns_enable(struct nvmet_ns *ns)
31{
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -040032 int flags = O_RDWR | O_LARGEFILE;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040033 struct kstat stat;
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -040034 int ret;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040035
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -040036 if (!ns->buffered_io)
37 flags |= O_DIRECT;
38
39 ns->file = filp_open(ns->device_path, flags, 0);
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040040 if (IS_ERR(ns->file)) {
41 pr_err("failed to open file %s: (%ld)\n",
Wei Yongjun81cf54e02018-05-31 01:55:58 +000042 ns->device_path, PTR_ERR(ns->file));
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040043 return PTR_ERR(ns->file);
44 }
45
46 ret = vfs_getattr(&ns->file->f_path,
47 &stat, STATX_SIZE, AT_STATX_FORCE_SYNC);
48 if (ret)
49 goto err;
50
51 ns->size = stat.size;
52 ns->blksize_shift = file_inode(ns->file)->i_blkbits;
53
54 ns->bvec_cache = kmem_cache_create("nvmet-bvec",
55 NVMET_MAX_MPOOL_BVEC * sizeof(struct bio_vec),
56 0, SLAB_HWCACHE_ALIGN, NULL);
Wei Yongjun1367bc82018-05-31 11:41:30 +000057 if (!ns->bvec_cache) {
58 ret = -ENOMEM;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040059 goto err;
Wei Yongjun1367bc82018-05-31 11:41:30 +000060 }
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040061
62 ns->bvec_pool = mempool_create(NVMET_MIN_MPOOL_OBJ, mempool_alloc_slab,
63 mempool_free_slab, ns->bvec_cache);
64
Wei Yongjun1367bc82018-05-31 11:41:30 +000065 if (!ns->bvec_pool) {
66 ret = -ENOMEM;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040067 goto err;
Wei Yongjun1367bc82018-05-31 11:41:30 +000068 }
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -040069
70 return ret;
71err:
72 ns->size = 0;
73 ns->blksize_shift = 0;
74 nvmet_file_ns_disable(ns);
75 return ret;
76}
77
78static void nvmet_file_init_bvec(struct bio_vec *bv, struct sg_page_iter *iter)
79{
80 bv->bv_page = sg_page_iter_page(iter);
81 bv->bv_offset = iter->sg->offset;
82 bv->bv_len = PAGE_SIZE - iter->sg->offset;
83}
84
85static ssize_t nvmet_file_submit_bvec(struct nvmet_req *req, loff_t pos,
86 unsigned long nr_segs, size_t count)
87{
88 struct kiocb *iocb = &req->f.iocb;
89 ssize_t (*call_iter)(struct kiocb *iocb, struct iov_iter *iter);
90 struct iov_iter iter;
91 int ki_flags = 0, rw;
92 ssize_t ret;
93
94 if (req->cmd->rw.opcode == nvme_cmd_write) {
95 if (req->cmd->rw.control & cpu_to_le16(NVME_RW_FUA))
96 ki_flags = IOCB_DSYNC;
97 call_iter = req->ns->file->f_op->write_iter;
98 rw = WRITE;
99 } else {
100 call_iter = req->ns->file->f_op->read_iter;
101 rw = READ;
102 }
103
104 iov_iter_bvec(&iter, ITER_BVEC | rw, req->f.bvec, nr_segs, count);
105
106 iocb->ki_pos = pos;
107 iocb->ki_filp = req->ns->file;
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -0400108 iocb->ki_flags = ki_flags | iocb_flags(req->ns->file);
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400109
110 ret = call_iter(iocb, &iter);
111
112 if (ret != -EIOCBQUEUED && iocb->ki_complete)
113 iocb->ki_complete(iocb, ret, 0);
114
115 return ret;
116}
117
118static void nvmet_file_io_done(struct kiocb *iocb, long ret, long ret2)
119{
120 struct nvmet_req *req = container_of(iocb, struct nvmet_req, f.iocb);
121
122 if (req->f.bvec != req->inline_bvec) {
123 if (likely(req->f.mpool_alloc == false))
124 kfree(req->f.bvec);
125 else
126 mempool_free(req->f.bvec, req->ns->bvec_pool);
127 }
128
129 nvmet_req_complete(req, ret != req->data_len ?
130 NVME_SC_INTERNAL | NVME_SC_DNR : 0);
131}
132
133static void nvmet_file_execute_rw(struct nvmet_req *req)
134{
135 ssize_t nr_bvec = DIV_ROUND_UP(req->data_len, PAGE_SIZE);
136 struct sg_page_iter sg_pg_iter;
137 unsigned long bv_cnt = 0;
138 bool is_sync = false;
139 size_t len = 0, total_len = 0;
140 ssize_t ret = 0;
141 loff_t pos;
142
143 if (!req->sg_cnt || !nr_bvec) {
144 nvmet_req_complete(req, 0);
145 return;
146 }
147
148 if (nr_bvec > NVMET_MAX_INLINE_BIOVEC)
149 req->f.bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
150 GFP_KERNEL);
151 else
152 req->f.bvec = req->inline_bvec;
153
154 req->f.mpool_alloc = false;
155 if (unlikely(!req->f.bvec)) {
156 /* fallback under memory pressure */
157 req->f.bvec = mempool_alloc(req->ns->bvec_pool, GFP_KERNEL);
158 req->f.mpool_alloc = true;
159 if (nr_bvec > NVMET_MAX_MPOOL_BVEC)
160 is_sync = true;
161 }
162
163 pos = le64_to_cpu(req->cmd->rw.slba) << req->ns->blksize_shift;
164
165 memset(&req->f.iocb, 0, sizeof(struct kiocb));
166 for_each_sg_page(req->sg, &sg_pg_iter, req->sg_cnt, 0) {
167 nvmet_file_init_bvec(&req->f.bvec[bv_cnt], &sg_pg_iter);
168 len += req->f.bvec[bv_cnt].bv_len;
169 total_len += req->f.bvec[bv_cnt].bv_len;
170 bv_cnt++;
171
172 WARN_ON_ONCE((nr_bvec - 1) < 0);
173
174 if (unlikely(is_sync) &&
175 (nr_bvec - 1 == 0 || bv_cnt == NVMET_MAX_MPOOL_BVEC)) {
176 ret = nvmet_file_submit_bvec(req, pos, bv_cnt, len);
177 if (ret < 0)
178 goto out;
179 pos += len;
180 bv_cnt = 0;
181 len = 0;
182 }
183 nr_bvec--;
184 }
185
186 if (WARN_ON_ONCE(total_len != req->data_len))
187 ret = -EIO;
188out:
189 if (unlikely(is_sync || ret)) {
190 nvmet_file_io_done(&req->f.iocb, ret < 0 ? ret : total_len, 0);
191 return;
192 }
193 req->f.iocb.ki_complete = nvmet_file_io_done;
194 nvmet_file_submit_bvec(req, pos, bv_cnt, total_len);
195}
196
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -0400197static void nvmet_file_buffered_io_work(struct work_struct *w)
198{
199 struct nvmet_req *req = container_of(w, struct nvmet_req, f.work);
200
201 nvmet_file_execute_rw(req);
202}
203
204static void nvmet_file_execute_rw_buffered_io(struct nvmet_req *req)
205{
206 INIT_WORK(&req->f.work, nvmet_file_buffered_io_work);
207 queue_work(buffered_io_wq, &req->f.work);
208}
209
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400210static void nvmet_file_flush_work(struct work_struct *w)
211{
212 struct nvmet_req *req = container_of(w, struct nvmet_req, f.work);
213 int ret;
214
215 ret = vfs_fsync(req->ns->file, 1);
216
217 nvmet_req_complete(req, ret < 0 ? NVME_SC_INTERNAL | NVME_SC_DNR : 0);
218}
219
220static void nvmet_file_execute_flush(struct nvmet_req *req)
221{
222 INIT_WORK(&req->f.work, nvmet_file_flush_work);
223 schedule_work(&req->f.work);
224}
225
226static void nvmet_file_execute_discard(struct nvmet_req *req)
227{
228 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
229 struct nvme_dsm_range range;
Sagi Grimberg1b72b712018-07-11 12:43:16 +0300230 loff_t offset, len;
231 u16 ret;
232 int i;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400233
234 for (i = 0; i <= le32_to_cpu(req->cmd->dsm.nr); i++) {
Sagi Grimberg1b72b712018-07-11 12:43:16 +0300235 ret = nvmet_copy_from_sgl(req, i * sizeof(range), &range,
236 sizeof(range));
237 if (ret)
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400238 break;
239 offset = le64_to_cpu(range.slba) << req->ns->blksize_shift;
240 len = le32_to_cpu(range.nlb) << req->ns->blksize_shift;
Sagi Grimberg1b72b712018-07-11 12:43:16 +0300241 if (vfs_fallocate(req->ns->file, mode, offset, len)) {
242 ret = NVME_SC_INTERNAL | NVME_SC_DNR;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400243 break;
Sagi Grimberg1b72b712018-07-11 12:43:16 +0300244 }
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400245 }
246
Sagi Grimberg1b72b712018-07-11 12:43:16 +0300247 nvmet_req_complete(req, ret);
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400248}
249
250static void nvmet_file_dsm_work(struct work_struct *w)
251{
252 struct nvmet_req *req = container_of(w, struct nvmet_req, f.work);
253
254 switch (le32_to_cpu(req->cmd->dsm.attributes)) {
255 case NVME_DSMGMT_AD:
256 nvmet_file_execute_discard(req);
257 return;
258 case NVME_DSMGMT_IDR:
259 case NVME_DSMGMT_IDW:
260 default:
261 /* Not supported yet */
262 nvmet_req_complete(req, 0);
263 return;
264 }
265}
266
267static void nvmet_file_execute_dsm(struct nvmet_req *req)
268{
269 INIT_WORK(&req->f.work, nvmet_file_dsm_work);
270 schedule_work(&req->f.work);
271}
272
273static void nvmet_file_write_zeroes_work(struct work_struct *w)
274{
275 struct nvmet_req *req = container_of(w, struct nvmet_req, f.work);
276 struct nvme_write_zeroes_cmd *write_zeroes = &req->cmd->write_zeroes;
277 int mode = FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE;
278 loff_t offset;
279 loff_t len;
280 int ret;
281
282 offset = le64_to_cpu(write_zeroes->slba) << req->ns->blksize_shift;
283 len = (((sector_t)le16_to_cpu(write_zeroes->length) + 1) <<
284 req->ns->blksize_shift);
285
286 ret = vfs_fallocate(req->ns->file, mode, offset, len);
287 nvmet_req_complete(req, ret < 0 ? NVME_SC_INTERNAL | NVME_SC_DNR : 0);
288}
289
290static void nvmet_file_execute_write_zeroes(struct nvmet_req *req)
291{
292 INIT_WORK(&req->f.work, nvmet_file_write_zeroes_work);
293 schedule_work(&req->f.work);
294}
295
296u16 nvmet_file_parse_io_cmd(struct nvmet_req *req)
297{
298 struct nvme_command *cmd = req->cmd;
299
300 switch (cmd->common.opcode) {
301 case nvme_cmd_read:
302 case nvme_cmd_write:
Chaitanya Kulkarni55eb942e2018-06-20 00:01:41 -0400303 if (req->ns->buffered_io)
304 req->execute = nvmet_file_execute_rw_buffered_io;
305 else
306 req->execute = nvmet_file_execute_rw;
Chaitanya Kulkarnid5eff332018-05-23 00:34:39 -0400307 req->data_len = nvmet_rw_len(req);
308 return 0;
309 case nvme_cmd_flush:
310 req->execute = nvmet_file_execute_flush;
311 req->data_len = 0;
312 return 0;
313 case nvme_cmd_dsm:
314 req->execute = nvmet_file_execute_dsm;
315 req->data_len = (le32_to_cpu(cmd->dsm.nr) + 1) *
316 sizeof(struct nvme_dsm_range);
317 return 0;
318 case nvme_cmd_write_zeroes:
319 req->execute = nvmet_file_execute_write_zeroes;
320 req->data_len = 0;
321 return 0;
322 default:
323 pr_err("unhandled cmd for file ns %d on qid %d\n",
324 cmd->common.opcode, req->sq->qid);
325 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
326 }
327}